Risk-Aware Prediction-Optimization Integrated Day-Ahead Dispatch for Integrated Energy System

Abstract

Accurate source-load forecasting is essential for the day-ahead dispatch of integrated energy system (IES). However, minimizing statistical forecasting errors does not necessarily lead to better operational performance. To address this mismatch, this paper proposes a risk-aware prediction-optimization integrated method for day-ahead dispatch of an IES. The proposed framework considers photovoltaic generation, electrical load, and heating load forecasting as upstream tasks, and the electro-thermal dispatch problem as the downstream optimization task. A dual-based gradient approximation method is developed to capture the marginal impact of forecast errors on the dispatch cost. Furthermore, a spectral risk measure is introduced to emphasize high-loss samples and reduce the tail risk of forecast-driven dispatch. Case studies show that the proposed method reduces the mean decision loss by 56.5% and the CVaR-95 decision loss by 38.8% compared with the forecast-then-optimize baseline. These results demonstrate that decision-oriented source-load forecasting can improve the economic effectiveness of IES dispatch.