Mobile emission inventories are constructed by multiplying a pollutant emission factor by a travel activity (e.g., number of trips, vehicle miles travelled, etc.). To create emission rates, vehicles are tested on dynamometers using driving cycles, or speed–time traces. The process currently used to create the driving cycles is deterministic. However, if we examine the data and data collection techniques closely, it is clear that an observed speed, v(t), represents one of the many possible values that true speed, V(t), may take on at a given time t. With an ordered set of random variables {V(t)} and associated probability distributions, driving cycles should be defined by a stochastic process. In this study, we propose a new approach for constructing driving cycles using Markov process theory. The new approach not only provides an important statistical foundation for drive cycle estimation, it also overcomes several key limitations of the current driving cycle construction methodologies. For example, we use a maximum likelihood estimation (MLE) partitioning algorithm that enables us to associate a segment with a specific modal operating condition (e.g., cruise, idle, acceleration or deceleration), which, in turn, preserves finely resolved driving variability. We apply the new method to the data used to construct EPA's new regulatory facility-specific driving cycles. Comparisons with these cycles indicate relatively similar global results (e.g., average speeds) under uncongested conditions. However, the new cycles tend to contain a higher frequency of small scale acceleration and deceleration modal events than are represented in the EPA cycles. For congested conditions, in addition to greater frequencies of acceleration and deceleration modal events, the new cycles tend to have higher speeds and harder accelerations. Overall, the improvements in the new method represent significant advances in the development of stochastic driving cycle construction methods.
