An Adaptive Load Baseline Prediction Method for Power Users as Virtual Energy Storage Elements

Abstract

The high proportion of renewable energy connected to the power grid leads to insufficient regulation capacity. Physical energy storage system can provide fast regulating capacity, but, it is seldom used on a large scale in power system due to its cost. The virtual energy storage system which aggregates a variety of flexible load resources can also achieve the same effect as physical energy storage. The scheduling of virtual energy storage depends on the accurate prediction of its power baseline. This paper analyzes the multi-dimensional factors that affect the baseline of virtual energy storage elements, including temperature, date attributes and electricity price. Considering the above factors, an adaptive baseline prediction method based on BP neural network, SVR and LSTM neural network algorithm is designed. Numerical analysis shows that the error of the proposed method is only about 20% of that of the method actually used in the demand response of a city in China.