TRAFFIC SIMULATION BASED ON CAS THEORY

摘要

In this paper, we present a decentralized multi-agent based dynamic traffic simulation system intuited by Complex Adaptive System (CAS) theory with SWARM platform, which is a multi-agent software platform for the simulation of complex adaptive systems. Swarm allows people to develop traffic systems from bottom up: starting from a collection of individual vehicles and traffic signals, that follows specific behavioral rules, the interaction of theses individual agents are simulated on a road map, thus leading to emergent patterns of traffic dynamics. In our simulation, vehicles, roads and traffic signals are defined as agents, who make independent decisions about their actions. A two-dimensional Cellular Automaton (CA) is used to simulate multi-intersection dynamic traffic flow and the user can interactively manipulate control parameters of the vehicles. In addition, a combined algorithm of reinforcement learning and genetic algorithm (GA) is applied to self-organized control of the traffic signals. We focus on the abstract problem and report on some interesting characteristics of the emergent structures produced by the algorithm. Furthermore, we compare the traffic signal control approach introduced in this paper with the traditional macroscopic control method. Using experimental method it can be concluded that the performance of our approach is much superior to the traditional control method in that it can overcomes the drawbacks such as huge data transfer and communication, accurate traffic model and so on.