Development of a predictive vehicle exhaust volatile organic compounds (VOCs) emission model based on machine...

Vehicle exhaust emission is a major source of volatile organic compounds (VOCs), and accurately quantifying emission is crucial in investigating their contributions and formulating effective control strategies. Current widely used emission models usually do not consider the variability of engine operating conditions, leading to great uncertainty in emission estimation. In this study, we developed a multilayer perceptron (MLP)-based machine learning model to predict the vehicle VOCs emissions. The model was trained on a dataset comprising real-time VOCs emission rates and six vehicle operating parameters (vehicle speed, acceleration, engine revolutions per minute, spark advance, absolute throttle position and intake air temperature) characterized by ion molecule reaction mass spectrometer (IMR-MS) and on-board diagnostics (OBD) system. The results indicated that our model exhibited satisfactory predictive ability for different vehicles, with R² values ranging from 0.57 to 0.75. Absolute throttle position and engine revolutions per minute were identified as the key parameters that can influence the performance of emission model, with the correlation coefficients of 0.9 and 0.64, respectively. Higher errors (RMSE > 45) were generated for vehicles with a greater distribution in the high emission values. Existing studies have primarily focused on modeling conventional vehicular pollutant emissions. This study represents the pioneer attempt to characterize the vehicular VOCs emissions by machine learning and OBD parameters. OBD systems in modern vehicles allow external devices to access real-time data. Therefore, we could quantify vehicle VOCs emissions more conveniently without measuring them by sophisticated instruments.