DQN BASED DYNAMIC DISTRIBUTION NETWORK RECONFIGURATION FOR ENERGY LOSS MINIMIZATION CONSIDERING DGS
Seong-Ll Lim, L. F. Nishimwe H., Seon-Ho Yoon · IET conference proceedings. · 2021
In a distribution network with a high penetration rate of solar photovoltaic (PV), curtailment of solar PV's output is inevitable to keep the network stable resulting in a high loss of renewable energy. Without network reinforcement, dynamic distribution network reconfiguration (DNR) that hourly controls the network topology by controlling sectionalizing and tie switches can reduce solar PV curtailment as changing the power flow in the network. This paper aims to minimize total energy loss by using DNR. Power loss is defined as the amount of curtailed output of PV and line losses from power flow. We formulate this problem as a Markov decision process (MDP) and use a deep Q-network (DQN) to solve it. DQN algorithm is a data-driven approach for MDP, so it does not require topology information in the dynamic DNR problem, i.e., a model-free characteristic. Furthermore, we adopt dropout to reduce overfitting the training data. A case study shows a performance improvement of the proposed DQN based dynamic DNR algorithm in terms of the total energy loss using a 33-bus distribution test feeder.