Carbon price prediction based on LsOALEO feature selection and time-delay least angle regression

Accurate carbon price prediction can help governments to improve the environment and reduce carbon emissions at a lower cost. To improve the prediction accuracy of carbon price, many scholars have attempted to use multiple factors to predict the carbon price. This method improves carbon price prediction, but also brings problems such as increased complexity and decreased interpretability of prediction models. When analyzing the influencing factors of carbon price, scholars tend to overlook the coupling relationship between feature selection and time delay, resulting in redundant or missing influencing factors. To solve the above problems, we innovatively propose a carbon price prediction model based on LsOALEO feature selection and time-delay least angle regression (LsOALEO-FSTDLARS). The model is mainly composed of two parts: (1) Based on LARS, the prediction model is constructed by integrating feature selection and time delay, and the feasibility of the model is proved theoretically. (2) On the basis of lioness optimization algorithm, the local escape operator is used to improve the global search ability of lioness optimization algorithm, and the algorithm is used for feature selection and time delay optimization. Finally, we conduct experiments on the data sets of the carbon markets in Guangdong, Shanghai, Hubei, Tianjin and China, and compares the proposed method with various other prediction methods. The results of empirical analysis show that compared with the best performing single and combined forecasting methods, LsOALEO-FSTDLARS has an average reduction of 31.05% and 1.30% in RMSE, 46.62% and 5.37% in MAPE, and an average increase of 31.39% and 0.94% in R2 respectively. All of the above indicate that the LsOALEO-FSTDLARS model has higher prediction accuracy, stronger generalization ability and more stable effect, and can be used as an effective tool for carbon price prediction.

Automated summary

Accurate carbon price prediction is crucial for governments to improve the environment and reduce carbon emissions more cost-effectively. Many scholars have tried to predict carbon prices using multiple factors, but this approach has increased complexity and decreased interpretability. To address these issues, we propose a carbon price prediction model based on LsOALEO feature selection and time-delay least angle regression (LsOALEO-FSTDLARS). Experimental results demonstrate that this model outperforms other prediction methods in terms of accuracy, generalization ability, and stability.