A regional electricity price prediction method based on Transformer and Graph Neural Networks

Electricity price prediction plays a crucial role in optimizing energy trading and improving market efficiency. However, existing models struggle to simultaneously capture the complex temporal and spatial dependencies of electricity prices in power markets. To address this, we propose TransGraph-Opt, a novel model that integrates Transformer for temporal feature extraction, Graph Neural Networks (GNN) for spatial dependency modeling, and PCGrad optimization to alleviate gradient conflicts in multi-modal data. Experimental results on the PJM Interconnection Market Data and PMU Measurements of IEEE 39-Bus Power System Model datasets demonstrate that TransGraph-Opt outperforms traditional models in terms of MSE (0.3121 vs. 0.3783), MAE (0.2483 vs. 0.2554), and RMSE (0.5592 vs. 0.6132), highlighting its superior predictive accuracy. This work provides a robust framework for integrating heterogeneous data sources, offering promising applications in large-scale electricity market prediction and further advancements in smart grid technologies.