Advanced statistical and machine learning methods for multi-step multivariate time series forecasting in predictive maintenance

Abstract

The accurate prediction of failure events is of central interest to the field of predictive maintenance, where the role of forecasting is of paramount importance. In this paper, we present and compare some advanced statistical and machine learning methods for multi-step multivariate time series forecasting. Regarding statistical methods, we considered VAR, VMA, VARMA and Theta. The machine learning approaches we selected are variants of the Recurrent Neural Network model, namely ERNN, LSTM and GRU. All the considered methods have been evaluated in terms of accuracy, using 5 public datasets. As an extra contribution, we have implemented the multivariate version of the Theta method.