伙伴选择问题的多Agent强化学习演化博弈方法

Multi Agent reinforcement learning evolutionary game algorithm for partner selection

为了解决可供选择的企业伙伴组合规模日益扩大的问题,根据企业结盟伙伴选择招、投标问题的特点,建立了实现竞标费用和拖期惩罚费用之和最小的非线性整数规划模型.利用多Agent的强化学习思想和协调机制,在演化博弈算法的基础上,提出了多Agent强化学习演化博弈算法.将算法在多个不同规模的仿真实例上与遗传算法和演化博弈算法进行了对比分析,研究结果表明,该方法在处理规模较大的伙伴选择问题上,计算速度和达优率两方面的综合性能优势明显.

The formation of modem enterprise alliance solid network is changing the partner selection in great project from independent enterprises to combined enterprises. By analyzing the optimization objective of subprojects and the core competence of tender enterprises, a nonlinear integer-programming model of the combined enterprises partner selection was developed. Based on the multi Agent reinforcement learning, coordination mechanisms and evolutionary game algorithm, a multi Agnet reinforcement learning evolutionary game algorithm was designed. A case study was carried out to compare the proposed algorithm with the genetic algorithm and evolutionary game algorithm. The results suggest that the proposed method is effective in both running time and performance for the large-scale partner selection of enterprise alliance.