Modelling of a Non-Linear Dynamic System Using Long Short-Term Memory

Long Short-Term Memory (LSTM) is a Recurrent Neural Network (RNN) that overcomes typical neural network constraints. Because of its ability to record temporal dependencies and solve nonlinear equations, long-term data can be simply managed. LSTM is intended to alleviate the problems of RNN's vanishing gradient and exploding gradient difficulties. In this paper, we used LSTM to solve nonlinear plant equations with a gradient descent-based back propagation approach and retrieved the plant's output as well as other performance measures including Average Mean Square Error (AMSE) and Total Mean Average Error (TMAE). When the output of an LSTM is compared to that of a feed-forward Neural Network (FFNN), the LSTM outshines the FFNN. All the parameters of both models, such as iteration count and learning rate are kept constant.