Long Short-term Memory Network Prediction Model Based on Fuzzy Time Series

Abstract

This paper proposes a long short-term memory network (FTS-LSTM) prediction model based on fuzzy time series to improve the prediction accuracy of time series. First, the fuzzy C-means clustering FCM algorithm is used to classify the time series to form a fuzzy time series and obtain the membership matrix. Second, the LSTM net-work prediction model is constructed, and the FTS-LSTM network prediction model is proposed. The previously obtained membership is used as the full connection. The weight of the layer and its membership as the weight remain unchanged. This FTS-LSTM network prediction model not only considers the non-linearity and non-stationarity of the time series, but also resolves the inherent uncertainty and ambiguity of the data. Simulation results show that the FTS-LSTM network-based prediction model has faster training speed, higher prediction accuracy, and better prediction effect on time series with large ambiguities.