A novel carbon price prediction model combines the secondary decomposition algorithm and the long short-term memory network

Abstract

Carbon trading is regarded as an important measure to reduce carbon emissions. To provide more accurate carbon prediction results for policymakers and market participants, a hybrid carbon price prediction model combines empirical mode decomposition, variational mode decomposition, and long short-term memory network is proposed. The empirical analysis was conducted based on the actual data of all eight carbon market pilots in China. According to the results of empirical analysis, several main conclusions can be summarized. First, the prediction accuracy and robustness of the proposed model are optimal in comparison experiments. In the Beijing carbon market, the MAPE, RMSE, and R2 of the proposed model improved by 63.98%, 66.07%, and 12.24%, respectively, compared with the worst model. Second, the secondary decomposition can effectively improve the prediction accuracy. In the Beijing dataset, the combination of empirical mode decomposition and variational mode decomposition improved the MAPE, RMSE, and R2 values of the model by an average of 35.52%, 46.57%, and 8.94%. Third, the carbon market in Hubei province is relatively mature, while the carbon market in Tianjin is relatively low in maturity. The study can make a theoretical and practical contribution to the literature within this realm.