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

arXiv: https://arxiv.org/abs/2503.11001 · HTML: https://mlsanity.com/paper/2503.11001/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 14 Mar 2025.

## Built on by (most cited first, 2 of 2)

- [Decision-Focused Scenario Generation and Selection for Efficient and Robust Grid Dispatch](https://mlsanity.com/paper/2607.05830/) (2026)
- [Decision-Focused Continual Learning for Seaport Power-Logistics Scheduling: Generalization across Varying Tasks](https://mlsanity.com/paper/2511.07938/) (2025)

## Abstract (arXiv)

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.