Study on a Mutation Operator in Evolving Neural Networks
进化神经网络中的变异算子研究

In order to solve two difficult problems of premature convergence and slow searching speed of genetic algorithms in evolution neural network, a heuristic mutation operator is presented in this paper. Adaptive probability of mutation and heuristic mutation points selected is applied in it. When no evolution appears after many generations, the range of search will be extended by increasing probability of mutation, and a fine search will be started. The experiments of XOR problem demonstrate that the operator has fine adility of speedy convergence and maintains the diversity of the population automatically.