Multivariate adaptive regression splines models for vehicular emission prediction

Abstract

Background: Rate models for predicting vehicular emissions of nitrogen oxides (NO X) are insensitive to the vehicle modes of operation, such as cruise, acceleration, deceleration and idle, because these models are usually based on the average trip speed. This study demonstrates the feasibility of using other variables such as vehicle speed, acceleration, load, power and ambient temperature to predict (NO X) emissions to ensure that the emission inventory is accurate and hence the air quality modelling and management plans are designed and implemented appropriately. Methods: We propose to use the non-parametric Boosting-Multivariate Adaptive Regression Splines (B-MARS) algorithm to improve the accuracy of the Multivariate Adaptive Regression Splines (MARS) modelling to effectively predict NO X emissions of vehicles in accordance with on-board measurements and the chassis dynamometer testing. The B-MARS methodology is then applied to the NO X emission estimation. Results: The model approach provides more reliable results of the estimation and offers better predictions of NO X emissions. Conclusion: The results therefore suggest that the B-MARS methodology is a useful and fairly accurate tool for predicting NO X emissions and it may be adopted by regulatory agencies.