Aninterpretable artificial intelligence-based carbon price forecasting model for China under the perspective of multi-source data fusion

ZHU Mengrui; WANG Minggang

doi:10.3969/j.issn.1002-9605.2025.6.mtjjyj202506003

Abstract：Accurate prediction of carbon price is of great significance for formulating effective climate change policies,optimizing resource allocation,and promoting low-carbon technological innovation and economic transformation.As an emerging market,China's carbon market presents unique challenges in terms of policy design,market structure,and data base,and the complexity of its price prediction is much higher than that of the mature carbon market,and traditional methods have certain limitations in balancing accuracy and interpretability.To this end,this paper focuses on national carbon market price prediction from the perspective of multi-source data fusion and proposes an interpretable AI prediction model that incorporates unstructured features.Specifically,on the basis of traditional predictors,we comprehensively mine the key factors affecting carbon prices by including unstructured features such as public concern,policy uncertainty,and network topology indicators as predictor variables.Various structured machine learning and deep learning algorithms are evaluated and compared to explore the most suitable techniques for modeling high-dimensional nonlinear carbon price data.Finally,the contribution of each influencing factor to carbon price forecasting in the national carbon market is analyzed using Shapely additive interpretation.The results show that adding unstructured features such as policy uncertainty as well as network topology information can improve the prediction accuracy of the model,and the overall contribution of crude oil price among the influencing factors is ranked first,and all four policy uncertainty indices are ranked in the top six,which is more important than the majority of the traditional structured predictors.