A Transformer-Based Multi-Entity Load Forecasting Method for Integrated Energy Systems

Energy load forecasting is a critical component of energy system scheduling and optimization. This method, which classified as time-series uses prior features inputs to forecast future loads. Unlike traditional single-target scenario, an integrated has hierarchy many correlated consumption entities prediction targets. Existing data-driven approaches typically interpret entity indexes suggestive features, fail adequately represent interrelationships among entities. paper, therefore, proposes neural network model named Cross-entity Temporal Fusion Transformer (CETFT) that leverages cross-entity attention mechanism inter-entity correlations. The enhanced module capable mapping the relationships within time window informing decoder about in encoder concentrate on. In order reduce computational complexity, shared variable selection networks are adapted extract from different A data set obtained 13 buildings on university campus used case study verify performance proposed approach. Compared comparative methods, achieves smallest error most horizons buildings. Furthermore, importance, temporal correlations, building relationships, patterns analyzed with networks, therefore rich interpretability verified.