A Distributed Reactive Power Optimization Control Method for Distribution Networks Using Improved MPC and Deep-Learning-Based Prediction

Optimal reactive power regulation is essential for maintaining voltage stability, reducing line losses, and improving power supply quality in distribution networks. To address the issues of traditional static var generators(SVGs)—including slow dynamic response, limited compensation accuracy, and poor scalability of centralized control—this paper proposes a distributed SVG reactive power optimization control method that integrates improved model predictive control(MPC) with Long short-term memory(LSTM)-based prediction. The proposed approach uses LSTM to forecast short-term node voltage trends and embeds these predictions into the MPC optimization framework, enabling proactive control and dynamic feedforward correction. In addition, a voltage sensitivity matrix is introduced to construct a distributed cooperative optimization model, allowing each SVG node to achieve coordinated control under local communication conditions. Simulation results demonstrate that the method effectively reduces voltage fluctuations, improves power factor, and enhances power quality under scenarios with load variations.