多智能体系统中子域适应度评估的合作协进化协作

Cooperative co-evolutionary collaboration algorithm based on sub-domain fitness evaluation in multi-agent system

为了克服合作协进化算法在解决复杂多智能体系统协作问题时存在的适应度函数难以建立和协作行为难以达到全局最优等问题,提出1种子域适应度评估的合作协进化算法来实现异构多智能体系统中智能体的自适应协作。该算法将复杂问题域模型分解成相互影响较小、较易求解的子问题域模型,在子问题域模型之间并行使用合作协进化算法来完成智能体协作行为的进化,有效降低适应度评估的复杂度。在子问题域进行合作协进化时,在适应度函数中引入环境因子影响矩阵,将其他子问题域的影响信息映射到该子问题域中的个体适应度评估中,从而引导种群向全局优化方向进化。ECJ系统中的仿真实验结果验证了其有效性。

In order to overcome the difficulties of establishing the fitness evaluation function and obtaining the global optimal decision when the cooperative co-evolutionary algorithm is applied to solve the collaboration in complex multi-agent system, a cooperative co-evolution algorithm with sub-domain fitness evaluation was introduced to deal with the adaptive collaboration of heterogeneous multi-agent system. In the proposed algorithm, the complex problem domain model was decomposed into some sub-domain models with less interaction. The results show that the evolution of agents＇ behavior is completed by parallel cooperative co-evolution among the sub-domains, which reduces the complexity of fitness evaluation function effectively. While the cooperative co-evolution is executed in a sub-domain, a matrix of environmental impact is introduced to map other sub-domains＇ influence information to the fitness evaluation function in this sub-domain, to achieve the global optimization of population evolution. The proposed algorithm is proven to be effective in ECJ simulation platform.