A Weighted Predict-and-Optimize Framework for Power System Operation Considering Varying Impacts of Uncertainty (2503.11001v1)

Abstract: Integrating prediction and optimization enhances decision-making quality by yielding near optimal solutions. Given that prediction errors associated with multiple uncertainties have varying impacts on downstream decision-making, improving the prediction accuracy of critical uncertainties with significant impacts on decision-making quality yields better optimization results. Inspired by this observation, this paper proposes a novel weighted predict-and-optimize (WPO) framework for decision-making under uncertainty. Specifically, we introduce an uncertainty-aware weighting mechanism into the predictive model to capture the relative importance of each uncertainty for specific optimization tasks, and introduce a problem-driven prediction loss (PDPL) to quantify the suboptimality of weighted predictions for downstream optimization as compared to perfect predictions. By optimizing the uncertainty weights to minimize the PDPL, WPO framework enables adaptive uncertainty impact assessment and integrated learning of prediction and optimization. Furthermore, to facilitate weight optimization, we construct a surrogate model to establish the mapping between weights and PDPL, where multi-task learning and enhanced graph convolutional networks are adopted for efficient surrogate model construction and training. Numerical experiments on modified IEEE 33-bus and 123-bus systems demonstrate that the proposed WPO framework outperforms traditional methods by achieving a much smaller PDPL within acceptable computational time.