Multivariate Time Series Forecasting with Transfer Entropy Graph

Multivariate Time Series (MTS) forecasting is an essential problem in many fields. Accurate results can effectively help making decisions. To date, MTS methods have been proposed and widely applied. However, these assume that the predicted value of a single variable affected by all other variables, ignoring causal relationship among variables. address above issue, we propose novel end-to-end deep learning model, termed graph neural network with Granger causality, namely CauGNN, this paper. characterize information introduce causality our model. Each regarded as node, each edge represents casual between In addition, convolutional filters different perception scales are used for time series feature extraction, to generate node. Finally, adopted tackle structure generated MTS. Three benchmark datasets from real world evaluate comprehensive experiments show method achieves state-of-the-art task.