Centralized optimization scheduling of active distribution network-microgrid based on deep reinforcement learning

As an effective renewable energy grid connection method, microgrids (MGs) are increasingly being integrated into active distribution network (ADN). However, with the rapid expansion of MGs scale, model-based energy management methods cannot meet the requirements of real-time scheduling. Therefore, a centralized optimization framework for MGs and ADN is proposed to achieve real-time energy management of ADN and MGs. Firstly, the objectives and constraints are integrated to establish a centralized optimization scheduling model for ADN and MGs. Then, describe the energy scheduling problem as Markov Decision Process (MDP) and transform it into deep reinforcement learning (DRL) tasks for offline learning and online real-time scheduling. Finally, the computational efficiency is verified in the modified IEEE33-bus and the performance is compared with the Deep Q-Network (DQN), Genetic algorithm (GA) and Mixed integer second-order cone programming (MISOCP) approaches.