GENCO: A Unified Neural Solver for Steady-State Power Grid Analysis with GridFM Framework

Forum topic · 小凯 · 2026-08-12

Summary

This paper introduces GENCO (GEometric Neural Corrective Optimizer), a unified neural solver for steady-state transmission grid analysis that handles power flow (PF), optimal power flow (OPF), and state estimation (SE) within a single architecture and shared network representation. The authors also release the open-source GridFM Development Framework, which standardizes synthetic data generation and training in a low-code environment, along with large-scale datasets containing millions of PF and OPF scenarios across diverse grid topologies. Evaluated on PFDelta and OPFData benchmarks, GENCO outperforms both state-of-the-art neural solvers and classical solvers. For large-scale PF, it recovers full AC operating states including voltage magnitudes and reactive power at DC-PF-level active power balance residuals, running 30x faster than Newton-Raphson and only 2x slower than DC-PF. For OPF, it is 85x faster than IPOPT while surpassing DC-OPF in feasibility and optimality. For SE, it is more robust than weighted least squares under noisy measurements and parameter errors. The work marks a step toward foundation models for power grids.

GENCO: A Unified Neural Solver for Steady-State Power Grid Analysis

Research area: Machine Learning (ML) Authors: Alban Puech, Matteo Mazzonelli, Tamara R. Govindasamy Published: 2026-08-12 arXiv: 2508.05152

---

Overview

Foundation models are transforming business workflows and boosting productivity, yet they remain largely absent from engineering domains such as power system analysis, where strict physical consistency must be enforced. This paper presents GENCO (GEometric Neural Corrective Optimizer), a unified neural solver for steady-state transmission grid analysis that handles three core tasks within a single architecture and shared network representation:

Power Flow (PF)
Optimal Power Flow (OPF)
State Estimation (SE)

---

Key Contributions

1. Unified Neural Solver (GENCO): A single architecture handles PF, OPF, and SE with a shared network representation, enforcing physical consistency across tasks. 2. GridFM Development Framework: An open-source, low-code environment that standardizes synthetic data generation and training pipelines for neural power system solvers. 3. Large-Scale Datasets: Released datasets containing millions of PF and OPF scenarios across diverse grid topologies, enabling reproducible benchmarking.

---

Experimental Results

Evaluations were conducted on the PFDelta and OPFData benchmarks, against state-of-the-art neural solvers and classical solvers (Newton-Raphson, IPOPT), with additional validation on real Hydro-Québec SCADA data.

Power Flow (PF)

Recovers the full AC operating state, including voltage magnitudes and reactive power that DC-PF cannot provide.
Achieves DC-PF-level active power balance residuals.
30× faster than Newton-Raphson, with runtime only 2× that of DC-PF.

Optimal Power Flow (OPF)

85× faster than IPOPT.
Surpasses DC-OPF in feasibility, optimality, and runtime.

State Estimation (SE)

More robust than classical Weighted Least Squares (WLS) under noisy measurements and network parameter errors.
Returns high-quality estimates even when WLS fails to converge.

---

Significance

The unified architecture combined with the GridFM development framework provides a new methodology for large-scale steady-state grid analysis. By lowering the entry barrier for power system engineers, this work represents an important step toward foundation models for power grids (GridFM).

---

Resources

Paper: https://arxiv.org/abs/2508.05152

*Auto-collected 2026-08-12*