多策略并行学习的异构粒子群优化算法

Heterogenous particle swarm optimization algorithm with multi-strategy parallel learning

针对标准粒子群优化(PSO)算法在复杂问题上收敛速度慢和早熟收敛的缺点,提出了一种多策略并行学习的异构PSO算法(MHPSO)。该算法首先从种群多样性和跳出局部极值的角度提出了两种新学习策略(局部扰动学习策略和高斯子空间学习策略),并将这两种策略与MBB-PSO策略融合组成高效稳定的策略池。其次提出了一种简单有效的策略更换机制,指导粒子迭代寻优中何时更换学习策略。基准测试函数的实验结果表明,改进的粒子群优化算法在求解精度和收敛速度上得到极大的提高。与一些改进PSO算法(如自适应的粒子群优化(APSO)算法等)相比,所提算法具有更优良的寻优性能。

The standard Particle Swarm Optimization （PSO） suffers from the premature convergence problem and the slow convergence speed problem when solving complex optimal problems, so a Heterogenous PSO with Multi-strategy parallel learning （MHPSO） was presented. Firstly two new learning strategies, named local disturbance learning strategy and Gaussian subspace learning strategy respectively, were proposed to maintain the population＇s diversity and jump out from the local optima. And an efficient and stable strategy pool was constructed by combing the above two strategies with the existed one （MBB-PSO） ; Secondly, a simpler and more effective strategy change mechanism was proposed, which could guide particles when to change the learning strategy. The experimental study on a set of classical test functions show that the proposed approach improves the solution accuracy and convergence speed greatly, and has a superior performance in comparison with several other improved PSO algorithms, such as APSO （Adaptive Particle Swarm Optimization）.