基于进化规划的神经网络设计方法

Neural Network Design Method Based onEvolutionary Programming

利用进化规划同时进化神经网络的结构和权值．为了避免在网络中形成信号的冲突和竞争，用连接规则决定神经元之间的连接方式．在进化规划中，变异分为结构变异和权值变异两种．提出在静态网络中引入延时权概念来构成时序网络，从而将静态网络的设计方法推广到时序网络的设计．最后给出了仿真结果．

A neural network design method is proposed, in which evolutionary programming is used to evolve both the architecture and weights for a static neural network. In order to avoid the signal conflict and competition, the connection rule is presented to decide the connectivity in the network. The mutation of evolutionary is divided into two parts: structure mutation and weight mutation. By adding delay links to a static neural network a recurrent network can be formed, by which the static network design method can be extended into the recurrent network. The simulation results are give to illustrate the efficiency of the proposed method.