Research On Application Of Recurrent Neural Network Based On Improved Activation Function

Artificial neural network(neural network for short)is a parallel algorithm imitating biological neurons,which has been widely used in many different fields such as science and engineering.Recurrent neural network is an important branch of neural network.In recent years,recurrent neural network plays an indispensable role in the online solution of time-varying matrix equations.At present,the research of recurrent neural network mainly focuses on the improvement of its model structure,so as to continuously optimize its convergence and robustness.It is found that the performance of recurrent neural network can be improved by improving its activation function,so improving the activation function of neural network has become one of the topics worth studying.Time-varying problems can be solved by traditional numerical iteration methods,but the traditional numerical iteration methods are usually serial calculation,and the amount of calculation is particularly large.Therefore,using neural networks with parallel computing capability and low computational complexity to solve time-varying problems has become a common method in academia and industry.This paper first introduces the structure of recurrent neural network,convergence theorem,and two kinds of classical model structures(gradient neural network model and zeroing neural network model)derived from recurrent neural network for solving time-varying problems.Then,based on the previous research,an improved activation function is proposed,which can further improve the convergence speed and enhance the robustness of the recurrent neural network.The convergence,stability and robustness of recurrent neural network based on improved activation function are analyzed and proved strictly.In this paper,the recurrent neural network based on the improved activation function is applied to solve time-varying Sylvester equations of different order,time-varying nonlinear equations and trajectory tracking control of various types of mechanical arm,and a series of simulation experiments are carried out by Matlab software.The simulation results show that compared with the neural network based on the traditional activation function,the recurrent neural network based on the improved activation function proposed in this paper has faster convergence speed,higher accuracy and stronger robustness,and effectively enhances the performance of the recurrent neural network in solving time-varying problems..