Yang Deng
College of Electronics and Information Engineering, Shanghai University of Electric Power, Shanghai 200090, ChinaYang Junjie
1. College of Electronics and Information Engineering, Shanghai University of Electric Power, Shanghai 200090, China; 2. Shanghai Dianji University, Shanghai 201306, ChinaHu Chenyang
College of Electronics and Information Engineering, Shanghai University of Electric Power, Shanghai 200090, ChinaCui Dan
College of Electronics and Information Engineering, Shanghai University of Electric Power, Shanghai 200090, ChinaChen Zhaoguang
College of Electronics and Information Engineering, Shanghai University of Electric Power, Shanghai 200090, China1. College of Electronics and Information Engineering, Shanghai University of Electric Power, Shanghai 200090, China; 2. Shanghai Dianji University, Shanghai 201306, China
TM715
Aiming at the problem of low prediction accuracy caused by randomness, fluctuation and nonlinear factors of power load, a short-term load prediction model based on least squares support vector machine (LSSVM) optimized by variational mode decomposition (VMD) and Sparrow search algorithm (SSA) was proposed. In this method, the original load time series was decomposed into the intrinsic mode function (IMF) and residual component (Res) of different frequencies by VMD. Then, different LSSVM prediction models were established for each component and parameters were optimized by SSA. Finally, the final prediction results were obtained by combining the predicted values of each component. Taking two groups of real data from The University of Mons in Belgium and a certain area of Henan Province in China as examples, the prediction results were compared with the predicted values of LSSVM, VMD-LSSVM and SSA-LSSVM models, and the MAPE values of the two groups of data proposed in this paper were 1.5016% and 4.765% respectively, far lower than those of other models. The results show that the combined prediction model in this paper has some advantages in prediction accuracy.
