Data-Driven Intelligent Decision-Making Method for Distribution Network AVC Master Station

As modern power systems evolve, with the expansion of distribution networks and the increasing complexity of their operating environments, traditional Automatic Voltage Control (AVC) methods struggle to cope with dynamic load fluctuations and the integration of distributed energy resources. These challenges hinder timely voltage regulation and reduce overall system reliability. There is an urgent need to introduce data-driven intelligent decision-making approaches. This paper proposes a data-driven intelligent decision-making method for the AVC master station, based on the integration of multi-model load forecasting and reinforcement learning optimization. Load and voltage forecasts are generated using XGBoost and LSTM models, with a weighted ensemble strategy employed to enhance prediction accuracy. A deep Q-network (DQN) combined with a prioritized experience replay mechanism is introduced to build an intelligent decision-making framework, enabling online learning and adaptive regulation, while transfer learning techniques are used to improve the system's generalization capability. In various distribution network simulation scenarios, the proposed intelligent decision-making system significantly outperforms traditional methods in terms of load forecasting error, voltage regulation accuracy, response speed, and energy efficiency. The load forecasting error is reduced to 3.1 %, and the voltage compliance rate is improved to 95.2 %. These results demonstrate the effectiveness of the proposed approach in enhancing operational performance under both steady and dynamic conditions. The data-driven approach markedly enhances the intelligence and operational reliability of the distribution network AVC master station, providing a scalable and adaptive solution for future smart grid control systems with closed-loop “perception-predictiondecision” capabilities.