Optimal dispatch of grid-connected microgrids considering the uncertainty of renewable energy generation based on deep reinforcement learning

In recent years, the rapid development of renewable energy power generation, mainly including wind power and photovoltaic power, contributes significantly to solving the energy crisis and environmental pollution. However, renewable energy generation has the characteristic of random fluctuations, which influences the stability of microgrid system. To reduce the negative impacts, optimal dispatch of grid-connected microgrids considering the uncertainty of renewable energy generation based on deep reinforcement learning is developed in this study. The proposed optimal dispatch framework consists of an upper-level and a lower-level optimization problem. The energy optimal set points of different distributed generators in the upper-level dispatch method are optimized with the minimal operational cost of the grid-connected microgrids, and a demand response strategy based on dynamic electricity pricing is employed to minimize the electricity cost of the consumers in the lower-level dispatch, and the chance constraint is transformed into a mixed-integer linear programming to simplify the solution of the optimization dispatch method. To obtain the energy optimal set points of different distributed generators in the upper-level system, an improved deep reinforcement learning algorithm is designed for the load demand and renewable energy generation. Finally, the experimental results illustrate that the proposed method has more accurate prediction results for load power and renewable energy output when comparing with traditional dispatch models based on probability density functions, the economic benefits of the grid-connected microgrid system and users have been also improved.