Study on dynamic recursive neural network structure and learning algorithm

Abstract

In order to solve the present problem of dynamic recursive neural network such as slow learning speed, low model accuracy and bad application result, several new dynamic recursive network are put forward based on its structure. The approach of neural network automatic design based on the integration of self-adaptive evolutionary strategy and improved backpropagation algorithm is also advanced to realize the rapid evolution of network structure, weights and self feedback parameter in the same time. The actual application of system modeling shows that the advanced dynamic recursive network and learning algorithm is feasible and perfect.