Revisiting Deep AC-OPF

2025-08-30

AI Summary

Recent learning-based methods for AC optimal power flow (AC-OPF) claim substantial speedups and high accuracy, yet their practical benefits remain inadequately validated against strong, interpretable baselines. Method: We conduct a systematic empirical evaluation of machine learning approaches for AC-OPF, introducing OPFormer-V—a lightweight, Transformer-based model predicting voltage magnitudes and angles—and rigorously compare it against high-performance linear regression and other strong baselines. Contribution/Results: While OPFormer-V outperforms DeepOPF-V, its relative gains are marginal; crucially, simple linear models achieve comparable accuracy across most test cases while accelerating inference by an order of magnitude. Our analysis reveals that many existing learning-based AC-OPF methods lack sufficiently rigorous baseline comparisons, casting doubt on the necessity of complex deep architectures. We advocate for adopting interpretable, easily deployable strong baselines in evaluation protocols—providing both methodological reflection and concrete guidelines for future research in learning-based power system optimization.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Optimization
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Recent work has proposed machine learning (ML) approaches as fast surrogates for solving AC optimal power flow (AC-OPF), with claims of significant speed-ups and high accuracy. In this paper, we revisit these claims through a systematic evaluation of ML models against a set of simple yet carefully designed linear baselines. We introduce OPFormer-V, a transformer-based model for predicting bus voltages, and compare it to both the state-of-the-art DeepOPF-V model and simple linear methods. Our findings reveal that, while OPFormer-V improves over DeepOPF-V, the relative gains of the ML approaches considered are less pronounced than expected. Simple linear baselines can achieve comparable performance. These results highlight the importance of including strong linear baselines in future evaluations.

Problem

• Evaluating ML models for AC-OPF voltage prediction accuracy
• Comparing transformer-based models against simple linear baselines
• Assessing performance claims of deep learning approaches

Innovation

• Transformer-based model for bus voltages
• Comparison with linear baseline methods
• Improved accuracy over DeepOPF-V model