基于机器学习的中国碳市场风险预警研究  

Research on the risk warning of China's carbon market based on machine learning

作　　者：李哲 马羽珊 Li Zhe;Ma Yushan(School of Business,Nanjing Normal University,Nanjing 210023,China)

机构地区：[1]南京师范大学商学院,江苏南京210023

出　　处：《环境科学与管理》2025年第2期15-20,共6页Environmental Science and Management

基　　金：国家自然科学基金资助项目(71901124)。

摘　　要：碳排放权交易是实现碳达峰与碳中和的核心工具,因此建立碳价风险预警模型至关重要。研究选取能源价格、经济状况、国际碳价、环境因素、公众关注和不确定性六个方面的15个因素作为预警指标,并使用马尔科夫区制转换模型对碳价波动进行风险等级划分。随后,构建了极端梯度提升、随机森林、决策树等机器学习模型,并用SHAP方法解释模型。以广东碳价为例,XGBoost模型的样本外准确率达到81.95%,SHAP解释结果显示,碳价波动的主要风险因素来自国际碳市场、经济状况和能源价格。研究表明,该预警框架预测性能良好,为碳价风险管理提供了决策依据。Carbon emission trading is a fundamental mechanism for achieving peak carbon emissions and carbon neutrality,making the development of a carbon price risk warning model of paramount importance.This paper selects 15 factors from six aspects including energy prices,international carbon prices,environment factors,public attention,uncertainties as warning indicators,and the Markov regime-switching model is used to classify the risk level of carbon price.Finally,machine learning models including extreme gradient boosting(XGB),random forest(RF),decision tree(DT)models are trained,and the SHAP method is used to explain the model.Taking the carbon price in Guangdong as an example,the XGB model achieves the highest accuracy of 81.95%,and the SHAP interpretation results show that the main factors affecting the stability of the carbon price in Guangdong come from the international carbon market,economic conditions,and energy prices.The study demonstrates that this risk warning framework exhibits robust predictive performance and provides a comprehensive basis for carbon price risk management decisions.

关 键 词：碳排放配额 机器学习 风险预警 SHAP 

分 类 号：X22[环境科学与工程—环境科学]