含速度变异算子的粒子群算法

提出了一种新型的PSO算法——含速度变异算子的粒子群算法（PSOVMO）．该算法在进行变异时的变异对象是搜索速度（ν），而不是通常情况下的位置（Х）．其方法是，设置一个随迭代的进行按指数级数减小的临界速度．在变异开始到整个搜索循环结束之间的每一次迭代中，只要第i个粒子在d维上的搜索速度的绝对值│νi,d│大于此时的临界速度，就以一定的概率重新初始化νi,d：让νi、d 随机分布在区间[-Vmax，Vmax]上，从而通过位置迭代公式将原本聚集的粒子均匀地“驱赶”到前一位置的周围，达到变异的目的．通过对4个多峰测试函数所做的对比实验，表明PSOVMO优于原始的PSO，也优于按传统方法变异的PSO．

A PSO algorithm with velocithy mutation operator

A new Particle Swarm Optimization (PSO) algorithm PSO with velocity mutation operator (PSOVMO), was presented. In the algorithm, the searching speed (v) rather than the position (x) was used as the mutation object. A critical velocity was set in every iteration and the value of the critical velocity decreased according to exponential series in every iteration. In every iteration starting from the mutation to the end of the whole searching loop, as long as |v_(i,d)|, the ith particles absolute value of searching velocity on the dimension d, is greater than the critical velocity, the value of v_(i,d) was reset according to some probability, making v_(i,d) distributed stochastically in the range of [-v_(max), v_(max)], through the position updated formula (x_(i,d)=x_(i,d) v_(i,d)), originally converged particles were driven to around the former position, so the mutation was gotten. The contrastive experiments of four multi-peak testing functions Ackley, Schaffer f6, Griewank and Rastrigin, indicated that PSOVMO is greatly superior to original PSO as well as PSO with mutation via common methods.