A dual-stage attention-based Bi-LSTM network for multivariate time series prediction

In the context of the big data era, time series data present the characteristics of high dimensionality and nonlinearity, which bring great challenges to the prediction of multivariate time series (MTS). In view of the problems of insufficient feature extraction of MTS data and insufficient long-term dependency characteristics, we propose a novel model for MTS prediction named a dual-stage attention-based bidirectional long short-term memory (DABi-LSTM). Specifically, the input attention mechanism gives important driving series greater weight, effectively extracting the features of the driving series. Secondly, the bidirectional long short-term memory network (Bi-LSTM) network extracts the MTS features in two directions. Furthermore, the combination of the attention mechanism, Bi-LSTM and long short-term memory (LSTM) network can effectively solve the problem of insufficient long-term dependence in MTS prediction. Experiments based on typical datasets of finance, environment and energy determine the optimal window size and hidden size of the model, and demonstrate that the model gets the state-of-the-art performance for single-step prediction and multistep prediction compared other methods.