Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae...

25
Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics and Mathematical Modelling Technical University of Denmark Viking Workshop October 2005

Transcript of Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae...

Page 1: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Short term forecast of travel times on the Danish highway network based on TRIM data

Klaus Kaae AndersenThomas Kaare Christensen

Bo Friis Nielsen

Informatics and Mathematical Modelling Technical University of Denmark

Viking Workshop October 2005

Page 2: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Background/Motivation

• The high way network in the Copenhagen area is equipped with double induction loop detectors. From this measurement system data on each passing single vehicle (e.g. speed and length) is collected. The data is referred to as the TRIM data

• Travel time estimates and travel time forecasts are calculated based on the TRIM data and given to the public by means of TV, radio and the Internet

• The travel time estimates and forecasts are based on data on an aggregated level

Page 3: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Study objectivesThe study objectives are

• Develop statistical models with aim to identify single vehicles between induction loops from the TRIM data. Hereby, the travel time between loops can be estimated directly

• Estimate the variability in travel times

• Validate the accuracy and precision of the current methods for travel time estimation and prognosis

• Develop new methods as an alternative to the currently used methods

Page 4: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Outline of the presentation• The Data

• Modelling method- Pattern recognition methods based on statistical

methods, dynamical modelling EWMA

• Results- Percentage recognized vehicles, accuracy and precision,

implementation

• Comparison with existing methods

• Future work

Page 5: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

TRIM data

For each passing vehicle the system measures:

- Time- Speed- Length- Gap…The travel times can not be estimated directly from these data!!

Page 6: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Data

• We are considering a stretch of the high way network (Jægersborgvej to Hillerødmotorvejen)

• Data is TRIM data

1) Empirical data (measurements from TRIM)2) Simulated data (from the program VISSIM)

Page 7: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Empirical data

- Time- Speed- Length- Gap…

Page 8: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Simulated data

• Data from the same stretch has been generated using the micro simulation program VISSIM

• The simulated data match the empirical data well

• The advantage in using simulated data is that each passing vehicle can be identified – thus the modelling method can be validated!!

Page 9: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Induction-loop data

- Time- Speed- Velocity- Gap…

Loop 1 Loop 2

Page 10: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Modelling method

• The aim is to develop a statistical method for identification of vehicles at successive loops and hereby allow for a direct estimation of the current travel time between loops

• The focus is on: - Pattern recognition of non-typical vehicles (non

typical in terms of vehicle lengths)- Develop a statistical measure for how well a

vehicle ’match’ another vehicle- Develop a dynamical model for estimation of

forecasting purposes

Page 11: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Pattern recognition

• The vehicles are identified from the measured lengths

• The identification has to be unique, thus it is not possible to match every passing vehicle:

1) Sequences of vehicles (e.g. 4 passing vehicles in a sequence)

2) Long vehicles (i.e. Big trucks)

Page 12: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Identification based on sequences

• Assume, that 4 vehicles are registered at loop 1 having the lengths L={L1, L2, L3, L4}

• Subsequently, 4 vehicles are registered at loop 2 having the lengths L’={L1’, L2’, L3’, L4’}

• The corresponding ‘match’ (denoted M) is calculated as:

4

1

4

1

2' )(

ii

iii

L

LLM

Page 13: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Identification based on sequences

• If M is less than the statistical measurement precision U, the sequence is defined as a ’possible match’

• For each passing sequence a possible match is searched for (within a given time window)

• If only one ‘possible match is found’ then this is defined a a ’unique match’

• From unique matches the travel time S is calculated

Page 14: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Problems and error sources

• It is not possible not identify a sequence if vehicles within the sequence change lane in between loops

• It is not possible to identify large vehicles or sequences if the measurement error is too large

• A non-correct sequence may falsely be classified as a correct sequence

Page 15: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Estimation of travel time

• The travel time at time i, Ti, is estimated using an Exponentially Weighted Moving Average (EWMA) model:

iii STT 1ˆ)1(ˆ

is the forgetting factor andSi is the current estimated travel time from the identification of non-typical vehicles. Ti-1 is the travel time at time (i-1), thus the travel time 1 time instant before

Page 16: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

EWMA and time window

time

Estimated Travel Time

StatisticalUncertainty

Page 17: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Results(simulated noise-free data)

Number in sequence

2 3 4 5 6 7

Correctly identified

2029 1603 897 581 409 269

Identified in total

3284 1887 970 611 424 281

%Correct

61 % 84 % 92 % 95 % 96 % 96 %

Page 18: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Results

0 50 100 150

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

Identified vehicles

Minutes from 6 AM

Est

ima

ted

Tra

vel T

ime

Page 19: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Results

5 6 7 8 9 10 11 12

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

Time of day

Tra

vel T

ime

ewmatrue

Page 20: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Results

5 6 7 8 9 10 11 12

01

23

45

Time of day

Tra

vel T

ime

ewmatrue

current

Page 21: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Computational speed and implementation

• Implementation of the algorithm is simple

• The estimation of travel time is only necessary when there is congestion

• Time needed for computation is not relevant (the algorithm is very fast for given and small intervals)

Page 22: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Future work: forecasts

Short term forecast

Pattern recognitionEWMA

Historical data:

Travel times, accidents, weather etc

Long term forecast

Page 23: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Travel time for longer distances

• For longer distances it must be expected that fewer sequences can be identified

• Methods for interpolation and estimation of travel time for longer distances will be validated

Page 24: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

14 15 16 17 18 19 20

05

10

15

Time of Day

min

ute

s

Page 25: Short term forecast of travel times on the Danish highway network based on TRIM data Klaus Kaae Andersen Thomas Kaare Christensen Bo Friis Nielsen Informatics.

Conclusion

• The proposed method makes it possible to identify non-typical vehicles from loop to loop

• Thus it is possible to estimate the travel time between loops with reasonable precision

• Future work includes estimating the time for longer distances and validating the method using camera data