NOx emission estimation and prediction for newly registered motor cars using type approval data and machine learning techniques

Abstract

Nitrogen Oxides (NOx) are among the most harmful pollutants emitted by internal combustion engine vehicles, significantly contributing to urban air pollution and respiratory health issues. Accurate estimation and prediction of NOx emissions are essential for developing reliable emission inventories and informing policy decisions. This study proposes a methodology that utilizes type approval data manufacturerprovided emission factors under standard testing conditions to estimate and predict NOx emissions from vehicles in Sri Lanka. A comprehensive Vehicle Emission Test (VET) dataset from 2016 to 2022 was compiled, including attributes such as vehicle make, model, engine capacity, fuel type, and manufacture year. Data prior to 2016 were excluded due to inconsistencies. Type approval emission factors, typically based on Euro standards, were matched to each vehicle to estimate emissions. A Random Forest machine learning model was also developed to predict NOx output. The results offer insights into annual NOx emission trends and highlight key contributors by vehicle type and manufacturer. This approach demonstrates how standardized emission data can be leveraged to address data gaps in emission inventories, especially in developing regions.