The increasing penetration of renewable energy sources and deregulated electricity markets poses new challenges for short-term power system management, particularly in ensuring both economic efficiency and network security. Traditional Security-Constrained Unit Commitment (SCUC) formulations rely on DC power flow approximations that focus on active power balance, often neglecting reactive power feasibility and voltage constraints. This oversight can yield suboptimal or practically infeasible schedules, especially in stressed networks with high renewable variability.

To address these limitations, this paper presents a Benders' decomposition framework for SCUC that explicitly incorporates reactive power constraints and energy losses. The master problem, a Mixed-Integer Linear Programming (MILP), handles unit commitment decisions, active power dispatch, energy reserves, demand bids, and N-1 transmission security via linearized PTDF(Power Transfer Distribution Factor)/LODF(Line Outage Distribution Factor) approximations with iterative constraint generation. It enforces time-coupled generator constraints (ramp rates, min up/down times) and replaces the active power balance with slack inequalities to account for losses.

Temporally decoupled AC slave subproblems (Non-Linear Programming, NLPs), which can be solved in parallel, receive fixed active generation/demand and commitment from the master. Each slave optimizes nodal voltages and reactive dispatch to minimize active/reactive imbalances via slack variables, subject to full rectangular AC power flow equations, voltage bounds, and slack bus conditions. Dual multipliers from these balances generate separate optimality cuts for active (capturing losses) and reactive power, weighted by a tuning factor Alpha and iteratively appended to the master objective following the classic Benders' scheme.

Benders' decomposition generates timestep-specific feasibility cuts from simplified AC slaves that enforce active/reactive balance under local reactive control, bypassing network constraints. This formulation avoids intractable full AC-SCUC MINLPs on large grids, with slaves remaining feasible by construction (no feasibility cuts are needed). Heuristic convergence is obtained despite AC non‑convexity (since classical Benders’ decomposition provides no convergence guarantees for non‑convex problems) and post-processing includes base-case AC checks for voltage/branch violations and contingency flow validation under reactive bindings. Implemented in Pyomo (Gurobi/IPOPT), tests on IEEE 9/118-bus systems and ACTIVSg500 (24h, 1000 cases) demonstrate 10-45x speedups vs. holistic AC-SCUC MINLPs (unsolvable beyond 9-bus within 1h). Results show near-zero optimality gaps, rapid reactive balance (<4 iterations), stable commitment/dispatch after 5 iterations (1-5% variation). Future work examines initialization sensitivity, multi-cut strengthening, and nonlinear transmission constraints on larger networks.