Abstract: To improve the performance of genetic algorithm with machine learning method,a multi-strategy parallel genetic algorithm based on machine learning was proposed.The parallel thought was used to accelerate the evolutionary process of genetic algorithm,and the K-means clustering algorithm was applied to divide the initial population into multiple clusters.Then  the similar individuals were evenly distributed to different subpopulations to ensure the diversity and uniformity of the subpopulations.In the process of evolution,the sub-populations were allowed to communicate with each other,and excellent individuals were used to replace poor individuals in other populations to improve the overall quality of the population.The reinforcement learning that could autonomously perceive the environment was introduced to realize the self-learning of the important parameter's crossover probability in genetic algorithm,so that the crossover probability adapted to the evolution process based on experience.The function experiment verified the superiority and stability of the multi-strategy parallel genetic algorithm based on machine learning.

Key words: machine learning, genetic algorithms, reinforcement learning, K-means clustering algorithm, parallel computing

摘要： 为了改善遗传算法的性能,提出一种基于机器学习的多策略并行遗传算法,使用机器学习方法改善遗传算法性能。首先,利用并行思想加速遗传算法进化过程,使用K均值聚类算法将初始种群划分为多个簇,然后将相似个体均匀分配给不同的子种群,保证子种群的多样性和均匀性;同时,在进化过程中,使子种群间相互通信,使用优秀个体替换其他种群中的较差个体,提升种群整体质量。然后,引入能自主感知环境的强化学习,实现遗传算法中重要参数交叉概率的自学习,使交叉概率根据经验适应进化过程。最后,通过函数实例测试验证了基于机器学习的多策略并行遗传算法的优越性和稳定性。

关键词: 机器学习, 遗传算法, 强化学习, K均值聚类, 并行计算