A risk-based scheduling optimization strategy with explainability enhanced multi-scenario photovoltaic forecasting

Abstract

Photovoltaic (PV) power generation, as a crucial technology for addressing energy transition and climate change, is increasingly becoming essential for power systems. However, its inherent randomness and uncertainty challenge stable operation and economic efficiency. To address these, this paper proposes a risk-based scheduling optimization strategy with explainable multi-scenario photovoltaic forecasting to mitigate these uncertainties. First, by employing an enhanced Copula function and ISODATA clustering, multiple joint output scenarios are generated to capture PV uncertainty. Building on this, a Stacking regression model is utilized to improve forecasting accuracy, while Shapley Additive Explanations (SHAP) explainability analysis is incorporated to enhance the transparency and decision-making of the model. Furthermore, to optimize the dispatch strategy for PV generation, this paper introduces the GlueVaR risk measurement method, which combines the benefits of Value at Risk (VaR) and Conditional Value at Risk (CVaR), thereby refining risk management and increasing the reliability of decision-making. Case studies demonstrate that the proposed strategy enhances PV forecasting reliability, with the R² reaching 0.86, and improves model explainability through SHAP-based analysis. In addition, the GlueVaR-based risk scheduling reduces potential risk by approximately 6 %, while maintaining a balance between power system economy and stability.