An Analytical Approach to Distribution Network Reconfiguration Based on Deep Reinforcement Learning

This paper proposes a deep reinforcement learning (DRL) framework based on the dueling deep Q network (Dueling DQN) algorithm for distribution network reconfiguration (DNR), aiming to minimize power losses and voltage deviations. Since the training process in DRL relies heavily on trial-and-error, accurate environment modeling is essential. Therefore, this study comprehensively investigates various DRL modeling strategies, including different state representations and reward function formulations. Individual DRL agents are trained for each configuration to determine the most effective modeling approach for the DNR problem. The proposed methodology is evaluated on the modified IEEE ۱۶-, ۳۳-, and ۶۹-bus distribution networks. Furthermore, the impact of uncertainties in distributed generation and load demand is analyzed to demonstrate the flexibility and robustness of the proposed method.

کلیدواژه ها:

deep reinforcement learning ، Distribution network reconfiguration ، Various modeling ، Analytical approach