一种新型的动态递归神经网络及其全自动设计算法

A NEW DYNAMIC RECURSIVE NEURAL NETWORK AND FULLY AUTOMATIC DESIGN ALGORITHM

本文基于非线形自回归滑动平均模型 NARMA模型和前馈神经网络建模的思想 ,提出一种输入层与输出层神经元递归的动态递归神经网络 ;基于进化计算中遗传算法和进化策略与自寻优 BP算法的不同结合方式 ,提出两种动态递归神经网络全自动高效设计算法 ,实现了网络结构、权重和自反馈增益同时优化学习 。

Based on the modeling idea of non-linea l auto-regr essive moving average model and feedforward neural network, the new dynamic recu rsive neural network with the input and output neuron recursion is proposed. Bas ed on the different combined ways to the genetic algorithm, evolutionary strateg y and auto-optimal Back Propagation algorithm, the two fully automatic design a lgorithms for the dynamic recursive neural network are also advanced to realize high learning speed and simultaneous optimization learning of network struct ure, weights, and self feedback parameters. The result of the real application s hows that the new network and design algorithms are effective.