基于演化博弈惩罚机制的多智能体协作稳定性研究

针对复杂、动态环境中多Agent协作的稳定性问题,提出了一种基于博弈论及惩罚机制的协作方法,通过效用函数来选择最优策略,实现均衡协作;为了提高协作的稳定性与成功率,引入惩罚机制,通过不断调整惩罚系数来维护多Agent协作的稳定性,并在形成协作团队时,充分考虑参与协作的Agent的信誉值。仿真结果表明,该方法能有效地降低任务完成时间,避免Agent在动态协作中随意退出,提高协作效率及协作稳定性。

Research on multi-agent cooperation stability based on the punishment mechanism of evolutionary games

The coordination stability problem in complex environments is one of the key problems in the research of multi agent cooperation. We present a multi agent cooperation stability method on the basis of game theory methods and punishment mechanism. To maintain the stability of multi agent cooperation and achieve a balanced cooperation, a punishment is introduced and continuous adjustment of the penalty factors is performed. Agent credit values are fully considered when the cooperation team is formed. Simulation results show that the proposal can not only reduce task completion time effectively, but also avoid agent exits in the dynamic cooperation, thus improving the cooperation efficiency and stability.