Multi-Energy Load Forecasting for Integrated Energy Systems based on Causality Anlysis and Multi-Task Learning

Accurate load forecasting is paramount for achieving economic dispatch and ensuring the efficient operation of integrated energy systems. This paper delves into the intricate web of relationships between loads and weather factors, unveiling causal links between them that exhibit substantial correlations. Our investigation culminates in the introduction of an innovative multi-energy load forecasting methodology, grounded in causality analysis and Multi-gate Mixture-of-Experts (MMoE). Firstly, weather features strongly correlated with loads are identified using the Pearson correlation coefficient. The inverse information entropy causal inference method is them employed to refine this selection, isolating weather factors that exert a pronounced causal influence on load variations. Secondly, an MMoE model is constrcucted utilizing expert sub-networks and gating units to capture variations in the coupling characteristics among multienergy loads. Performance verification, conducted using publicly available datasets, demonstrates that the proposed model, enriched with causality analysis and MMoE, significantly enhances the accuracy of multi-energy load forecasting.