Study on Ultra-Short Term Power Load Forecasting Based on Local Similar Days and Long Short-Term Memory Networks

Abstract

Power load forecasting is the basis of planning and economic operation of power system. Accurate load forecasting is helpful to improve the safety and economic benefits of power network operation. Because the load data have the characteristics of seasonality, periodicity, non-linearity and time series, this paper proposed a method for ultra-short term load forecasting of power system based on local similar day and long short-term memory (LSTM) networks. Firstly, the local similarity model was used to select the historical days with similar load variation characteristics as training samples, and then LSTM method, which is suitable for dealing with the time series problems, was used to predict the ultra-short term power load. Finally, based on the actual data of a power plant in Liaoning Province, the power load of four typical days in spring, summer, autumn and winter of 2018 was forecasted by generalized regression neural network (GRNN), Elman neural network, LSTM and the method proposed in this paper. The results of these four methods were compared, and the results show that the proposed method has higher prediction accuracy.