摘要
精准有效地预测碳排放量有助于推动低碳经济的发展。基于2000—2023年时间序列数据,选取国内生产总值(gross domestic product, GDP)、能源消耗、城镇化水平和人口数量等核心驱动因素作为预测指标,分别构建灰色模型(grey model, GM)GM(1,5)、反向传播神经网络(back propagation neural network, BPNN)模型和岭回归模型(ridge regression model, RRM)进行实证分析。实证分析结果表明,3种单一模型的平均相对误差(mean relative error, MRE)分别为5.06%、0.44%和1.02%。为进一步提升预测精度,采用误差倒数法确定最优权重系数,构建了GM-BPNN-RRM变权组合预测模型。结果显示,组合模型的平均相对误差降至0.40%,其预测性能优于各单一模型。
Accurate and effective carbon emission prediction can help promote the development of low-carbon economy.Based on time series data from 2000 to 2023,core driving factors including gross domestic product(GDP),energy consumption,urbanization level,and population size were selected as predictive indicators.The grey model(GM)GM(1,5),back propagation neural network(BPNN)model,and ridge regression model(RRM)were constructed respectively for empirical analysis.The empirical analysis results demonstrate that mean relative error(MRE)of these individual models are 5.06%,0.44%,and 1.02%,respectively.To further enhance the prediction accuracy,GM-BPNN-RRM variable-weight combination prediction model was developed using the reciprocal error method to determine the optimal weight coefficients.Results show that the combination model achieved a reduced mean relative error of 0.40%,demonstrating superior predictive performance compared to individual models.
作者
王娟
李学鹏
WANG Juan;LI Xuepeng(School of Data Science,Tongren University,Tongren,Guizhou 554300,China;College of Material and Chemical Engineering,Tongren University,Tongren,Guizhou 554300,China)
出处
《北京信息科技大学学报(自然科学版)》
2025年第5期93-98,共6页
Journal of Beijing Information Science and Technology University(Science and Technology Edition)
关键词
GM(1
5)
反向传播神经网络
岭回归模型
误差倒数法
变权组合模型
GM(1,5)
back propagation neural network(BPNN)
ridge regression model(RRM)
reciprocal error method
variable-weight combination model
