QUANTIFYING THE STUPIDITY OF THE SMART HOMEPER OLOF HEDEKVIST

Internet of Things in the Smart Home, March 21‐22, 2016

Stupid or Smart Home? You are not stupid if you read everything on the Internet… …But you are stupid if you think all of it is true.

How can we use that awareness in the Smart Home?

National Metrology Insitute of Sweden There is an error in every measured value… …but it can still be saved as a correct value.

“The voltage is 4,87 V…

… ± 0,1%...

… with 95% confidence”

Measurements in the Smart Home IoT in Smart Home is about measuring and controlling In and outdoor temperature to control heating Open data on outdoor temperature in the vicinity

Energy consumption at different times of day

Illumination through windows or from lamps

Positioning of objects or guests

…

Who cares about accuracy when the value is digital?

Stupidity of the Smart Home The Smart Home measures and controls relevant parameters in the home Comfort for residents

Optimizing / minimizing energy consumption

Minimizing waste

Poor data results in poor output Worse if control unit presumes data is good

The Smart Home if Stupid if measured values are presumedcorrect to last digit.

Smart Home needs Quality of Open Data IoT Open Data at thingful.net

Open data available in abundance Low value if accuracy unknown

Low value if time reference is unknown

Smart Homes can make customizedweather forecasts if data reliable

There is a difference between quality ofdata and high quality data.

Accuracy, Trueness and Precision Standardized definitions in ISO 5725‐1

Accuracy describes the closeness of a measurement to the true value Consists of Precision and Trueness

True value in reference to SI units, or other internationally defined units.

High precision when all values are closely together.

Poor precision benefits from averaging multiple values

Trueness describes a systematic error in measurement No improvement from averaging multiple samples Can be improved from averaging samples from multiple sensors Correction from calibration necessary Can decline over time

Accuracy in household energy‐meters Energy Provider measures total energy into the house, Etot Private measurements on each phase, E1‐E3

Etot ≠ E1 + E2 + E3 If Etot > E1 + E2 + E3 where does the energy go?

If Etot < E1 + E2 + E3 Where does the energy come from?

Truth: Etot is correct within 0,1% If accuracy of private meters are included, there are 

no inconsistencies

Master energymeter

Private energymeter 1

Private energymeter 2

Private energymeter 3

Etot

E1

E2

E3

True data, or derived. Examples True data: the parameter actually measured Voltage

Number of occurences

GNSS coordinates

Low level derived parameters (sensor detects parameter and outputs unambiguous voltage) Acceleration

Air pressure

Temperature

Irradiation

True data, or derived. Examples High level derived parameters (input from multiple sensors calculates parameter) Distance walked (using step counter)

Altitude change (without GPS)

Illumination

Accuracy vs resolution Illumination Unfiltered Si‐detector as lux‐meter, up to 150% error under LED light Can not be calibrated since LED spectrum is not specified

“IoT” device presents light level with 5 digits, no accuracy given Should read: 1300 lux ± 20% (2σ)

Illumination never better than 1‐2%

Note: Luxmeter is manually set at x10

Example, temperature sensor High quality Pt100‐sensors with precision better than 0,1°C trueness better than 0,4°C with linear approximation 0‐100°C Requires low probe power to avoid self heating 100mV * 1 mA = 100 µW 100 µV error corresponds to 0,4°C

Assumptions: Pt100 temperature sensor: accuracy better than 0,4°C Feed with constant current, I = 1mA Measure voltage over 100 Ω resistor Measure voltage over sensor Voltmeter 8 bit with 200 mV full scale, 1% accuracy Resistor 5%

Result at 30°C, uncertainty ±2 °C

Time logg Accurate data at unknown time is useless

Typical XO have timing stability 10‐5, differs up to 8 minutes one year after synchronization

Synchronization with reliable ntp‐server/s

Used ntp‐server/s should never be hard‐coded Popular servers becomes overloaded

Upgrades may change both DNS and IP

Even reliable servers fail sometimes

Quality measures in data Study of air quality in Umeå kommun

Linked Oped Data

Proposition of MetaData format associatedwith all Data.

DCAT‐AP (Data Catalogue vocabulary –Application Portal) Specification for describing public sector datasets in Europe

Name Type Description

Method Class The method used to make the measurement or observation

Equipment Class The equipment used to make the measurement or observation

measurementAccuracyType

Property Describing the type of accuracy in the measurement or observation

… … …

Conclusion All data transmitted from an IoT device should include, at least: Estimated data value

Time of sampling last value

Estimated accuracy As a value or a function

Time since last calibration, if necessary for function

Confidence level

Time of last synchronization

Estimated accuracy of time stamp.

All data received to an IoT device should at least: Take into account the accuracies of the received data and operate accordingly