Dynamic Equivalent Model of Active Distribution Network Using Hierarchical Double Deep Q Network

The modeling of active distribution networks (ADNs) has become more and more difficult due to the increasing distributed generation from renewable energy sources and electric vehicle (EV) loads. To address the accuracy problem in simulating power flow in ADN, an equivalent ADN model, including equivalent wind turbine, asynchronous motor, EV loads, and daily loads with internal identification of EV loads, is proposed. Regarding parameter identification, a hierarchical double deep Q network (HDDQN) is presented by improving sample selection and behavioral value function evaluation in deep reinforcement learning through the dyadic network, prioritized empirical replay, and Huber loss function strategies. In HDDQN, the first layer establishes an equivalent model, the second layer determines the model weights, and the third layer determines the weights of EVs. The dynamics of the original ADN are tracked in layers with high accuracy. The simulation results show that the equivalent model obtained by the HDDQN is more accurate and applicable than those obtained by the traditional DDQN and DQN.