生态学报2026，Vol.46Issue(8)：3939-3953,15.DOI:10.20103/j.stxb.202506121494
基于LightGBM模型改进的交通运输业碳排放预测研究
Research on carbon emission prediction of transportation industry based on LightGBM model improvement
摘要
Abstract
The growth rate of carbon emissions from the transportation sector has been significant,and it has now become China's third largest source of carbon emissions.Since existing transportation carbon emission prediction models cannot accurately reflect the characteristics of growth rate changes,this paper employs transportation carbon emission data from 30 provinces across China between 2006 and 2023 as a sample.It proposes an improved carbon emission prediction model for transportation sector based on LightGBM to further scientifically and reasonably predict China' s transportation carbon emissions,analyze growth characteristics,and identify potential causes.This study analyzes six aspects:population factors,technological level,economic level,greening level,transportation level,and interprovincial differences.Root Mean Square Error(RMSE),Mean Absolute Percentage Error(MAPE),Mean Absolute Error(MAE),and Coefficient of Determination(R2)are used as evaluation metrics to compare the predictive performance of different machine learning algorithms.We combined superior algorithm models to enhance overall forecasting accuracy and utilized the composite prediction model to investigate regional heterogeneity in carbon emissions across provinces,and SHAP values were employed to interpret the combined prediction model and analyze the input features.The results show that the joint prediction model based on the LightGBM algorithm,Boruta feature selection algorithm,and genetic algorithm hyperparameter optimization performs exceptionally well,with 1.23,5.14%,0.92 and 0.989 respectively in the four evaluation indicators of Root Mean Square Error(RMSE),Mean Absolute Percentage Error(MAPE),Mean Absolute Error(MAE),and Coefficient of Determination(R2).This indicates exceptionally high predictive accuracy and model fit.From a machine learning perspective,it has been demonstrated that composite models outperform single models in predicting transportation carbon emissions.The features are ranked in descending order of importance as follows:provincial differences＞value-added of the tertiary industry＞carbon emission intensity＞urban green space area＞population size＞car ownership＞highway mileage.Provincial differences have the greatest impact on China's transportation carbon emissions,making them a key variable in transportation carbon emissions prediction.Other influencing factors include the added value of the tertiary industry.This ranking reveals the varying degrees of influence exerted by different factors on China' s transport carbon emissions,facilitating precise research into the growth trends of distinct influencing factors.Additionally,scenario analysis based on model-driven projections indicates that significant heterogeneity exists in the pathways to peak carbon emissions for the transport sector across China's provinces,autonomous regions and municipalities.Although 12 provinces,municipalities and autonomous regions have begun to exhibit declining trends in transport carbon emissions around 2030,enabling them to achieve carbon peaking on schedule,while transportation carbon emissions in another 18 provinces,municipalities,and autonomous regions are still rising,posing a risk of delayed peak carbon emissions.Therefore,it is necessary to implement early interventions and strengthen regionally differentiated control strategies in areas where carbon emissions remain on an upward trajectory.This will effectively curb the growth momentum of carbon emissions from transport to ensure the achievement of the carbon peaking goal for national transportation on schedule.关键词
交通碳排放/机器学习/特征选取/参数寻优Key words
transportation carbon emission/machine learning/feature selection/parameter optimization引用本文复制引用
刘战豫,毛若男,张宇飞..基于LightGBM模型改进的交通运输业碳排放预测研究[J].生态学报,2026,46(8):3939-3953,15.基金项目
河南省哲学社会科学教育强省项目(2025JYQS0244) （2025JYQS0244）
河南省软科学项目(252400412014) （252400412014）
河南省高等学校重点科研项目软科学(24A630014) （24A630014）
