Abstract: Aiming at the shortcoming of the standard PSO algorithm, that is easily plunging into local minimum, we oppose a new multi-mutation particle swarm optimization algorithm(MMPSO). In each iteration, the position of particles which is satisfied the mutation condition are mutated with many functions,and each function is endowed a probability. The probability distribution is relied on the specific optimization problem.The experimental results show that the MMPSO enhance the global searching ability and the probability of successful searching, and overcome the standard PSO’s liability to convergence to local optimum. It is also superior to single mutation particle swarm optimization algorithm(SMPSO).

Key words: particle swarm, mutation, function optimization

摘要： 针对基本PSO算法存在易陷入局部最优点的缺点，本文提出了一种新型的PSO算法——混合变异粒子群算法。在每次迭代中，符合变异条件的粒子，以多种变异函数方式进行变异，而这些变异函数被赋予了一定概率，概率的划分取决于特定的优化问题。对几种典型函数的测试结果表明：在变异函数概率分配设置合适的情况下，混合变异粒子群算法增强了全局搜索能力，提高了搜索成功率，克服了基本PSO算法易于收敛到局部最优点的缺点，也明显优于单变异粒子群算法。

关键词: 粒子群, 变异, 函数优化