Characterizing on-road particle number emissions from container diesel trucks using interpretable machine learning

Abstract

The rising number of containerized diesel trucks (CDTs) significantly increases particulate matter emissions. However, research on number concentration of particulate matter from CDTs under real-world driving conditions remains limited. This study combines field measurements with interpretable machine learning to systematically analyze the particle number (PN) emission patterns of CDTs. The study collected emission data from a China V CDT on real-world roads in Shanghai using a Portable Emission Measurement System (PEMS) and extracted multidimensional features—encompassing environmental, engine, and driving characteristics—for preliminary analysis. This initial analysis found that high PN emissions are concentrated during the acceleration phase. And PN emissions exhibit a second-order polynomial relationship with vehicle specific power (VSP). Subsequently, among multiple machine learning models, the eXtreme Gradient Boosting (XGBoost) model demonstrated optimal performance in predicting PN emissions for different road scenarios. The model identified engine power, air-fuel ratio, and VSP as the three core influencing factors, with their importance varying by road type: engine power was dominant on highways (43 %), while VSP was primary on non-highways (53 %). The study also used partial dependence plots to analyze the non-linear effects of these key factors. Additionally, the PN emission factor for non-highways (1.91 × 10¹² #/km) is more than twice that of highways (9.80 × 10¹¹ #/km).