采用二次强化学习策略的多目标粒子群优化算法

Multi-objective Particle Swarm Optimization Algorithm With Quadratic Reinforcement Learning Strategy

针对多目标粒子群算法在进化后期易出现早熟收敛、种群多样性丢失的问题,本文提出采用二次强化学习策略的多目标粒子群优化算法SslMOPSO.首先利用无速度多目标粒子群框架,通过向所有个体历史最优学习实现粒子的第一次强化学习;其次将分解策略融入多目标粒子群算法中,使粒子向指定数量邻居的均值学习,实现粒子的第二次强化学习,增强算法跳出局部最优的能力,提高种群的多样性;最后分别在具有两目标和具有三目标的七个基准测试函数上进行仿真实验,结果表明,所提算法获得的非支配解集较对比算法具有较好的分布性,表现出较好的搜索性能.

Aiming at the problem of premature convergence and loss of population diversity in multi-objective particle swarm optimization algorithm in late evolutionary stage,this paper proposes a multi-objective particle swarm optimization algorithm SslMOPSO with quadratic reinforcement learning strategy.Firstly,a speed-free multi-objective particle swarm framework is adopted to achieve the first reinforcement learning of all particles through the best learning history of all individuals.Secondly,the decomposition strategy is integrated into the multi-objective particle swarm In the algorithm,particles are learned from the mean of the specified number of neighbors to realize the second reinforcement learning of the particle,which enhances the ability of the algorithm to jump out of the local optimality and improve the diversity of the population.Finally,the simulation experiments are carried out on seven benchmark test functions with two targets and three targets respectively.The results show that the non-dominated solution set obtained by the proposed algorithm has a better distribution than the comparative algorithm and shows better search performance.