Development And Application Of Driver-agent Traffic Flow Mixed Simulation System

Vehicle population in large and medium-sized cities is increasing dramaticallyevery year due to the acceleration of urbanization process. Problems like trafficcongestion, traffic accident and environment pollution gain increasing attention fromboth governments and researchers. Using virtual-driving system and microscopic trafficsimulation system to conduct studies like analysis of traffic flow or assessing theeffectiveness of traffic control policy has become a hot topic in traffic research field. Inthis paper, a creative driver-agent traffic flow mixed simulation system is proposed bycombing the two subsystems mentioned.Data communications between subsystems are crucial to successful developmentof the mixed system. This paper introduces the contents, types and conversion methodsof transmitted data in detail. To guarantee data reliability and accuracy, specialcommunication protocol is proposed. Serial communication technique is used to realizedata transmission.The performance of mixed system is improved from four aspects. By applyingstepwise displacement, change-lane behaviors of vehicles become smooth. By addingdifferent car models alternately to the mixed system, the turning lights will flash whenvehicles change from one lane to another. This research also amends the microscopicsimulation system to make sure car operated by the real driver will always be shown onthe interface. Finally, system time correction guarantees the consistence between systemtime and real time.Validation of microscopic simulation model is conducted using the mixed systemat the end of this research. By analyzing the driver’s real position data and predictionsfrom car-following model, this paper concludes that car-following model accuratelydescribes drivers’ following behaviors at5%tolerance. Further analysis indicates thattraffic density affects the accuracy of car-following model. Comparison between realand predicted lane changing data shows that accuracy of lane-changing model is verylow. More work is needed to identify real cause of the low accuracy.