Paper Overview
Research Area: ML Authors: Muhy Eddin Za'ter, Anna Van Boven, Bri-Mathias Hodge, Kyri Baker Published: 2026-04-23 arXiv: 2604.21891
Abstract
Maintaining instantaneous balance between electricity supply and demand is critical for reliability and grid stability. System operators achieve this by solving the unit commitment (UC) task, a high-dimensional, large-scale mixed-integer linear programming (MILP) problem that is rigorous and heavily constrained by the physical laws of the power grid. As grids integrate variable renewable energy and new technologies such as long-duration energy storage, UC must optimally address multi-day time horizons and may need to be solved at higher frequency. Consequently, traditional MILP solvers increasingly struggle to compute solutions within tightening operational time limits.
To bypass these computational bottlenecks, this paper proposes a new framework that uses a Transformer-based architecture to predict unit commitment schedules over a 72-hour horizon. Since raw predictions in high-dimensional spaces often yield physically infeasible results, the pipeline couples the self-attention network with deterministic post-processing heuristics that systematically enforce minimum up/down times and minimize excess capacity.
Finally, these refined predictions are used as a warm start for a downstream MILP solver, with a confidence-based variable fixing strategy to substantially reduce the combinatorial search space. Validation on a single-bus test system shows that the full multi-stage pipeline achieves 100% feasibility and significantly accelerates computation time. Notably, in approximately 20% of test instances, the proposed model reaches feasible operating plans with lower total system cost than relying on the solver alone.
--- *Auto-collected on 2026-04-27*