提出了一种进化策略求解Hopfield神经网络的方法.该进化策略分三个阶段,即第一阶段只在较小区间上求出局部优化解;然后,在此基础上,由第二阶段求出较大区间上的局部优化解;最后由第三阶段求出全局优化解.同时采用Hopfield神经网络动态方程指导第一阶段的局部进化策略的进化方向,因而大大加快了优化搜索速度.在分阶段的进化策略中,其第一阶段只需搜索较小区间,第二和第三阶段的搜索则建立在其前一阶段的基础上,因此可以采用较小的遗传群体规模,从而明显地提高了求解Hopfield神经网络的速度,减少了计算内存的需求.

The method of applying evolutionary strategies to converge Hopfield neural network is proposed in this paper. Genetic search of evolutionary strategies(ES) consists of three consecutive processes. ES performs local search in some small domain in the first process,and the search domain grows in the second process. Then ES performs the global search in whole domain in the third process. We use the dynamic equation to indicate the ES search direction to speed up the local search speed obviously. The ES search domain is small in the first process, and the second and the third process are based on the optimization results of their previous processes,therefore the convergence speed of the proposed method is much faster and it needs less memory space.