神经网络结构及其权值优化的遗传算法

Genetic Algorithm for Optimization of Neural Network Structure and Weight Distribution

神经网络结构及权值优化属两级进化方法。即把结构和权值进化分级进行,并在两级进化过程中采用不同的编码方式和适应度函数,但都使用改进遗传算法—反向变异算子。反向变异算子可按需要搜索的方向搜索,且不会造成早熟收敛。仿真结果表明,该算法取得了预期的效果。

Simultaneity optimization for structure and weight distribution of neural network belongs to two class evolutionary method. First step, neural network structure is evolved; second step, neural network weight distribution is evolved, and use different code mode and fitness function in process of two class evolutionary, and modified genetic algorithm---reverse mutation operation was used. Reverse mutation operation can search according to the expectant direction, and do not bring premature convergence. Emulation result shows expectant optimization efficiency is obtained with genetic algorithm.