基于NeuralProphet-LSTM模型的碳价预测研究  

Research on Carbon Price Prediction Based on NeuralProphet-LSTM Model

作　　者：蔡远航 冯建新[1] 王艳青 李婉君[2] 丁元明[1] 胡越 CAI Yuanhang;FENG Jianxin;WANG Yanqing;LI Wanjun;DING Yuanming;HU Yue(Key Laboratory of Communication and Network,Dalian University,Dalian 116622,Liaoning Province,China;Dalian Institute of Chemical Physics,Chinese Academy of Sciences,Dalian 116023,Liaoning Province,China)

机构地区：[1]大连大学通信与网络重点实验室,辽宁省大连市116622 [2]中国科学院大连化学物理研究所,辽宁省大连市116023

出　　处：《全球能源互联网》2025年第2期239-249,共11页Journal of Global Energy Interconnection

基　　金：大连大学学科交叉项目(DLUXK-2023-ZD-001)。

摘　　要：随着人类活动的不断扩展,温室气体的排放量也在持续增长,加剧了碳环境容量的稀缺程度,提高了对碳排放权进行定价的强烈需求。碳市场交易价格作为发挥碳市场功能的核心要素,关乎碳市场的稳定运行和碳减排效率。碳市场交易价格的准确预测对有效开展碳资产投资和寻求最低碳减排成本具有重要的意义。为此,提出一种基于NeuralProphet-LSTM(long short-term memory,长短期记忆)模型的新型碳价格预测方法:首先使用NeuralProphet对碳价序列进行趋势、季节性效应、事件和节假日效应以及自回归效应的模块分解并初步预测;之后使用其预测结果计算残差放入LSTM中进行更深层次的信息挖掘;最后将LSTM对残差的预测通过组件加法与NeuralProphet预测结果组合,完成碳价序列信息的融合。针对欧盟碳市场和中国湖北碳市场进行预测,结果显示该模型的预测性能超过了其他模型,展现出较高的应用价值。With the expanding of humanity’s activities,the continuous escalation in greenhouse gas emissions has exacerbated the scarcity of carbon environmental capacity,thereby intensifying the demand for precise carbon emission rights pricing.The transaction price in the carbon market,as a pivotal element driving the functionality of the carbon market,is crucial for the stable operation of the carbon market and the efficiency of carbon emission reductions.Accurate forecasting of carbon market transaction prices is of paramount importance for the effective investment in carbon assets and the pursuit of the lowest carbon emission reduction costs.Consequently,this paper proposes a novel carbon price forecasting method based on the NeuralProphet-LSTM(long short-term memory)model.Initially,the NeuralProphet model is utilized to decompose the carbon price series into trends,seasonal effects,event and holiday effects,and autoregressive effects for preliminary forecasting.The forecast results are then used to calculate the residuals,which are inputted into the LSTM for deeper information mining.Finally,the LSTM’s prediction of the residuals is combined with the NeuralProphet forecast results through component addition,completing the integration of carbon price series information.Forecasts for the EU carbon market and China’s Hubei carbon market demonstrate that this model’s forecasting performance surpasses other models,showing high application value.

关 键 词：碳价预测 人工智能 混合模型 NeuralProphet LSTM 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] X196[自动化与计算机技术—计算机科学与技术]