Short-term electricity consumption forecasting method based on empirical mode decomposition of long-short term memory network

Abstract

Accurate electricity consumption forecasting can improve the efficiency of grid dispatching and effectively guarantee the stable operation of the power system. Electricity consumption forecasting is important for the analysis of customer-side electricity consumption behavior, but the instability of electricity consumption sequences poses difficulties for forecasting. Therefore, an improved combination of integrated empirical modal decomposition (EMD) and long short-term memory network (LSTM) is proposed for customer-side electricity consumption forecasting. This paper starts from the idea of blind source separation and independent prediction, and firstly decomposes the original electricity consumption data into several inherent mode functions (IMFs) with different frequencies and amplitudes by empirical modal decomposition (EMD), and then uses LSTM to extract features and make temporal prediction for each IMF component one by one with machine learning intelligent algorithm., and finally obtains end-user-side short-term electricity consumption prediction results by accumulating multiple target prediction results. Compared with direct forecasting, the proposed EMD-LSTM independent forecasting model is able to identify the characteristics of each frequency component of electricity consumption data, and its error is reduced by about 15% on average, thus achieving the goal of improving the accuracy of load forecasting in short-term electricity consumption forecasting scenarios.