Data-Centric Optimization: An End-to-End Strategy for Power Systems

作者：Yanzhi Wang, Jianxiao Wang, Jie Song · 发表于：Engineering · 年份：2025 · DOI：10.1016/j.eng.2025.09.028 · 研究领域：Smart Grid Energy Management、Electric Power System Optimization、Optimal Power Flow Distribution

In power systems, data-driven optimization has historically focused on model refinement rather than the quality of input data, thus creating a critical performance bottleneck for decision-making. We address this by proposing data-centric optimization (DCOpt), a framework that embeds a dynamic feature extraction mechanism within a decision-aware, end-to-end learning architecture. DCOpt systematically identifies and prioritizes data features that maximize downstream economic efficiency. Validated on a day-ahead electricity dispatch task using 2023 independent system operator (ISO) New England data, our approach improves decision efficiency by 8.9 % over traditional model-centric methods. Our work demonstrates that strategic data curation, not just model complexity, is key to unlocking the full potential of artificial intelligence (AI) in complex system engineering.