Multi-timescale optimal control strategy for energy storage using LSTM prediction–correction in the active distribution network

Abstract

The daily output of wind power is inversely proportional to the load demand in most situations, which will lead to an increase in peak-to-valley difference and fluctuation. To solve this problem, this study proposes a long short-term memory prediction–correction-based multi-timescale optimal control strategy for energy storage. First, the proposed strategy performs a long short-term memory (LSTM) prediction on the power of wind power and load. Then, it establishes a predictive planning model to improve the effect of peak shaving and the operating income of energy storage. Finally, it uses the method of online correction of power lines for peak shaving to further optimize the energy storage power according to the error between the residual energy of energy storage and the planned residual energy in the actual peak shaving process. By comparing with traditional strategies, the proposed strategy is found to be significantly better than the constant power strategy and the power difference strategy in the peak shaving effect and operating income.