基于改进遗传算法的前向神经网络

针对遗传算法的主要算子———交叉算子 ,设计了新的交叉算子 ,使个体尽可能地分散在整个解空间 .在具体交叉操作中 ,产生随机个体参与交叉以更好地搜索新的解空间 .并提出了组合变异策略 ,假如对变异后个体隔代保护策略 ,构造了一个有效的改进遗传算法 .利用该改进遗传算法 ,构造了前向进化神经网络 .它综合了改进遗传算法优良的全局寻优性能和前向神经网络的非线性映射能力 .

Feed-forward NN based on improved genetic algorithm

A new crossover operator is designed, according to its prominent position in genetic algorithm (GA), to make sure that the individuals scatter, in the whole solution space, as even as possible. In practice, the crossover operator is implemented, as to search new solution space better, with randomly-produced individual involved. Strategies of combinatorial mutation and the protect of an improved GA(IGA) were introduced. Examples show that the IGA is effective. On the base of the IGA, a feedforward evolutional neural network algorithm, which melts together the distinguished global search of IGA and the strong nonlinear mapping of NN, is constructed to solve nonlinear mapping problem.