GENCO - A Unified Neural Solver Embedded in a Development Framework for Steady-State Grid Analysis
This work addresses the computational inefficiency of traditional power system analysis methods and the lack of a unified, physically consistent architecture in existing neural solvers. It proposes GENCO, a unified neural solver that, for the first time, enables end-to-end solution of power flow (PF), optimal power flow (OPF), and state estimation (SE) within a single geometric deep learning framework. By integrating AC power grid physics and differentiable optimization, GENCO guarantees physical consistency of its solutions. The accompanying open-source GridFM framework facilitates low-code data generation and training, along with the release of a large-scale synthetic dataset. Experiments demonstrate that GENCO solves PF 30× faster than the Newton-Raphson method with higher accuracy than DC-PF, achieves OPF solutions 85× faster than IPOPT while yielding superior objective values, and maintains high robustness and convergence in SE under measurement noise and parameter errors.