A Distributed Reactive Power Cooperative Control Method Based on Hybrid Optimization Framework: Newton-Particle Swarm Federated Learning Fusion

The widespread access to distributed grids and the high penetration of renewable energy expose the shortcomings of traditional reactive power optimization methods in handling non-convex characteristics, data privacy, and solution space limitations. This paper proposes a Newton-Particle Swarm Federated Learning hybrid algorithm architecture. The proposed framework strictly constrains the power flow equations and equipment operating boundaries through a three-objective optimization model. At the same time, a boundary-node information interaction mechanism is designed to enhance global convergence efficiency. Numerical simulations based on a 58-node system demonstrate that the method reduces line losses by 3.81% and improves voltage deviation optimization by 37.55%. Additionally, the convergence time of the central server exhibits a U-shaped relationship with the aggregation period. Compared with traditional methods, the framework avoids local optima through algorithm fusion, balances data security with global optimization efficiency, and provides an innovative coordination strategy for decentralized control architectures in modern storm-resilient power networks.