A deep learning based voltage model predictive control method for distribution networks

This paper proposes a control method combining model predictive control and deep reinforcement learning technology to address the voltage control challenges caused by the volatility and unpredictability of renewable energy. Firstly, this paper proposes a collaborative optimization operation control strategy for distribution networks that considers the optimal long-term control effect. This strategy comprehensively considers the current and future renewable energy output situation, obtaining the lowest total cost of distribution network operation and control over a long period of time. Secondly, this paper utilizes deep learning techniques to consider the potential temporal patterns of renewable energy output, and uses this method to generate real output scenarios that may occur in the future, and jointly constrain the solution of reactive power optimization strategies for distribution networks, thereby improving the accuracy and rationality of voltage control strategies. Finally, this paper applies the above theoretical results to the IEEE-33 node example for verification, and compares the experimental results with the effects of using other scenario generation methods, demonstrating the effectiveness and superiority of the proposed method.