This study aims to identify high emissions of CO2 and NOX from gasoline passenger cars based on PEMS data by introducing a concept of emission state, and investigate their correlations with driving behaviors. The clustering approach of K-means++ was employed to classify the instantaneous mass emission value and emission rate under various road types, respectively. A novel identification indicator (i.e., the ratio of change rate and growth rate of instantaneous emissions) was proposed as the basis for dividing each emission state. Subsequently, three matrices (i.e., probability matrix, value matrix, and identification matrix) were constructed to reflect relationships between emission states and emission rates under each road type. Moreover, driving scenarios of CO2 and NOX high emission were investigated and compared by machine learning models with SHAP explanation and ordered logistic models. The empirical results indicate that the identification indicators of CO2 and NOX high emissions are 2.49 g/s2 and 3.66 mg/s2 on the freeway, 2.98 g/s2 and 2.12 mg/s2 on the primary road, and 2.77 g/s2 and 2.05 mg/s2 on the secondary road. Within the same ranges of driving behavior parameters on the freeway, the occurrence probability of CO2 high emission state is higher than that of relatively high emission state, while an opposite trend is observed for NOX emissions. Interestingly, despite NOX and CO2 show similar emission characteristics on the primary and secondary road, the driving behaviors corresponding to high emissions of NOX and CO2 present significant disparities. Generally, the acceleration is the primary determinant of CO2 high emissions, while both acceleration and deceleration are significant contributors to NOX high emissions. The findings of this study recommend that long periods of high-speed travelling should be avoided on the freeway. Frequent and abrupt changes in acceleration and deceleration should be minimized on the primary and secondary road, respectively.

中文翻译：

---

一种基于汽油乘用车 PEMS 数据的 CO2 和 NOX 高排放状态识别新方法——来自驾驶行为的洞察


本研究旨在通过引入排放状态的概念，基于 PEMS 数据识别汽油乘用车的高排放 CO2 和 NOX，并探讨它们与驾驶行为的相关性。采用 K-means++ 的聚类方法对不同路型下的瞬时质量排放值和排放率分别进行分类。提出了一种新的识别指标 （即瞬时发射的变化率和增长率的比值） 作为划分每种发射状态的基础。随后，构建了三个矩阵（即概率矩阵、值矩阵和识别矩阵）来反映每种道路类型下排放状态和排放率之间的关系。此外，通过机器学习模型与 SHAP 解释和有序 logistic 模型，对 CO2 和 NOX 高排放的驾驶场景进行了调查和比较。实证结果表明，高速公路上CO2和NOX高排放的识别指标分别为2.49 g/s2和3.66 mg/s2，一级公路上2.98 g/s2和2.12 mg/s2，二级公路上2.77 g/s2和2.05 mg/s2。在高速公路上驾驶行为参数的相同范围内，CO2 高排放状态的发生概率高于相对高排放状态的发生概率，而 NOX 排放则呈相反的趋势。有趣的是，尽管 NOX 和 CO2 在主干道和次干道上表现出相似的排放特性，但对应于 NOX 和 CO2 高排放的驾驶行为存在显着差异。一般来说，加速度是 CO2 高排放的主要决定因素，而加速和减速都是 NOX 高排放的重要因素。 这项研究的结果建议，应避免在高速公路上长时间高速行驶。在主干道和次干道上，应分别尽量减少加速和减速的频繁和突然变化。