基于决策树的城市在用车环检首检结果预测模型研究

在用车环检是机动车环保监督管理的有效手段，本文通过建立环检结果预测模型以达到事前管理的目的.研究使用通过简易瞬态工况法的数据库，采用决策树算法，建立了环检数据库数据预处理规则，证实了整数编码可用性和运算负荷小的优点，并采取过采样方法解决数据不平衡问题，通过ROC曲线确定子模型的超参数，建立了环检结果预警模型.结果显示，等级1级以上级别可涵盖90%的不合格车辆，累计行驶里程数、车龄对环检结果影响最大.研究表明，该方法复用性较强，有利于通过易获取的信息预测环检结果，可有效支持机动车排放分级分类管理和政策制定.

Modelling the initial emissions inspection of urban in-use vehicles with decision trees

In-use vehicle emissions inspection is an effective approach to environmental supervision and management of vehicles. In this study, a predictive model of vehicle emissions inspection was developed for proactive management. On the basis of the short-transient-loaded-mode database, the decision-tree algorithm was employed to establish the data preprocessing rules for the database of vehicle emissions inspection, which confirmed the feasibility of integer coding and low computing cost. The oversampling method was adopted to deal with the imbalanced data for training the predictive model, where the hyperparameter was tuned with a receiver operating characteristic (ROC) curve. The results show that 90% of the unqualified vehicles could be identified as level-1 or above by the predictive model, while the cumulative mileages and vehicle age were the most influential factors. This universal method is valuable for predicting the results of vehicle emissions inspection on the basis of readily available information, which can support vehicle-emission classifications and policy making.