Multi-agent systems allow the simulation of complex phenomena that cannot be easily described analytically. As an efficient tool, the agent-oriented traffic models have emerged to investigate vehicular traffic behaviors. In this article, a new agent-based traffic simulation model is proposed for solving the traffic simulation problems. A vehicle with the driver is represented as a composite autonomous agent in this model. Different from the classical car-following principles, vehicle-agent moving approaches are proposed instead of particle-hopping techniques. This model defines reasonable acceleration and deceleration rates at any certain condition. In this simulation, this can offer a natural, even cognitive way to follow the leader and switch lane. The simulation results have verified that this model has achieved more realistic results without loss of accuracy than those obtained from the cellular automata traffic models. This model can provide better simulation performance than the traditional vehicle-following models which are used to investigate the nonequilibrium traffic flow. A comparison of the real flow with the simulated freeway flow and lane capacity highlights the validness of this model.
microscopic traffic model, vehicle-following model, multi-agent, cellular automata, acceleration Citation:Dai J C, Li X. Multi-agent systems for simulating traffic behaviors.Microscopic vehicular traffic models focus on the study of the interaction between vehicles and investigate the synthesis characteristics of complex traffic phenomena. At present, there are several different types of traffic simulation models: vehicle-following (VF) models, Cellular Automata (CA) models and the multi-agent (MA) models. Vehicle-following models are based on Newtonian dynamics. In Newtonian mechanics, the acceleration may be regarded as the response of the particle to the stimulus it receives in the form of force which includes both the external force as well as those arising from its interaction with all the other particles in the system [1]. Pipes exploited the Follow-the-leader model in the 1950s [2]. After his heels, Gazis [3,4], Newell [5,6], Edie [7], Gipps [8], Bando [9], Treiber [10] proposed new VF models. Since the vehicular traffic is a nonequilibrium and complex system, classical vehicle-following models are not adequate to capture the essential features of lane changing, overtaking, speed limit and unanticipated parking. While particle hopping approaches usually formulated using CA are powerful techniques to investigate microscopic traffic. CA models have further been studied in recent decades [1,11-20], but CA models do not incorporate realistic driver and vehicular behavior. Vehicles are modeled as particles having unrealistic erratic acceleration and deceleration rates. And vehicles accelerate independent of their velocity and have speed jumps of about 27 km/h in 1 s [11]. Vehicles tend to change speed abruptly; they can come to a stop from a maximum speed of 135 km/h in 1 s [11]. In STCA (Stochastic Traffic Cellula...