提出了一种基于遗传算法的前馈神经网络的自动化设计方法 (genetic m ultilayer neural network,GMNN ) ,用以同时完成对网络结构空间和权值空间的搜索。该算法利用模拟退火算法、 BP算法和小生境技术来加快算法的收敛速度 ,改善解的性能。初步实验结果表明 ,该方法的收敛速度较快 ,由此得到的神经网络的泛化能力也较好 ,能够达到根据训练样本自动优化设计多层前馈式神经网络的目的。 A Genetic Algorithm based feedforward neural network (GMNN) is proposed to automatically search the network structure space and weight space simultaneously. The algorithm uses simulated annealing algorithm, BP algorithm and niche technology to speed up the algorithm convergence rate and improve the performance of the solution. The preliminary experimental results show that the convergence speed of this method is fast and the neural network obtained by this method is more generalized and can automatically optimize the design of multi-layer feedforward neural network based on training samples.