1.Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China;2.Institute of Electrical Engineering, Chinese Academy of Sciences, Beijing 100190, China

The stochastic volatility of renewable energy generation and the time series coupling characteristics of energy storage operation control bring many challenges to the energy management and optimal operation of microgrid, which becomes a hot issue for academic research. This paper proposes an energy management and optimization strategy for microgrid based on improved dueling deep Q network (DQN) algorithm. The strategy adopts a multi-parameter operation exploration mechanism and an optimally designed neural network structure to learn the environmental information such as the power output of distributed renewable energy, the electricity price of energy trading market and the state of electric load, and applies the learned strategy to microgrid energy management and optimization. The simulation results show that the performance of the energy management and optimization strategy for microgrid based on the improved dueling DQN algorithm is better than the scenario-based stochastic programming algorithm, the DQN algorithm and the dueling DQN algorithm.