Two-level optimal scheduling strategy of demand response-based microgrids based on renewable energy forecasting

Abstract

Abstract Considering the influences caused by the uncertainty of renewable energy generation (REG) and load on the stable operation of microgrid (MG), a two-level optimal scheduling strategy, including upper-level model and lower-level model, of demand response-based MGs using improved deep reinforcement learning (DRL) is proposed in this study. In the two-level optimal scheduling strategy, energy optimal set points of different distributed generators in the upper-level model are optimized with the objective of the minimal operational cost of the MG, demand response based on dynamic electricity pricing mechanisms is employed to minimize the electricity cost of the consumers in the lower-level model, and the opportunity constraint is transformed into a mixed-integer linear programming to simplify the solution of the optimization scheduling model. To deal with the uncertainty of the renewable energy and load, a freshness priority experience replay deep reinforcement learning (FPER-DRL) is developed to deploy the DRL prediction model for prediction of REG and load power. Finally, the experimental results illustrate that compared with traditional scheduling models based on probability density functions, the proposed method in this paper has more accurate prediction results for load power and renewable energy output, the economic benefits of MG and power users have been also improved.