Inference of road traffic congestion from sensor events

Transcription

1 Inference of road traffic congestion from sensor events Nick Gould and James Cheng, Manchester Metropolitan University William Mackaness, University of Edinburgh

2 Background UK Department for Transport - T-TRIG grant Aim - make better use of road sensor data Currently historic reporting on the state of strategic routes Quasi real-time? Inspired by Rude and Beard (2012) - High level event detection in spatially distributed time series Can high level events be detected from low level, primitive events? Traffic congestion Storms

3 Identifying primitive events from a sensor parameter Body event Initiating event Terminating event time

4 Major event Y Sensor for parameter A Sensor for parameter B Rate of change and its direction X

5 Our use case Rude and Beard Two indicators of congestion Vehicle speed Vehicle density (no. vehicles / km / lane) Simulated Etihad football stadium, Manchester, UK, January 13 th 2016 Data Sources Journey times on links Bluetooth sensors on traffic lights Vehicle counts at points Induction loops buried in road Indirection indicator Non-recurrent congestion

6 City Centre Bluetooth sensors used

7 City Centre Example Journey Time sensor pair

8 Abnormal journey times - towards stadium

9 19:45 21:35 13 th January typical days Abnormal journey times - detail

10 City Centre Sample volume sensors

11 City Centre Sample volume sensors

12 Example volume - towards stadium

13 Example volume - away from stadium

14 Identifying primitive events parameter Body event Initiating event Terminating event time journey time Exceptional Sequence study day data - exceptional value study day data - un-exceptional value 1 s.d. above mean of typical days Sequence = 3 or more consecutive points outside 1 s.d. slot

15 Local regression Δjt Quadratic Δt

16 Identifying clusters of primitive events Spatio-temporal clustering (Birant & Kut, 2007)

17 Limitations Identifying sequences No quasi real time Need to know earlier No directions Generalised links as points (mid-points) Want formalisation not visualisation not going to keep checking a map when driving. No context

18 Context What does Journey time of 102 seconds on the link between X and Y in the SE direction tell us? Relative values Magnitude Location Direction Reclassified Journey time - very low, low, normal, high, very high in relation to typical days Location - very near, near, far - in relation to stadium Direction - towards, away from, neutral - in relation to stadium

19 Classifying journey time magnitudes journey time magnitude very low low none none high very high 2 standard deviations below mean 1 standard deviation below mean Mean of journey time readings on typical days for this link at this time slot 1 standard deviation above mean 2 standard deviations above mean

20 Processing in R Using DAISY generate dissimilarity matrix magnitude, direction, location Using AGNES generate hierarchical clusters for each 10 minute time slot Kaufmann and Rousseeuw, 2005

21 Alert? Abnormal journey time clusters 15-5 minutes before kick off

22 Abnormal journey time clusters minutes after full time

23 Next steps To respond to congestion we need diagnosis Cause and characteristics of congestion (Lécué et al. 2012) Report on current state of network is not enough Congestion caused by football match that will start in 5 minutes is better We need context of congestion Context is part of the semantics of a domain Semantics can be encapsulated in an ontology (Kavouras & Kokla, 2008).

24 Defining a football match Extend the Transport Disruption ontology (Corsar et al. 2015) Landmark isa Event Football Stadium isa isa Concert occursat occursat hasa Activity isa Football Match isa isa Public Event hasa causes Congestion Attractor Published Start Time Published End Time EventBrite API (Lécué et al. 2012) Greater Manchester Road Activities Permit Scheme (GMRAPS)

25 Next steps other events Unpredictable

26 Next steps other events Unpredictable?

27 Next steps sink hole in Mancunian way Location? Direction?

28 Summary Semantic approach - productive Favoured by TfGM richer descriptions Cross system High journey times in Tokyo is different from high journey times in Manchester Allows for diagnosis?

29 Acknowledgements Department for Transport Transport for Greater Manchester

30 References BIRANT, D. & KUT, A ST-DBSCAN: An algorithm for clustering spatial temporal data. Data & Knowledge Engineering, 60, CORSAR, D., MARKOVIC, M., EDWARDS, P. & NELSON, J The Transport Disruption Ontology. The 14th International Semantic Web Conference. Bethlehem, Pennsylvania. KAUFMAN, L. & ROUSSEEUW, P. J Finding groups in data: an introduction to cluster analysis, Hoboken, New Jersey, John Wiley & Sons. LÉCUÉ, F., SCHUMANN, A. & SBODIO, M. L Applying Semantic Web Technologies for Diagnosing Road Traffic Congestions. In: CUDRÉ-MAUROUX, P., HEFLIN, J., SIRIN, E., TUDORACHE, T., EUZENAT, J., HAUSWIRTH, M., PARREIRA, J. X., HENDLER, J., SCHREIBER, G., BERNSTEIN, A. & BLOMQVIST, E. (eds.) The Semantic Web ISWC 2012: 11th International Semantic Web Conference, Boston, MA, USA, November 11-15, 2012, Proceedings, Part II. Berlin, Heidelberg: Springer Berlin Heidelberg. LLAVES, A. & KUHN, W An event abstraction layer for the integration of geosensor data. International Journal of Geographical Information Science, 28, RUDE, A. & BEARD, K High-Level Event Detection in Spatially Distributed Time Series. In: XIAO, N., KWAN, M.-P., GOODCHILD, M. & SHEKHAR, S. (eds.) Geographic Information Science. Springer Berlin Heidelberg.