摘要
为从全生命周期角度分析铁路无砟轨道碳排放,研究量化不同阶段铁路无砟轨道碳排放的差异,并提出一种自动机器框架与可解释人工智能(Explainable Artificial Intelligence, XAI)相结合的无砟轨道碳排放预测及特征分析模型。首先,划分无砟轨道全生命周期不同阶段,构建无砟轨道全生命周期碳排放计算系统;其次,选择AutoGluon自动机器学习模型,筛选特征变量,进行全局特征排序,结合XAI进行局部解释;最后,以某铁路无砟轨道为例进行研究。结果显示,全生命周期碳排放量为124.48万t。其中,工程物化阶段、运营维护阶段、拆除阶段的碳排放占比分别为42.84%、57.11%、0.05%,人工、材料、机械产生的排放占比分别为0.7%、95.8%、3.5%。AutoGluon-XAI模型结果表明,相比于随机森林等传统机器学习模型,AutoGluon预测精准度更高,综合性能最优;在全局解释中,重要性排前4的特征变量为更新周期、地段类型、轨道结构类型、地基条件,均为影响无砟轨道全生命周期碳排放的重要因素;在局部解释中,分类变量的不同特征呈现出不同的贡献效应,路基地段、板式无砟轨道等分类特征对碳排放正向促进效应较显著,而隧道地段、石质地基等分类特征则对碳排放负向抑制效应较显著。
This paper aims to analyze the carbon emissions of railroad ballast tracks from a whole life cycle perspective.It quantifies the differences in carbon emissions at various stages and proposes a prediction and feature analysis model for ballast track carbon emissions that combines an automated machine learning framework with Explainable Artificial Intelligence(XAI).First,the various stages of the ballast track life cycle are delineated to construct a carbon emission calculation system for the entire life cycle.Next,the AutoGluon automated machine learning model is employed to select feature variables,conduct global feature ranking,and integrate with XAI for local interpretation.Using a railroad ballast track as a case study,the total life cycle carbon emissions amount to 1244800 tons.Of this total,the carbon emissions from the engineering phase,operation and maintenance phase,and dismantling phase account for 42.84%,57.11%,and 0.05%,respectively.Additionally,emissions generated by labor,materials,and machinery contribute 0.7%,95.8%,and 3.5%,respectively.The results from the AutoGluon XAI model indicate that,compared to traditional machine learning models such as random forest,AutoGluon achieves higher prediction accuracy and optimal overall performance.In the global explanation,the top four feature variables,ranked by importance,are renewal cycle,lot type,track structure type,and foundation condition.These factors significantly influence the carbon emissions of ballast tracks throughout their entire life cycle.In the local interpretation,the varying values of the top four characteristic variables demonstrate different effects on carbon emissions.Specifically,categorized features such as a renewal cycle of 35 years,road base section,and slab ballast track exhibit more significant positive contributions to carbon emissions.Conversely,categorized features such as tunnel lots and stone foundations have a more pronounced negative effect,inhibiting carbon emissions.
作者
鲍学英
韩通
BAO Xueying;HAN Tong(College of Civil Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)
出处
《安全与环境学报》
北大核心
2025年第6期2431-2440,共10页
Journal of Safety and Environment
基金
中国国家铁路集团有限公司科技研究开发计划实验室基础研究项目(L2023Z001)
甘肃省自然科学基金项目(23JRRA918)。
