Delft University of Technology Faculty of Electrical Engineering, Mathematics and Computer Science Man Machine Interaction Group Dynamic Routing using Ant-Based Control by Adriana Camelia Suson Supervisor: Prof. Dr. Drs. L. J. M. Rothkrantz Delft, August 2010 Dynamic Routing using Ant-Based Control Thesis submitted in partial fulfilment of the requirements for the degree of MASTER OF SCIENCE in MEDIA AND KNOWLEDGE ENGINEERING by Adriana Camelia Suson born in Braşov, Romania Media and Knowledge Engineering Department of Electrical Engineering Faculty of Electrical Engineering, Mathematics and Computer Science (EEMCS) Delft University of Technology Dynamic Routing using Ant-Based Control by Adriana Camelia Suson Department: Man Machine Interaction http://mmi.tudelft.nl/ Student Number: 1391968 Date: 23 August 2010 Committee Members: Member: Prof. Dr. Drs. L. J. M. Rothkrantz, TUDelft Member: Dr. Ir. P. Wiggers, TUDelft Member: Ir. H. Geers, TUDelft Member: Bsc. B.Tatomir, TUDelft Abstract Currently most car drivers use static routing algorithms based on the shortest distance between start and end position. But the shortest route is different from the fastest route in time. Because existing routing algorithms lack the ability to react to dynamic changes in the road network, drivers are not optimally routed. In this thesis we present a multi-agent approach for routing vehicle drivers using historically-based traffic information. The general workings of our solution bears strong similarities with Ant Based Control (ABC) and AntNet, but an important modification has been made, namely the adaptation of ant-like agents for spatio-temporal routing. The dynamic routing algorithm proposed, routes self-interested drivers on an intersec- tion to intersection basis via the fastest path between a proposed source and a destination. For this to happen, a time-expanded graph encodes variable road network costs. Ant-like agents are launched in this graph. They use a technique of collective learning based on locally dependent pheromone tables. Finally, we report results obtained for part of The Netherlands’ GIS-based road net- work. In the established experiment setting, the new ABC makes a positive difference for drivers. An important reduction of the travelling time was observed in 53% of the cases. The experimental results also showed that ABC clearly outperforms Static Dijkstra’s algorithm and Dynamic Dijkstra with updates algorithm. Contents Contents i List of Figures v List of Tables vii List of Symbols xi 1 Introduction 3 1.1Problemsetting................................. 3 1.2Problemstatement............................... 8 1.2.1Purpose.................................. 8 1.2.2Statementofobjectives......................... 9 1.2.3Scope................................... 10 1.3Societalrelevanceofthethesis......................... 11 1.4Researchchallenge............................... 12 1.5Organisationofthethesisdocument...................... 12 2 Related work 15 2.1Taxonomyofroutingsystems......................... 15 2.2Dijkstra’salgorithm............................... 19 2.3 A swarm intelligence approach to routing . 20 2.3.1 The class of ant “inspired” algorithms . 20 2.3.1.1Featuresofantalgorithms.................. 22 2.3.1.2 Application of ACO to static problems . 24 2.3.1.3 Application of ACO to dynamic problems . 26 2.3.2 Theclassofbee“inspired”routing.................. 30 2.3.2.1Beecolony’scharacteristics.................. 30 2.3.2.2Applicationsofbeecolonies.................. 30 2.3.3Hierarchicalnature-inspiredrouting.................. 32 2.4Trafficmodelsforsimulationofrouting.................... 33 2.4.1Simulatingtrafficflow.......................... 34 2.5Conclusionofliteratureoverview....................... 36 3 Model Design 39 3.1Generaldesignaspects............................. 40 3.1.1 Centralized versus decentralized technical architecture . 40 3.1.2Traveltimemodels........................... 42 3.1.3Choosingthealgorithm......................... 45 3.1.4 Analytical-based model versus simulation-based model . 45 3.1.5 Output representation and update management . 46 3.2Resourcesforthecurrentroutingsystem................... 46 i ii CONTENTS 3.2.1Spatialnetworkformulation...................... 47 3.2.2Temporalnetworkformulation..................... 50 3.3Algorithmicapproach.............................. 52 3.3.1Descriptionofanexample....................... 52 3.3.2Routingtable.............................. 56 3.3.3Solutionconstruction.......................... 57 3.3.4Updatesoftheroutingtable...................... 57 3.4 Travel time estimation methodology for a road segment . 59 3.5Vehiculartraffic................................. 59 3.5.1 Modelling vehicular traffic . 60 3.5.2Constraintsofnetworkloading..................... 60 4 Implementation details 63 4.1Developmenttool:Quintiq........................... 63 4.1.1Motivation................................ 63 4.1.2Quintiqenvironment.......................... 64 4.1.3Quintiqanddatabases.......................... 65 4.1.4QuintiqandPTV............................ 66 4.2Systemarchitecturaldesign.......................... 67 4.2.1Chosenarchitecture........................... 67 4.2.2UMLdiagrams.............................. 69 4.3Algorithmpseudo-codedescription...................... 77 4.3.1ABCalgorithm.............................. 77 4.3.2Dynamictraveltimeestimation.................... 78 4.4Implementationissues.............................. 78 4.4.1Vehiclegeneration............................ 78 4.4.2 Modelling the future . 78 4.4.3Parameters................................ 79 4.5Systeminterfacedescription.......................... 79 5 Evaluation of the routing algorithm 85 5.1Experimentalmethodology........................... 85 5.1.1Generalexperimentalsettings..................... 86 5.1.2Datapreprocessingandfiltering.................... 86 5.1.3 Routing algorithms used for comparison . 90 5.2Pathselection:adaptivitytest......................... 90 5.3Travelingtime:effectivenessanalysis..................... 91 5.4 Pheromone evolution: stability test . 93 5.5Globalimprovement............................... 97 5.6Analysisofthealgorithmrun......................... 97 5.6.1Averageageoflivingforwardant.................... 97 5.6.2Numberofforwardants......................... 98 6 Conclusions and Future Work 101 6.1Summary.....................................101 6.2 Reached goals . 101 6.3 Description of the findings . 103 6.4Futureworkorpossibleextensions......................103 Bibliography 105 A Class Diagram 113 B Results of the adaptivity test 117 CONTENTS iii CPaperfromProceedings of ANTS 2008 121 DPaperfromProceedings of the 2009 Winter Simulation Conference 125 List of Figures 1.1 Traffic congestion in The Netherlands on A9 at the off-ramp towards N205. .... 4 1.2 Dynamic Routing Information Panel (DRIP) on the ring of Rotterdam. 4 1.3 Minutes of delay on a DRIP for drivers heading towards Amstel intersection. 4 1.4MapdisplayoftrafficcongestioninTheNetherlands............... 5 1.5 Portable GPS car navigation system of the Dutch manufacturer Tom Tom. 5 1.6 Mobile Millenium traffic-monitoring system based on GPS cellular phones. 5 1.7 Causes of travel time loss on German highways, adapted from [54]. 7 1.8Schematicviewofthesisorganisation....................... 13 2.1 Example of routing guidance architectures according to [74]. 17 2.2 Shortest path finding capability of ant colonies. 21 2.3 Process organisation of the Ant Colony Optimization Metaheuristic. 26 2.4 Layered routing model of BeeJamA accordingto[72]............... 32 3.1 Technical infrastructure of decentralised routing system. 41 3.2 Technical architecture of centralisedroutingsystem............... 42 3.3 Network for instantaneous and actual route choice calculation. 44 3.4Potentialgraphofthenode-linknetwork..................... 48 3.5 Geographical positioning of the locations of intersection Amstel. 49 3.6Anode-per-intersectionnetworkflowdatamodel................. 50 3.7 Delay distribution for the segment Holendrecht Oudenrijn from A2. 51 3.8 A three-dimensional time-expanded model of the costs of a road network [35]. 51 3.9 Time aggregated graph of the costs of a road network. 51 3.10Exampleofantbehaviour.............................. 53 3.11 Decision factors of the forward ant at each intersection. 53 3.12 The forward ant creates the backward ant at the destination. 53 3.13Detailedexampleforwardantbehaviour...................... 54 3.14Detailedexampleofbackwardantbehaviour................... 55 3.15 Example of probabilities updated by backward ant. 56 3.16Exampleoftheroutingtable’sstructure...................... 56 3.17Conservationofflowinanetwork......................... 61 4.1ArchitectureofQunitiqEnvironment,version4.2.5................ 65 4.2QuintiqEnvironment’sinteractionwithdatabases................ 66 4.3 Architecture of the routing system as it was implemented in Quintiq. 68 4.4 Use case diagram illustrating main functionalities of our routing system. 69 4.5Usecasepackageforloadingdata......................... 70 4.6 Package representing the dataset selection use cases. 70 4.7Extendinfrastructurerelatedusecases...................... 71 4.8Representationofthesimulateusecases...................... 71 4.9 Manipulate data package of use cases. 71 v vi LIST OF FIGURES 4.10SimplifiedClassDiagram.............................. 75 4.11 Simplified Sequence Diagram for interactions at the server level. 76 4.12Theorganisationoftheuserinterfacecomponents................ 81 4.13 The Routing Form belonging to the current implementation.