基于学习因子自适应改变的粒子群算法研究

The research of PSO based on the adaptive changes of acceleration coefficients

标准粒子群算法的学习因子是固定值,但是研究发现这种取法却并不合适,会影响到算法的性能.本文通过研究得到以下结论:为了保证粒子群算法搜索到更广阔的空间以及粒子的收敛性,不管是调整单个学习因子还是两个同时调整,学习因子c1对应的函数都应该先凹后凸,而c2对应的函数应该先凸后凹;绝大多数情况下两个因子一起调整会比只调整一个要好;两种调整策略同样都是c1对应的函数先凹后凸、而c2对应的函数先凸后凹的情况时,非对称性调整优于对称性调整.

The acceleration coefficients of standard PSO are fixed numbers,but the research showed it is not appropriate because the performance of this algorithm would be destroyed.Three conclusions had been drawn in this paper：To assure the wider search range and the convergence of the particles,whether you changed only one acceleration coefficient or both acceleration coefficients,c1 should begin with concave and end with convex and c2 was conversely;Generally speaking,PSO performed better when changed both acceleration coefficients at the same time than only changed one of them;Non-symmetric adjustment was better than symmetric adjustment when two strategies both were the first circumstance.