A Deep Reinforcement Learning Framework for Optimal Distribution Network Structure

Distribution network reconfiguration is one solution for optimal utilization of distribution networks under different conditions. With the advent of distributed generation and electric vehicles, distribution networks are facing various challenges, each of which requires the detection of an optimal distribution network structure. In this paper, one of the most popular deep reinforcement learning algorithms, the Double Deep Q Network, has been employed to reconfigure IEEE ۳۳- and ۶۹-bus distribution networks. This modeling has been investigated in two analyses: normal and daily. An innovative method is utilized in this paper to detect radial constraints and non-isolation among buses. The effects of uncertainty in distributed generation, such as photovoltaic panels, and variable loading, such as commercial, industrial, and residential load profiles, have been evaluated in the daily analysis of this paper. This modeling aims to minimize power losses and voltage deviation. This modeling demonstrates that the suggested approach surpasses previous approaches in reducing power losses and voltage deviation.

کلیدواژه ها:

distributed generation ، Distribution Network Reconfiguration ، Deep reinforcement learning ، Photovoltaic Panels