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

TL;DR

The paper addresses how diverse uncertainties impact power-system optimization differently and proposes Weighted Predict-and-Optimize (WPO), which embeds uncertainty-aware weights into the predictive loss and introduces a problem-driven prediction loss (PDPL) to measure downstream decision suboptimality. A differentiable surrogate model based on an enhanced graph convolutional network (GCN) maps weights to PDPL, enabling gradient-based weight optimization, while Multi-Task Learning (MTL) accelerates data generation for training. Empirical results on modified IEEE 33-bus and 123-bus distribution networks show that WPO consistently achieves the lowest PDPL, reducing decision loss by about 55% on average, and demonstrates scalability and interpretability across varying risk profiles. The framework offers a general, end-to-end approach to tailor predictions to downstream optimization goals, with potential extensions to probabilistic predictions and stochastic optimization.

Abstract

Prediction deviations of different 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. Motivated by this observation, this paper proposes a novel weighted predict-and-optimize (WPO) framework for decision-making under multiple uncertainties. Specifically, we incorporate an uncertainty-aware weighting mechanism into the predictive model to capture the relative impact of each uncertainty on specific optimization tasks, and introduce a problem-driven prediction loss (PDPL) to quantify the suboptimality of the weighted predictions relative to perfect predictions in downstream optimization. By optimizing the uncertainty weights to minimize the PDPL, the proposed WPO framework enables adaptive assessment of uncertainty impacts and joint learning of prediction and optimization. Furthermore, to facilitate weight optimization, we develop a surrogate model that establishes a direct mapping between the uncertainty weights and the PDPL, where enhanced graph convolutional networks and multi-task learning are adopted for efficient surrogate model construction and training. Numerical experiments on the modified IEEE 33-bus and 123-bus systems demonstrate that the proposed WPO framework outperforms the traditional predict-then-optimize paradigm, reducing the PDPL by an average of 55% within acceptable computational time.

Figures (11)

• Figure 1: Diagrams of (a) traditional predict-then-optimize, (b) integrated predict-and-optimize, and (c) proposed weighted predict-and-optimize.
• Figure 2: Structure of surrogate model constructed by enhanced GCN.
• Figure 3: Model structure: (a) single-task learning and (b) multi-task learning.
• Figure 4: Modified IEEE 33-bus DN with DGs and uncertain loads.
• Figure 5: Node-wise prediction loss under two weight settings.