“…Thus, the research of deep time series forecasting models becomes prevalent. Starting from RNN [3], [4], [5], [24], [57], [68], [69], popular networks which are successful in other research fields are successively applied to time series forecasting, like CNN [1], [2], [18], [22], [52], GNN [9], [10], [70] and Transformer [6], [7], [31], [36], [37], [41], [71]. They are mainly built upon the hypothesis that time series are causal, auto-regressive and stationary.…”