Robust photovoltaic power output forecasting and ramping capacity estimation for smart energy systems

The transition to decarbonized energy systems has been accelerated by the rapid deployment of photovoltaic (PV) power generation. Accurate forecasting technology of PV power output is crucial for maintaining grid stability and effectively optimal operation considering ramping requirements. However, traditional forecasting methods often struggle with unreliable weather predictions and regional discrepancies in forecast accuracy, leading to suboptimal performance of operation results. In this paper, a novel PV power output forecasting model that combines an enhanced temporal convolutional network (TCN) with pre-trained heuristic invariant risk minimization (PHIRM) was proposed to address these challenges. The enhanced TCN captures long-term temporal dependencies in the data, while IRM improves the model's generalization under distributional shifts caused by regional variations in weather forecast errors. Additionally, a method for estimating the ramping capacity of power systems based on the predicted PV power output was also proposed. Case studies on a modified IEEE 33-bus distribution system demonstrated that the proposed method significantly improved forecasting accuracies compared to traditional approaches. The results also indicated that effective ramping capacity estimation results could enhance the integrating capability of PV generation for power grids. • A novel PHIRM-TCN model is proposed for robust PV power forecasting, achieving 30.2% lower MAE compared to conventional TCN methods under weather forecast uncertainties • Pre-trained heuristic invariant risk minimization effectively addresses domain shifts across regions, improving model generalization while significantly reducing training time • Integrated ramping capacity estimation framework achieves 0.14% error for 15-minute intervals, enabling reliable flexibility assessment for smart energy systems with high PV penetration