Parameter Identification of Wiener Model Based on LSTM Neural Network

This paper discusses a modeling and identification scheme of the Wiener model using long short term memory (LSTM) neural network. The Wiener model is composed of a dynamic linear block and a static nonlinear block, in which the linear block is modeled by ARX model, the nonlinear block is modeled by LSTM neural network, and special test signals composed of Gaussian signal and real wind power generation data are employed to realize parameter identification of each block for the Wiener model. Firstly, the input-output of Gaussian signal is used to estimate the dynamic linear block by adopting correlation analysis theory. Then, the input-output of random signal is used to identify the nonlinear block using stochastic optimization method of adaptive momentum. Simulation results show that the scheme proposed can effectively identify the Wiener model based on LSTM neural network and predict the power of wind power generation accurately.