基于进化策略的动态递归神经网络建模与辨识

Evolutionary Strategy Based Dynamic Recursive Neural Network Modeling and Identification

提出一种采用进化策略实现动态递归神经网络结构、权重和自反馈增益同时进化的学习算法 ,以及自适应进化机制。与改进 BP算法相结合 ,各取所长 ,形成集成化动态递归神经网络建模辨识算法。实际应用结果表明 ,所提出算法不仅明显提高了动态递归网络模型辨识算法的收敛速度和精度 ,而且实现了动态递归网络的全自动优化设计。

The learning algorithm of the dynamic recursive neural network based on evolutionary strategy is presented, which realizes the evolution of network construct, weights and self feedback coefficient of the dynamic recursive neural network together. The adaptive evolutionary method is also advanced. The identification algorithm integrating the forward evolutionary algorithm and improved BP algorithm for the dynamic recursive neural network model is formed. The result of the applicatoin shows that the advanced algorithm not only improves the learning speed and model precision, but also realizes the fully automatic optimization design for the dynamic recursive neural network.