摘要： 为解决传统遗传算法早熟收敛和收敛速度慢的问题，提出一种基于强化学习的多策略选择遗传算法MPSGA。通过使用不同的选择策略将整个种群划分为3个子种群并分别进化，能提高种群的多样性，有效避免遗传算法的早熟收敛问题。将种群的多样性和算法的运行机制相结合，根据种群多样性的变化运用强化学习算法动态地优化各子种群间的比例参数，从而将种群多样性保持在合适的范围，一定程度上解决了收敛速度和全局收敛性之间的矛盾。实验结果表明，该算法在收敛精度和搜索效率上都表现出较好的性能。

关键词: 遗传算法, 多策略选择, 强化学习, 种群多样性, 比例参数

Abstract: A new multiple policy selection Genetic Algorithm(GA) based on reinforcement learning is proposed to avoid the premature convergence and low speed of convergence. The whole population is divided into three sub-populations and each of them evolves respectively by using several different selection policies, which improves the diversity of population and avoids the premature convergence effectively. Population diversity is associated with the running mechanism of the algorithm, and the parameters of the sub-populations are optimized dynamically using reinforcement learning according to the variance diversity, which can maintain the population diversity in the appropriate range and it solves the contradiction between convergence speed and global convergence to a certain extent. Experimental results show that the algorithm has a high performance in precision of convergence and search efficiency.

Key words: Genetic Algorithm(GA), multiple policy selection, reinforcement learning, population diversity, proportion parameter

中图分类号: 

• TP301.6