Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Predicting AC Optimal Power Flows: Combining Deep Learning and Lagrangian Dual Methods
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use OPF-DNN instead of the DC approximation | AC-OPF prediction on NESTA networks · NESTA power networks, total average | load-flow solution distance to AC-OPF (%): 4.5733 -> 0.2124 | Helped |
| Add OPF constraint penalties to the baseline DNN (ablation) | OPF-DNN · NESTA power networks, total average | load-flow solution distance to AC-OPF (%): 2.3706 -> 1.3424 | Helped |
| Add a hot-start state to the constraint-aware model (ablation) | OPF-DNN · NESTA power networks, total average | load-flow solution distance to AC-OPF (%): 1.3424 -> 0.3793 | Helped |
| Use Lagrangian dual updates to learn constraint multipliers (ablation) | hot-start, constraint-aware OPF-DNN · NESTA power networks, total average | load-flow solution distance to AC-OPF (%): 0.3793 -> 0.2124 | Helped |
| Learn constraint weights as trainable parameters rather than with dual updates (ablation) | hot-start, constraint-aware OPF-DNN · NESTA power networks, total average | active generation distance to AC-OPF (%): 0.0198 -> 0.0721 | Hurt |
| Use OPF-DNN instead of the DC approximation | AC-OPF prediction on NESTA networks · NESTA power networks | average speedup vs. AC-OPF: 15.2x -> >!10^4$x | Helped |
| Train OPF-DNN with hot-start datasets differing by 1%, 2%, or 3% (ablation) | OPF-DNN · NESTA power networks | load-flow solution cost distance to AC-OPF (%): 0.2124; 0.1608; 0.1358 | No clear effect |
Closest papers by meaning
- PF∆: A Benchmark Dataset for Power Flow under Load, Generation, and Topology Variations (NeurIPS 2025 · similarity 0.86)
- Inverse Optimization via Learning Feasible Regions (ICML 2025 · similarity 0.82)
- PowerGraph: A power grid benchmark dataset for graph neural networks (NeurIPS 2024 · similarity 0.82)
- PowerPM: Foundation Model for Power Systems (NeurIPS 2024 · similarity 0.81)
- A Deep Generative Learning Approach for Two-stage Adaptive Robust Optimization (ICLR 2025 · similarity 0.81)
- Predicting Lagrangian Multipliers for Mixed Integer Linear Programs (ICML 2024 · similarity 0.79)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.