Reinforcement learning based dual population evolutionary algorit 2027 Appli

The increasing penetration of wind and photovoltaic generation introduces significant uncertainty and security challenges to the operation and scheduling of modern power systems. Many existing optimal power flow (OPF) studies typically treat renewable uncertainty, load variability, and contingencies separately, so the influence of these factors on the feasible region and on the trade-offs among economy, network efficiency, and voltage security remains poorly characterized. Moreover, most adopt a single constraint-handling strategy, which limits their search capability in the resulting strongly constrained problem. To close these gaps, this paper makes two contributions. First, a renewable-integrated multi-objective optimal power flow (MOOPF) framework is developed to model probabilistic wind and PV generation and to evaluate three groups of cases: baseline MOOPF cases, representative-load MOOPF cases, and selected-contingency security-constrained cases. The resulting formulations are nonconvex constrained multi-objective optimization problems. Second, a reinforcement-learning-based dual-population evolutionary algorithm (RLDPEA) is proposed to solve this problem. Two co-evolving populations adopt the ε-constraint method and a constraint-ignoring strategy, enabling effective navigation of both feasible and infeasible regions, while embedded Q-learning adaptively selects parents and operators from archive feedback to balance feasibility, convergence, and diversity.