Robust modeling and forecasting of diesel particle number emissions rates

This paper develops predictive models for high-frequency particle number emissions rates, analogous to the models that have been developed for gaseous emissions and particulate mass. Data from diesel buses under real-world driving conditions is used and predictive models are based on engine operating variables, vehicle kinematic variables, vehicle specific power, and gaseous mass emissions rates. Particular focus is devoted to estimation and forecasting that is robust to outliers and asymmetric error distributions. The models based on a combination of vehicle kinematic variables and gaseous emissions offer good data fits when compared to models based on engine operating variables. Furthermore, least absolute value minimization leads to superior out-of-sample predictive accuracy compared to conventional, least squares minimization.

► This paper develops predictive models for high-frequency particle number emissions rates, analogous to the models that have been developed for gaseous emissions and particulate mass.
► Linear models based on a combination of vehicle kinematic variables and gaseous emissions rates achieve satisfactory predictive performance for particle number emissions rates.
► Carbon dioxide and nitrous oxides mass emissions rates are effective predictors of particle numbers, in contrast to vehicle specific power.