CONTRACTOR REPORT

.SAND82-7114 Unlimited Release UC-63

Detailed Residential Electric Load Determination

The BDM Corporation 1801 Randolph Road NE Albuquerque, NM 87106

Prepared by Sandia National Laboratories Albuquerque, New Mexico 87185 and Livermore, California 94550 for the United States Department of Energy under Contract DE-AC04-76DP00789

Printed June 1984

Issued by Sandia National Laboratories, operated for the United States Department of Energy by Sandia Corporation. NOTICE: This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Govern­ment nor any agency thereof, nor any of their employees, nor any of their contractors, subcontractors, or their employees, makes any warranty. ex­press or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, prod­uct, or process disclosed, or represents that ita use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise, does not necessarily_ constitute or imply its endorsement, recommendation, or favoring by the United States Government, any agency thereof or any of t heir contractors or subcontractors. The views and opinions expressed here­in do not necessarily state or reflect those of the United States Government, any agency thereof or any of their contractors or subcontractors.

Printed in the United States of America Available from National Technical Information Service U.S. Department of Commerce 5285 Port Royal Road Springfield, VA 22161

NTIS price codes P,inted copy: A06 Microfiche copy: AO!

SAND82-7114 Unlimited Release Printed June 1984

Detailed Residential Electric Load Determination

The BDM Corporation 1801 Randolph Road NE Albuquerque, NM 87106

Abstract

Distribution Category UC-63

Data on residential loads has been collected from four residences in real time. The data, measured at 5-second intervals for 53 days of continuous operation, were statistically characterized. An algorithm was developed and incorporated into the modeling code SOLCEL. Performance simulations with SOLCEL using these data as well as previous data collected over longer time intervals indicate that no significant errors in system value are introduced through the use of long-term average data.

THE 8DM CORPORATION

FOREWORD

This final report, BDM/A-82-299-TR-Rl, is submitted by The BDM Corporation (BDM) documenting work performed for Sandia National Labora­tories under Contract 62-3977.

BDM wi shes to thank the fo 11 owi ng ut il i ty compani es whose exce 11 ent cooperation made this work possible:

The Georgia Power Company - Atlanta, GA Public Service Electric and Gas Company - Newark, NJ Public Service Company of New Mexico - Albuquerque, NM

and the homeowners in Albuquerque for their hospitality.

THE 80M CORPORATION

TABLE OF CONTENTS

Chapter Page

I EXECUTIVE SUMMARY 1-1

A. INTRODUCTION 1-1 B. PROGRAM OVERVIEW 1-9

1. Overview 1-9 2. Utility Surveys/Direct Measurement 1-9 3. Statistical Load Model 1-11 4. Performance Comparisons 1-12

II LOADS AND MODELS 11-1

A. INTRODUCTION II-I B. LOAD MOOELING 11-4

1. Modeling Approach II-4 2. Modeling Strategi·es II-8

III SURVEY OF UTILITIES AND RESIDENTIAL SELECTION 111-1

A. INSTANTANEOUS LOAD DATA INVENTORY III-I B. SELECTION OF PARTICIPATING UTILITIES AND

RESIDENCES FOR DIRECT MEASUREMENT 111-4 C. RESIDENTIAL CHARACTERISTICS 111-4

IV DIRECT MEASUREMENT IV-l

A. DATA ACQUISITION SYSTEM - OVERVIEW IV-l B. DETAILED RESIDENTIAL ELECTRICAL LOAD DATA

ACQUISITION SOFTWARE IV-5 C. SYSTEM OPERATION IV-5 D. SYSTEM CALIBRATION AND CHECKOUT IV-6 E. RESIDENTIAL LOAD MONITORING IV-7 F. RESIDENTIAL APPLIANCE CHARACTERIZATION . IV-8

V STATISTICAL MODEL DEVELOPMENT V-I

A. INSTANTANEOUS LOAD DATA BASE V-I

1. Data Base Development Procedures V-I • 2. Data Base Description V-2

vii

THE BOM CORPORATION

Chapter

VI

A B C D

TABLE.OFCONIENTS (Concluded)

B. STATISTICAL LOAD MODEL

1. Modeling Methodology 2. Load Model Development 3. Load Model Implementation into SOLCEL

PERFORMANCE COMPARISONS

APPENDICES

LOAD DATA BASE ACCESS AND FORMAT SOLCEL3 EXECUTION PROCEDURES PROFILE PLOTS . .. RESIDENTIAL LOAD MONITORING5YSTEM SCHEMATIC WIRING DIAGRAMS

viii

Page

V-6

V-6 v-a

V-23

VI-l

A-I B-1 C-l

0-1

THE BOM CORPORATION

Figure

I-I

I-2a

1-2b

1-3

II-I

II-2

IV-l

IV-2

IV-3

V-la

V-lb

V-Ie

V-ld

V-2

V-3

V-4

V-5

V-6

VI-l

LIST OF FIGURES

Program Summary

Sample Instantaneous Load Profiles

Sample Instantaneous Load Profiles (Concluded)

PV Performance Versus Array Size for Various Probabilities of an Active Profile

Time-Averaged Versus Instantaneous Load Data

Instantaneous Load Profile

Detailed Residential Load Data Acquisition and Load Monitoring System

JEM/HP-85

Esterline-Angus

Newark Data Collection Summary

Atlanta Data Collection Summary

Albuquerque #1 Data Collection Summary

Albuquerque #2 Data Collection Summary

Modeling Methodology

Periodic Regression Example with Three Harmonics

Plot of Predicted (P) and Observed (0) Load Versus Time

Plot of Energy Above Average (ABVAVE) Versus Quartile (QRTL)

Plot of Energy Above Average (ABVAVE) Versus R-Squared (RSQR)

Three Possible Configurations of Load Versus PV Output

ix

Page

1-3

1-5

1-6

1-8

II-2

II-7

IV-2

IV-9

IV-9

V-3

V-3

V-3

V-3

V-7

V-9

V-ll

V-13

V-14

VI-2

THE BDM CORPORATION

Figure

VI-2

VI-3

VI-4

VI-5

LIST OF FIGURES (Concluded)

Annual PV Performance Versus Probability of an Active Profile for Different Array Sites

Monthly PV Performance Versus Probability of an Active Profile for 1 kW Array

Monthly PV Performance Versus Probability of an Active Profile for 4 kW Array

PV Performance Versus Array Size for Various Probabilities of an Active Profile

x

VI-7

VI-9

VI-lO

VI-12

THE 8DM CORPORATION

LIST OF TABLES

Table Page

I-I RESULTS FROM SOLCEL SIMULATIONS ANNUAL DIRECT FRACTION 1-4

I II-I SUMMARY OF UTILITY TELEPHONE SURVEY III-2

III-2 RESIDENTIAL CHARACTERISTICS I II-5

IV-I DATA COLLECTION BY RESIDENCE IV-7

V-I LOAD DATA BASE SUMMARY V-4

V-2 SAMPLED INTERVALS CATEGORIZED BY HOUSE, SEASON AND WEEKDAY/WEEKEND V-5

V-3 BREAKDOWN OF INTERVALS BY CRITERIA FOR ACTIVE PROFILES V-I7

V-4 CLASSIFICATION TABLE DISPLAYING DISTINCT PROFILES V-I9

V-5 PROFILES BY SEASON AND WEEKDAY/WEEKEND V-22

VI-l RESULTS FROM SOLCEL SIMULATIONS VI-5

VI-2 REDUCTION IN PV DIRECT FRACTION COMPARED TO PROBABILITY OF AN ACTIVE LOAO VI-8

THE BDM CORPORATION

A. INTRODUCTION

CHAPTER I

EXECUTIVE SUMMARY

A residential load profile is the pattern of electric use in a

residence over a period of time. Util ities routinely monitor residences

and use these profiles to analyze the demand for electricity by a

particular class of user. A profile is created by monitoring the load

over a time interval. When a utility constructs a profile, the time

i nterva 1 used is usually between 15 llii nutes and 1 hour. Wh il e th i s

i nterva 1 is adequate for uti 1 i ty requi rements, it is too 1 arge to reveal

the instantaneous behavior of the load. Before the completion of this

program. it was generally assumed that use of this long-time constant

data in residential PV system simulations significantly underestimated

the ; nteract i on between a PV systein and the ut il ity and produced 1 arge

errors in calculating the fractions of energy provided by an array to a

load, or that which was purchased from or sold back to a uti 1 ity. The

measurement of the instantaneous behavior of a residential load was

required so that more accurate models of residential usage and PV system

performance may be developed to predict the energy demanded by a

res i dence and that wh i ch is provi ded by a PV array or by the ut il ity.

The Detail ed Resident i a 1 El ectri c Load Oetermi nat i on Program was

initiated to measure residential electric loads at instantaneous inter­

vals and to determine the effects of using instantaneous load data in

annual residential PV power systems performance simUlations.

Prior performance predictions of residential PV power systems had to

be conducted using models that could accept only hourly time-averaged

load data which does not reveal the instantaneous load behavior caused by

residential appliance usage. Other studies of the effects of instan­

taneous loads have used statistical approaches involving the complete

characteri zat i on of the res i dent i all oad based on the probab il it i es of

various househo 1 d app 1 i ance usage for average durat ions and ind i vi dua 1

1-1

THE 80M CORPORATION

appl iance load requirements. These statistical studies have assumed a high occurrence of instantaneous load profiles and, consequently, signif­icant impacts of instantaneous loads on annual residential PV system performance.

Other attempts to direct measure loads at instantaneous i nterva 1 s had been taken on very small samples, on one residence for 3 days. While these data did revea 1 greater PV system ut i1 ity interact i on at sma 11 er monitori ng i nterva 1 s, the samp 1 e size was too 1 i mi ted to support firm conclusions.

The objectives of the Detailed Residential Electric Load Determina­tion Program were to measure real, instantaneous load data, and to develop a statistical model based on these measured data to predict residential electric demand. Residential load data are currently being collected at 15-minute intervals by utilities throughout the nation. With this large data base the statistical model can be used to predict the instantaneous load demand for PV systems in any area of the country. For performance predi ct ions the stat i st i ca 1 model was incorporated into SOLCEL, a Sandia National Laboratories-developed (SNLA) computer simula­tion code designed specifically for the analysis of PV systems.

Residential electric load patterns were directly measured at 5-second intervals on four residences selected in the northeast, southeast, and southwest regions of the country. A summary of the program is shown in figure 1-1. A load data base conSisting of 5,080 complete, continuous, 15-minute intervals, was established (1,270 hours or 53 days of data). Load data were collected for a 2-week period during the summer and again for a 2-week period in the winter on each of the selected residences. To date, this data base is the largest source of instantaneous load data available.

A major finding of this program was the limited occurrence of active load profiles. An active profile exhibits a high energy level above the 15-minute average, high squared multiple correlation (which means the model adequately explains variability), and usually occurs between 5:30 AM and midnight. Inactive profi les, in contrast, are steady state

1-2

THE 8DM CORPORATION

RESIDENCE RESIDENCE RESIDENCE RESIDENCE NEWARK ALBUQUERQUE 1 ALBUQUERQUE 2 ATLANTA

'liT T JEM

362.0 HRS

MONITORED

~

JEM JEM

329.5 HRS 305.0 HRS

JEM

274.5 HRS MONITORED

/ DATA REDUCTION AND ANALYSIS

15-SEC LOAD DATA ANALYZED FOR EACH 15-MINUTE INTERVAL

ACTIVITY OF INTERVALS EXAMINED

ACTIVE INTERVALS WERE CLUSTERED INTO GROUPS I ALGORITHM DEVELOPED AND INCORPORATED INTO SOLCEL

ALGORITHM SELECTS AN ACTIVE INTERVAL BASED ON FREQUENCY AND PROBABILITY OF OCCURRENCE

I PROBABILITY OF AN ACTIVE INTERVAL ALLOWED TO VARY FROM 0 TO 1

I ANNUAL PV SYSTEM PERFORMANCE SIMULATIONS USING 15-SECOND DATA.

Figure 1-1_ Program Summary

1-3

• RESIDENCES SELECTED BY UTILITY AS TYPICAL CUSTOMER

• .JOULE ELECTRONIC METER (JEMI MEASURES TRUE RMS VOLTAMPERES (PRODUCT OF CURRENT AND VOLTAGEI

• SUMMER AND WINTER MONITORING

• 5·SECOND INTERVAL DATA CONTINUOUSLY COLLECTED IN COUNTS OF KVAh

• ONLY 20% OF INTERVALS SHOWED SIGNIFICANT ACTIVITY DIFFERENT FROM 15-MIN AVERAGE

.,3 DIFFERENT CLUSTERS WERE IDENTIFIED

• USED 13 ACTIVE INTERVALS, ONE SELECTED FROM EACH CLUSTER

• SELECTION CRITERIA­ACTIVE INTERVAL OCCURS BETWEEN S:30 AND 24:00 AND HAS ENERGY ABOVE AVERAGE OF.06 KVA

• O-LOAD COMPLETELY INACTIVE -15·MIN DATA ADEQUATE

• I-LOAD ALWAYS ACTIVE FOR EVERY INTERVAL WHICH MEETS THE SELECTION CRITERIA

• ONE FULL YEAR OF ACTUAL lS·MINUTE MONITORED LOAD DATA USED. ONE FULL YEAR OF HOURLY INSOLATION DATA USED.

BDM/A-82-299-TR-R 1

THE 80M CORPORATION

loads that closely follow the l5-minute time-averaged data and occur

mostly during the night. mid-morning and afternoon. Sample active and inactive profiles are shown in figures I-2a and I-2b~

After the load data were statistically characterized. only 734 i nterva 1 s out of the 5.080 i nterva 1 sin the data base possessed the

characteri sti cs (descri bed above) stati st i ca 11y ident ifyi ng them as an

active profile and differentiating them from the 15-minute time-averaged

load. Th i s 1 imited occurrence of act i ve profil es substant i a 11y reduced

the probability that an active profile would occur when the statistical

model was exercised and annual system performance simulations conducted.

In the test cases that were run for performance comparisons using the instantaneous load mode 1. the PV array supp 1 i ed 33.3 percent of the

energy demanded by the residence on an annual basis. This represents the

most accurate prediction of the actual PV fraction. These results were

then compared with the results obtained using IS-minute and hourly time­averaged data.

As can be seen from table I-I. less than a 2 percent dec1 ine is

obtained using hourly data when compared to the use of instantaneous

data. Even operating the model with twice the probability of the

occurrence of an active profile produces little difference in the amount of energy provided by the array.

TABLE I-I. RESULTS FROM SOLCEL SIMULATIONS ANNUAL DIRECT FRACTION

ANNUAL PERCENT LOAD SUPPLIED

MODEL 15 MIN HOURLY PV DIRECT FRACTION .333 .339 .346

MODEL WITH

TWICE PROB.

.327

This comparison did not address the effects of array size or of

increasing the probability of an active load beyond 0.4. When array size

1-4

.... I

\J1

ALBUQUERQUE 2

MONDAY 1/18182

10

,~

01

~ 6111111 ""I.

INSTANTANEOUS (15 sec) LOAD PROFILE '" .,

---:...L..-'---'--~ __ ~ ___ ~_~~_~_"_~_-L_

12:00 6mln 13:00

ALBUQUERQUE FRIDAY 1/1/82

~

,

. -

I

'" .., '" -

I r---J{

L-_ It'"" U W .............. L..!- l.Ju -~ ~

,18:30

TIME·AVERAGED (15minl LOAD PROFILE

." f;

1----

~]rJ1:llC-~ .. 3 ~:'r= _~ __ .~~~. __ ~~_~~Lo'''h--4d _ ....... " I

14:00 15:00

TIME OF DAY

~

t~ 1°' r·. j •• ,

,- a.

4IIItI+~ J \ -J P..-~~~=-] U~~.JJIUI

17:30. 18:30 19:30

TIME OF DAY

BDM/A-B2-299-TR-RI

Figure I-2a. Sample Instantaneous Load Profiles

-i ::r: m OJ 0 s: ()

0 :::0 ""0 0 :::0

~ 0 Z

t--I 1

0'\

NEWARK,NJ

FRIDAY 7/17/81 10

h; .. . INSTANTANEOUs (16 sec,. I lLOADPROFIL~ __ ..•

or. .. N

~ 61 ! (~~~~~~~~~~:0C16mlnl _~ _ rf'--<c'- \ ~1 ~

.... ~--".LfL=r--L~ __ ~-'tr

or, .,. 1-)

Ji\ -t9 0 rJ_~~-

~-.--"-.-.~~-.~- _ .. --~-1~~~·--~·-~-·-~- ~-~~··---·--~,:~4:F DAY Ii min 13:46

NEWARK, NJ WEDNESDAY 2/3/82

~

10

.'>

'" (T,

f- ....

18:30

• • ...... --"-~

-".

'" '" "

.....,.".... ~ --....,

2Ili'"

n n

'" J';

OJ

.J.

121:3(;.

TIME OF DAY

J1,.,.

f~ " OJ

~~ ~'7"? ~ t-~~ 22:30

BDM/A-82-299-TR-Rl

Figure I-2b;. ~ample Instantaneous Load Profiles (Concluded)

-i :z: m CD 0 s: (')

0 ::0 -0 0 ::0 » -i 0 Z

THE 8DM CORPORATION

is cons i dered and the probab il i ty of an act i ve load is increased, on 1 y for very small array sizes and unrealistically active loads do instan­taneous load data produce a significant decline in the PV fraction. This effect is shown in figure I-3. The probability of an active load is increased from 0, or IS-minute load data, to 1, or a continuously active load. The observed probability of an active load was 0.2 obtained from the measured data. A probability of 0.4 was considered an extreme active case for load variability. The probability was extended to 1 to deter­mine sensitivities. Array size was increased from 1 to 9 kW.

The conclusion can thus be made that IS-minute time-averaged data are adequate for predicting the annual performance of a residential photovoltaic system for realistically sized arrays and load activity. Instantaneous data are not required.

The Detailed Residential Electric load Determination program has produced several unique products. These are

(1) A statistical model incorporated into SOlCH that can convert time-averaged load data already available from util ities into instantaneous data for performance simulation and analysis.

(2) A total of 53 days of continuous, instantaneous residential load data collected on four residences in different regions of the country incorporated into a data base and available on the SNlA computer.

(3) An improved SOlCEl simulation capability using 15-minute time­averaged load data and the instantaneous load model. This allows the use of SOlCEl with standa'rd time-averaged data currently collected by utilities.

(4) A re 1 i ab 1 e, remote-s i te, instantaneous load monitoring system that can be mounted directly on a residence or utility pole and transmit load data via telephone lines. The system can be used for additional residential or commercial monitoring.

I-7

THE 80M CORPORATION

.7

. 6

.5

.4

PV DIRECT FRACTION .3

.2

.1

0 0

ANNUAL AVERAGE

INSTANTANEOUS DATA SIGNIFICANT FOR ONLY VERY SMALL ARRAY SIZES .

TYPICAL MEASURED ACTIVITY PROBABILITY 0.2

2 4 6

ARRAY OUTPUT (kWl

8 10

BDMI A-82-299-TR-R 1

Figure 1-3. PV Performance Versus Array Size for Various Probabilities -of an Active Profile

1-8

THE 80M CORPORATION

B. PROGRAM OVERVIEW

1. Overview

Included in this phase of the program were the following four tasks:

(1) A survey of utilities and other sources to determine the avail­ability of instantaneous residential electric load data

(2) A direct measurement program of residential electric loads at

instantaneous intervals if load data from utilities were unavailable

(3) The development of a statistical model. based on the direct

measured data, which could be used to predict the behavior of instantaneous loads using time-averaged load data

(4) PV system performance comparisons using the instantaneous model, and time-averaged load data

Chapters II through VI contain detailed information on these tasks. In addition, several appendices present information on the load data base

access and format, SOLCEL execut i on procedures, profil e plots, and the monitoring system schematics.

Two additional tasks involving powerline waveform quality moniLring and PV system-utility interactions were also conducted under this program. These will be reported separately.

2. Utility Surveys/Direct Measurement

A comprehensive survey was completed of the two largest utili­ties in each state of the northeast, southeast, and southwest regions of

the country. A tota 1 of 51 util it ies were contacted to determine the

extent of residential load monitoring programs, sample size, and sampling interval. The utility survey revealed that most of the utilities have

instituted residential load monitoring programs for reporting load data

under Public Utilities Regulatory Policy Act of 1978 (PURPA). Time­

averaged intervals of 15 to 30 minutes are considered adequate by the

uti 1 ities for reporting purposes. No sources of instantaneous load data

were identified from utilities, utility organizations. or government research projects.

1-9

THE 8DM CORPORATION

A direct measurement program was then initiated to develop a

sample of instantaneous residential electric loads on typical residences

in these regions. The Public Service Company of New Mexico, Georgia

Power, Atlanta, Georgia, and the Public Service Electric and Gas Company

of Newark, New Jersey, were selected for participation in the program.

Residences were selected by the utility to obtain a typical residence

from their current residential monitoring program. Selection was made on

the basis of electrical consumption that approximated the sample mean for

residences from util ity load data base. Four residences were instru-

mented, two in Albuquerque and one each in Atlanta and Newark.

For the instantaneous load data collection, a hardware package

capable of measuring loads at intervals of approximately 5 seconds was ~

developed. The data acquisition. hardware consisted of two major

sect ions, the measurement equ i pment and the computer contro 11 er.

Measurement equipment was located at each of the four monitored houses.

A weatherproof box built to NEMA specifications housed the complete

measurement system consisting of current loop transformers connected to a

Scientific Columbus joule Electronic Meter (JEM) where data were measured

in units of kVAh. After digitization within the JEM, the accumulated

data were collected at regular intervals by the computer/controller via

telephone 1 ines. The computer /contro 11 er was an HP-85 desktop computer

chosen for both its data-collection and analysis capabilities. Data

accumulated in the HP-85 were stored on 5-inch floppy disks.

Data collection software included data acquisition, analysis, and

transfer programs. An ana lysi s program plotted the data on the HP-85 I s

CRT and a hard copy was obtained from a built-in thermal plotter. In

addition, a data transfer program transferred the data to the SNLA

computer network. Samp 1 e load plots obtained from the HP-85 pri nter are

shown in figures I-2a and I-2b. The sample plots shown in figure 1-2a

show a sample of an active load profile. This type of profile, however,

was present in only 734 of the 5,080 intervals in the load data base.

1-10

THE BDM CORPORATION

More typical of the data base is the sample load profile for the Newark

residence (2/3/82) shown in figure 1-2b where the instantaneous load and

time-averaged load are not statistically different.

3. Statistical Load Model

The statistical load model was then developed to serve as a

mathemati ca 1 operator that converts rea l-time-averaged data into i nstan­

taneous data. Originally. the model was planned to take the form of a

multiple linear regression prediction equation. However. analysis has

shown that the instantaneous profiles are driven by particular appli­

ances. Si nce a 1 arge samp 1 e of instantaneous app 1 i ance loads wou 1 d be

required to adequately model their input to the load profile (much larger

than feasible for this project). an alternative modeling methodology was developed.

Each I5-minute interval was characterized by periodic regres­

sion analysis (sums of sine and cosine curves). The coefficients of

these regression equations characterize the I5-minute profiles and

classify and distinguish the significantly different profiles. Cluster

analysis and classification analysis were used to group profiles

together. The real I5-minute interval of load data closest to the group

average was selected to preserve the instantaneous nature of the data.

Thirteen representative profiles were selected from the load data base for inclusion into the model.

The statistical model and the load prediction scheme were

structured into an algorithm to predict the type of instantaneous profile

to use based on the I5-minute time-averaged load. This algorithm was

incorporated into the SOLCEL photovoltaics analysis computer code.

SOLCEL performs the system simulation using a full year of real load

solar and weather data to calculate the percent of energy provided to the

residence by a PV array. by the battery storage system, and by the

utility. Prior performance simulations using SOLCEL could only be

performed using hourly load data. The code wi 11 now accept hourly and

I5-minute load data. The SOLCEL code with the statistical model was

exercised to determine the effect of instantaneous loads versus time­averaged loads for calculative PV fractions.

I-11

THE 8DM CORPORATION

4. Performance Comparisons

SOLCEL simulations were performed using the statistical model

to ca 1 cu 1 ate the annual performance of a res i dent i a 1 PV system. These

simulations were performed to examine the sensitivities of the PV

fraction to variations in the time-averaged load data, array size, and

probabil ity of occurrence of an act i ve profil e. Annual performance was

simulated using 1 year of real load data at a IS-minute interval and a

year of solar radiation and weather data for Albuquerque, New Mexico~

The simulation was performed using IS-minute interval data with the

instantaneous load model and I-hour interval data. Simulations were also

performed using IS-minute interval data with the instantaneous load model

but with double the probability of an active profile occurring.

The performance comparisons are shown in table I-I. The use of the

instantaneous model, which gives the most accurate prediction, produces a

PV fraction of 33.3 percent. The use of IS-minute load data produces a

dec 1 i ne of 1 ess than 1 percent in the fraction of the load met by the

array. Using hourly load data the PV fraction decl ines by less than

2 percent. The model using twice the probabil ity of an active profile

produces a decline in the PV fraction by 1.9 percent.

The sensitivity of the PV fraction to variations in the probability

of the occurrence of an active profile and the size of an array is shown

in figure 1-3. This figure shows that the PV fractions change as both

the probability of an actiye load and array size are varied. Very little

change occurs in the PV fraction between i nact; ve loads, 0 probabil ity,

and extremely active loads, 0.4 probabil ity (twice the observed proba­

bility from the measured data). Only for the very small array (1 kw) and

an unrealistically active load (probability of 1) do instantaneous data significantly affect PV fractions.

1-12

THE 80M CORPORATION

A. INTRODUCTION

CHAPTER II

LOADS AND MODELS

The designer of a photovoltaic (PV) system must be able to predict

the total energy that will be produced by an array and the fractions of

energy that will be used directly by the load or purchased from (or sold

to) the utility. These performance predictions are required so that a

system can be properly sized to meet the demand and so that the economic worth of the system can be determi ned. In the past, the simulation models used for these performance predictions required the use of hourly

time-averaged load and insolation data because no sources of

instantaneous data were available. It was generally thought that the use

of instantaneous load data, either statistically developed or directly

measured, would produce more accurate estimates of PV fractions, that is,

the amount of energy supplied by an array to a load. The consensus prior

to the initiation of this program was that the use of hourly time­

averaged data to simulate the annual performance of a PV system in a

residential application would produce errors in calculating these fractions of energy.

The issue of using instantaneous load data versus time-averaged data

to predict energy fractions can be simply illustrated by figure II-I.

This figure shows a hypothetical PV system power output, assumed to be

constant, and a load at three different time intervals: instantaneous,

15 minute, and 1 hour. As the time interval becomes larger, the amount

of variability in the load and the PV fractions decline until, at the

longer time interval, little inferaction takes place between the PV

array, the residential load, and the utility. Two approaches were

subsequently undertaken to determine the effect of using instantaneous

data in the s imu 1 at i on of the annua 1 performance of a res i dent i a 1 PV system.

The first approach was purely statistical, involving the proba­

bilistic distribution of residential appliance usage. By estimating the

II-l

THE BOM CORPORATION

LOAD

POWER ........... ''''1 PV AVERAGE OUTPUT

o 60

TIME A (INSTANTANEOUS)

LOAD

POWER ~~~.;,;~ PV AVERAGE OUTPUT

o 15 30 45 60

TIME B (15 MIN AVERAGE)

POWER PV AVERAGE OUTPUT LOAD

o 60 TIME

C (1 HOUR AVERAGE)

BDMIA-B2-299-TR-R1

Figure II-I. Time-Averaged Versus Instantaneous Load Data

II-2

THE BDM CORPORATION

power used by a standard complement of residential appliances and the

probability of appliances being on or off at a specific time, a

statistical distribution was constructed and synthetic instantaneous

profil es deve loped. These profil es were app 1 i ed to res i dent i all oad PV

systems interactions and the PV fractions calculated. This probabilistic

approach led to a higher calculation of the activity of a load and

therefore produced more significant effects on PV fractions when

comparing smaller time interval data. This work, by Sillman (ref. 1) at

the Solar Energy Research Institute, and Lathrop (ref. 2), Clemson

University, was completed prior to the detailed measurement of instanta­

neous loads and represented the best characterization of instantaneous

load/PV system interaction available at the time.

The second approach was to measure directly the load patterns of a

variety of residences to determine the actual behavior of an instanta­

neous load. Hart (ref. 3), at the Northeast Residential Experiment

Station, measured at 5-second intervals the array output and load demand

for 3 days on one prototype PV residence and a separately-monitored

residential load. Recognizing the small scope of his sample, Hart

calculated a large variation in the amount of energy sold back to the

util i ty, depending upon the day. Hi s report recommended co 11 ect i ng more

load data during different seasons and from a variety of residences to

determine the significance of instantaneous data on PV system/util ity

interaction. Hart's work was also completed prior to the publication of

the results obtained in the Detailed Res'idential Electric Load Determina­

tion Program (DRLD) and reported here.

As part of the DRLD program, four residences in three regions of the

country were instrumented and the load measured at 5-second i nterva 1 s.

Each residence was continously mo"f'litored for 2 weeks during the summer

and winter to obtain seasonal load variations. The load data were then

stat i sti ca lly analyzed and incorporated into a probabi 1 i st i c model and

performance predictions conducted to determine the amount of energy

supplied by an array to a load.

11-3

THE 80M CORPORATION

Several major conclusions have emerged from this effort. These conclusions, briefly summarized here, are discussed in greater detail in the following chapters.

First, the measured activity of the loads was less than previously believed. Active profiles had a low occurrence in the measured data sample--less than 20 percent. Again, an active profile exhibited a high energy level above the 1S-minute average, a high squared multiple corre­lation, and occurred between S:30 a.m. and midnight. Inactive profi les closely followed the 1S-minute time average. (Refer to figures I-2a, I-2b).

Second, varying the time constant from 1-hour time-averaged load data to 1S-minute average load data to instantaneous data produced a very sma 11 dec 1 i ne in the fraction of the energy prov; ded by the PV array to the load on an annual basis (refer to table I-I). Furthermore, these percentages were relatively insensitive to increasing or decreasing the size of the array or to varying the probability of the occurrence of an instantaneous load (refer to figure I-3). In the sensitivity analyses, the probabil iy of an active load was varied from 0 (load following the IS-minute average) to 1 (load always active). The array sizes were varied from 1 kW to 9 kW.

The conclusion drawn from these measured data is that, for realistically-sized arrays and normal load applications, the use of 1S-minute data is adequate for the annual performance calculation of the percentage of array power supplied to a load by the PV array.

B. LOAD MODELING

1. Modeling Approach

The goal of the DRLD project has been to develop the capability to predict instantaneous residential load for typical residences in the northeast, southeast, and southwest, and then to assess the need for use of shorter time intervals in the modeling and analysis of photovoltaic system performance. With most of the instantaneous variability of

II-4

THE 8DM CORPORATION

residential load due to major appliance loads and usage practices, except

for the instantaneous motor starts that have 1 itt 1 e or no effect on the

array output-load match-up, consideration was initially given to esti­

mating frequencies and durations of appl iance usage in combination with

average appl iance loads to form an instantaneous load simulation model.

This approach was rejected due to the subjectivity which would enter into

the estimation of the frequencies and duration of appliance usage. The

selected approach was to establish a 5-second interval data base of.

measured load data and to develop a prediction model of instantaneous

load from the collected data. The prediction model would then serve as a

mathemati ca 1 operator whi ch converts rea 1 15-minute time-averaged data into instantaneous load data.

The largest collection to date of near-instantaneous (5-second)

load data from typical residences was accumulated under this program.

From this data base, a statistical prediction model of instantaneous load

demand, operating on 15-minute time-averaged load data, was developed and

incorporated ·into the SOLCEL Photovoltaic Analysis Program. Fundamentally

this prediction model uses the frequency of occurrence in the data base

of act; ve instantaneous load profil es, in compari son to the 15-mi nute

time average and frequenci es of occurrence of the different types of

active profiles. These active profiles were characterized by periodic

regression analysis and categorized using cluster analysis and

contingency table layouts into 13 distinct profile types which match major appliance usage.

ClaSSifying the active profiles into distinct types supports

the conclusions that the variabl ity of residential load is driven by

major appliances. In fact, the different profile types can be matched to

specific appliances or to simulataneous usage of more than one appliance.

However, instantaneous load profiles for specific appliances are simply

not available beyond the limited appliance characterizations accomplished

during the data collection phase of this program on one residence. The

profiles obtained are presented in chapter IV and discussed briefly in

the following paragraphs. Further, given the many types, brands, and

II-5

THE 8DM CORPORATION

efficiencies of appliances, development of specific appliance load profiles would not be practical.

A sample profile of two major residential electrical appliances is shown in figure 11-2. This figure was obtained by plotting the directly measured S-second data collected on the second monitored residence in Albuquerque. For the purposes of clarity, the data have been plotted at lS seconds. These profiles are part of the distinct profile groups identified and incorporated into the statistical model. The figure shows the instantaneous profile and the lS-minute time­averaged profile which a utility would collect.

The power being consumed is shown on the abscissa in units of kVA. The time interval shown on the ordinate is 1 hour from (12:00 to 13:00 hours) divided into S-minute increments. Through the appl iance characteri zat ion, it was poss i b 1 e to i denti fy the two app 1 i ance cycl es shown in the figure. The first profile (on the left) is an electric dryer. The characteristic cycle time of the heating element in the dryer is approximately 2.S minutes. The second profile (on the right) is an electric oven cycling approximately every S minutes.

The S-second interval load data base collected for this program did not include any all-electric residences. The residences monitored were selected by the respective uti 1 ity companies as being in the most typical usage class which was diversified load. All four residences use electric appliances and two contain electric forced-air-conditioning. The only major app 1 i ances not e 1 ectri ca 11y powered in any of the

residences are space heating and water heating; all residences contained gas space heating and water heating. However, both space heating and water heating would be expected to have operation cycles similar to electric ovens and dryers (shown in figure II-2) and would add only additional electrical consumption to the overall base load rather than increase load variability on a IS-minute interval basis. By varying the probability of active profiles to a completely active load, the expected load variability of an all electric residence is fully accounted for.

II-6

THE 80M CORPORATION

ALBUQUERQUE 2 MONDAY 1/18/82

10

5

o N ...

12:00

INSTANTANEOUS (16 SEC) LOAD PROFILE

~ TIME AVERAGED (16 MINI

I

---6 MIN

Figure 11-2. Instantaneous Load Profile

II-7

" .

/1 U u

13:00

BDM/A-82·299·TR·Rl

THE 8DM CORPORATION

Major findings of the DRLD Program are that active profiles do not occur as frequently as expected, and that long-term photovoltaic system-load match-ups are not significantly affected by using 15-minute time-averaged load data. These results were determined by exercising the stat i sti ca 1 model through several SDLCEL simu 1 at ions (see chapter VI). However, these results differ from the findings of three other projects recent1y completed, all of which are for all-electric residences. Analysis using the SOLCEL/Statistical Model spanned the complete range of active profiles from 0 probability (load completely inactive) to 1 (load completely active) and array sizes. Although these results are not based on all-electric residences, the results are val id in the all-electric case for the reasons discussed above. The observed occurrence of active load profiles in the measured data was a probability of occurrance of .2. Doubling this probability to .4, a reasonable estimate of the load variability in an all-electric residence, had less than a 2 percent effect on PV fractions.

2. Modeling Strategies Both Sillman (ref. 1) and Lathrop (ref. 2) utilized the load

modeling strategy of estimating probabilities of appliance usage and average durations of use and e 1 ectri ca 1 demand. Si 11 man addressed the on-site use and sell-back potential for photovoltaic systems in residential applications. Lathrop analyzed the potential for load­management in residential applications. Both projects required the development of a load prediction scheme and used event paced simulations, with events being the use of individual appl iances determined from the estimated probabil ities. (As mentioned previously. this strategy was considered in the early stages of the DRLD project, but estimating the probabilities of appliance usage -was not considered feasible. Indeed, estimating typical appliance usage practices would require quite a large sampling of residences and particular appliances.)

Results from both projects indicated that PV fractions were significantly overestimated in particular situations using I5-minute time-averaged load data and therefore a smaller time interval was

II-8

THE 80M CORPORATION

required to accurately address their particular objectives. In using

time-stepped s imu 1 ati ons based on probabil i ti es for several app 1 i ances,

even slight overestimation of the, individual probabilities used can

result in large overestimation in the aggregate over long periods. Even

though Sillman varied the probabilities over a range, the developed load

profiles appear more variable than the majority of actual data collected

under the DRLD project. The same higher vari abil ity appears in the Clemson University project (ref. 2).

The load measurement work completed by Hart (ref. 3) suggests

that PV fractions and other PV system performance information may contain

large' errors using time-averaged load data. This project collected a

small sample of 5-second interval load data and array output da~a from a

residence in the Northeast Residential Experimental Station (NE RES).

From Hart I s sample, an approach using a few days of data led to results

showing a potentially large loss of ,information using time-averaged load data and array output data.

On a~ individual day, especially a day with large fluctuations

in insolation, the errors due to time-averaged data may indeed be large,

especi ally when percentage errors of small measures are bei ng observed.

In'the DRLD project the results are for an entire year over which many of

the individual days do contain somewhat large errors. Other days contain

very little error because the array either meets almost all of the load

or very little of the load. Secondly, in the NE RES data, for the "most

representative" worst day, the percent load satisfied ,by the PV array is

extremely small--8.54 percent--which resulted in the utility providing

most of the load requirements and very 1 ittle sell-back. Of the amount

so ld back to the uti 1 ity, any error due to time i nterva 1 appears 1 arge

due to the small total amount. Hie resu 1 tant error is 1 ess than 1 kWh,

but on a percentage basis the amount sold back to the utility--as

estimated from I5-minute interval data--is underestimated by 15 percent.

Further, when compari ng I5-minute i nterva 1 resu 1 ts to 5-second i nterva 1

resu 1 ts, the amount of energy bought from the uti 1 ity is underest imated

by only 3.3 percent, even though the amount is slightly less than 1 kWh.

II-9

THE 8DM CORPORATION

This leads to an overestimation ,of the PV fraction, with 15-minute

interval results showing the array provided 4.6 kWh compared to 2.7 kWh

for 5-second intervals, or 14.7 percent versus 8.5 percent. Again, the

error appears very large, but the actual amount of the error is extremely

small when an entire year is taken into account and where these results are seen as a "worst case."

A final issue is that the Hart results (ref. 3) include 5-second

interva 1 array output data whi ch are hi gh ly vari ab 1 e, perhaps due to

weather on that winter day in the northeast region. The DRlD model ing

used hourly insolation data for the year span of the simulations, and the

weather data were for Albuquerque, New Mexico, the southwest region.

II-10

THE 8DM CORPORATION

REFERENCES

1. Sanford Sillman, On-Site Use and Sellback in Residential Photovol­

taic Systems Based on Monitored Load Data, SERI/TP-254-l739, Golden,

Colorado, October 1982.

2. Jay W. Lathrop, "Investigation of Load Management Sfrategies for

Res i dent i a 1 Photovo 1 ta i c Systems," Proceed i ngs, Th i rd Photovo 1 ta i c

Systems Definition and Projects Integration Meeting, Sandia, 1982.

3. George W. Hart, The Effect of Data Sampl ing Rate on Util ity Inter­

action Calculations, DOE/ET/20279-2l2, Lexington, Massachusetts, October 1982.

II-ll/12

THE 80M CORPORATION

CHAPTER III SURVEY OF UTILITIES AND RESIDENTIAL SELECTION

A. INSTANTANEOUS LOAD DATA INVENTORY

A comprehens i ve telephone survey was conducted of the two 1 argest utilities in each state in the three regions of concern, the Northeast. Southeast. and Southwest. A total of 51 utilities were contacted by. telephone to determine residential load monitoring activity and the availability of instantaneous residential load data. Each utility was asked about load monitoring programs. including the sampling interval and the sample size of residences being monitored. A summary of the survey is shown in table III-I.

As can be seen from the table. many utilities are measuring residential load data directly. The utilities are required under Public Utilities Regulatory Policy Act (PURPA) of 1978 to gather and report representative load data for large rate classes. This requirement provides a very large data base of time-averaged load data that can be used in SOLCEL with the statistical model for residential PV system simulation.

The general conclusions of the survey were that the majority of the utilities are monitoring an average residential sample of approximately 300 homes at a time-averaged interval of 15 or 30 minutes which is adequate for PURPA requirements; however. no utility,was monitoring resi­dences at smaller intervals. Utilities did. however. indicate a strong interest in the resu 1 ts of the program and a wi 11 i ngness to cooperate with the direct measurement task.

In addition to the survey of'- utility companies, other sources were contacted to determine the ava 11 ab11 ity of instantaneous data. These sources included the Edison Electric Institute. the Electric Power Research Institute. the Association of Edison Illuminating Companies. the Federal Energy Regulatory Commission. and load recorder manufacturers. A 1 iterature search was initiated for residential electric load research

III-l

THE BOM CORPORATION

TABLE III-I. SUMMARY OF UTILITY TELEPHONE SURVEY

UTILITY COMPANY CONTACTED COMMENTS

1. ALABAMA POWER COMPANY 265 P/I 15 TAPE YES ·SOUTHERN CO. SERVICES

2. SOUTHERN ELECTRIC GENERATING COMPANY

3. ARIZONA PUBLIC SERVICE 15 COMPANY

4. TUCSON GAS & ELECTRIC 15 COMPANY

5. ARKANSAS·MISSOURI POWER NOT DOING MEASUREMENT

COMPANY CONSOLIDATING WITH ARKANSAS POWER & LIGHT

6. ARKANSAS POWER & LIGHT 260 P/I 15 TAPE YES

7. CONNECTICUT LIGHT AND POWER COMPANY

8. HARTFORD ELECTRIC LIGHT MONITORING BY CONNECTICUT

COMPANY POWER & LIGHT

9. DELMARVA POWER & LIGHT 700 P/I 15 TAPE

COMPANY

10 POTOMAC ELECTRIC POWER 250 P/I 15 TAPE YES

COMPANY

11. FLORIDA POWER & LIGHT 1.8 MIL PII 15 TAPE YES COMPANY,

12. FLORIDA POWER CORP. UNK PII 15 TAPE YES WILL BEGIN IN 1981

13. GEORGIA POWER COMPANY 300 15/30 TAPE YES ·SOUTHERN CO. SERViCES

14. SAVANNAH ELECTRIC AND 200 15 TAPE YES

POWER COMPANY

15. LOUSIANA POWER AND 354 15 YES

LIGHT COMPANY

16. GULF STATES UTILITIES 340 kW 15 YES

COMPANY

17. BANGOR HYORO·ELECTRIC NOT DOING MEASUREMENT

COMPANY

18 CENTRAL MAINE POWER CO, 337 15 TAPE YES

19. BALTIMORE GAS & ELECTRIC 225 kW 15 TAPE YES

COMPANY

20. POTOMAC EDISON COMPANY NOT DOING MEASUREMENT

21. BOSTON-EDISON COMPANY 15

22. NEW ENGLAND POWER CO. 250 kW 15 TAPE YES

23. MISSISSIPPI POWER CO. 200 kW 30 TAPE YES ·SOUTHERN CO. SERViCES

24. MISSISSIPPI POWER & UNWILLING TO ANSWER QUESTIONS

LIGHT COMPANY

25. PUBLIC SERVICE COMPANY 105 kW 30 TAPE YES

OF NEW HAMPSHIRE

26. GRANITE STATE ELECTRIC -NEW ENGLAND POWER CO. NOT DOING MEASUREMENT

27. PUBLIC SERVICE ELECTRIC kW 15 TAPE YES

AND GAS COMPANY

1II-2 BDMIA-82-299-TR-Rl

THE 80M CORPORATION

TABLE II I -1. SUMMARY Of UTILITY TELEPHONE SURVEY (Conel uded)

UTILITY COMPANY CONTACTED COMMENTS

"SUBSIDIARY Of

28. JERSEY CENTRAL POWER & NOT DOING MEASUREMENT AND LIGHT COMPANY "GENERAL PUBLIC UTILITIES

29. PUBLIC SERVICE COMPANY kW 15 TAPE YES OF NEW MEXICO

30. NEW MEXICO ELECTRIC NOT DOING MEASUREMENT SERVICE COMPANY

31. NIAGARA MOHAWK POWER 100 kW 15 TAPE CORPORATION

32 CONSOLIDATED EDISON CO. 400 kW 15 TAPE YES OF NEW YORK, INC.

33. AMERICAN ELECTRIC POWER 15

34. DUKE POWER COMPANY 220 15 TAPE NOT REALLY

35. CAROLINA POWER & LIGHT 400 15 TAPE YES COMPANY

36. OKLAHOMA GAS & ELECTRIC 714 kW 15 TAPE YES 15 kW P/V DEMO. COMPANY

37. PUBLIC SERVICE COMPANY 185 kW 15 TAPE YES OF OKLAHOMA

38. PHILADELPHIA ELECTRIC 390 kW 30 TAPE COMPANY

39. PENNSYLVANIA POWER & 320 kW 30 TAPE YES LIGHT COMPANY

40. BLACKSTONE VALLEY 282 15 TAPE YES ELECTRIC COMPANY

41. NARRAGANSETT ELECTRIC 250 kW 15 TAPE YES "NEW ENGLAND POWER SERVICES COMPANY COMPANY

42 LOCKHART POWER COMPANY NOT DOING MEASUREMENT

43. SOUTH CAROLINA ELECTRIC 300 kW 15 TAPE YES AND GAS COMPANY

... KINGSPORT POWER COMPANY 120 30 TAPE "AMERICAN ELECTRIC POWER SYSTEM

45 TENNESSEE VALLEY 500 kW 15 TAPE YES SAMPLE SIZE IS TOTAL IN

AUTHORITY COMMERCIAL, INDUSTRIAL, RESIDENTIAL

48. HOUSTON LIGHTING AND POWER 15 COMPANY

47. TEXAS POWER & LIGHT CO. 250 kW 15 TAPE YES

48 VERMONT ELECTRIC POWER NOT DOING MEASUREMENT

COMPANY INC.

49. VIRGINIA ELECTRIC & POWER 322 30 TAPE YES COMPANY

50. APPALACHIAN POWER CO. 170 kW 15/5 TAPE "AMERICAN ELECTRIC POWER COMPANY

51. MONONGAHELA POWER CO. REFERRED TO ALLEGHENY POWER SERVICE COMPANY

1II-3 BDM/A-82·299·TR·Rl

THE BDM CORPORATION

through the National Technical Information Service. Since the survey of

utility companies and other data sources did not uncover any sources of

instantaneous load data, the direct measurement program was begun.

B. SELECTION OF PARTICIPATING UTILITIES AND RESIDENCES FOR DIRECT MEASUREMENT

The Pub 1 i c Servi ce Company of New Mex i co (PNM), Albuquerque, New

Mexico; Public Service Electric and Gas Company (PSE&G), Newark, New

Jersey; and the Georgia Power Company (GP), Atlanta, Georgia were

selected for participation in the program. These utilities were selected

on the basis of their willingness to participate in the program, the

extent of their residential load monitoring programs, their involvement , in load management and load research, and their abil ity to provide regional representation.

Individual residences were selected by the uti 1 ities as typical of

residences in each service area and on the basis of electric consumption.

In addition, the selected residences were already part of the utility's

exi st i ng load mon i toring program so that para 11 e 1 uti 1 ity time-averaged

data cou 1 d be compared wi th the instantaneous data co 11 ected under the program.

C. RESIDENTIAL CHARACTERISTICS

PNM provided a list of 10 residences from their load monitoring

sample for consideration. These residences were selected by the utility

as being typical of their service area. In addition, time-averaged load

data for 1 year were available on.~ach residence for comparison with the

instantanous data. Each homeowner was contacted and asked to participate

in the load monitoring program. The two residences finally selected

represent a mix of new and older residential construction. However, as

is typical of residences in PNM's service area, both residences use

natural gas for domestic water heating and space heating and evaporative

III-4

THE 80M CORPORATION

air conditi oners for cooling. The i nstrumentat ion was direct ly mounted to the residences in Albuquerque because the cooperation of the i ndividua 1 homeowners was requ ired for equ ipment test and cali brat ion, residential appliance characterization, and the power line waveform quality measurements.

I n contrast, PSE&G and GP se 1 ected res i dences but did not want the homeowner involved in monitoring activities. The load monitoring instru­mentation was mounted directly on the utility service pole. The resi" dences in Newark and Atlanta, as is typical of residences in these areas, a 1 so use natural gas for domest i c hot water and space heati ng, but use central air conditioning.

The characteri st i cs of the four mon itored res i dences are shown in table III-2.

TABLE 1II-2. RESIDENTIAL CHARACTERISTICS

Albuquerque Albuquerque Atlanta Newark #1. #2

Space ft2 1,200 1,950 2,500-3,000 4,300* Age 25 2 16-20 Heating Gas Gas Gas Gas Cooling Evaporative Evaporative 3TN Central AC 6TN Central AC Domestic Hot Water Gas Gas Gas Gas Appliances Electric Range X X X Gas Microwave X Dishwasher X X X Clothes Dryer Electric Electric Gas Washer X X X X Refrigerator/

Freezer X X X X Freezer X X Color Television X X X

Occupants 3 4 3 4

*includes 1,300 ft2 basement

III-5/6

THE BOM CORPORATION

CHAPTER IV DIRECT MEASUREMENT

A. DATA ACQUISITION SYSTEM - OVERVIEW

The data acquisition system consisted of two major elements: an instrumentation package located at the monitored houses and a centralized data collection and analysis facility.

The instrumentation package at each house was a complete, automated, self-contained system. It measured the power being consumed by the house, conditioned the data, and sent it in digital form over telephone lines to the central faci1fty (see ffgure I-I).

This system configuration had several advantages. First, it allowed the instrumentation located at the monitored houses to be put in a weather-proof (NEMA 3) enclosure and mounted on the outside of the house. This proved to be a very "clean" installation whose appearance was not significantly different from the circuit-breaker boxes located on a typi ca 1 home. The enclosures, meter, modem, and computer contro 11 er are shown in figure IV-I.

The second advantage of this configuration was that it did not require the presence of test personnel at the monitored home to acquire data. This ensured that the activities and energy consumption of the home being monitored were not affected by the presence of test personnel and also made data co 11 ect i on more conveni ent at the sites remote from the data recording facility.

The current and voltage being consumed by the homes was measured by a Joule Electronic Meter (JEM). This is a single-function. two-element kVAh meter that provides both a dl~ital and an analog pulse output. It has two inputs for ac voltage and two inputs for ac current. The JEM instantaneous ly measures current and vo 1 tage. Each input is instanta­neously squared which produced a dc component proportional to the average square of the signal being sampled. After some filtering and amplifica­tion. the squared voltage and current proportional signals are multiplied

IV-l

..... < I

N

SWITCH BOX AND POWER SENSORS (CURRENT TRA~SFORMERS)

PACKAGED ENCLOSURES

,

JEM-l METER & MODEM

HP 85 DATA ACQUISITION SYSTEM @ BDM IN ABQ

Figure IV-I. Detailed Residential Load Data Acquisition and Load Monitoring System

-I :x: m OJ o s:: (j

o :0 "tI o :0 }> -I o Z

THE 8DM CORPORATION

and the product square rooted. The resultant signal is then proportional

to the square root of the product of the average squared current and

average squared voltage. This result gives true rms voltamperes. The

simplified equation of the function is:

Additional circuits provide an analog output signal and an input to an.

integrator which generates kVAh pulses. The voltage inputs of the JEM

were connected directly to the two hot wires of a standard three-wire,

single-phase residence service. Current inputs for the meter were

acquired using step.:.down current transformers. Because of the lO-amp

maximum current input of the JEM, two square-D model 5N201 current trans­

formers (each rated at 200:5 amperes) were connected in parallel to

monitor each of the hot wi res of a standard three-wi re power drop (see

appendix 0 for the system's schematic wiring diagrams).

TEK-COM modems were used on both the JEM and computer ends to inter­

face the equi pment wi th the phone 1; nes. A 20-mi 11 i amp current loop

interface was used to connect the JEM to the modem, and an RS-232C inter­

face connected the HP-85 to the modem. The modem connected to the JEM

was a 300/baud auto-answer modem that was able to answer the phone when

the modem on the HP-85 queried it. The modem on the HP-85 had an

intelligent dialing feature capable of storing five telephone numbers,

anyone of which could be dialed as instructed by the controlling

computer.

The modem JEM meter combination and switchgear were housed in a

weather-proof NEMA case mounted on the sides of the Albuquerque houses

and on telephone poles at the Afl anta and Newark houses. A NEMA 3-R

enclosure was used for the switchgear and a NEMA 12-R enclosure for the

JEM and modem. The enclosures were designed to provide an overall power

switch that permitted safe access to the power line connection points for

direct residential mount applications, a GFI power outlet for use with

test equipment, surge protection, temperature control for the enclosures

IV-3

THE 80M CORPORATION

containing the JEM and the modem units, and outlets which were used for

power 1 i ne waveform qual i ty monitori ng. The des ign met the Nat iona 1

Electric Code and state codes and used UL approved components.

A thermostatically controlled heating blanket combination was used

in the meter modem enclosure to maintain operating temperature- for the

instrumentation. Temperatures within the box were maintained at between

650 and gOoF. For the cases where the boxes had to be mounted in direct

sunlight, a flat metal plate covered the front surface of the box and was_

separated by an air space of approximately 1/2 inch. This resulted in a

thermal "chimney effect" that aided in carrying away excess heat that

could build up in direct sunlight or summer heat.

The autodialing modem connected to the HP-85 computer was capable of

storing and recall i ng up to fi ve telephone numbers ina non-vo 1 atil e

memory. Commands to the autodial ing modem, in addition to communication

with the JEM at the house being monitored, were completed through an

RS-232C interface. Termi na 1 emu 1 ator software was used to rna i nta i nand

update the telephone-number list within the modem.

The HP-85 computer/controller system was programmable in the Basic

computer 1 anguage. The HP-85 has 32k of i nterna 1 random access program

memory (RAM). The HP-85 also has a built-in cassette tape drive which

facilitates internal program storage. Collected data were plotted using

the HP-85's built-in graphics capability. In addition, a thermal printer

easily generated a hard copy either of text or graphics. Once collected,

data were transferred across the HP-IB bus to an HP dual floppy-disk

drive. Each of the 5-1/4 inch disks in the HP drive were capable of

storing 270k bytes of information, more than sufficient to contain a

single day's worth of data.

The computer/controller haraware was also used to effect the

transfer of collected load data to the Sandia National Laboratories com­

puting system for further analysis. -This transfer was accomplished

through the telephone lines to Sandia's dial-up network. Once

transferred to Sandia, the data were reformatted and stored on magnetic

tape.

IV-4

THE BDM CORPORATION

B. DETAILED RESIDENTIAL ELECTRICAL LOAD DATA ACQUISITION SOFTWARE

The data acquisition software was written on the HP-85 and controlled all processes involved in the data acquisition. A timing flag allowed the HP-85 to query the JEM meter for electrical load data at precise 5-second intervals. Occassionally. the JEM did not respond when queried because of noise in the telephone lines. This left the program and/or the RS-232C interface waiting for a response and tended to discon:: nect the HP-85. After some experience with this type of failure. a timing system built into the program corrected the problem by requerying the JEM until it responded. Each time the JEM meter failed to respond. a message was printed on the thermal paper giving the time that the failure occurred. Also. an audible tone issued roughly once every 5 seconds until the problem either was corrected or the noise stopped and the communications were re-established automatically.

The JEM meter was queried by five-character ASCII strings. A typical five-character string is llFll. which commands the JEM to read what is in its accumulating register and to store that value in a storage register. A string of llLll commands the JEM to transmit what is in the storage register. So. in order to read data, it is necessary first to freeze what is in the accumulating register and the storage register and then read the contents of the storage register.

C. SYSTEM OPERATION

When the HP-85 is turned on, the data-acquisition program is loaded from the built-in magnetic tape. When first executed. this program enters as the terminal emulation ·mode. This mode is used to establ ish communications through the auto-dialing intelligent modem with the JEM and to test-query the JEM. A manual check of data ensures that the computer system is receiving the data properly. Once it has been estab­lished that the data are being transmitted correctly, the program is put into the data-collection mode where the programmer is asked to calibrate

IV-5

THE 80M CORPORATION

the clocks within the HP-85 with the time of day and to establish a time

when data collection should begin.

The user was prompted at the end of each 24-hour period to load a

disk in the appropriate drive. For example, the first drive used by the

program is "Drive 0" while the second drive is labeled "Drive 1." At the

program start, the user is prompted to load an even-numbered disk in

"Drive 1." At the end of 24 hours, the program has completed writing

data to "Drive 0" and proceeds to write data to "Drive I" where it auto­

matically prints a message to load an odd-numbered disk in "Drive D." When the character string is sent to the JEM meter, the JEM responds

with an 80-character string of information. That information includes

both the kVA information and the kVAR or reactive power-information. The

HP-85 extracts the kVA from the BO-character stri ng and merges it with

time of day information. The accumulated kVA and time of day are merged

and integrated into each element of a 12-element array. When this array

is filled, once every minute, the data are dumped to the floppy disk.

D. SYSTEM CALIBRATION AND CHECKOUT

Prior to the installation of the load monitoring instrumentation at

a residence, all equipment was calibrated and tested. The JEM has an

analog output for kVA and a serial digital output for kVAh. Both of

these outputs were calibrated using a load resistor which was turned on and off for fixed periods.

The analog output was monitored using an Esterline-Angus strip chart

recorder. JEM digital readings and Esterline-Angus readings agreed to

within 0.4 percent when checked against cal ibrated loads. The square-D

current transformers were calibrated by PNM and were found to be accurate to within 1 percent.

The instrumentation was configured as it would be used in residen­

tial load monitoring and installed in a test bed, the 8DM Prototype

Development Laboratory, a 6,000-square-foot research facility. The

actua 1 load of the facil ity was then moni tored for fi xed test periods.

IV-6

THE 80M CORPORATION

The readings from the JEM meter were compared to the continuous output of

Esterline-Angus strip chart reporter, and the power measured by each

instrument agreed to within 0.4 percent. Actual load data were taken at

4-second i nterva 1 sand compari sons made of different samp 1 i ng i nterva 1 s

from 4 seconds to 1 minute to define the correct sampling interval for

residential monitoring. A problem was encountered initially with the

pulse output of the JEM (15,744 pulses per hour) which produced signifi­

cant round-off error when the data were plotted at a 4-second sampl ing

interval. In order to overcome this count error, the circuit boards of

the JEM were modified to quadruple the pulse output (62,976 pulses per

hour), thereby greatly improving accuracy. Although 4 seconds

represented the minimum sampling interval of the JEM, a 5-second interval

was selected to allow for data transfer within the HP-85.

E. RESIDENTIAL LOAD MONITORING

The two Albuquerque residences were instrumented using the JEM and

modem enclosure and switchgear, which was direct-mounted to the resi­

dence. In both Newark and Atlanta, the JEM (modem enclosure only) was

mounted to a utility service pole. Table IV-l summarizes the total data

collection by residence. A more detailed discussion of the load data

base by region, season, and weekday/weekend can be found in chapter IV.

TABLE IV-I. DATA COLLECTION BY RESIDENCE

Hours of Load Data Collection

ALB-l ALB-2 Atlanta Newark

Summer 215.5 224.5 149.5 190.5 Winter 146.5 104.0 155.5 84.0

Total 362.0 328.5 305.0 274.5

IV-7

THE 80M CORPORATION

Load data were collected through local phone lines. A data collec­

tion team was on-site in Newark and Atlanta to collect summer and winter

data. In Newark, the original residence selected for monitoring

exhibited very small variation in the monitored load due to the use of

natural gas for cooking and domestic water heating. The instrumentation

package was removed and installed at a larger residence by PSE&G. The

resi dence exh i bi ted considerable load vari abil i ty during the summer due

to central air conditioning, but in the winter exhibited little activit:£.

beyond small appliance usage due to use of natural gas for cooking, heating, and domestic hot water.

F. RESIDENTIAL APPLIANCE CHARACTERIZATION

Res ident i a 1 app 1 i ance profil es were character; zed at the two

A lbuquerque res idences to determi ne the signatures of speci fi c app 1 i­

ances. The characteri zat i on was conducted to prov i de both a means of

visually examining the load data plotted by the HP-85, (to determine what

appliances were on) and to calibrate the JEM against the more accurate

strip chart recorder. These app 1 i ance profil es were co 11 ected us i ng the

Esterline-Angus chart recorder connected to the analog output of the JEM.

During the appliance characterization, the JEM was simultaneously trans­

mitting digital load data to the HP-85 which was used for comparison and

calibration. The analog and digital appliance profiles are shown in

figures IV-2 and IV-3. The digital JEM data were sampled every 5 seconds

and then plotted every 15 seconds as shown in figure IV-2. The power measured by each device agreed to within 0.4 percent which verified the

calibration performed in the laboratory. During the test, all appliances

in the residence were disconnectea,' then connected and allowed to cycle

so that the Signature could be obtained. Both residences contained

electric ovens; only one residence contained an electric dryer. Each had

the normal complement of small appliances.

The Esterline-Angus sampled and plotted the load every 0.5 seconds

and was 'able to identify motor starts which were not seen in the JEM

data. These however did not prove significant in measuring the power consumed'.

IV-8

THE BDM CORPORATION

kVA

4.0

kVA

3.68

:xi .... '" .... '" ~ ... '" 0 0

'" 0

.Q

;;:

~ ~

" I-... ... .. ... V')

JEM Data/HP-85 Plot

JEM OATA RECORDED DURING TEST

Figure IV-2_ JEM/HP-85

" u " m >

" '" ... " 0. .. ...

ESTERLINE ANGUS (Strip Chart Output)

" " ... " .. "

.., " > U m

0 m " :IE: U,. ~ '" 1- ... ... " § L~ ~ ::: U " 0"

" .. u-" .. .. 0_ ... 3 ,. u",

Time (5 min Per Division) (Esterline Angus Recorder)

Figure IV-3_ Esterline-Angus

IV-9/10

BOM! A.a2·299-TR-R 1

:xi .... '" .... "' .. ... '" '" ... 0 ... ..

N

" " " .Q

" .. ... ;;: u "-",5 " .. m Q.m > " E:E 'U ... ... ~ m <C ... ~ ... '" .. , ..

" ... ... "

... I-... - " .. on .... Q. .. e e " .. " .. 0 - " "'''' ... I- ... ...

BOM/A.a2-299-TR-Rl

THE BOM CORPORATION

CHAPTER V

STATISTICAL MODEL DEVELOPMENT

A. INSTANTANEOUS LOAD DATA BASE

1. Data Base Development Procedures

Residential load data were collected at 5-second intervals via

the JEM, modem, and HP-85 configuration in the form of integer counts

(number of revolutions of the meter). These integer counts for up to

24 hours in duration were stored on floppy disks. Using the HP-85

terminal emulator capability and RS232 interface, contents of each floppy

disk were transferred to the Sandia Network Operating System (NOS) com­

puter and stored on files.

The transfer to the Sandia computer was over phone lines, and

interference and noise problems occurred that caused some holes and

extraneous characters to ari se in the data files. These errors were

detected through observation of the file listings and screenings with a

text editor. Most of these errors (i.e., one missing 5-second data

point) were easily corrected, with missing data items estimated by

comparison with surrounding data points. Some transferred files did

contain enough noise to require retransfer of the file. To complete the

verification of the transferred data, the files were read via a Fortran

program with a formatted read and also processed through a Biomedical

Computer Program, P-Series (BMDP) Statistical Package, program which read

the files alphanumerically and listed frequencies of symbol occurrences.

Once the data transfer and verifi cat i on were comp 1 eted. the

count measurements were converted" into kVA units and the data were

structured into data base format. A conversion constant had been previ­

ously derived which accounted for the time interval of the count, and a

kVA per count constant determined from calibration of the JEM. The

processing of the data was then accomplished with a Fortran program which

converted the data; broke the data out into I5-minute blocks with

V-l

THE 8DM CORPORATION

associ ated time of day, day of week, house, and season; and cal cu 1 ated other descriptive information.

Eight load data base files contain the instantaneous load data collected, broken out into separate files for each of the four residences for winter and summer. An example~of two continuous IS-minute intervals as they are confi gured in the load data base is shown in append i x A. Procedures for accessing the data base on the SNL computer under the NOS are also detailed in appendix A.

2. Data Base Description

A summary of the duration of the data sample for each residence is presented in figures V-la, b, c, and d. The figures contain the header records from the load data base pertaining to the contiguous blocks of lS-minute intervals of collected data. Data collected in Newark and Atlanta were not as continuous as in Albuquerque, due to phone line problems. This resulted in an increased number of blocks of continuous data each with somewhat smaller continuous durations.

The load data base contains 5,080 complete, and mostly contin­uous (as can be seen from the figures), IS-minute intervals (1,270 hours) of ..!i-second sampled data. Tables V-l and V-2 display a breakout of the sampled intervals by residence, season, and day of week. These tables indicate that during the monitoring period an adequate sampling of each region, season, and weekday/weekend was obtained. In most cases, data 'collection was planned to include two weekends to allow determination of whether weekdays and weekends display different instantaneous load patterns. Summer data collection in Newark resulted in data which displayed very steady, but varying, loads. Since steady loads are modeled by the IS-minute average, and prel iminary winter results from Newark displayed the same steady loads, the winter data collection period in Newark was shortened.

More emphasis was given to load data collection of the Albuquerque residences since monitoring could· be conducted without travel. In addition, all 96 quartiles (15-minute intervals in the day) were sampled, with an average of 53 l5-minute interval samples per quarti le.

V-2

THE 80M CORPORATION

NEhlAR-f( TH'-' 7/16/31 16:45-24: OO~ 29- QUARTILES. 68 - % ATLNTA-- FfH 7/ 3/81 19:1)0-2:4:1)0. 20 I~Ut1RTILES~ 77 - 96-NE~~ F!>I 7/17/91 24:00-16:30" 66- GlUARTILES,. 1 -~ ~TLNTA- SFt-T 7/ 4/81 24: 00-18:4.5, 75" QUFtRTILES, I - 75 NEWAP+( F!>I 7/17/91 16:4!!i-:4: I)O~ 29 t:;}UARTlLES,. 69 - % A-TLNTA- SA-T 7/ ...... 81 19= 0(1-24: 00, 20 I~IJAATILES., 77 -% NEWAA~ Sf>T 7/18/9-1 a:4:00-e4:00,. % QU~TILES,. 1 -% A-TLNTA- SUN 7/ 5/31 24:00-18:45~ 7'!i QUARTILES., I - 75 NEW_ SlJtt 7/19-/91 24: 00-16=30, 66 QUARTILES. 1 -66 A-TLHTA- THU 7/ 9-/91 17:30-24:·00, 26 QUFtfHILES,. 71 -. % NEWAPJo( SUN 7/1'9-/131 16:~-2""-=OO,. ago. QUt+PTILES,. 68 - % fHLNTA- ~I 7/10/91 24: 00-17: 1~!" &9- OOART ILES,. 1 -~ NEItlAAI( I'ID" 7/20/81 24101)-16-= 30, 66 QUFtfHILES,," 1 - 66 ATLHTA- FR-I 7/10/81 17:30-24-: 00,. 26 QUAR-T I LES, 71 -% HE, ... FtPoI( MON 7/20/91 16: 4!3-~ 00" 29- QUFtRTlLES,. ,;a-% FtTLNTA- SA-T 7/11/91 24-: 00-17: 15,. 69- QUARTILES, I - 69-HEWAAW nJE 7/21 .... ·~11 24.:00-16:30, % QUARTILES,.. 1 -66 FtTLNTlT SAT (/11/81 t7:30-e4-:00!" 26 QUAATILES. 71 - 96 Me~~ T'JE 7/21/13'1 1&:45-24:00, 29- QUFtRTILES,. 6&-% FtTLNTA- SUN 7/12/81 24: 00- Q: 30, 2 QUftRTILES~ 1 - 2 f'fe~ "'ED 7/22/91 24:00-24=00, % QUMTILES,. 1 -% FtTUiTFt- SUN 7/12 ____ 81 1: 1"-24' 00. ~1 QUAATILE-S, 6- - % NEt.Jf\~~ Tl#J 7/23/91 ~4: 00-16=30, 66 QUf,fHlLES,. 1 -66- FtTLHTA- !"tON 7/13/81 24-: 01)- 1= 00, 4- QUAA·TILES, 1 - 4 Nebl~1oE: THU 7/23."81 16: 4-'5-244 I) I), 29- QU~TlLES,. ,,8 -% NEbI~K F!>I 7/24.-'81 24'00-16:30" 660 QUFtRTILES,. I - " .. FtTUiTA- F'tDN 7/13/81 1.;. t~-2:4: OO~ 91 QUt1RTILES!" .; -%

FtTLNTA-- TUE· 7/14 .... 81 24:00- 1:00, 4- I'-UAA-TILES·, I - 4-Nel.JAR-1( Tl#J 1/;t.8/82 20: 15-24: 01), 15 QUftR.TILES,. 82 - % HElJFtP¥- FI>I 1/29-/82 24: 00- 2= 00" 8 QUFtfHILES,. 1 - 8 "'+EI.o:IAPW SITT 1/:30/82 4: 30-24-101),. 79 QUAR-TILES,. 19- - % I'tEI .• ARK SUH V"31/82 24: 00- 3: 1~,. 13 QUAATlLES~ 1 - 13 NEI.,jARK SUN 1/31/~2 12~30-24;QO!" ~ QUftR-TILES,. 51 -% NE",ARW I'IDN 1/32/92 24'00- 3'''', 15 QUAR-TILES,.· 1 - 15 t'lewFtPI( MIlI'f 2/ 1/82 9: 1~-19:4-'5,. 39 QUARTILES. 38 - 75 t'lE~AAI< TUE 2/ 2/92 21 :4!5-2of.: 00" 9-- QUMTlLES,. 88 - % NeWARf( I.ED 2/ 3/82 2 .... :00- ':10'1> 22 QUARTILES,. 1 -22 HelolFtP~ WED 2/ ·3/82 S>: 1'-14: 15'1> 20 QUFtRTlLES,. 39 - 57 r-IEI."AAK weD 2 ..... ~ve2 21:30-24: 00,. 10 QUMTILES,,· 87 - % HEWAAK THIJ 2/ ....... ·82 24: 00- 2; 00,. 9 QUAR-TILES,. 1 - 8 NEWARK SA--T 2/13/"82 12: 00-24: 1)·0,. 4S GUART ILES-,. --% NEltIA~1{ SUN 2/14,.00-82 24: 00- 1:.30, 6 QUARTILE-S,. 1 - ,; END OF FILE

A-TLNTA- MON 2/22/82 16: 00-2.4-: 01), 32 I;!UFtRTILES!" 65 - '?; F"tTLNTA- TUE 2/23 .... 82 24-: 00-15:4,5, 63 QUARTILES, I - 63 FtTLNTFt- TUE 2/23/82 16-:00-24;00, 32 QUAR-TJLES~ 65 -% F"tTLHTA- 'tiED 2· .... 24 .... 132 24: 00-15: 4-~!" 63 QUAATlLes. 1 - 6-3 ATLNTA- I.<JE-D 2/24/82 16:00-2:4:00, 32 j)UFtIHILES" ,,5 -'?;

A-TLNTA- THO 2/25/82 24: 00-15: "'''j., 63 QUARTILES, I - 63 ATLNTA- THU 2/25~"82 16: 00-24: 00, 32 QUFtRTILES, 65 -% FtTLNTA- FR-I 2/26 ..... 92 24: 00-2oW 00, % QUAATILES,. 1 -% FtTLNTA- SAT 2/27/::'2 24:00- O~ 30, 2 QUARTILES, 1 - " FtTLNTA- SA-T 2/27/82 0:45- 6:0(1- 21 QUARTILES, +- - 2 .... FtTLNTA· SAT 2/27/82 11: 15-24: 01), 51 QUAATILES, .... -% fHLNTA- SlIr+ 2/28 ____ 82 24:1)0-24:00, % QUARTILES, 1 -% FtTLNTA- MOti 2/29/82 24: 00- 9-: 4-'5, 39- QUAATILES,. 1 - 39-END OF FILE )7

>1

Figure V-lao Newark Data Collection Summary Figure V-lb. Atlanta Data Collection Summary

ALBU-l THIJ 10/1:5 ..... 31 21: 15-e4: 00, 11 QUARTILES, 96 - % At....B1J-1 FlU 10""-1&/81 24: 00-24: 00, 96- GlUAR-TILES, 1 -% ALBU-l SfH 10/17/81 24' 00-24: 00, 96 QUARTILES, I -% ALBU-l SUN 10/19/91 2. .... :1)0-21\00, 84 QUFtRTILEt, 1 - so. ALBU-l SUN 10/18/81 21:1~-24:00, 11 QUART I LES, 86-% ALBU-l MON 10/19-/81 24:1)0-21:00, "84- QUA-RTILES, I - 8 .. FtLBIJ-l MON 10/19/81 21: 15-2:4: 00, 11 QU~TILES,. 9" - % FtLBIJ-l rUE 10/20....-81 24: 00-2: .... : OO!" % aUAR-TILES, I -% ALBU-l weD t 0/21/81 24-: 00-24: 01), % QUAATILES, 1 -% FtLBU-l THIJ 11)/22/81 24-: 00-24: 00, % QUARTILES!" 1 -% FtLBU-l FlO I 10/2:3/91 24: OQ-e4: 00, 96- QUAFHILES, I -% ALBU-1 Sf>T 10,..24..-131 24: 00-21 :IS!" 8:5 QUFlRTJLES~ I - 95 ALBU-l I"JED 12/30/81 21:30-24: 00, 10 QUAPTILES,. 87 - % ALBU-l THIJ 12/31/131 24: 00-24: 00, % QUARTILES,. I -% PLBIJ-l F'-I 1/ 1/82 24: 00-24: 00, % QUARTILES,. 1 -% ALBIJ-t SA-T 1/ 2/82 24-:00-24:00, % QUFtRTILES" 1 -% ALBU-l SUN 1/ 3/82 24: 00-24= 00, % QUAATILES, I - 96 FtLBU-l I'IDN 1 .... 4/$2 24; 00-24: 00, 96 QU~TILES, I -% ITLBU-l TliE 1/ 5/82 24: 00-24: 00, 96 QUAATILES, 1 -% END OF FILE )?

Figure V-Ic. Albuquerque #1 Data Collection Summary

V-3

ALBIj-2 FlOI 9 .... 11/91 1'9-: 00-24: OO!" 20 QUt1RTILES, 77 -% ALBU-2 Sf>T 3-/12."8-1 24: 00-2:4: OO!" % QUAR-TILES, 1 -% FtL1W-2 SUN ~"'-13/81 24: oo-a4: 00, % QUARTILES,. I -% f>LBU-e MON 9-/14/81 24: 00-24-: OO~ % QUFtRTILES,. 1 -% ALBU-2 TUE 0;./15/81 24:00-24:00, % QUARTILES, I -% ~LBU-2 IJ;lE-D 9-/16-/$1 24: 00-24= 00., % QUAATI LES. 1 -% ALBU-2 THO) 9/17·"31 24: 00-24-:-00, S QUAATILes, 1 -% ~LBU-2 F'-I 9/19/91 24:00-24:00, % QUftfHILES, I -% FtLBIJ-2 SA-T 9-/19/91 24-: 00-24: 00, % QUARTILES, I -% FtLBU-~ ."SUN 9/20/8-1 24: 00-24: 00, % QUARTILES, I -% FtLBIJ-2 MON 9/21/81 24: 00- 3:30!" 14 QUFtR-TILES, I - 14-ALBIJ-a F'-I 1/15/82 17:4.'5-24:00, as QUARTILES, 72 -% FtLBU-2 S"'T 1/16/82 24:00-24:00, % QUFtRTILES, I -% ALBU-e SUN 1/17/82 24:01)- 1 :4-"3~ 7 QUA-RTILES, I - 7 At...lIU-2 SUN 1/17/92 12:00-a4:00!" 4S QUAATILES, 4~-% ALBU-e MIlN V18/92 24: 00-24: 00,. 96 QUARTILES, 1 -% ALBU-2 TUE 1/19,..'82 24: 00-24: 00, % QUAR-TILES, I -% ALBIJ-2 ~ED 1,..'20/92 24: 00-12= 00, 48 GlUFtRTILES,. 1 -49 E"I'tD OF FILE )7

BDMI A-82·299-TR-R 1

Figure V-Id. Albuquerque #2 Data Collection Surrmary

THE 8DM CORPORATION

TABLE V-I. LOAD DATA BASE SUMMARY

NEWARK ATLANTA ALBU-1 ALBU-2 TOTAL

15-min intervals 762 598 862 898 3120

SUMMER hours 190.5 149.5 215.5 224.5 780.0

15-min intervals 336 622 586 416 1960

WINTER hours 84.0' 155.5 146.5 104.0 490.0

15-min intervals 1098 1220 1448 1314 5080

TOTAL hours 274.5 305.0 362.0 328.5 1270.0

.

MON TUE WED THUR FRI SAT SUN TOTAL

15-min intervals 395 291 288 353 584 659 550 3120

SUMMER hours 98.75 72.75 72.00 88.25 146.0 164.75 137.5 780.00

IS-min intervals 316 296 205 214 225 392 312 1960

WINTER hours 79.00 74.00 51.25 53.50 56.25 98.0 78.00 490.00

IS-min intervals 711 587 493 567 809 1051 862 5080

TOTAL hours 177.75 146.75 123.25 141.75 202.25 262.75 215.50 1270.00

V-4

THE 8DM CORPORATION

SUMMER

WINTER

TABLE V-2. SAMPLED INTERVALS CATEGORIZED BY HOUSE, SEASON AND WEEKDAY/WEEKEND

NEWARK ATLANTA ALBU-l ALBU-2

IS-min intervals 571 240 586 514

WEEKDAY hours 142.75 60.00 146.50 128.50

15-min intervals 191 358 276 384

WEEKEND hours 47.75 89.50 69.00 96.00

15-min intervals 145 452 394 265

WEEKDAY hours 36.25 113.00 98.50 66.25

IS-min intervals 191 170 192 151

WEEKEND hours 47.75 42.50 48.00 37.75

V-5

TOTAL

1911

477.75

1209

302.25

1256

314.00

704

176.00

THE 8DM CORPORATION

B. STATISTICAL LOAD MODEL

1. Modeling Methodology

A model ing methodology involving the prediction of peaks (load

spikes). peak durations. and frequencies per I5-minute interval based on

the time-averaged load was initially proposed for this project. Previous

analysis of I-minute interval data (the shortest interval data available

at the time) conducted for the proposal displayed a strong positive cor~

relation of .97 between peaks and functions of the present and previous

time-averaged loads. However. the I-minute data were simulated. a fact

not known at the time, and proved not to be real istic when compared to

real data collected during the measurement phase of this project. In

fact. preliminary findings of the direct. measurement program showed that

the instantaneous load profiles. or patterns. were driven by major

appliances only. and that smaller appliances simply added to the base

load. The peak loads. durations, and frequencies were thus based on

particular appliance load profiles. Predicting which appliances were on

based on the time-averaged load would have been difficult to do. In

addition. instantaneous app 1 i ance load patterns other than those

developed during the pilot phase were not available.

The findings redirected the development of the statistical load

mode 1 i ng methodo logy. A new approach was formu 1 ated and i nvo 1 ved three

basic stages plus the final prediction:

(1) Instantaneous load profil e characteri zat i on for each I5-mi nute

interval

(2) Determination of the most prevalent profile groups based on

analysis of the characterization measures

(3) Representation of the prevalent profile groups

(4) Prediction of which profile group representation to use based

on the time average.

The modeling methodology is shown in figure V-2.

Periodic regression analysis is a curve-fitting technique,

similar to polynomial regression, appl icable for periodic data and was

V-6

<: I '-I

FINAL DATA 8ASE

5080 IS-MINUTE INTERVALS

,10

INTERVALS CLUSTERED AND DISCRIMINATED INTO 13 MAJOR DISTINCT PROFILES

MODEll NG STEPS

ANALYSIS DATA BASE

COMPUTER ROUTINES ... IS-SEC DATA FOR DATA

p

EACH IS-MIN INTERVAL REDUCTION TO BE MODELED

FREQUENCY ... PROBA8 I Ll TV ANALYSIS • DISTRIBUTIONS FOR

SELECTION OF PROFILES GIVEN TIME OF DAY, DAY OF WEEK, . SEASON

Figure V-2. Model ing Methodology

PERIOOIC REGRESSION ... ANALYSIS

CLUSTER AND DISCRIMINANT ANALYSIS

... •

COEFFICIENT DATA BASE

HARMONIC COEFFICIENTS FOR EACH INTERVAL

I

ALGORITHM INCORPORATING THE PROF I LES DEVELOPED AND IMPLEMENTED INTO SOLCEL

BDMI A-82-299-TR-R 1

-I ::r: m III o s:: ()

-0 :0 "'0 o :0

~ o z

THE BOM CORPORATION

used to define or characterize the instantaneous profile for each I5-minute interval in the data base. This analysis uses functions of sine and cosine curves. the coefficients of which characterize the instantaneous profile. Cluster analysis and classification analysis were used to group similar profiles. based on the coefficient values. Once a preci se separation of di st i nct profil e groupi ngs was determi ned. coefficient averages for each grouping were calculated. Based on several criteria. the real I5-minute intervals of 5-second load data having. coefficients closest to the group averages were determined. These real profiles were used to represent the distinct profile groups in order to maintain the instantaneous nature of the data. Frequency analysis was performed on the profile groups to investigate possible relationships of the groups to particular residences (regions). season. and day of the week; and to determine probabilities of occurrence to be used for prediction of the presence of the profiles. Finally. an algorithm was deve loped i ncorporati ng the profil e representations and pred i ct ion scheme. Thi sal gorithm was implemented into the SOLCEL photovo lta i cs analysis code to provide the capability for analyzing residential photovoltaic applications using predicted instantaneous load requirements.

2. Load Model Development a. I5-Minute Load Profile Characterization

An example of the regression analysis results and the coefficient estimates for a selected I5-minute interval is shown in figure V-3. Three pairs of sine and cosine functions of the timel ine variable. X.· were included in the periodic regression example to form three harmonics. The high squared multiple correlation coefficient of 0.936 underlined in the figure ind-icates the periodic regression model fits the data quite well, with only three harmonics. Coefficients for the sine and cosine pairs define the presence and magnitude of the corresponding harmonic in the equation. The V-intercept coefficient value is the I5-minute interval average load. Due to the orthogonality of the sine and cosine functions, individual coefficient estimates remain

V-8

THE BOM CORPORATION

the same, no matter how many other harmonics are in the equation. The

orthogona 1 i ty also resu lts in identi ca 1 standard errors of the coeffi­

cients in the equation. Coefficients for each sine and cosine pair, say

a and b respectively, can be combined to provide the amplitude, A, and

phase angle, e, of the sine curve harmonic by the following relations:

tan e = alb

-HUI:-'HPLE R O-.-9-6-f-1~o---MULTIPLE R-SQUARE 0.9362

.-STO.-€RROR-GF-€-5T,----G ... 1-1-54·---,--------------------

'NA-l:YSIS O-F-V·A~H-NGG.gE---------------------SUM OF SaUA~ES OF MEAN SQUARE F RATIO

--~R:EGRES-SION-2-ih 1It933 6-------- 6--- ~h69·1-556 -U9 ..... 92J-RESIOUAL 1.5083462 49 0.3078262£-01

VARIABLES IN EQUATION --------------5'ISTO~R-OR ST-{)-RSir-------VARI ABL E COEFFICIENT OF CJEFF COEFF F TO REMOVE

(V-INTERCEPT 1.232 ) ·GO·s-x- 5 0-..-2-56 - 0 .. ~-J.,3s-~--Q .. 2-1-95-9 .. 626-SINX 6 0.791 D.al3 0.860 569.068

-GG5-2X 1 -O--'-l-4lJ------Q-...lJ-33 -O-.-t-'iZ 1-7-.-3-ZQ-SIN2X 8 O.ZO~ 0.033 0.222 37.826

-GOS,lX 9----0.....02-3- -lJ ... 0.,3 3 O· • ..Q.J0----D-.·6~5-SIN3X 10 0.195 0.033 0.212 34.5U6

BDM/A-82-299-TR-Rl

Figure V-3. Periodic Regression Example with Three Harmonics

Regression analysis was performed on the 4-second interval data

collected during the initial stages of the direct measurement program.

V-9

THE BOM CORPORATION

One of the conclusions reached was that data could be aggregated to

16-second intervals for modeling. Doing so reduced the roundoff effects

in the data, adequately maintained the instantaneous profile information,

and advantageously reduced the amount of data to be analyzed by periodic

regressi on for each I5-mi nute interval. Each I5-mi nute i nterva 1

contained 56 intervals 16 seconds long, with the exclusion of the last

data point. This corresponds to the degrees of freedom shown in the figure.

A plot of the observed data (0) for the I5-minute interval

in the example and the predicted values (P) from the periodic regression

equation in the previous figure is presented in figure V-4. The plot

displays the adequacy of the periodic regression characterization of the instantaneous profile.

Prior to performing periodic regression on each of the

intervals in the load data base, the 5-second interval data were aggre­

gated into I5-second i nterva 1 load va lues, for the reasons previous ly

stated, producing 60 measurements per I5-minute interval. Since no prior

i nformat i on exi sted on how many harmoni cs were necessary to adequate 1y

model the highly variable load profiles, it was decided to use eight

harmonics in each periodic regression analysis. The use of eight

harmoni cs was cons i dered adequate for most i nterva 1 s and was also an

upper bound in consideration of the large volume of information which would require file storage.

Periodic regression analysis was performed on each of the

5,080 15-minute intervals in the load data base using International

Mathematical and Statistical Library (IMSL) subroutines. Coefficients

from each of the periodic regression equations and descriptive informa­

tion relating the house, season, day, date, time, minimum, maximum,

energy above average, squared multiple correlation, standard error of

estimate, and standard error of coeffi ci ent were concurrent 1y stored on

file for further analysis. With 72.5 percent of the equations having

squared multiple correlations (R2) greater than 70.0 percent, the

periodi c regress ion stage was cons idered to have comp 1 ete 1y accompli shed the goal of adequate profile characterization.

V-10

THE BOM CORPORATION

2.275 + N-~ + . /e'-¥~ · · • · · .

· , \, ·

· ~ · 1.925 + -':1 +

p · ~ · ~ E · ~\R r · 9 1~ '. ,: · h~ · T · \ · -tJ-

~ 1.575 + I + --I. · N · I I ·

· · --0-11-4;0 B · · -S E · · --R V 1.225 + +

-E-- . D · I · . •

OWER) · I · hDS;)

"i · · (P

I · I · . • 575D + + .

· · . · ~ ~ ~=9R:-ro-aqp-; · .~G' + +

· .~ if ~~ · --. I

· ~ 0 ~ ~ U 0' I) · --;J-~---il1.z'----<2 ... ~-----3&---..... ~a--- &D-

NOTE: This plot was generated on a standard printer. A width of 50 columns was specified and 56 data points were plotted, resulting in 6 overplots.

p-0---

BDM/A-82-299-TR-R 1

Figure V-4. Plot ,of Predicted {P} and Observed (0) Load Versus Time

V-ll

THE BDM CORPORATION

rntervals having active profiles in contrast to steady,

inactive load profiles which closely follow the I5-minute average were

then selected. Since inactive, or steady, loads are adequately modeled

by the 15-minute average load, no further modeling of these profiles was

necessary. Most profil es in the load data base were cons i dered to be

inactive and occurred during the night, mid-morning, and mid-afternoon.

Three criteria were used to determine whether profiles

were sufficiently active to require further modeling: time of day.­

energy above average, and the squared multiple correlation. The energy

above average measures the energy (area) below the instantaneous profile

and above the I5-mi nute average, and thus re I ates the act i vity of the profile. Active profiles would necessarily have high energy above average va lues. Squared multiple correlation measures the amount of

vari abil ity accounted for by the regress ion equat i on in contrast to a

hori zonta I line. Low corre I at ions i ndi cate that the time average va lues

(Y-intercept or constant term in the equation) is adequate for modeling and that the profile is inactive.

Plots of energy above average versus quartile (15-minute

time interval of the day) and squared multiple correlation for the 5,080

intervals are presented in figures V-5 and V-6 respectively. Also

indicated on the plots are the cut-off levels for the three criteria used to define active profiles:

(1) Quartiles greater than 22 (5:30 - 24:00) (figure V-5)

(2) Energy above average greater than .06 kVA (figure V-5, V-6)

(3) Squared multiple correlation greater than or equal to 70 percent (figure V-6)

The breakdown of the 5,080 intervals according to the three criteria is

shown in table V-3 (counts from the-plots do not exactly match the table

due to interval roundoff on the plots). rn chOOSing the cut-off levels

for the three criteria, some consideration was given to maintaining a

manageable data set for the remaining analYSis. The cut-off level of

0.06 kVA for energy above average was used since the intervals appeared

to split up at that point. Of the 60 intervals having energy above

V-12

<: I

w

A~ SUMMER B ~WINTER

J B

•• + •••• + ••••••••• x •••• + •••• + ••••••••• • ••• + ................ .

n • ----------·------·8

A

/I A

-i :I: m

.,~2 + _ .. ______ ._~-----~.; ____ _ . ___ 4_. ----------_._-----

co o s: ()

o :D ""0 o :D l> -i R A

QUARTILES GREATER THAN 22 (05:30 - 24:00 HRS) _

Figure V-So Plot of Energy Above Average (ABVAVE) Versus Quartile (QRTL)

o Z

ACTIVE PROFILES

ENERGY ABOVE AVERAGE GREATER THAN

Q" INACTIVE PROFILES (CORRESPONDING TO 15-MIN DATA)

BDM/A-82-299-TR-R 1

< I

.0::.

•• " ... + •••• + •••• + ........... +.,," .+.

·32 ...

ACTIVE PROFILES

---i I m OJ o ~ ()

o :D "tJ o :D

~ o z

•. l--------1L..A.--. • A. .1.-__ n RR- -.IIi R +

A 8 8 ____ '--_--'H<L-_--"~. .J!. R _~

II AA

Figure V-6; Plot of Energy Above Average (ABVAVE) Versus R-Squared (RSQR)

ENERGY ABOVE AVERAGE

GREATER THAN

.06 KVA

~ INACTIVE PROFILES (CORRESPONDING TO IS·MIN DATA)

BDM/A-82·299·TR·Rl

THE 8DM CORPORATION

average greater than .06 kVA and between midnight and 5:30, all but one of them were traced to the Newark residence in summer, probably indicating high usage of electric air conditioning (complete demographics on the Newark and Atlanta residences were not obtained and direct commu­nication with the homeowner was not allowed). Since these intervals were few in number and imprecisely defined, they were not included as active profiles. This decision led to choosing 5:30 as a cut-off for time since a lmost a 11 prof il es between mi dni ght and 5: 30 had energy above average­values less than 0.5 kVA and were observed to be low, steady loads. The 58 i nterva 1 s havi ng energy above values greater than 0.6 kVA and corre­lation less than 70 percent were found to be extremely varying loads, primarily from one of the Albuquerque residences and possibly due to the stove burner elements, which could not be adequately characterized by the eight harmonic equation.

The three criteria were used simultaneously to classify 734 profil es (out of the tot a 1 5,080 i nterva 1 s) as act i ve prof i 1 es (see table V-3). The coefficients and descriptive information for these intervals were stored on a separate file for the next analysis stage.

b. Determination of Distinct Profile Groups The goal of this analysis stage was to classify the active

instantaneous profiles into distinct groups or categories to determine the most prevalent profiles. This goal is based on the premise that the active instantaneous load profiles are due mostl'y to a few major appliances and that, when a particular appliance is in use, the resulting load profiles would all be similar. Even though the base loads may be different, resulting in different 15-minute averaged load values, the load profiles due to a particular appliance should be similar. In fact, each distinct profile group developed could probably be associated with either an individual or a combination of major appliance loads.

Load profiles were grouped according to various criteria and based on the coeffi ci ents of the characteri st i c per i od i c regress i on equat ions. Cl uster anal ys i sand cont i ngency tab 1 e 1 ayouts were used to examine the groupings resulting from the various criteria. In cluster

V-15

THE 8DM CORPORATION

TABLE V-3. BREAKDOWN OF INTERVALS BY CRITERIA FOR ACTIVE PROFILES

ENERGY ABOVE QUARTILE

AVERAGE

LE 22

LE .06

GT 22

LE 22

GT .06

GT 22

NOTE: GE - GREATER THAN OR EQUAL TO GT - GREATER THAN LE - LESS THAN OR EQUAL TO LT - LESS THAN

CORRELATION

LT 70 GE 70

313 801

1045 2069

0 60

58 734

analysis, the coefficient variables are recoded as belonging to specified intervals, and cases (profiles) having common codes for the coefficients are grouped together. The specification of the intervals for each coefficient and the number of coefficients included greatly affected the groupings.

Output from the cTuster analysis routine (Biomedical Computer Programs, P-Series, Program P3M) is quite lengthy and is not inc 1 uded here. The cluster ana 1ys i s effort requ i red a 1 arge number of trial and error analyses. Initially, the coefficient values for the sine and cos i ne functions were used to group the i nterva 1 s. Several prob 1 ems were encountered involving the common scale of the variables, symmetry of

V-16

THE 8DM CORPORATION

the variables about zero, and the large number of variables (up to 17) and cases (734). I twas deci ded to transform the coeffi ci ents into the harmoni camp 1 itude and phase angl e values and to reduce the number of harmoni cs used to cl uster the profil es. Resu lts from these ana lyses displayed better grouping of the profiles. Intervals for the amplitudes were reduced to four: low, medium, high, and very high. Phase angles were not included individually, but combined to form phase angle differences between pairs of harmonics when both were in at least the medium category. The phase angle for a harmonic defines where along the time axis the sine curve lies. This location was not considered impor­tant since it is dependent upon when exactly the data collection occurred and when appliances were turned on. However, when two or more harmonics were present in the profile characterization, at the medium or higher categories, the phase angle difference between the pairs of harmonics did aid in grouping the profi les. Phase angle differences were placed in three categories: in phase, between phase, and out of phase.

Several analyses using this structure led to a good grouping of the profiles. One conclusion reached was that only the first two harmonics were needed to group the profiles. The fi na 1 group i ng structure is summarized in a contingency table layout shown in table V-4. Displayed in the table are the number of 15-minute interval profi les in each group and the group classification number. Thirteen different profile groups were identified, with graups 10 and 12 being a combination of two somewhat s imil ar groups and group 13 bei ng somewhat of a "grab bag" group. All the profiles not falling into a distinct group were placed in group 13. With the distinct profile groups defined, remaining modeling steps were to form representations of the groups and to develop a prediction or selection strategy.

c. Profile Group Representations To construct the statistical model predicting the instan­

taneous load prof; 1 e, preci se defi nit i on of the predi cted profil es (and therefore the profile groups) was necessary. Mathematical formulation of the profiles using the sine curve (harmonic) ampl itude and phase angle

V-17

<: I LE .05 .....

(X)

2nd HARMONIC LE .10

GT .10

TABLE V-4. CLASSIFICATION TABLE DISPLAYING DISTINCT PROFILES

\ 0 W

M E D

H I G H

1ST HARMONIC AMPLITUDE

LE .05 LE .10 LOW MED

NUMBER NUMBER OF PROFILE OF PROFILE

PROF ILES GROUP PROFILES GROUP

52 1 136

IN 66 55 2 BTWN 49

OUT 13

IN 9 45 3 BTWN 8

OUT 1

IN = PHASE ANGLES IN PHASE BTNW = PHASE ANGLES BETWEEN PHASE OUT = PHASE ANGLES OUT OF PHASE

4

5 6

13

13 13 l3

LE .15 GT .15 HIGH EX. HIGH

NUMBER NUMBER OF PROFILE OF

PROFILES GROUP PROFILES

80 7 38

40 8 9 43 9 17 2 13 2

14 10 14 7 l3 27 1 l3 6

LE = LESS THAN OR EQUAL TO GT = GREATER THAN

PROFILE GROUP

11

13 12 13

10 12 l3

-I J: m to 0 s:: (')

0 ::0 "tJ 0 ::0

~ 0 Z

TOTAL IS 734.

THE 8DM CORPORATION

differences was considered, but found to smooth out the profiles. In view of how the SOLCEL PV analysis code operates and with the desire to maintain the instantaneous nature of the profiles, it was decided to use the interval of real instantaneous data which was closest to the average

. or centroid of the profile group. That is, real data profiles were used to represent the distinct profile groups.

Mathematical averages of the first four harmonic ampl i­tudes and the difference between the first two harmonic phase angles were. calculated for each profile group. From the interval profiles in a par­ticular group, the interval with amplitude and the phase angle difference closest to the averages for that group was determi ned. Closest was defined by several criteria: minimum sum of squared deviations, minimum sum of absolute deviations, and two different weighting schemes used to weight the deviations from the amplitudes more than the deviation from the phase angle difference. The representative intervals chosen matched on at least three of the above four criteria for all the groups.

Having determined the representative intervals for each of the 13 groups, the real load data for the intervals were extracted from the load data base and placed on a separate file. Plots of the 13 repre-

sentative load profiles are included in appendix C. • d. Profile Prediction

The final step of the load modeling task was to develop a ru 1 e for determi ni ng the presence of an active prof il e, and se 1 ect i ng which representative profile to use. Possible relationships to be included in the prediction of load profile were time of day, day of week, season, region (NE, SE, SW), and the 15-minute average load. These possible relationships were investigated through analysis of contingency tables and plots. If particular profiles occurred only at specific times of day, on weekdays or weekends, during a particular season, only in a

particular region, or only for selective time-averaged loads, a relationship probably existed. However, since the loads are appliance­driven, an argument could be made that the differences occurred only because of the use of different kinds of appliances, each with different

V-19

THE 8DM .CORPORATION

•

efficiencies, load requirements, and usage practices. With a sample of

only four households of appliances, determining precise relationships was

not feasible. In addition, no clear relationships in terms of time, day,

season, etc., as described above, were determined.

An example of the contingency table layouts used to

explore the relationships is given in table V-5. This contingency table

depicts the breakdown of all the 5,080 sampled intervals according to

profile group (inactive profiles defined as profile group O)~_

weekday/weekend, and season.

A basic approach involving frequencies of occurrence was

decided upon for the profile prediction. The frequency of occurrence of

active profiles out of the total possible intervals when an active pro­

file could occur was used to develop the probability of an active

profile. Since very few active profiles occurred between midnight and

5:30, these intervals were not included as possible active profiles.

Referri ng to the breakdown of the samp 1 ed i nterva 1 s presented in table

V-3, the first and third rows were excluded as being possible active

profiles. The resultant probability of an active profile was calculated to be

(58 + 734)/(1045 + 2069 + 734 + 58) = 0.203 Finally, given the presence of an active profile, the selection of 1 of

the 13 representative profiles was based on the frequencies of occurrence

of the profile groups out of the 734 total active profiles. Individual

profile group probabilities were calculated by dividing the number of

intervals in each group, given in table V-5, by the totai of 734.

The probabilities and decision criteria were formed into

an algorithm to be inserted into the SOlCEl PV ana 1ys i s code. Even

though profiles are predicted by this relatively simple probabi listic

prediction scheme, simulation results over an annual time period are

cons idered to be both real i sti c and informative. However, match ing the

predicted instantaneous load profiles real data for smaller time

durations, one day for example, would probably show large discrepancies.

V-20

THE BDM CORPORATION

TABLE V-5. PROFILES BY SEASON AND WEEKDAY/WEEKEND

P"CFlt 2 OAt ( S, S 2

p f) I ..:.t:Ji~~:: ~ wlrqrUI: TOTALS -------------------------------------::- • J W'TtC'IA, I 105('; J.C4~ 4346

.. [[KENO I 1 <)27 614 I

I.D WE!:: KD AY I 15 19 52 W:El<::Mfl I ; .'

I :2.8 iii '- t: ;( EJ "" I ~ ~~ 55

!';EEKt=:ND I 5 15 I

3.0 liGKDAY I 20 11 45 W •• L I(::~fl I J 5 r

if.C W::!([l Ai I .2, 5) 136 loOE~KENO I 31 23

I !}.D lo:EEKDAY I 35 10 66

Ii:: !(ENfl I 17 If I

E.C l.iE_KBA'l' I 13 U' 49 WEEKEND I 10 4

I 7.0 WEEK!)iI,Y I 35 16 80 WEE IE r:Hl 1 2'" 5

I B.e IiIC:KOAY I e .. 'f ~y 40

WEEKEND I 11 2 I

9.0 \iEE KO AY I 20 8 43 !C ICME! I 3 ti .-

r 16. WEt_l(!JA l' I 1~ n

28 "'-WEEKEND I 10 3 -"-

II. WEEKDAY I 16 4 38 wCE1(2/11e I 15 3

I 12. W[EK1AV I 7 1 q

44 ~~

wrEKENO I 22 3 I

13 .. IJEEKDH I 21 19 58 W::E'E:~e I 15 8

------------------------------------- 5080

v- 21

! I

! !

THE 8DM CORPORATION

3. Load Model Implementation into SOLCEL

SOLCEL is a SNL-developed computer simulation code designed specifically for the analysis of PV systems. The code currently is capab 1 e of accepting annual load data in hourly or I5-mi nute i nterva 1 s and perform the simulation accordingly. Using the annual load data and solar/weather data, the program simulates the annual performance of specified PV and battery storage systems; computes the load requirements met by the PV system, battery storage, and the uti 1 ity; and performs ao. economic analysis.

An algorithm incorporating the statistical load model predic­tion scheme and the representative profiles was implemented into the SOLCEL code. Real load data defining the profiles were included as deviations from the I5-minute time averaged load. SOLCEL still steps through the simulation at 15-minute intervals, but if an active profile is predicted, based on a random draw and comparison to the active profile probability, all computations are performed at I5-second interval steps.

Annual 15-minute interval load data are required to utilize the statistical load model. In stepping through the simulation, the I5-minute load value is used if (1) the time is between midnight and 5:30 A.M., (2) the time-averaged load is less than 500 W, or (3) the probabil­ity criterion for an active profile is not met. Otherwise, the active profiles are used. Based on another random draw, one of the 13 profiles is chosen. For active profiles, the algorithm uses the chosen profi le deviations and the current time-averaged load to calculate the 15-second load values for that 15-minute interval. SOLCEL then steps through at 15-second intervals and calculates the PV provided load, battery storage provided load, and utility provided load. With the added capability of the instantaneous load statistical-model, SOLCEL is capable of providing complete and realistic annual PV system performance measures. Appendix B describes how to access and execute the version of SOLCEL containing the statistical load model.

V-22

THE 8DM CORPORATION

CHAPTER VI

PERFORMANCE COMPARISONS

Photovo1taic (PV) array residential performance is usually measured

by the percentage of the PV output that meets the load demand compared to

that sold to the utili ty. The load that is not met by the PV array is

bought from the util ity. The instantaneous output of the array and the

instantaneous load are, therefore, crucial in the exact calculation of PV.

performance.

Simulation programs have been used to determine PV performance as a

function of Typical Meteorological Year (TMY) solar/weather data. Errors

in performance calculations have been known to exist when time-averaged

load data are used, where time-averaged load refers to load demand

averaged over a given interval (Le., 15 minutes). These errors in

performance calculations are in the estimates of PV output sold to and

the energy bought from the utility. The amount sold and bought are

always underestimated when the average data are used. Stud i es over

short-time intervals have shown serious errors when comparisons are made

between time-averaged and instantaneous load data. Percentage errors

over short i nterva 1 sin the amount sold back to the utili ty have been

shown to be as high as 50 percent. But, the percentage error is not as

important as the magnitude of the error. For example, if the true amount

sold was 20,000 units and a 5 percent error (1,000 units) was made, a

1 arger error resu 1 ts than if the true amount so 1 d was 6, 000 un i ts and a

15 percent error (900 units) was made. Also, the combination of the

above two data points result in only a 7 percent error. Since PV arrays

operate over long time periods, the important issue is the magnitude of

the error over long i.e., annual, periods of time.

To further define the problem, consider figure VI-I. Three possible

situations exist for PV output and instantaneous load data:

(1) Figure VI-la shows the case for which the maximum load demand

is below the PV array output. For this case, 100 percent of

the load is met by the PV array, and no error (amount bought

VI-1

THE 80M CORPORATION

__ PV AVERAGE OUTPUT

LOAD

POWER

al

LOAD

POWER

- - - - - - - - PV AVERAGE OUTPUT

bl

LOAD

POWER PV AVERAGE OUTPUT

cl

BDM/A-82-299-TR-Rl

Figure VI-I. Three Possible Configurations of Load Versus PV Output

VI-2

THE 8DM CORPORATION

from or sold to the util ity) would be generated during time­

averaged load data.

(2) Figure VI-Ib shows the case for which the minimum load demand

is greater than the PV array output. Again. no error would be

generated using time-averaged load data.

(3) Figure VI-Ic shows the third case for which the load demand

crosses the PV output. For this case. power is being sold to

and bought from the ut i 1 i ty duri ng the time i nterva 1. Us i ng.

time-averaged load data would result in errors in the

calculation of PV performance.

Assuming the normal variation expected from a residential load. these

errors are maximized over a given time interval by having the time­

averaged load demand equal to the time-averaged PV output. Raising or

loweri ng the time-averaged load from the time-averaged PV output wi 11

decrease the error (i.e •• the error in the amount sold will decrease as

the time-averaged PV output is shifted above or below the time-averaged

load). This concept can be proven mathematically by considering areas

under the curve.

USing the above concept. the error generated by using a time­

averaged load to predict PV array performance can be restated as follows.

The error generated by using a time-averaged load is a function of the

numerical difference between the time-averaged load and time-averaged PV

output. Large errors will be generated when the numerical difference is

small and small errors will be generated when this difference is large.

For an annual PV performance. the issue becomes how often the t ime­

averaged PV output becomes re 1 at i ve ly equal to the time-averaged load

demand.

Large errors occur only when the time-averaged load is approximately

equal to the time-averaged PV output. The annual frequency of occurrence

of approximately equal time-averaged load and PV output will depend on

how closely matched the annual PV output is compared to the annual load

demand. When these are closely matched. one would expect this approxi­

mate equality to occur more frequently. This frequency of occurrence is.

VI-3

THE 80M CORPORATION

therefore, the key issue in the relative size of error in predicting PV performance, and this frequency of occurrence is largely a function of PV array size relative to load demand.

To determine the size of a PV array that would cause a high frequency of approximately equal time-averaged loads and PV outputs, computer simulations were run. These simulations used actual 15-minute time-averaged load data measured from a res i dence, vari ous array si zes, the statistical load model, and Albuquerque TMY solar/weather data.

8DM has previously obtained i-year duration sets of 15-minute interval residential load data monitored by the Public Service Company of New Mex i co for three Albuquerque res i dences. These res i dences were of the same class as those monitored in the Detailed Residential Electric Determination Load program. Data for one of the residences were selected and used as the input 15-minute load data, along with Albuquerque TMY solar/weather data, for SOLCEL/statistical-load-model simulations.

SOLCEL simulations were performed to produce and examine the sensi­tivity of annual and monthly performance measures to variations in the probability of the presence of an instantaneous load and to examine the effects of various array sizes. The sensitivity is a function of the frequency of occurrence of approximate ly equal time-averaged load and PV output. The purpose of this sensitivity analysis was to estimate the effects of the variables on PV direct fraction. If changes in the values of the probabi 1 ity of an active profi le and/or the PV array size .cause significant changes in the PV direct fraction, the PV direct fraction is sensitive to the variables. If changes in the variables have little effect on the PV direct fraction, the PV direct fraction is relatively insensitive to the variables. Also, the PV direct fraction may be sensitive only in subranges of the entire range examined for the variables. Thus, this analysis quantified the effects on PV direct fract i on of gi ven changes in array size and probab il i ty of an act i ve profile. Thirty (30) simulation runs were completed varying the size of the PV array from 1 kW to 9 kW and the probability of an active load from 0.0 to 1.0. The probability of an active load of 0.0 implies 15-minut~

VI-4

. -

THE 8DM CORPORATION

averaged load data were used. The probabi 1 ity of 1.0 imp 1 i es an active profile was used. If the time was between 5:30 and 24:00 and the average 15-mi nute power was above .06 kVA. then the 15-mi nute average load was used. Both monthly and yearly totals of PV array performance were accum­ulated for each run.

The results of the simulations showed PV direct fraction was relatively insensitive to the probability of an active profile except for impractical small array sizes. This result is contrary to previous bel iefs that estimation of PV direct fraction would be highly sensitive to load activity.

Initially. four SOLCEL simulations were performed to produce annual performance measures for comparison of

(1) One-hour interval load data and time step (2) Fifteen-minute interval load data and time step (3) Fifteen-minute interval loa,d data, using statistical load model

with probability of active profile set to default (0.192) (4) Fifteen-minute interval load data, using statistical load model

with probability of active profile set to 0.4

The last simulation. with the probability of an active profile approxi­mately double the value determined from collected data, was performed to investigate the sensitivity of the annual performance measures to increasing the probability.

The results for the four initial simulations are displayed in table VI-l in the form of annual percentages of load met by the 4 kW PV system. As expected, with smaller time i nterva 1 s and the use of instantaneous profiles from the load model. the percent load satisfied by the PV system declines.

TABLE VI-i. RESULTS FROM SOLCEL SIMULATIONS

MODEL 15 MIN

PV DIRECT FRACTION .333 .339

VI-5

MODEL WITH HOURLY TWICE PROBABILITY

.346 .327

THE BOM CORPORATION

To further determine the sensitivity of PV performance measures,

additional SOLCEL simulations were run. Figure VI-2 illustrates a plot

of the results of the simulation runs, varying array size and the proba­

bility of the occurrence of an active profile, and also shows annual

results expressed in PV direct fraction, where PV direct fraction equals

PV AC to LOAD Total PV AC

PV direct fract ion is shown as a funct i on of the probab il ity of an

active load where each line represents a different size PV array and the

value of M is the slope of the 1 ine (i.e., I = 1000 is a 1000 Watt PV

array and M = .134 implies a 13.4 percent reduction in the PV direct

fraction when the probability of an active load is increased from 0.0 to

1.0). When the probabi 1 ity of an active load is 0.0, the simulation is

run using I5-minute time-averaged load data. The sol id vertical 1 ine

represents the observed probabil ity of an active load in the data from

the four monitored residences, and the dashed line shows where the

probability is greater than twice the observed value. Load activity

beyond twice the observed value can be considered unrealistically active.

This graph demonstrates the annual sensitivity of PV performance to array

size and probability of an active load. The probability of an active

load causes an· increase in the frequency of the occurrance of approx i­

mate 1y equal PV direct fract ion (i ncrease infrequency of approximately

equal PV output and load demand). This increase in the frequency of

occurrence is small for appropriately-sized PV arrays and larger PV

arrays (4 kW or larger). Also from figure VI-2, as PV array size

decreases, the effect of the proaabi1ity of an active load increases.

Table VI-2 depicts the reduction of PV direct fraction for given PV array

size and probability of an active load. For the observed probability of

an active load and a 4 kW array, only a 0.6 percent reduct ion is seen

when compared to using I5-minute time-averaged load data. Even the 1 kW

array shows an insignificant reduction of 2.7 percent.

VI-6

THE 80M CORPORATION

PV DIRECT FRACTION

AC TO LOAD TOTAL AC

.7

.6

.5

.4

. 3

.2

.1

o

.......

-

-

-

-

o

-ANNUAL AVERAGE

--- ---------- ------ - --- ----- - - -- --- - -- - - . -- ----- - - ----- - - - - - -

- OBSERVED

. , .2 .4 .6 .8 1.0

PROBABILITY OF ACTIVE PROFILE

1= 100bw m=-_134

1= 2000W m = -_078

I =3000W m = -_048

I =4000W m = -.031

1= 6000W m = -.017

1= 9000W m = -_009

BOM/A-82-299-TR-Rl

Figure VI-2. Annual PV PerfQ.r:mance Versus Probability of an Active Profile for Different Array Sites

VI-7

THE 80M CORPORATION

TABLE VI-2. REDUCTION IN PV DIRECT FRACTION COMPARED TO

PROBABILITY OF AN ACTIVE LOAD

ARRAY SIZE OBSERVED TWICE OBSERVED EXTREME CASE (kW) (.2 prob) (.4 prob) (1.0 prob)

9 0.2 0.4 .9 6 0.3 0.7 1.7 4 0.6 1.2 3.1 3 1.0 1.9 4.8 2 1.6 3.1 7.8 1 2.7 5.4 l3.4

The significance of having approximately equal average load demand

and average PV output is displayed· in figures VI-3 and VI-4. Figures

VI-3 and VI-4 show monthly PV direct fraction as a function of

probability of an active load for a 1 kW and a 4 kW array, respectively.

Each line represents the value of the PV direct fraction for the months

of January and April, and M is defined as in figure VI-2 (M = .l34

implies a 13.4 percent reduction in the PV direct fraction where the

probability of an active load is increased from 0.0 to 1.0). For the

1 kW array, a PV direct fraction near 1.0 for the 15-minute averaged-load

data (probability of an active load is 0.0) implies the I5-minute

averaged-load demands are approximately equal to the PV output. Thus, as

the activity of the load is increased, errors in the PV direct fraction

become 1 arger. For the 1 arger 4 kW array, the array output is 1 arger

than the load demand, and as the activity of the load is increased,

relatively small errors occur.

When PV direct fraction is plotted versus the array size

(figure VI-5), the insenSitivity of the PV direct fraction to the proba­

bility of an active profile becomes apparent.

Three lines are displayed that represent the probability of an

active profile of 0.0, 0.4, and 1.0. Very little change is seen in the

VI-8

THE BDM CORPORATION

1.0

. 9

.8

.7

.6

PV DIRECT .5 FRACTION

.4

.3

.2

.1

0

0

MONTHLY (1 kW)

EXTREME CASES CAN BE OBSERVED FOR SHORT INTERVALS AND SMALL ARRAYS' .

-- -- ------ ------- -JAN. m=-.1aa APRIL m = -.249

~OBSERVED

.2 .4 .6 .8 1.0

PROBABILITY OF ACTIVE PROFILE

BDM/A-82·299-TR·Rl

Figure VI-3. Monthly PV Performance Versus Probability of " an Active Profile for 1 kW Array

VI-9

THE BOM CORPORATION

.6 -

.5

.4

PV DIRECT .3 FRACTION

.2

.1

o

-

.

-

o

MONTHLY (4 kW)

LARGE VARIATIONS DO NOT OCCUR FOR BIGGE R ARRAYS

--- .... ---- ------- - - JAN. m = -.062

--- ----- - APRIL m = -.035

I- OBSERVED

, , I

.2 .4 .6 .8 1.0 PROBABILITY OF ACTIVE PROFILE

BDMIA-82-299-TR·Rl

Figure VI-4. Monthly PV Performance Versus Probability of an Active Profile for 4 KW Array

VI-10

THE BDM CORPORATION

PV direct fract i on between the probabi 1 ity of an act i ve prof il e of 0.0

and 0.4 which represent I5-minute time-averaged data and the extreme case

of realistically active loads. Only for the 1 kW PV array is the PV direct fraction significantly affected.

The s imu 1 ati on resu 1 ts support the concl us ion that for typi ca 11y­

si zed arrays and observed probabi 1 i ty of an active load, instantaneous

load data are not significant in the calculation of annual PV direct

fraction. Figure VI-5 shows graphically that for only extremely small.

arrays, where the load demand would approximately equal the PV output, do

instantaneous loads become significant in the calculation of PV performance.

VI-l1

THE BOM CORPORATION

PV DIRECT FRACTION

.7

. 6

.5

.4

.2

ANNUAL AVERAGE

INSTANTANEOUS DATA SIGNIFICANT FOR ONLY VERY SMALL ARRAY SIZES .

TYPICAL MEASURED ACTIVITY PROBABILITY 0.2

o+-----~----~----~----~--~ o 2 4 6 8 10

ARRAY OUTPUT (kW)

BDM/A-82-299-TR-R1

Figure VI-5. PV Performance Versus Array Size for Various Probabilities of an Acti ve ProfiJ~

VI-12

- .

THE BOM CORPORATION

APPENDIX A LOAD DATA BASE ACCESS AND FORMAT

This appendix describes the procedures necessary to access the instantaneous load data base. The deve 1 opment and extent of th i s data base are discussed in chapter IV.

A. ACCESS TO LOAD DATA BASE FILES

Collected instantaneous (5-second interval) load data have been partitioned into eight files by residence and season and stored on duplicate public TFILE tapes on the Sandia NOS computer. Their Volume Seri a 1 Numbers (VSNs) are 13138 and 18409 and can be accessed by the following NOS command:

TFILE,R,13I38,lfnI,lfn2, ••• where lfnN = file name to be retrieved from the tape. The files will appear in the user's direct access permanent file space (required), with the same file names, within 30 minutes after the command is entered. Table A-I lists the pertinent files contained on these tapes and their contents.

B. LOAD DATA BASE FILE FORMAT

For each fil e, the 5-second i nterva 1 load data are blocked into I5-minute quartiles with 15 lines of 12 measurements, 1 minute per line. Figure A-I illustrates the format of a load data base fi leo A header record precedes each quartile of data and contains the quartile time, number from 1 to 96 (midnight to 12:15 AM is number I), 15-minute average in kVA and kVAh, and the energy above average in kVAh (the area below the instantaneous values and above the average). Contiguous blocks of quartil es per day are aggregated by header records whi ch precede the header and data of the first quartile in the block. This header record depicts the corresponding residence, day of week, date, time duration of

A-l

THE 8DM CORPORATION

TABLE A-I. LOAD DATA BASE FILE LIST

FILE NAME DESCRIPTION

NWRKDAT Newark, summer, 5-second interval load data ATLDAT Atlanta, summer, 5-second interval load data ALBIDAT Albuquerque No. 1, summer, 5-second interval load data ALB2DAT Albuquerque No. 2, summer, 5-second interval load data NWKDATW Newark, winter, 5-second interval load data ATLDATW Atlanta, winter 5-second interval load data ALBIDTW Albuquerque No. 1, winter, 5-second interval load data ALB2DTW Albuquerque No. 2, winter, 5-second interval load data

/

A-2

:z,. I

W

~EWAP< THU 7/16/81 1~145-24100, 29 ~UARTIlES, 6~ - Q~ 16:45-17:(,(; :lPTl ~~; ·~V'=: 3.Cl3 KVA O~-.76-KVii~r;f·ij(~'fY-AB1VE AVE:~-03KVAH-

_...:;.3..-:..21 3.21 3.533.21 3.213.213.53 3.n 3.21 3.21 3.53 3.2:;..1 __ _ 3 .21 3 • 5 3 3 • ~ 1 3. 21 3. 21 '3 • 53 3 • 21 3 • 2 1 3. 5 ":\ 3. 21 3. 21 3 • 21 3.53 3.21 1.21 3.51 3.21 3.21 3.53 3.21 3.21 3.n 3.53 3.21 2.89 3.21 2.~9 2.81 3.21 2.H9 3.21 2.e93.21 2.89 3.21 2.~9 3.21 2.B~ 3.21 2.~9 3.21 2.89 3.21 2.8'1 3.21 3.7.1 2.1.19 3.21. __ _ 2.8~~r-2.89 3.21 2.E~9 3.21 2.89 3.21 ?;1i-9-3.?l--2;aq----·3:Z1 2.89 3.21 .?.81 3.21 2.89 3.21 3.21 2.B9 3.21 2.89 3.21 3.21 __ _

--2.69 3.21 2.H 3.21 2.f.9 3.21 2:89---3-.21 2.f'9 3.?1 2.89 3.21 2.89 3.21 2.89 1.21 2.B9 3.21 2.89 3.21 ?f'Q ~.21 2.89 -.:2:...: • ...;.f'.-.;9 __ _ 3.21 2.C9 3.21 2.89 3.21 2.89 3.21 2.f9 2.PQ-3-.-n ·-2;69 2.e9 3.21 2.89 3.21 2.89 3.21 2.89 3.21 2.89 3.21 2.'39 3.21 2.89 2.89 3. 21-2."[1"1"" 3.21 2. &9 3.21 2.99 2;-6"92. e9-Z--;09-t~-89-·2·;-1i-9-----2.25 2.89 2.57 2.89 2.69 2.57 2.89 2.57 2.P.Q 2.89 2.57 2.e~9~ __ _ 2.252.812.577..892.892.57 2.B9 2.;7 ?89 2.89 2.57-2.89 2.89 2.57 I 2.139 2.57 2.89 2.89 2.57 2.89 :>.89 2.25 2.A9 2.f9:--,.~.,..,--

1710L-17:15 JRTl 69; ~VEI 2.65 KVA OR .71 KVA~; ENE~GY ~qQVE AV~: .02 KVAH 2.69 3.21 3.21 3.21 3.21 3~21 3.21 3.21 ~.21 3.n 2.89 2 •. ..;;8,.,.9 __ 2.57 2.89 ~.a9 2.57 2.89 2.57 2.89 2.89 ~eq Z.~Q 2.69 2.1.19 2.57 2.89 2.139 2.57 2.89 2.89 2.89 2.57 ?A9 2.1'1') 2.89 2.57 __ _ 2.B9 2.89 2.i9 2.89 2.!11 2.69 2.89 2.fj~.57 2.-Bq-~&92.89 2.89 2.6~ 2.139 2.~9 2.57 2.89 2.89 2.57 2.69 2.89 2.57 2.89 2.6~ 2.89 2.57 2.89 2.6~ 2.89 2.51 2;89 2.89 2.89 2.89 2.69 2.57 2.89 2.69 2.89 2.57 2.89 2.89 2~69 2.57 2.S9 2.89 2.57 2.89 2.57 2.99 2.89 2.57 2.89 2.57 2.89 ?Aq 2.57 2.89 2.89 2.57 2.69 2.89 2.89 2.89 2.69 2.89 2.57 2.89 2.8Q 2.57 2.8~9~ __ 2.89 2.57 2o':l~ Z·.57 2.89 2.89 2.57 2.1'9 ~.57 2-;-B~ 2.89 2;57 2.81 2.57 2.89 2.89 2.57 2.89 2.69 2.57 2.89 2.89 2.57 2.89 2.89 2.89 2.89 2.89' 2.57 2.89 2.89 2.57 2.89 2.1.19 2.57:.---,2;;..:.'-;6"'9-----2.8~ 2.57 2.89 2.57 2.89 2.89 2.57 2.89 2.8Q 2.57 2.89 3 •. ~21~ ____ _

-r.t9 3.Zr--z.~9 ~.89 3.21 2.89 3.cr-r.e9 3.-Zr-r.1\'1 Z.B9~-21

2.89 3.21 2.39 2.9? 3.21 2.69 3.21 2.69 2.e9 3.21 2.89 2.89 -,1,.....7r7:-T-I5~7: 3(1 ORT[ 701 ~V:I 3.25 KVA O~ .81 I<VAH; ENEqGYftl?'lve Ave: • 02 KVAH

Figure A-l. Load Data Base Example

-i I m CD o ~ n o ::c ." o ::c ~ o z

THE 80M CORPORATION

the block of contiguous quartiles, the number of quartiles in the block, and the block dUration in terms of quartile numbers. According to the example in figure A-I, 29 quartile header records and the corresponding I5-minute durations of load data measurements would be read until the next block indicator header record would be encountered. Table A-2 describes the format for the different lines of the files comprising the load data base.

A-4

· -

. .

THE 80M CORPORATION

TABLE A-2. LOAD DATA BASE FILE FORMAT DESCRIPTION

COLUMN LINES* NUMBER DESCRIPTION FORMAT

1 1-6 Residence: A6 NEWARK ATLNTA ALBU-l ALBU-2

1 8-10 Day of Week: A3 (or A2 if necessary - the MON, TUE, WED, first two characters uniquely THU, FRI, SAT, SUN define the day)

1 12-20 Month, Day, Year 3(12,1X)

1 21-32 Beginning and End Time 4(12, IX) (hrs., min.) of Contiguous Block of Quartiles

1 34-35 Number of Quartiles in 12 the Block

1 48-54 Beginning and End Quartile Numbers for the Block

12, 3X, 12

2 1-12 Beginning and End Time (hrs., min.) of Quartile

4(I2,IX)

2 18-19 Qu art il e Number 12

2 27-31 15 Minute Time Average (kVA)

FS.2

2 41-44 15 Minute Time Average F4.2 (kVAh)

2 70-73 Energy Above Average (kVah) F4.2

3-17 1-72 180 S-Second Interval Load Measurements Comprising the IS-Minute Interval

14( 12F6.2!), 12F6.2

(I-Minute Duration Per Line)

Repeat lines 2-17 for the number of quartiles in the contiguous block (as read from line 1, columns 34-35), ~hen read new header record with line 1 format and do the same.

LINE 1 (BLOCK HEADER) FORMAT: (A6,lX,A3,lX,3(I2,lX),4(I2,lX),IX,12,12X, I2,3X,I2) .

LINE 2 (QUARTILE HEADER) FORMAT: (4(I2,IX),SX,I2,8X,FS.2,9X,F4.2,25X,F4.2)

LINES 3-17 (LOAD DATA) FORMAT: (14(12F6.2/),12F6.2)

*Description given for first 17 lines in figures A-I.

A-5/6

. .

THE BDM CORPORATION

APPENDIX B SOLCEL3 EXECUTION PROCEDURES

This appendix briefly describes the procedures necessary to access and execute the version of SOLCEL which includes the statistical load model (SOLCEL3). A more complete discussion of SOLCEL is given in references 1 and 2 below.

Files required for execution of SOLCEL3 are (1) SOLCEL3 source code

(2) A year's duration of Typical Meteorological Year (TMY) weather data

{3} A year's duration of I5-minute load data {4} Input data

The following NOS commands are used to access the SOLCEL3 source code: compile the source code to create the relocatable binary (SOLBIN); attach the fil es containing Albuquerque TMY weather data (ABQTMY), 15-mi nute load data for an Albuquerque residence (PNM02Q), and input data; set up a math function library (FXMATH); execute the program; and print the results file:

ATTACH,SOLCEL3 DEFINE,SOLBIN/CT=PU FTN,I=SOLCEL3,L=FTNLST,R=3,OPT=0,B=SOLBIN ATTACH,TAPEl=ABQTMY ATTACH,TAPE2=PNM02Q GET,TAPE5=SOLINP LIBRARY, FXMATH SOLBIN PRINT,TAPE6

The ABQTMY, PNM02Q, and SOLINP fil es were used to produce the resu lts presented in Table V-I of chapter V. These files, along with the source code file (SOLCEL3), are stored on duplicate TFILES (VSNs 13138 and 18409) and can be accessed by the command presented in appendix A. Except for the input data file (SOLINP), all these files require direct access permanent file space. On a previous project, 15-minute load data for durations of a year or more were collected from utilities and stored

8-1

THE 80M CORPORATION

on the SNL computer system. These load data were collected on residences similar to those used for development of the statistical load model and are appropri ate for use wi th the mode 1. For further i nformat i on and access details on these files, see reference 2.

SOLCEL3 can be executed using IS-minute time-steps with or without the statistical load model. In addition, the code can be executed using I-hour time-steps, requiring a year's duration of hourly load data, however, the statistical model cannot be exercised using I-hour time",,_ steps.

Input data specifying one of these options is included in the SOLCEL3 input data file (SOLINP), which is read under the NAMELIST convention. A listing of this file is presented in Figure B-1. The TIMSTP and PROB variables, in the second to last line, control the time­step and statistical load model. The PROS variable specifies the overall probability of an active profile, for which instantaneous load profiles would be used (see chapter IV). For execution of SOLCEL3 at hourly or IS-minute time-steps, without the statistical load model, TIMSTP would be set at 1.0 or 0.25 (in hour units) and PROB should be set to 0.0. To execute the statistical load model, PROB can be set to a desired value (between 0.0 and 1.0) or to the value realized through the analysis (0.192), which is the default. The default value will be used if PROB is omitted or set to the value. Utilizing the statistical load model requires TIMSTP to be set to 0.25.

C DETRILED RESIDEHTIRL LDRD ," 4~ W ARRA-Y r 60Kldl+ BRTTE~'" CAPACI TV ,~ 151"1!H TII"IE AYERA6E LDRD DFHA-1BA-T~'" NDP~I-O.NDP~I=O.NBC=24U.NPBC=4.HSBC=&O. BTI"IP~-IOO •• BTi'IP!'!X"1-00.S­

$END , 1CDSTHS IECDN=IS 1DCAC PYTT=4000 •• IH¥~TRc3$ 1DPTl'tZ , 1DRIENT ID~INT=2S­$PHlJTD IPye~:l 1P~TN& P~!DT=.F.S 1SDLDRT ISKIP=I.TII"ISTP=.25.PROBc.I92S :sTE1'!P CL=I.21~CW=40. 41$

Figure B-1. SOLINP File Listing

B-2

. .

THE BDM CORPORATION

REFERENCES

1. SOLCEL2, A USER'S MANUAL, Sandia Report SAND79-1785, 1979.

2. SOLAR ENERGY/BATTERY STORAGE FINAL REPORT, BDM/A-81-058-TR, June 12, 1981.

B-3/4

. .

·'-

THE 8DM CORPORATION

APPENDIX C PROFILE PLOTS

Plots of real load data for the I5-minute intervals which were

selected to represent the 13 distinct profile groups are presented in

Figures C-I through C-13. The horizontal axis is the I5-minute interval

timeline, broken out into the 60 15-second intervals for which the asso­

ciated electric load values, in kVA, are plotted. These plots display

the actual data used in the statistical load model. See chapter IV for a

complete discussion of these profile groups and the statistical load model.

C-l

THE BOM CORPORATION

LOAD (kVA)

••••• + •••• + ••••••••• + •••• + •••• + •••• ~ •••• 4 •••• +.~ •• ~ ••• •••••••••

• • •

2.75 + +

~ •

· • ~ · teL .•

~ .. ; ~

• • · • · · ..

• • •

• · ,~_ ..... l'

• .

• • •

1 .15 .. +

· y

• •

· · • •

.1. .2~ + .. • • • • • V V v VVVYV • •

• •••• + ' •••• + •••• + •••• + •••• + •• )( • + .......... + •••• + .......... + •••• + •• .. 1'1 -:s

10 31) 40

TIME (15 SECOND INTERVALS)

Figure C-1. Plot of Load Versus Time for Profile Group 1

C-2

.

. .

, .

THE 80M CORPORATION

LOAD (kVA

••••• +x ••• + ••••••••••••••••••• + •••••••••••••• + •••• + •••• + •••• + ••

• 2.RO + +

• · ... A · ./ V V •

• .., At.

+ Y kAA •

Y V\; · · • , - • 2.1 0 • ..

· · • •

· • .~ ~ •

· • •

· •

) • 1 .4 () .. ..

· • '" ~. • . (\/ \

T V-\ · • , no: • . ~

.

" · •

· \ • • ,

.?~ C .. +

\ -· · • ...

· y \ •. • ~j ~ 1"1 '" · V •

• •

••••• + •••• + •••• + •••• + •••• + •••• + ........... + ••••••••• + •••• X •••• + •• -------''"'''-r. -----ll""s!-. ----~2""':----¥'---·---4~-"''"~----''''!!-~ c~ .• __

10 lie

TIME (15 SECOND INTERVALS)

Figure C-2. Plot of Load Versus Time for Profile Group 2

C-3

THE 80M CORPORATION

LOAD (kVA )

.~ . -,-

-~ -c'

. -:

2~~

.V

., .~;

-' . ' ..

• '; 0

••••• + •••• + •••• • •••• • •••• • •••• + •••• x •••• • •••• + •••• + •••• + •••• + •• . ,.. • · A · y

f\. T

· • · · · I\j ~ • · ~V· 1 · .. +

· • · • · · · • · · • · • · · · · .. .. · • · · · • · ·

T

• • · • · • · · ~ ..

• · •

• y •

, . · • · 'i • · • . • .. ..

\. yy- • · AJ\/v ~~ ·

, . . -- · ••••• + ••••••••• + •••• + •••• + •• x.+ •••••.••. + •••• + •••• + ••••••••• 4 ••

'; . 1: 3 'j 111 jr)

TIME (15 SECOND INTERVALS)

Figure C-3. Plot of Load Versus Time for Profile Group 3

C-4

. .

THE 80M CORPORATION

LOAD (kVA

• •••• + ..... + •••• '" ..... • +. 'II •• + •• 'II. + ........... + ..... + ••• 'II ••••• .f ....... .

~,,~lJ-~+-------------------------------------------------------------------------.+---

-y 0

· · • •

;?pr • +

· - • •

· • ., __ A ~ •

· · • · · · .~ •. , !j y +

• · •

· · • )

1.7 c I

· 0 _.

V\ • · •

"\ 11 • 1 .fJ. ;: + +

· · • · · 0

· •

· • •

• "fryC + 1. +

• ~;++"\ \;+\ r-\ ~ · - ·

· y y y y y · • a 1:':-- +

••••• + ••••••••• + •••• + •••• +" ..................... )(+ •••• +.y ......... + •• " 1 'j 25 :3 " 4-9 -¥.---------------

TIME (15 SECOND INTERVALS)

Figure C-4. Plot of Load Versus Time for Profile Group 4

C-5

THE 8DM CORPORATION

LOAD (kVA )

. " . ,

~ .1 :;

..:... . , .. ~

2 .. q~

., ~ .

1.75

.. ..

1 .05

••••• + •••• + ................ 4 •••• + .......... + ..... +.X •• + ............ y .... .

· " · · [ · · · I

.. · • · .~ · · r· k • · l I L • ..

J .. · l·~···· V • • Ji- ... • · V V · · •

~

· • · • • • · • .. .. · • · · · • · • .. · • · • · • · · .. .. · • y • · • · •

. .

· • •

· · · .. A A. .. . .. - · .x ... + ••••••••• + ........... + •••• + ................ + •••• + •••• + ...... ..

" ·0 1'5 z ... c·

10 21 3D

TIME (15 SECOND INTERVALS)

Figure C-5. Plot of Load Versus Time for Profile Group 5

C-6

THE 8DM CORPORATION

LOAD (kVA )

"

3.')

,

,-' .. "

~ .R

1.,

, "

• ':;0

••••• + •••• ~~~ •• ~ •••• ~ •••• 4 ••••••••••••••••••• + •••••••• •••• x •••• • · X •

· • • • · • • - • .. ..

• · • • · · · •

• · •

• .. +

• · · • •

· • I

· Y .

• · · + t\ ~ + ,

\ • • -" '--'--" .-- fo- r- 0 .-

·A11...L.f.. ~I ~-' J\ • ...

\ \ l \-· • •

• • . -,------..

l + ~ • "- •• 1

· Y'I V\.I "4 · • ..... .. • + ............ + ' •••• 4)( .... + ..... + ............. + ................... + ........ ..

I'j •

1 {] 2 !J 30 40 ')0

TIME (15 SECOND INTERVALS)

Figure C-6. Plot of Load Versus Time for Profile Group 6

C-7

THE 8DM CORPORATION

LOAD (kVA )

3.!'O

I.tc'

2.1tC

-.

;: .1 C

1.4 C

• J "_ .

...... + •••• +- .............. + ...... + ..... + ............. + ....... + ..... + ... )( ............. ..

y

• · A • · \ • + +

· V • · \ · · . \ • · A AAA A A • , ~\.{ YVY\..I\J.}+;

?

· · · • · \ • · ~ • + .. · • · • · • · · · · · • · • · · + .. · • · • · · • •

· •

· · • •

+ +

• · \ AM.AAI'Vv\Nv • · < - ·

+ • Y .. ... + ....... )( ..... -+ ........ -+,. .............................. + ........ + ......................... + .. ..

• "r :rt: =;=;

1 ~ 40 S()

TIME (15 SECOND INTERVALS)

Figure C-7. Plot of Load Versus Time for Profile Group 7

C-8

THE 80M CORPORATION

LOAD (kVA )

••••• + •••••••••••••• + ••••••••• + •••• +~ •••••••• + •••• + ••• • + •••• + •• .... 0 )( •

• · I\J\. • . • ·VY· •

,,~ • I:) +

· · · • · • , • y •

• · •

• r~ • t.~ .. ...

· · 0

· · .. c -· • I

· • •

4.5 + ...

· · • .-· \. •

" " ........ ...... · ·Y'y "\ A It f" · · \+I. W \c.-,-~ ~\ . • ... "-" •

3 .~, ... YY'4 + .1.

· y-o

• (] ,.

••••• + •••• + ••••••••••••••••••• + ••••••••• 4 •••• + •• X.+ ••••••••• y •• t:" ! ,. 2'; 31:"",-- If '1 Fro: .-----

10 30

TIME (15 SECOND INTERVALS)

Figure C-8. Plot of Load Versus Time for Profile Group 8

C-9

THE 80M CORPORATION

LOAD (kVA )

, .

5 .. ~::

~

-. r.: :: .. ~

, " -

1 .. 5

1 -. -

................................ + ••••••••• + .................. + ............. . (

+

· · · · · · 1\ · \.J •

· • · • y • · • .. +

· • -\ · \ f' I' •

· • \. J \.t.v f T

· ... . . . . · • · • · V · • ..

• · •

• · · · •

· · · · .. A +

• J\. . I \ · · ~ · • LIY yy ~yy •

· y +

................ "., ................. • •••• ..f .......... + ....... x ........... .. E - .

11l 30 Ii' 50

TIME (15 SECOND INTERVALS)

Figure C-9. Plot of Load Versus Time for Profile Group 9

C-10

THE BDM CORPORATION

LOAD (kVA )

( (\ (", ., ....

~ ,-

if .~,c

~ ~~

--

~.oo

_ 'J .., "

1 .5 n

--, .. -

• ................ + .......... + ••• , .... -••• of ..... + •••• + ..... + • • " ............ X

· • • · ~ +

• · \. ·

\. /" • · - 44 • --· •

· • •

+ +

· · • · · •

· y

· • • • · + +

• • •

· · · · •

• · · + •

-- ~--

· • -- •

· · -. I • • • • • I • • • • . , . I • • • • · YVYYYVVY\..IY¥'ifVYY\.:IY\..!Y'\.. •

• •••• + ......... y ........... •• x ....... + .................. + •••• + ............. .

10

TIME (15 SECOND INTERVALS)

Figure C-10. Plot of Load Versus Time for Profile Group 10

C-ll

THE 8DM CORPORATION

LOAD (kVA

, ) ,

-..

r, .1~

.~

j.C

-

2.4

:>- .. '

1 .2

••••• + ••• ~ ••••• t ••••••••• + •••• • ••• 4 ••••• + •••• + •••• + •• ~~ •••• y •••

•

(\ ?

· • · j \ • · ~ · · .~ r ry • • k +

· \/ '\I'r • · • · • • · •

· • · • · • · · • + • · · • · · · •

+

· • · • · • · . · • +

• · ~ • · I · · ·

T

· • · -~

• · · +.AA/

~ . • +

•

• ••• y ... •• x •••• ................ + •••• + ••••••••• + •••• + ..... + ............ . . - 2S 3- PC

10 '30

TIME (15 SECOND INTERVALS)

Figure C-ll. Plot of Load Versus Time for Profile Group 11

C-12

, .

THE 80M CORPORATION

LOAD (kVA )

j. 2~'

q , E f',

"( ... 75

-.!; 9

2 ,2~,

1 '" ,

• -!Se

••••• + •••• • •••• • ••••••••• + •• 4.+ •••• +~ ••• + •••• + •••• + ••• X ••••• + ••

•

~ - ,

· · • · • . • + + , • · y . · · ~ • , .. · t+J · , · · •

• • +

0

· •

· • • I

· • 0

o,N •

+ +

· · •

· •

· • • • . .. • •

· '!YVYV\ •

+ \ .. - . V . . -

-•

----------.----------------------------------------------------------------------------.------•• Y •• + •••• + •••• +X ........ l1li ...... + .......... + •••• + ............... + ••

1" 25

'" TIME (15 SECOND INTERVALS)

Figure C-12. Plot of Load Versus Time for Profile Group 12

C-13

THE 80M CORPORATION

LOAD (kVA )

H'

l .2

-.

7 n , .

~ .

1 • ,0

-

•••••••••• + ••••• w ••• + ••••••••• + ••••••••• + •••• + •••••• ••• + •••• ~ ••

• +

· · · ~ A

V · -· y · · A A ·

N J ~ N /I 'Y\ - -

· · · \.t1 r-l ~ - y ~

, ~ .. · .. · 1 V

. • + +

· • · • .

· • · •

· · · • · • · · + +

· • · .. · • · · 1

· .. · • • •

• + + .. • " 1 J • .

\A V • · ..

· · • .. •

••••• + •••• + •••• + ••••••••• + •••• + •••• + •••• + •••• + •••• + •••• • ••• x+ •• ------------~G,.~------~l~S~------~?~.~~.--~----~~·-~----,----+~~s------~----------­

'l()

TIME (15 SECOND INTERVALS)

Figure C-13. Plot of Load Versus Time for Profile Group 13

C-14

. '.

THE BDM CORPORATION

. '

APPENDIX D RESIDENTIAL LOAD MONITORING SYSTEM

SCHEMATIC WIRING DIAGRAMS

D-l!2

, '

THE BDM CORPORATION

~ ~

/lS&- METER.

Jr-- CONDU \,. '6E E , "OTE I @

HOUSE INTERFACE UNIT (ZOO_.AMP'=>~~c;. PIoo,C.ONNE.C.T 'SWITC.H) ( .... OUNTE.D OI'LOVT5IDE. Vi/':,LL .. UTII",!TY ENT~Nc;.J;:

RE"'"PENTIAL LOAD~ING, S'(STEi'<I

SCHE>-IATIC WIRING DIAGoR.Ai'<I

i ~ ~-------------------------.------~~=L==tL==~-~·~~~~&

Figure D-la, Residential Load Monitoring System Schematic Wiring D-3/4 Diagram--House Interface Unit

THE BOM CORPORATION +

•

/()T£S:

'. All OOTSIO£ WIIIIHCi TO BE IN LIQUIO·Tt~T CONDUIT AND All WIRING TO B£ IN ACCORDANCE WITH H.E.C. AHCl sun: COO£$,

,. All WOAK SHAlL BE UECUT£O IN A GOOO WOU'Wt.!K£ HAHNER.

J. OEBURR ALL I«)lES AT LEAST 1/32- "IH. NlO lISE lIQlJIO TAPE so WIRES WILL HOT SHORT OUT. .. AlL WIRING HAS TO BE STMHO£O MD T't'PE TW OR EQUIYAlENT. .. USE HEAT SHRINK nJalJ«j ON ALL SPLICES. - ,. ALL WIRING WHERE POSSIBLE wiLL BE IN SoI'IRAl wRAP TUBING, EXCEPT 200 AI1P SERYICE WIRING {nilS WrLL tolD WilliNG BUNDlE TOGETHER AND HA.Y[ LESS CHANCE Dr SHOIITlfrIi OUT).

7. COYER IS USED WHEN J[H AHD I'ODEH UIIU IS HOT USED (COYER HAS NEOPRENE GASI([T). WHEN THE J[M AHD 11)00. UNIT ARE USED NEl WILL BE CC»IIECTEO InTH LIQUID-TIGHT CONOOIT IIITH HUTS AT BOTtI ENDS (4· LONG). TJlE CONOUIT "'ILL HAvE A PlUG OF NEDf'R[N[ R\lIIU IN IT so HEAT WILL HOT LEAK OUT. THE PlUG WILL HAY[ A HOTCH IN IT SO WIRtNG CAN RUN BY AND HAK[ A SEAl. USE 2" CDHOUIT. .. USE G[ RUBSER GlUE B[TWEEN J-80J MD S1o'IT(Jf [NClOSURE. AND 8[1\IE[H OUTlET 80X AND SWITCH EHCLOSUR[ TO ,"I:E A SEAl. .. DO NOT GROIJIrjD VffH 100000IHG SCRE\/ IF NEUTRAL IS ALREADY GROUND IN

-i> HOusE.

10. THE AC OISCOIf([CT swrrCH ENCLOSURE TO BE !'(UITEO TO HOUSt SOOt THAT HINt,"L DAMAGE wILL OCCUR TO IIJUS[ WHEN SWITCH IS REI'I)VED.

II. THE lIQUlD- TIGHT CONDUlI WILL HAVE A 1/2· H'l,E IN IT. HAlF WAY. so TELEPHONE CAIl[ CAli' BE COHNECTED TO HOOEM .... IT.

12. rOR VIR( COlOR (JrIOT SERVlC[ WIAING):

!!!:!.A2.L~..nOR RF CURR[NT 'R08[S

HOT ,1 - BlACK PUT AATCHIHG CABLE HOT .2 - SLUE O£SIGftATOR l..A8EL5 ON NEUTRAL ~ \MITE CO·A,l AND ASSOCIATEO

I'OW'ER LINE CURRENt MOfCITOIi

HOI 11 • BLACK RETURN - IIHU( HOT ,2 • BLUE R[TURH - WHITE

..

~ ~

I I I I ......"oe.o.flOOO

f

, IZ6-1lJ .wH(frIOl COHfC(CfOII;, 12' nun 50 1'1111 SOCkEr .. WIlli KlOO AHO a,.NoP

" .-ll 4((' ,..[TAl NJT iliUM LOCK WASHEII " .. 6-ll 1£' ""AI. MIT wITH LOCK WASH£II " " 4_40 W' f1[TAl NUT wlnt lOCK W~SH[R " , 81»! !10K !II[Of'R[fiI( IUBER I/Z-O II I 11'" D.O. .. , 8DH 51( NEOPRENE 1W88[R 1/4-0 I SOL I S-H " , 91191-ll [AJO" COItPQAA1'lON. Rf CURREHT PROlE ]I

8HI TO 81'ttl

• SN201 SQISAA[ 0 COf"IPANT. CWIRENT TRANSfORMR " , 31 .. ~. CONNECTOR. 8He TO " AMPlER .. fEIW.£ 8NC TO " "-UG

• 31·011 NtfH[1tOl 8NC. CAlLE [NO PLUG ,. , 31-22001 VG - 492 NtPH[PIJl BNC. AWIER BlIrW.. MIT. JACk TO JAU " , 126-218 IWHEHOl. COfIHECTOfI:. 126 SERIES 5 PIN JJ

SOCKET WITH LOCKIHG niP , JOf174 NYLON WAlDOtt. I'(IlU Pn'lOflC CDHH[CTOJt IJ(SIGHU/SERV'ICE lZ eMO( UP fROM KIT) KIf

, 5CXl" 1,4101, 11 RESISTOft (Ill ,,"0 1t]1 " , SIC, SW. IS R£SISTOR (Rl AND 1t5) JO , lTS·IOS'A . VOlTR£l. T[RMI~ STIli' " • lAG 1116 NtfI FAST BLOW fUSE '" , l4200c limE FUSE PANEL K)UNT FOR lAG J:rPE OF FUSE 11

2 IfOJ-A BUSWNIf, 1/'1 LINE FUSE HOLDER FOR 3'" TYPE 16 Of FUSE -

J' 3/S" (IEFQtlE SMA INKfMC HeAT SHRINWLE TUlING 15

l' 1/4· CUFORE SHRINKING . HEAT SHRINKAlILE l1J8ING 14

J' 1/8- (BEFORE SHRINKING HEAT SHRINKABLE WIliNG 2J

". 511-40 AlPHA. SPiRAl llRAP TUBING II

" ,,. . ALI'f<A EXTENSION CORD. 3 WIRE IB AWG, SJRANO£D 11

5' RG-S8A/U COAX CABLE. STRANDE!) 10

IS' 12 AWG WIRE T't'P[ l\/. IU~;. WIRE, STRANDED BLUE " IS" 12 ..wG WIRE T't'PE TW. BlDG. WIRE, STRANDED BLACK " IS' 12 AWG lURE T't'P[ N, 8LDG. WIRE. STRANDED WttITt 11 11" IS' 22 MIG HOOK-UP WIRE - . 300V HIN. STRAHO(O, ttOQK-UP WIRE BLUE " , IS" II AJiG HOOK-UP wiRE I 300'1 HIN. STRAN[J(O, KOOK-lIP VIAE BLAtt:: " IS' 22 MIG HOOK-UP wIRE JOOV I'tIN. STR."ND£O HOOK-lIP wiRE \/HIlE " AI' 10QRI2AilGIYP[l\I TYP[ TIl BLDG. VIRE STAMOE!) " AI' WIRE 200 NiP SERVICE wIRE 12

, 5PA·l00 DALE AC ARRESTOR " , PANEL I'I[TAL PANEl FOR INSIDE or J-80l. 8ll't 51: 10

, COYER 4- X 4" I'1[TAL (OVER V ITH GASKET. 1Ill't 5K • AI' cOt\'oun MD FITTINGS 1'1[TAI. RA!NTlGHT. COHoon FOR UTIUT'f ENTAMCE WIRING • , COftI)lJIT AHO rIlTl'"iS "ETAI. ... LONG 2' 0.0. COHoon W!fH FITTINGS RAIHTIGHT , 1 1D22]8 ",(rAl. CIRCLE A.V. PROOUCIS. J-eox. RAIHTlGHT .·X."X4" , 1 AC SOCKET II!TII GrJ AI; OUTlET VIlli GFI IIREAKER 120 'lAC .1 ~ AMPS • IIREAt::(R !IWSI rn fT£H nU)

, OUTLET COYER "'TAl AAINTIGHT COVER FOR. 2 OIITt[fS !MI FIT ITEM Il) • , OUTlET lOX "'TAl JAIHT!GHT OIITtET lOX fOR: 2 OUTLETS J

, "'" 100 5US~ FUS[, OH[-Tl~ 2S0 yAt OR LESS 11200 AMP'S ,

1 Oll .. NflII SQlMIt[ 0 toMPANT. NEM_]R, WETY SWITCH..f.(HEIW. , (19 1!"16'H X Il 1/ .. " ooTY. l WIRE. 2 SlADE. 2 FUSE. 2"0 YAC f'200 N9'S x 7 1/"·0) (l] hI.)

000.118)'0 ,. .... flOO. .... ,[ ..... , OCliC'lOlf'noot -"" tIll Of .... lI".A' _.- .~ _f(""l

./'/ ..J.e.. Ru~~E.L-L .+.~,

pICtCIO'

./" ~ '(,!~JI~ b ..... V _ .. AU.LVG..&. ...... l_IOOUIl.- .... fti!OJ.IHOot. "fU

... 00 ....... "'''''AGI'' ~E~I DE.NTIAL. LQA.D MO,,"ITORING, ""Y~TEM

""-'L 'OCHE:M"'T'C \II'R'NG, DIAGoto:"'M

~! •• h_~I'o -sii. ID1-""02DOOI5~ r ....... .oAt ... _.

I ... u ....... ~ .. 0A0CIIf1C1...,.. ..

-----CAu: NONE 1-' Z. .. z.

- ,- i 1 ri Fl gure D 1 b. ReJlcent a Load Monlto ng System Schematic Wiring Diagram--House Interfate Unit (Concluded)

I 1 l ~

I' 1

D-5/6

I !

I ! !

•

THE BOM CORPORATION

) t:: "

;1 I ,

! I 'I ~I i rr-~~o~fZ :L<~Li~I~\lQl0]:;

- - + - + t., ":IQO ~C>I Zeo

,-<;-1-1'<:11'6 ZCJ y-~+-~,.

lO >0 .. .r. .. !>+

" ZZ

HOTES:

1. AlL OOTSUI[ VIRING TO IE IN UQOID--TlGKT COHOOn ""0 ......... WIRING TO BE IN ACCOROAHC( VITtI II.[.C. AKJ STAT[ COo[S.

2. All WOftK SHot.ll !£ U(CUTtO IN A GOOO WORkf'Wt.IK£ M.t.NHER.

1. DOllAR All HOlES AT LEAST 1/32- /'lIN. AND US{ lIQtJID TAPE TO 'R(YENT WIRES FROM SHORTING OUT.

4. All WIRI~ MUST IE 5TRAHOEtI MO TrP[ TV Ol EQOIVAlOlT.

s. .. ,. ..

U$[ 101 ~IJIK lUll,,,, 011 ALL Sl'llas.

All 'ilIAING WHUE I'OSS .. E Will It: II S#IIIN.. 11M' noel",. ueE" 200 NtIf S(AVIC[ WIRING (11115 wlU. tol.O WItIING BUNDLE TOOEn4ER AND HAW[ LESS C*1tC£ Of SHOIt" .. OUT).

nI( JIM-1OOOt [NClOAll( WILL K MXMlID TO 'I)Ur(1'tMG 'lAT[ TO tNSU«E NINIPW.. twWOE TO tee[ MN [NQ.0Sl*: IS R[M)YED.

WHEN THE JEM-fODDiI _T IS USED IT .... " IE Q)M(CTtO WI"" LIQUID­TtGKT COMOUtT WITH .rs AT II01M E~ (4- lOfC). nt( COKlUIT WIU MAYE ... !'lUI> IW ~ Al.NU; tN " 10 MEAT WIU IIIJf UAI: 0IIf. TME 'Ullii Wtll I[ wrtH[O so wtll .. CM U IV MO fIW.( A SEAl. ul( 2- COMIUIf.

(

9. COVU; IS USED Mil 8 AND MODO! UNIT IS fIIOT USEO (conK HAS NEOfItENE GA.'KET).

10. THE LIQUID-TIGHT COIClUIT WilL HAVE ... 1/2- I()l[ IN IT TKfIOUGH ONE WALL, so TElEPHOM: CML£ CAN IE F[O IN AId) CONNECTED TO jl(j0EM UNIT.

11. fOR WIRE COLOR (M3T SERVICE WIRIIiIG):

YOL TAGE QtlOR

~-:;.-'8UCI'.­

NOT 2 • RUE PlEUTMl ... ITE

CUIU\E1fT K>fIlTOII:

HOT 1 • ILACI R£1lJIIIjrIj • !oMIT( Kn' 2 • IlUE ROlJIVC • 1t1flTE

125 ","EA ,UtS , Nfl lO 01 1M[ lIt$211 (OMEClOfI lIAr,[ CONII(CTlDN !oM[It[ CONNECTOII: ts SOlIO[O TO (ldlT IOAAD (lKSI[J( or I'IXID'! UNIT).

TO ~WITCH ENC.L05URE

.

I I

(NOlOOCPea H~ .1-,) [""/"": _l' .. UM.YO." .......... IM.IOIJlM)IA._., •• 1'-' nyu:

__ ....... RE'::>IDENT"\L LO~D MONITORING,

~Y~TEM

t-_____ t--1I""'·L ,~C.HE.MA.TIC. WIRIN<;' DIAG,RA.M

! ~~,. o,.~ ::'--.. - ~. I U'}_.... r-' e J,-----,r-----,--+-------ir----f:r-il:",.i\'l~~ .. -~-;if~~ ....... ~· -i 5230B 0 02.Doo1576 J I J"'I('(--"-iL-________________________________________________________________________________________________________ -----------r---------------------------tl======~~~=~ .. ~:~~:~-~------------Ji_-_-_-_-_-_-_-_-_-_-_-_-_-lt_-_-_~ ____________ ~~~::~'_N~O~N~E~ _______ lL:~=: _____ ~=_ ____ ..J

t Flgure D-le. Residential Load Monitoring System Sehemptie Wiring Di agram--JEM/ Modem Ene 1 osure

D-7!8

i I· I

•

• •

• •

GENERAL REPORTS DISTRIBUTION SYSTEMS PROJECTS

DIS TR IBUT ION:

TID-4500-R66, UC-63a (551)

R. H. Annan (25) US Department of Energy Forrestal Bldg. 1000 Independence Ave. SW Washington, DC 20585 Attn: M. B. Prince A. Bulawka

M. Pulscak S. J. Taylor

V. N. Rice A. D. Krantz L. Herwig

US Department of Energy Attn: Robert D. Jordan, Director Division of Active Heating and Cooling Office of Solar Applications for Bldgs. Washington, DC 20585

US Department of Energy Attn: Michael D. Maybaum, Director Division of Passive and Hybrid Office of Solar Applications Washington, DC 20585

Jet Propulsion Laboratory (15) 4800 Oak Grove Drive Pasadena, CA 91103 Attn: R. V. Powell (4)

R. Ferber K. Volkmer W. Callaghan R. Ross R. S. Sugimura S. Krauthamer A. Lawson E. S. Da vi s

Jet Propulsion Laboratory Solar Data Center MS 502-414 4800 Oak Grove Drive Pasadena, CA 91103

SERI, Library (2) 1536 Cole Boulevard Bldg. #4 Golden, CO 80401

Dist 1

SERI (3) 1617 cole Blvd. Golden, CO 80401 Attn: Walter Short

R. DeBlasio

Florida Solar Energy Center 300 State Road 401 Cape Canaveral, FL 32920 Attn: Henry Healey

EPRI P. O. Box 10412 Palo Alto, CA 94303 Attn: Roger Taylor

House Science and Technology Room 374-B Rayburn Building Washington, DC 20515

Energy Laboratory 711 Virginia Rd. Concord, MA 01742 Attn: M. Russell

E. C. Kern

Office of Technology Assessment U. S. Congress Washington, DC 20510

New MexJco Solar Energy Inst. (3) New Mexico State University Attn: J. F. Schaefer P. O. Box 3S0L Las Cruces, NM 88003

Aerospace Corporation (2) Attn: S. Leonard

E. J. Simburger P. O. Box 92957 Los Angeles, CA 90009

Alternative Sources of Energy Attn: Larry Stoiaken Milaca, MN 56353

Arco Solar, Inc. Attn: J. Arnett 20542 Plummer Ave. Chatsworth, CA 91311

California Energy Commission Attn: E. Quiroz 1111 Howe Avenue Sacramento, CA 95825

Carbone Investment & Management Corp. Attn: Mr. Robert C. Carbone 570 Dwight Place Berkeley, CA 94704

General Electric Co. Attn: E. M.Mehalick Advanced Ener gy Programs P. O. Box 8661 Philadelphia, PA 19101

Georgia Institute of Technology Attn: S. I. Firstman College of Engineering Atlanta, GA 30332

Hirst Company Attn: Mr. Carrol Cagle P. O. Drawer 1926 Albuquerque, NM 87103

Mass Design Attn: Gordon Tully 138 Mt. Auburn st. Cambridge, MA 02138

Raymond J. Bahm 2513 Kimberley Ct. NW Albuquerque, NM 87120

Solarex Corporation Attn: Marth Bozman 13353 Piccard Drive Rockville, MD 20850

Dist 2

Sandia National Laboratories: 2525 R. P. Clark 6200 V. L. Dugan 6220 D. G. Schueler 6221 M. K. Fuentes 6221 T. D. Harrison 6221 D. F. Menicucci 6221 M. G. Thomas ~;2.5) 6223 D. Chu 6223 G. J. Jones 6223 T. S. Key 6223 H. N. Post 6223 J. W. Stevens 6224 D. E. Arvizu 6224 E. C. Boes 6224 M. W. Edenburn 6224 A. B. Maish 6224 R. B. Siegel 3141 C. M. Ostrander (5) 3151 W. L. Garner (3) 8424 M. A. Pound 3154-3 C. H. Dalin (25)

for DOE/TIC (Unlimited Release)

••