基于高阶矩冲击机器学习Multi-LSTM模型的中国碳价预测  

China's Carbon Price Prediction Based on Machine Learning Multi-LSTM Model from the Perspective of High-Order Moment Impact

作　　者：云坡 陈江华[1] 唐文之 YUN Po;CHEN Jiang-hua;TANG Wen-zhi(School of Economics and Management Hefei University,Hefei 230601,China)

机构地区：[1]合肥学院经济与管理学院,安徽合肥230601

出　　处：《安徽师范大学学报（自然科学版）》2022年第1期6-12,共7页Journal of Anhui Normal University(Natural Science)

基　　金：教育部人文社科研究青年基金项目(21YJC790152);安徽省社科创新发展课题攻关项目(2021CX028).

摘　　要：碳价预测是以市场手段推动污染物减排、促进碳市场发展的关键。本文聚焦从高阶矩(偏度-峰度)视角研究市场非对称信息和极端因素对碳价的动态冲击。使用湖北碳市场2014.4.28—2021.2.26时间序列数据,基于实验法构建Multi-LSTM机器学习模型检验预测效果。研究发现,Multi-LSTM模型的碳价预测误差RMSE、MAE、MAPE仅为2.65、1.74、2.43,市场预测准确度为0.88,具有显著的预测精确和稳定性。表明考虑高阶矩冲击的碳价预测机理有效性和合理性得到证明,Multi-LSTM模型能够进行有效的碳价预测。Carbon price prediction is the key to promote pollutant emission reduction and carbon market development by market means.The paper focuses on the dynamic impact of market asymmetric information and extreme factors on carbon price from the perspective of high-order moments(skewness-kurtosis).Using the time series data of Hubei carbon market from April 28,2014 to February 26,2021,a multi-LSTM machine learning model based on the experimental method is constructed to test the prediction effect.It is found that the carbon price errors RMSE,MAE and MAPE of Multi-LSTM model are only 2.65,1.74 and 2.43,and the market prediction accuracy is 0.88,which has significant prediction accuracy and stability.It shows that the effectiveness and rationality of carbon price prediction mechanism considering high-order moment impact is proved and the Multi-LSTM model can predict the carbon price effectively.

关 键 词：碳价预测 高阶矩 机器学习 Multi-LSTM模型 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]