A novel carbon price forecasting method based on model matching, adaptive decomposition, and reinforcement learning ensemble strategy

Carbon price is closely related to energy conservation and emission reduction cost, and carbon price forecasting is conducive to improving and stabilizing market trading mechanisms. Considering the matching relationship between the prediction model and the carbon price market, a decomposition ensemble carbon price forecasting method based on reinforcement learning model fusion is designed. Firstly, considering the problem of disharmony between different base models and data patterns, a model matching strategy is developed to match the appropriate prediction base model for each carbon trading market. Secondly, the empirical wavelet transform (EWT) algorithm is used to decompose the carbon price into several subseries, which reduces the complexity of the original data and improves the prediction effect of each base model. Finally, for each carbon price market, the Q-learning algorithm of reinforcement learning is used to integrate the prediction results of the selected multiple base models into the final price forecasting of the corresponding market. The model is verified in the seven carbon trading markets, and experiments show that the model has superior and stable prediction performance compared with other methods. The mean absolute percentage errors of the Hubei, Beijing, Shenzhen, Chongqing, Tianjin, Shanghai, and EU markets are only 0.3515%, 1.2335%, 1.3388%, 1.2128%, 0.3229%, 0.2418%, and 0.4094%, respectively. It shows that the EWT method and reinforcement learning ensemble strategy do improve the prediction performance, and the proposed model can be used as a feasible tool for price assessment and management in carbon price markets.