A hybrid forecasting model of carbon emissions with optimized VMD and error correction

Abstract

Global carbon emissions have increased dramatically in recent years, leading to a spate of extreme weather events worldwide. So, it is very important to establish a scientific carbon emissions forecasting model for environmental protection and human health. Aiming at the nonlinear and nonstationary of carbon emissions, a forecasting model with optimized variational mode decomposition (VMD) and error correction, including neural network estimation time entropy (NNetEn), VMD optimized by dingo optimization algorithm (DVMD), back propagation optimized by the hunter-prey optimizer (HPOBP), complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), error correction (EC) and least squares support vector machine optimized by African vultures optimization algorithm (AVOALSSVM), named DVMD-NNetEn-CEEMDAN-HPOBP-AVOALSSVM-EC, is proposed. To simplify the description, the name of the proposed model will be abbreviated as DNCHAE. Firstly, original data is decomposed into intrinsic mode functions (IMFs) by DVMD, and then IMFs are divided into low-complexity components and high-complexity components by NNetEn. Secondly, the sum of high-complexity components is decomposed by CEEMDAN, and HPOBP is used to forecast the decomposition results. Then, AVOALSSVM is used to forecast the low-complexity components, and all the forecasting results are combined to obtain the preliminary forecasting results. Finally, error correction is done to reduce the error, and final forecasting results are obtained. Carbon emissions from China and Russia are forecasted to validate the model's forecasting effect. Experiments show that DNCHAE has a good forecasting effect on carbon emissions. The MAE, RMSE and MAPE of China are 0.1624, 0.2048 and 0.0055 respectively, and the forecasting effect is far better than that of other forecasting models.