Daily electricity consumption combined forecast based on variational mode decomposition and sample entropy

Abstract

Owing to the daily electricity consumption sequence is time-varying, nonlinear and vulnerable to random factors, a host of single model can't meet the demands of high-precision forecasting. For this problem, this paper proposed a new combined prediction algorithm. Firstly, the original electricity data is decomposed into Intrinsic Mode Function (IMF) with bandwidth by Variational Mode Decomposition (VMD) algorithm and the feature information of itself extracted simultaneously. At the same time, in order to accelerate the speed of prediction and reduce workload, the IMF decomposed by Sample Entropy (SE) is reconstructed into the trend component, the periodic component and the random component according to the self-similarity analysis. Then, according to the characteristics of each component we applied the Least Squares Support Vector Machine based on Maximum Correlation Entropy Criterion (MCC-LSSVM) and Auto-Regressive Moving Average Model (ARMA) to forecast components respectively and add the components to obtain the final predicted value. Finally, the daily electricity consumption data of a Wanda Plaza in Guangzhou are tested and compared with other algorithms. The results show that the daily electricity consumption hybrid forecasting algorithm based on VMD-SE can achieve higher prediction accuracy.