Grid physics-informed and time-adaptive stacked learning for real-time electricity price forecasting: a short-term to mid-term approach

Driven by the rapid transition towards high-proportion renewable energy and market-oriented reforms, China's real-time electricity markets face unprecedented price volatility. Accurate forecasting of these critical prices remains challenging as existing models often neglect physical grid constraints and robust anomaly handling, exposing market participants to substantial financial risks. To address these gaps, we propose a novel grid physics-informed time-adaptive stacked learning model for short- to mid-term real-time electricity price forecasting. Considering physical constraints, it incorporates a security-constrained unit commitment simulation to emulate the real-time market clearing process, enhancing spike prediction by capturing real-time supply-demand balance through accurate commitment unit capacity forecasting. A dual-stage time-segment robust outlier correction factor with outlier labeling is introduced to accurately distinguish and label true price anomalies, which promotes precise trend extraction by variational mode decomposition and substantially improves both stable and extreme-period forecasts. Furthermore, a time-adaptive stacked learning framework accounting for distinct fluctuation patterns across different time periods is developed, dynamically weighting base models via feedback mechanisms to generate targeted predictions, ensuring robust performance from 24-h to 168-h horizons. Validated by Hubei provincial data, the proposed model achieves mean absolute errors of 37.42 CNY/MWh (5.24 USD/MWh) and 92.08 CNY/MWh (12.89 USD/MWh) for 24-h and 168-h predictions, respectively, outperforming all state-of-the-art benchmarks by at least 34.25% and 9.24%. For extreme price forecasting, the model achieves root mean square error of 43.09 CNY/MWh (6.03 USD/MWh) for spikes and 53.21 CNY/MWh (7.45 USD/MWh) for troughs, reducing errors by up to 71.91% and 57.45%. Overall, the proposed model provides a robust and accurate solution for multi-horizon real-time price forecasting, offering valuable insights for highly volatile electricity markets and practical guidance for market participants in risk management and bidding strategy optimization.