一种用于BP网络优化的并行模拟退火遗传算法

Parallel simulated annealing genetic algorithm for optimizing BP neural network

针对模拟退火算法和遗传算法存在的不足,提出了并行模拟退火遗传算法,并用于3层BP神经网络优化。在适应度函数中引入模拟退火机制,采用排序、最优保存策略选择算子、启发式交叉和多点非均匀变异改进遗传算子,利用模拟退火算法产生新解增加搜索方向,并结合并行进化思想对经典遗传算法进行改进。通过对英文字母识别的仿真实验,表明该方法全局搜索能力、局部搜索能力和收敛速度都优于经典遗传算法。

A Parallel Simulated Annealing Genetic Algorithm （PSAGA） was given for the optimization of 3 levels BP neural network. Simulated annealing （SA） method was applied in fitness sealing, genetic operator was improved by ranking selection which copied the fittest, heuristic crossover and multi nonuniform mutation, and SA was used as the state generator. The idea of parallel evolution was combined into PSAGM. Simulation to recognition of English letters proved PSAGM was better than simple ~enetic algorithm in global search, local search and speed of convergence.