Multi-Step-Ahead Prediction with Long Short Term Memory Networks and Support Vector Regression

Abstract

Long-term machine condition prognosis is very important for condition-based maintenance. With the rapid development of deep learning, we use long short term memory (LSTM) network with error correction based on support vector regression (SVR) to realize multi-step-ahead time series prediction in this paper. As the parameters of trained LSTM model cannot be updated in the forecasting stage, the SVR model is used to predict the evolution of the residual error in LSTM to update the model parameters. The proposed LSTM-SVR model is validated and compared with LSTM and SVR model using Mackey-Glass time series and TE benchmark data. The experimental results show the proposed method is effective and can achieve a considerable multi-step-ahead forecasting performance.