多Agent系统中强化学习的研究

As the growth of the agent technology, its application scope is extended continuously. Now it has some successful examples in the fields of industry, business, medicine and entertainment. Following are the problems existing in the current research of Multi-Agent systems. First, how the agents in multi-agent systems study and adapt to the dynamic environment when they only have limited knowledge. Second, how agents in MAS cooperate and negotiate to accomplish tasks effectively. Therefore, the learning ability, adaptive ability and cooperative ability are the keys of the MAS research. As a special frame of machine learning, reinforcement learning can learn from environment by interacting with it even knows little about it. This project does an in-depth research on some theories and key technologies of reinforcement learning in MAS. It accomplishes the prospective goals and obtains remarkable achievement on the key technologies. The innovations of this project are that it gives a reinforcement learning arithmetic based on MAS, an adaptive negotiation model of MAS, a reinforcement learning arithmetic based on BP neural network, a function evaluating arithmetic of reinforcement learning, a learning frame of MAS and a method quickening convergence of reinforcement learning. Since the environment of emulate robots World Cup football match ----Robocup is dynamic, real-time, oppositional and uncertain, it is a challenging multi-agent environment. We study this environment and use our achievements into it, which is proved effective. These achievements can be used in fields like intelligent control, intelligent robots and so on, and will enhance the intelligent behavior ability of the intelligent systems.We have published more than 10 papers on the core journals and academic conferences, some of which have been admitted by EI. 5 PH.ds and 5 Masters participate this project. And the achievements have been used in RoboCup software environment.

研究多Agent系统中强化学习的有关理论和技术，将研制一个强化学习模型和提高强化学习樟菜俣鹊姆椒ǎ约昂凸槟裳啊⑸窬绾鸵糯惴ㄏ嘟岷系挠泄丶际酰⒐乖煲桓龆郃gent强化学习系统。本项目的研究成果可应用到智能控制、智能机器人、个人数字助手和Internet上的智能信息检索等应用领域，能大大提高智能系统的智能行为能力。