摘要
新闻数据涵盖了与碳价格密切相关的政策、经济和能源等信息,对碳价格的影响具有时效性。为量化新闻影响力的衰减程度,基于词频统计和指数衰减对新闻数据提取特征,提出了1种新闻影响力衰减时间序列的计算方法,新闻的衰减效应更准确地反映新闻对碳价格的影响程度。为提高预测精度,构建了融合新闻影响力衰减的碳价格多元分解集成预测模型,运用噪声辅助多元经验模态分解方法对碳价格和新闻数据进行多元分解,基于样本熵重构分量,使用机器学习方法对分量进行预测,加和集成得到预测结果。以湖北省碳价格为例进行实证分析。结果表明:新闻影响力指数衰减方法能有效刻画新闻与碳价格的相关性,多元分解集成模型表现出优异且稳定的预测性能。
The news covers information closely related to carbon prices,including policies,economics,and energy.Its impact on carbon prices is time-sensitive.To quantify the degree of news influence attenuation,this paper proposes a method for calculating attenuated news influence based on word frequency statistics and exponential decay,which extracts features from news data.The decay influence of news more accurately reflects the extent of its influence on carbon prices.In order to improve prediction accuracy,this paper constructs a multivariate decomposition-ensemble prediction model of carbon prices incorporating news influence exponential attenuation.It applies noise-assisted multivariate empirical mode decomposition method to decompose the data into several subcomponents.Then the subcomponents are reconstructed based on sample entropy.Finally,machine learning methods are applied to predict the subcomponents,which are aggregated to obtain the prediction results.A case study of empirical analysis is conducted using carbon prices of Hubei Province.The results show that the news influence exponential attenuation can effectively portray the correlation between news and carbon prices.The proposed multivariate decomposition-ensemble model shows excellent and stable prediction performance.
作者
张大斌
黄均杰
凌立文
胡焕玲
ZHANG Dabin;HUANG Junjie;LING Liwen;HU Huanling(College of Mathematics and Informatics,South China Agricultural University,Guangzhou 510642,China)
出处
《河南科技大学学报（自然科学版）》
CAS
北大核心
2024年第1期51-61,M0005,M0006,共13页
Journal of Henan University of Science And Technology:Natural Science
基金
国家自然科学基金项目(71971089,72001083)
广东省自然科学基金项目(2022A1515011612)。
