Optimization Model for Reactive Voltage Control in Active Distribution Networks Based on Transformer Based Deep Reinforcement Learning

Aiming at the challenges of insufficient data feature extraction, limited model generalization ability, and low computational efficiency in traditional deep learning methods for reactive voltage control (VRC) in active distribution networks (ADN), a Transformer based Deep Reinforcement Learning (DRL) VRC for AND model is proposed. Firstly, the necessity of VRC in active distribution networks based on ADN coordinated control and its optimized operation mode was analyzed. Then, by introducing a transformer into the traditional DRL model, a VRC optimization model based on transformer was designed to effectively improve ADN performance. Finally, a comparative analysis was conducted between the proposed method and three other methods under the same database conditions through numerical examples. The experimental results show that under the same experimental conditions, the proposed method has the smallest root mean square error and the highest correlation coefficient, with the highest degree of data fitting. The minimum value is 0.0289, and the maximum correlation coefficient is 0.996, which meets the requirements of VRC in ADN. Its performance is superior to other comparative methods.