Multi-channel fusion graph neural network for multivariate time series forecasting

Yanzhe Chen, Zongxia Xie

Abstract: Highlights•Graphs are constructed to extract the relations of multivariate time series.•A time-varying channel obtains the instantaneous fluctuations in time series.•The global-stable channel gets the steady feature by an additional loss function.•The effectiveness of the proposed model is verified with traffic and engine datasets.