A Long-Short Term Memory Network for Chaotic Time Series Prediction
Margarita Terziyska, Zhelyazko Terziyski, Yancho V. Todorov · 2021
This paper demonstrates the development of an Long-Short Term Memory (LSTM) network and its application for predicting Chaotic Time Series (CTS). LSTM is a Deep Learning (DL) architecture and a type of Recurent Neural Network (RNN). Its ability to memorise past inputs make it extremely suitable for studying data sequences, such as CTS. To test the performance of a developed LSTM, several simulation studies were realized in a Matlab environment. The developed DL architecture was used to predict well-known standard CTSs, namely Mackey-Glass (MG), Rössler, and Lorenz. The obtained results were compared with Adaptive N euro Fuzzy Inference system (ANFIS) and Distributed Adaptive Neuro-Fuzzy Architecture (DANF A), previously developed by the authors. The comparison is made on the basis of Root Mean Squared Error (RMSE). It was found that the proposed LSTM structure is able to generalize the generated series and it has higher accuracy than the other two models.