Assessment of on-road High Nox Emitters by Using Machine Learning Algorithms for Heavy-duty Vehicles

Abstract

Abstract The aim of this study was to develop a model structure and to train a model based on chassis dynamometer datasets and subsequently use the trained model in conjunction with portable emission measurement system (PEMS) datasets in order to identify vehicles as possible high-NOx emitting vehicles. The long-short term memory (LSTM) model developed based on a single reference vehicle (i.e., Vehicle A) dataset was applied to the entire 12 diesel vehicle PEMS datasets in order to identify high-NOx emitters. The results showed that the vehicles that were manually identified as high emitting vehicles (i.e., control subjects) were also identified by the LSTM model to exceed real-world NOx emissions. Similarly, a random forest (RF) model was developed for a reference CNG vehicle (i.e., Vehicle N) and subsequently applied to 11 CNG vehicles with a 0.2 g/bhp-hr NOx regulation limit using PEMS data in order to identify any possible high NOx emitting vehicles. The results showed that the vehicles that were manually labeled as high emitters were also identified by the RF model to exhibit high real-world NOx emissions beyond any properly working vehicle. The prediction results show that high NOx emitting vehicles had a ratio of predicted versus measured NOx emissions that were lower than unity.