DYNAMIC ODME FOR AUTOMATED VEHICLES MODELING USING BIG DATA

Transcription

1 DYNAMIC ODME FOR AUTOMATED VEHICLES MODELING USING BIG DATA Dr. Jaume Barceló, Professor Emeritus, UPC- Barcelona Tech, Strategic Advisor to PTV Group Shaleen Srivastava, Vice-President/Regional Director (PTV Group Americas)

2 INTRODUCTORY REMARKS Connected vehicle systems and autonomous vehicles likely to be major game changers in traffic and mobility. No longer a question of if, but of when, in what form, at what rate. And through what kind of evolution path operational regimes in which vehicles are connected to each other and to the infrastructure, and augmented with autonomous capabilities. (Hani Mahmassani, Workshop 134: Emerging Needs for Improving Simulation Models, TRB 96 th Annual Meeting, Washington, January 8, 2017)

3 CONNECTED & AUTONOMOUS VEHICLES

4 AUTONOMOUS VS CONNECTED VEHICLES

5 SINGLE VEHICLE APPLICATIONS & COOPERATIVE APPLICATIONS

6 WILL AV REPLACE CURRENT AUTOMOTIVE TECHNOLOGIES FOR INDIVIDUAL MOTORIZED MOBILITY? OR, WILL MOSTLY BE USED FOR COLLECTIVE MOBILITY?

7 VISUALIZATION OF SHARED SELF-DRIVING CAR SIMULATION FOR LISBON

8 A SELF-ORGANIZING SYSTEM OR BETTER EXTERNALLY ASSISTED?

9 COOPERATIVE DRIVING WITH THE HELP OF V2X COMMUNICATIONS Source: D. Jia, D. Ngoduya, Enhanced cooperative carfollowing traffic model with the combination of V2V and V2I communication Transp. Res. B, March 2016 Source: L. Zhao, J. Sun, Simulation Framework for Vehicle Platooning and Car-following Behaviors under Connected-Vehicle Environment, Procedia - Social and Behavioral Sciences 96 ( 2013 )

10 TESTING VACS BY MICROSOPIC SIMULATION ACC string-stability ACC traffic efficiency From: Ntousakis, I.A., Nikolos, I.K., Papageorgiou, M.: On microscopic modelling of adaptive cruise control systems. 4th Intern. Symposium of Transport Simulation (ISTS 14), 1-4 June 2014, Corsica, France. Transportation Research Procedia 6 (2015), pp

11 ACC/CACC: STABILITY/EFFICIENCY Macroscopic simulation of traffic flow (spatio-temporal evolution of traffic density) close to an on-ramp using the GKT model, combined with a novel ACC/CACC modeling approach. Left: manual cars; Middle: ACCequipped cars; Right: CACC-equipped cars. Source: Delis, A.I., Nikolos, I.K., Papageorgiou, M.: Macroscopic traffic flow modeling with adaptive cruise control: development and numerical solution. Computers & Mathematics with Applications, 2015

12 HIERARCHICAL+ TM Network Traffic Control Link Control Link Control Connect VACS and TM communities for maximum synergy TM remains vital while VACS are emerging Overlapping link controllers? Share of control tasks? V2I V2V Papageorgiou, M., Diakaki, C., Nikolos, I., Ntousakis, I., Papamichail, I., Roncoli, C. : Freeway traffic management in presence of vehicle automation and communication systems (VACS). In Road Vehicle Automation 2, G. Meyer and S. Belker, Editors, Springer International Publishing, Switzerland, 2015, pp

13 EQUIPPED VEHICLES, V2V & V2I BECOME RICH DATA SOURCES TO SUPPORT MANAGEMENT AND GUIDANCE Autonomous vehicles rely on knowing the roadway they are traveling on, changes to the roadside such as new development or construction will require the type of real-time exchange of information that CV technology provides including valuable information about the road ahead allowing rerouting based on new information such as a lane closures, or congestion growing. ATKINS Autonomous vs connected vehicles what s the difference? (Suzanne Murtha 02 Oct 2015 )

14 SENSORS, FUNCTIONS AND SURROUNDING AWARENESS OF PROBE CARS DEPICTED IN RED- 4 VEHICLES DETECTED IN FRONT AND 3 BEHIND -X Y X Each of the vehicles measures each half second and stores its position, heading, speed, and acceleration, as well as distances and relative speeds to visible surrounding vehicles captured by the radars. -Y

15 Traffic Data Center V2V TRACKED EQUIPPED VEHICLES AWARE OF Space x A 1 2 SURROUNDING VEHICLES TRAJECTORY RECONSTRUCTION & TRAFFIC STATE ESTIMATION 1 4 A 2 3 GPS Equipped V2V 8 B Unequipped f d fl (a) (x f, y f ) (x l, y l ) f d fl (b) (x f, y f ) (x l, y l ) Time t l l f (x e, y e ) (c) d eo (x o, y o ) o Bluetooth Equipped VW equipped 1 The relative distance d eo between the equipped and the observed car The relative speed v eo between the equipped and the observed car 4 6 The map-matched Traffic Data Center A 5 V2V position (x 8 e,y e ) and speed 2 v 3 7 B e of the equipped car. GPS Equipped Unequipped Space x A 1 Time t 2 Source: L. Montero, J. Barceló et al., A case study on cooperative car data for traffic state estimation in an urban network, Paper , 95 th TRB Annual Meeting 2016, Compendium of Papers.

16 TRAFFIC DATA ANALYTICS (Extracting the most useful & valuable information from traffic measurements) Dealing with heterogeneous traffic data from varied technological sources (conventional detectors, Bluetooth, GPS, cooperative & autonomous vehicles ): - Data filtering, completion and fusion techniques - Processing huge amounts of data (Big Data Ad hoc Data Base Management Techniques) Filtering and completion techniques Kernel Smoothing Methods, Kalman Filter & traffic flow based models to identify and remove outliers And to supply missing data Data Fusion Techniques Kernel Smoothing Methods Machine Learning Traffic Models Dynamic Flow Models OD Estimation

17 CONCEPTUAL APPROACH TO AN ADAPTIVE AREA WIDE CONTROL STRATEGY BASED ON THE NETWORK FLOW DIAGRAM Origin r LARGE URBAN OR METROPOLITAN AREA Alternative recommended route GATE-IN Congestion Destination s QUEUE URBAN AREA TO MANAGE GATE-OUT Input flow rates (k) B C Critical Point in the managed area Real-time Traffic Data Measurements from sensors Output flows n(k-1) A Allow access Restrict access Estimation algorithm for n k ADAPTIVE FLOW CONTROL STRATEGY Figure 6 Potential use of the Network Fundamental Diagram to support Active Area WideTraffic Management Strategies M. Keyvan-Ekbatani, M. Papageorgiou, V. L. Knoop, Comparison of On-Line Time-Delayed and Non-Time-Delayed Urban Traffic Control via Remote Gating, TRB 2015 Annual Meeting, Paper

18 TRAFFIC DATA ANALYTICS: DYNAMIC OD ESTIMATION Identification of time-dependent mobility patterns in terms of Origin-Destination (OD) Matrices Exploiting ICT measurements Origin Destination τp t ij number of trips from Origin i to Destination j in time period for purpose p State equations AR(r) on deviates: g r ( k+ 1) = D( w) g( k w+ 1) + w( k) w= 1 Observation equations: z ( k) A1U 1(k) = A2U2(k) E(k) D(w) transition matrices describing the effects of previous OD path flow deviates g ijc (k-w+1) on current flows g ijc (k+1) T T v1(k) g(k) + v2(k) = F(k) g(k) + v(k) v3(k) First block: deviates of observations at sensor locations Second block: conservation flows for each time interval k Initialization g = Dg k k+ 1 k k k k T P k+ 1 = DPk D + W k k ( I G F ) P k+ 1 P k+ 1 = k+ 1 k+ 1 k+ 1 G k T ( F P F R ) k T k+ 1 = Pk + 1Fk + 1 k+ 1 k+ 1 k+ 1 + k g k+ 1 k k+ 1 = gk dk 0 KF recursive dynamics k ( z( k + ) F g ) d k+ 1 = Gk+ 1 1 k+ 1 k+ 1 Ad Hoc Kalman Filter to estimate the time dependent OD MLU OD Path id OD path links OD pair ICT sensor id Entry id MLU OD Path id OD path links OD pair ICT sensor id =(1,8) 6, 1, =(2,8) 7,1, =(1,8) 6,2,3, =(2,8) 7,2,3, =(1,8) 6,2, =(2,8) 7,2, =(1,9) 6,1,3, =(2,9) 7,1,3, =(1,9) 6,2, =(2,9) 7,2,4 2 Entry id J.Barceló, L.Montero, M.Bullejos, M.P. Linares, O. Serch (2013), Robustness and computational efficiency of a Kalman Filter estimator of time dependent OD matrices exploiting ICT traffic measurements. TRR Transportation Research Records: Journal of the Transportation Research Board, No. 2344, pp

19 DATA COLLECTION FROM AUTONOMOUS/CONNECTED VEHICLES Assumption: travel times T rq of drivers departing from origin r during time interval t going through POI q follow a distribution (not stationary under congestion), no matter the selected path. Approximate travel time distributions by discrete distributions with bin proportions updated according to collected on-line ICT data.

20 EKF APPROACH FOR NETWORKS (III) : FLOW ESTIMATES AND ERROR CORRECTIONS (SHALEEN SRIVASTAVA, 2010) Traffic flow at a location Flows y(t) Assume Gaussian distributed measurements Model simulation (virtual detectors) traffic flow Measurement (real detectors) traffic flow Flows measurement from the model at t1: Mean = z1 Variance = σz1 Optimal estimate of traffic flows: ŷ(t1) = z1 Variance of error in estimate: σ2x (t1) = σ2z1

21 EKF APPROACH FOR NETWORKS (III) : FLOW ESTIMATES AND ERROR CORRECTIONS (SHALEEN SRIVASTAVA, 2010) So we have the prediction ŷ-(t2) Detector data measurement at t2: Mean = z2 and Variance = σz2 Need to correct the prediction by model due to measurement to get ŷ(t2) Closer to more trusted measurement linear interpolation Corrected mean is the new optimal estimate of traffic flows (basically we have updated the predicted flows by model using detector data) New variance is smaller than either of the previous two variances

22 EKF APPROACH FOR NETWORKS (III) : FLOW ESTIMATES AND ERROR CORRECTIONS (SHALEEN SRIVASTAVA, 2010) If measurement is preferred: - Measurement error covariance decreases to zero - Weights residual more heavily than prediction If prediction is preferred: - Prediction error covariance decreases to zero - Weights prediction more heavily than residual

23 MEASURING THE QUALITY OF THE ESTIMATES ESTIMATED (KF APPROACH) VS TARGET FLOWS IN OD PAIRS FOR A 15 MINUTES INTERVAL Estimated vs Target OD Flows - 1h (veh/h 500 Demand Set 2: Estimated vs Target OD flows - 1h y = 1,04x - 2,9644 R² = 0,9387 Barcelona s Central Business District (CBD), Eixample, 2111 sections, 1227 nodes 120 generation centroids, 130 destination centroids (877 non-zero OD pairs) 116 Loop detector Stations & 50 Bluetooth Antennas Target OD flow Estimate OD flow OD Pair

24