A Driver Behavior Decision Model Based On Multi-Agent Systems

Traffic safety concerns millions of households and the problem about how to provide a safe and efficient traffic environment has gotten high attention from countries all around the world. In the transportation field, the artificial intelligence has been applied to the traffic research for many years. Some new technique has been introduced to this field for extending the ITS, such as expert systems, data mining, intelligent networks and etc. However, these studies only focused on the physical modification of road infrastructures and the improvement of control systems, but ignored driver's behaviors which play the important role in the driving process. Therefore, these studies cannot solve the problems we are confronted with. And so it has become a new challenge for experts to study the driver's behavior model.Because of its unique properties, the agent has been used to model the driver's behavior and decision-making process. By using the agent, we can easily analyze the driver's behavioral patterns in various road conditions, and find out the high-risk behavior and accident-prone condition for supporting the police's management. It can also improve traffic efficiency and reduce the accident. Though a lot of work has been done on this issue, there is no efficient solution reported to model the driver's decision-making process in the dynamic, complex and mutual traffic environment. This paper will focus on it and provide a new solution.1 Design the opened MAS-based Driver Behavior Decision Model.According to the behavior characteristics of the driver, we design a driver agent model, and figure out the agent's actions set and states set. Based on the driver agent model, we propose a MAS-based Driver Behavior Decision Model.2 Design the hybrid decision algorithm and reinforcement learning algorithms for supporting the DBDMWith this hybrid algorithm, we design the driving agent, which can manner more like human reasoning and decision-making and adapt to complex traffic environment. Finally, we improve the reinforcement learning algorithm so that it can be used in the DBDM.3 Design and Implementation of a simulation system based on MASBased on the previous model design, we develop a simulation system to verify our model's reliability and validity. Experiments show that the algorithm is reasonable, and can be used to support the DBDM.