A MULTI-AGENT APPROACH TO SIMULATE AND ANALYZE THE DRIVERS' ROUTE CHOICE BEHAVIOR UNDER THE IMPACT OF REAL-TIME INFORMATION

摘要

Agent Technology is a rising simulation approach in the field of artificial intelligence, and is also an extension of Object-oriented technology. Nowadays, this technology is applied widely in various research fields. Similarly, in micro simulation of urban traffic, Agent Technology plays a crucial role too. By constructing the Agent model of behavioral individuals in traffic systems, specifying the circumstances in which the individuals live, and adopting the advanced Agent simulation technology, we can reproduce the process of the formation of a complex and ordered system which emerges in the competition, interaction and cooperation between the Agents and the circumstances, also among Agents themselves. Thus, by changing the circumstances in which Agents inhabit, we can understand the mechanism of system formation under different decisions made by Agents which are influenced by the changes of their living circumstances, and furthermore, we can achieve better control of the traffic system. An SP (Stated Preference) survey of the drivers' route choice behaviors under real-time information was conducted in a congested commuting corridor in ShenZhen of China which has the fixed origination-destination (OD), and only a few alternative routes can be chosen by the drivers. In this survey, five different scenarios of traffic information were presented to the subjects, which were: no information, qualitative information, quantitative information, predictive information and descriptive information. On the basis of this survey, the drivers' route choice behaviors under each type of traffic information were modeled by GEE (Generalized Estimation Equation) technology, which could account for the correlations arising in this study because each subject made multiple choices. Subsequently, the two-layer Agent model of the driver-vehicle unit was developed. The two layers are Tactical layer and Strategic layer, which describe the tasks of driving (i.e., accelerating, decelerating, changing the lane) and answers for the route choice behavior under different types of information respectively. Finally, a simulation was conducted under each information scenario in the StarLogo environment which was developed for the Multi-Agent simulation by MIT. In each simulation scenario, the overall flow of the road system was used to evaluate the impacts of each type of information. Also, we studied the proportion of the uses provide with the information. Finally, the validity of this micro simulation model was checked by plotting the flow-density (Q-D) curve, which was compared to the classical one.