一种新的改进粒子群优化算法

摘要

In order to overcome inherent weaknesses of premature and partial convergence on Particle swarm optimization algorithm, this paper presents a regional selection particle swarm optimization algorithm.Aeerording to the areas where each particle exists, while the adaptive value is less than the best fitness value, it is re-initialized on the basis of region, so that the algorithm has strong capacity of global convergence and dynamic self-adaptive.Experimental results show that the new algorithm can not only greatly improve the global convergence ability and enhance the accuracy of convergence, but also avoid the premature convergence effectively, and find the global optimal solution.%研究神经网络优化问题,为了进一步解决粒子群优化算法本身存在的早熟和局部收敛的问题,提高神经网络训练精度,提出了一种区域选择粒子群算法(Regional Selection Particle Swarm Optimization,RSPSO).算法根据每个粒子所在区域不同,在每个粒子所在区域内,当适应值小于最佳适应值时,依据所在区域,重新进行初始化,从而使算法具有更强的全局收敛性和动态的自适应性.通过对几种典型的测试函数进行仿真结果表明改进算法具有更好的收敛精度,改善了优化性能,并且能够更有效避免早熟收敛问题,寻找到全局最优解.