A Causation‐Guided Data‐Driven Model for Electricity Consumption Forecasting of Industrial Chains

Industrial electricity consumption is a crucial indicator for evaluating the development situation of industries. Thus, the accurate forecasting of this indicator can support the analysis of industrial trends. With the systematisation of industries, the power usage of upstream, midstream, and downstream links in the industrial chain demonstrates intricate coupling effects. However, current research on electricity consumption forecasting primarily focuses on external influencing factors, while the internal interaction within the chain is not considered. Therefore, a causationguided data‐driven model for power use forecasting of the industrial chain is proposed innovatively in this paper. First, the industrial chain structure is adopted as the analytical basis. Convergent cross mapping (CCM) and partial cross mapping (PCM) are introduced to analyse causal relationships among different sectors in the industrial chain. Besides, grey correlation analysis (GCA) is used to quantify the correlation between external influencing factors and industries. On this basis, a data‐driven model integrating the temporal convolutional network (TCN) and the long short‐term memory (LSTM) is proposed to forecast power consumption in the chain. Its inputs are designed to be weighted by the quantified results of the causal and correlation analysis, covering the internal power usage and external factors. Furthermore, the time window of the input is adjusted according to the causal time lag. The proposed approach can enhance the accuracy of industrial electricity consumption forecasting by leveraging causation. Finally, in the case study, the proposed approach is validated in the electric vehicle industrial chain.