Project overview Phases I & II Adaptive learning node - PowerPoint PPT Presentation

Organic Traffic Control (OTC 3 ) J. Branke, J. Hähner, C. Müller-Schloer, H. Prothmann, H. Schmeck, S. Tomforde SPP 1183 Organic Computing | 11 th colloquium Munich | October 7 th /8 th , 2010 by Sharon Drummond

Project overview Phases I & II Adaptive learning node controller Collaborative control of traffic signals in urban road networks • Decentralised coordination • Hierarchical coordination Simulator EA Signal plan optimisation Regional Manager Observer Controller Observer LCS Signal plan selection SuOC 2

Project overview Phase III Route guidance and driver information • Communicating traffic lights • Variable Message Signs • Decentralised routing  Minimise travel times  Prevent congestions  Improve robustness wrt incidents Karte (c) OpenStreetMap (und) Mitwirkende, CC-BY-SA 3

Agenda 1. Decentralised routing – Distance vector routing – Link state routing 2. Test scenarios and results 3. Observer/controller refinements 4. Summary and outlook 4

1. D ECENTRALISED ROUTING by Shannon Kokoska 5

Routing components • Located at intersections • Estimate turning delays (based on current flow / green time) Destination Next hop Delay 18s 25s • Manage routing tables (for incoming sections) • Communicate routing data – Distance vectors – Link states 6

Distance vector routing (DVR) Routing components • Process local destinations ( ) • Compute turning delays … 10 16 • Create routing entries S4 12 • Advertise destinations A | S4 | 46 A | S3 | 36 A | S5 | 44 and distances … … Dest. Next Delay • Receive routing data S3 S5 A | S2 | 32 A S1 20 – Add new routes to … 20 14 S2 advertised destinations 18 S1 – Update distances if A | S1 |18 A advertised distance is shorter 7

Link state routing (LST) Link Delay S2  S1 20 … Routing components S2 • Determine turning delays S1 • Communicate delay changes – Link state advertisement – Network flooding • Create weighted network graph from received link states • Compute shortest paths using Dijkstra’s algorithm 8

Routing in road networks Compared to Internets • • Road networks limited in size Separate networks for road traffic and routing traffic • Intersections • Routing protocols differ in Separate queues / routing – Computational effort tables for incoming links • – Communication cost Turnings – Capacity influenced by  More important for traffic lights Internets – Non-linear cost relations 9

2. T EST SCENARIOS AND RESULTS 10

Experimental evaluation Scenario Manhattan network • 25 intersections ( ) • 20 destinations ( ) • 120 road segments ( ) Signalised intersections • Variable Message Signs (VMS) • 4-phased signal plans • Organic traffic lights Traffic demand … 3800-7600 veh/h (equally distributed among destinations) 11

Experimental evaluation Test case I • 75% compliance rate • Undersaturated demand • No incidents Reductions DVR | LST Travel time 5.0% | 3.4% Stops 3.1% | 2.6% Fuel/CO 2 5.8% | 4.6% No routing DVR LST Travel time Stops 12

Experimental evaluation Test case II • 75% compliance rate • Highly saturated demand • No incidents Reductions DVR | LST Travel time > 50% Stops > 30% Fuel/CO 2 > 35% No routing DVR LST Travel time Stops 13

Experimental evaluation Test case III • 75% compliance rate 1. 2. 3. • Undersaturated demand • 3 blocked roads No routing Routing 1. 2. 3. Reductions DVR | LST Travel time 32% Stops 12% Fuel/CO 2 19% No routing DVR LST Travel time Stops 14

3. O BSERVER / CONTROLLER REFINEMENTS In cooperation with by Ryan Wick 15

Observer/controller refinements Level 1: On-line learning Level 2: Evolutionary optimisation XCS-RC Handling of noisy, simulation-based fitness estimations, e.g.: • Improved discovery component • Simulated duration vs. estimation • Rule generalisation by inference quality  Improved learning rate  Reduced population size MP20 Perf. XCS Popsize XCS Perf. XCS-RC • Distribution of simulation time Popsize XCS-RC Cross-project publication • N. Fredivianus, H. Prothmann, and H. Schmeck. XCS Revisited: A Novel Discovery Component for XCS. Accepted for 8 th Int. Conf. on Simulated Evolution And Learning , 2010. • Presented at SPP-miniworkshop on Learning Classifier Systems 16

Organic Network Control Network simulator EA Reliable broadcast protocol for Parameters (NS-2 / Omnet++) • Delays MANETs Parameter optimisation • Buffer sizes • Increasing network load • Interval lengths • Dynamic environments: • Counter Observer LCS Static protocol configuration Parameter selection works well on average • Better performance due to: Network protocol instance – On-line adaptation SuOC – Context-aware protocol settings Cross-project publication S. Tomforde, E. Cakar, and J. Hähner. Dynamic control of network Fitness: +8% protocols – A new vision for future self-organising networks. In Proc. of the 6th Int. Conf. on Informatics in Control, Automation and Robotics – Intelligent Control Systems and Optimization, pages 285-290, 2009. 17

4. S UMMARY AND OUTLOOK 18

Summary and outlook Decentralised routing Outlook • 1. Regional routing Extension of organic traffic lights – Variable Message Signs – No car-to-x communication • Adapted Internet protocols – Local communication – Local traffic data • Routing improves robustness  Reduce communication effort – Highly saturated demands and computational cost – Road works – Accidents 2. Hierarchical routing – … • Network-wide traffic prediction • Incorporate external goals • System vs. user optimum 19

Selected publications 2010 S. Tomforde, H. Prothmann, J. Branke, J. Hähner, C. Müller-Schloer, and H. Schmeck. Possibilities and limitations of decentralised traffic control systems. In IEEE World Congress on Computational Intelligence , pages 3298-3306. IEEE, 2010. H. Prothmann, J. Branke, H. Schmeck, S. Tomforde, F. Rochner, J. Hähner, and C. Müller-Schloer. Organic traffic light control for urban road networks. International Journal of Autonomous and Adaptive Communications Systems , 2(3):203- 225, 2009. • H. Prothmann and H. Schmeck. Evolutionary algorithms for traffic signal optimisation: A survey. In Proc. of mobil.TUM 2008 - 2009 2009 - International Scientific Conference on Mobility and Transport, 2009. • S. Tomforde, H. Prothmann, F. Rochner, J. Branke, J. Hähner, C. Müller-Schloer, and H. Schmeck. Decentralised progressive signal systems for organic traffic control. In Proc. of the 2nd IEEE International Conference on Self-Adaption and Self-Organization (SASO 2008) , pages 413-422. IEEE, 2008. • H. Prothmann, F. Rochner, S. Tomforde, J. Branke, C. Müller-Schloer, and H. Schmeck. Organic control of traffic lights. In Proc. of the 5th International Conference on Autonomic and Trusted Computing (ATC-08) , volume 5060 of LNCS, pages 219-233. Springer, 2008. ATC08 BEST PAPER AWARD • J. Branke, P. Goldate, and H. Prothmann. Actuated traffic signal optimization using evolutionary algorithms. In Proc. of the 6th European Congress on Intelligent Transport Systems and Services (ITS07) , 2007. 2006 - 2007 • • F. Rochner, H. Prothmann, J. Branke, C. Müller-Schloer, and H. Schmeck. An organic architecture for traffic light controllers. In Informatik 2006 – Informatik für Menschen , volume P-93 of LNI, pages 120-127. Köllen Verlag, 2006. • • J. Branke, M. Mnif, C. Müller-Schloer, H. Prothmann, U. Richter, F. Rochner, and H. Schmeck. Organic Computing – Addressing complexity by controlled self-organization. In Post-Conference Proce. of the 2nd International Symposium on Leveraging Applications of Formal Methods, Verification and Validation (ISoLA 2006) , pages 185-191. IEEE, 2006. 20

Summary and outlook Decentralised routing Outlook • 1. Regional routing Extension of organic traffic lights – Variable Message Signs – No car-to-x communication • Adapted Internet protocols – Local communication – Local traffic data • Routing improves robustness  Reduce communication effort – Highly saturated demands and computational cost – Road works – Accidents 2. Hierarchical routing – … • Network-wide traffic prediction • Incorporate external goals • System vs. user optimum 21