基于多粒度级联森林的高排放重型柴油车辆的识别方法  

Identification of high-emission heavy-duty diesel vehicles based on multigrained cascade forest

作　　者：廖琳蔚 杨卓倩 杨鸿泰[1] 韩科 LIAO Linwei;YANG Zhuoqian;YANG Hongtai;HAN Ke(School of Traffic and Logistics,Southwest Jiaotong University,Chengdu 611756,China;School of Economics and Management,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]西南交通大学,交通运输与物流学院,成都611756 [2]西南交通大学,经济管理学院,成都610031

出　　处：《交通运输工程与信息学报》2024年第4期166-181,共16页Journal of Transportation Engineering and Information

基　　金：中央高校基本科研业务费专项资金项目(2682023CX044,2682023ZTPY012);自然科学基金面上项目(72071163);四川省国际科技合作项目(24GJHZ0342)。

摘　　要：机动车尾气排放已成为主要的空气污染来源。车载诊断系统(on-board diagnostics,OBD)作为重要的机动车排放监管工具,可以获取与氮氧化物排放相关的关键信息。然而,由于OBD系统存在数据缺失和数据质量不高的问题,难以准确评估车辆NO_(x)排放水平并有效筛查高排放车辆。本文提出了一种基于多粒度级联森林(multi-Grained Cascade Forest,gcForest)模型的高排放重型柴油车筛选方法。首先,使用Gumbel分布对重型柴油车辆的NO_(x)/CO_(2)数据进行概率分布对象拟合,以确定高排放阈值并标记高排放记录;其次,采用熵值法和多重共线性检验确定最优特征子集,并使用合成少数过采样技术(Synthetic Minority Over-sampling Technique,SMOTE)处理高排放样本和清洁样本比例不平衡问题;最后,构建gcForest模型用于分类排放超标数据。实验结果表明,该模型在识别高NO_(x)排放重型柴油车辆方面具有有效性和适用性。该方法提升了利用OBD数据识别高排放车辆的可行性,为精准监管机动车排放提供了可靠的数据支撑。Vehicle exhaust emissions are a major source of air pollution.On-board diagnostic(OBD)systems are important regulatory tools for vehicle emissions because they can directly access key information related to nitrogen oxide(NO_(x))emissions.However,owing to data unavailability and quality issues in OBD systems,accurately assessing the NO_(x) emission levels of vehicles and effectively screening high-emission vehicles are challenging.This study proposes a method for screening highemission heavy-duty diesel vehicles(HDDVs)using the multigrained cascade forest model.First,the Gumbel distribution is used to fit the probability density distribution of the ratio to determine the high-emission threshold and label high-emission records.Subsequently,the multicollinearity test is performed in conjunction with the entropy method to determine the optimal feature subsets.Next,the synthetic minority oversampling technique(SMOTE)is used to address the imbalance between highemitting and clean samples.Finally,the multigrained cascade forest model is constructed to classify data with emissions that exceed the standards.Comparative-analysis experiments verify the effectiveness and applicability of the model in identifying high NO_(x) emissions from HDDVs,thus enhancing the feasibility of identifying high-emission vehicles and providing reliable data support for the precise regulation of vehicle emissions.

关 键 词：信息技术 高排放车辆识别 多粒度级联森林模型 重型柴油车 车载诊断系统 Gumbel分布 

分 类 号：X734.2[环境科学与工程—环境工程]