Research on Energy Management Strategy of Microgrid Based on Improved Deep Q network Algorithm

The random volatility and uncertainty of renewable energy generation pose significant challenges to the energy management and optimal operation of microgrids, making this a hot topic in academic research. This paper proposes a microgrid optimization scheduling algorithm based on improved deep reinforcement learning algorithms (deep Q network (DQN), double DQN (DDQN) and adversarial DQN algorithm ). An improved neural network structure with a priority experience replay mechanism is adopted to learn environmental information, including the power output of distributed renewable energy, electricity prices in the energy trading market, power load, and battery status. The economic performance of microgrid system operation and the reduction of battery degradation are taken as the objective functions, and the learned strategy is used to manage and optimize the energy of microgrids. The simulation results show that the microgrid energy management and optimization strategy based on the improved competitive deep Q network algorithm with priority experience replay has better performance , effectively reduces the frequency of deep discharge of batteries, and prolongs the battery life.