A Markov process approach to driving cycle development

Air Quality mobile emissions inventories are constructed by multiplying a pollutant emissions rate by a travel activity (e.g., number of trips, vehicle miles traveled, etc.). To create emission rates, vehicles are tested on dynamometers using driving cycles, or speed-time traces. The data used to create the driving cycles have traditionally been treated as deterministic. However, if we study the data and data collection techniques closely, an observed speed-time profile is not deterministic. Rather, an observed speed, ν(t), represents one of the many possible values that true speed, V( t), may take on at given time t. With an ordered set of random variables {lcub}V(t){rcub} and associated probability distributions, driving cycles should instead be defined as a stochastic process. We propose a new approach to developing a driving cycle using Markov process theory. This new approach not only provides the theoretical foundation currently lacking in this field, but also incorporates several other key factors that should have been included but are currently missed in driving cycle construction methodologies.; We apply the new method to the EPA's 3-city chase car driving data (i.e., Baltimore, Spokane, and Los Angeles) used to create the EPA's MOBILE6 facility-specific driving cycles, and compare the new driving cycles with the EPA driving cycles. We conclude that many more accelerations and decelerations are represented in the new cycles than in the EPA cycles and an overall improvements in the new method of matching both the global and local driving characteristics of sample driving data. We also apply the method to the LA92 driving data used to create CARB's Unified Cycle. The new driving cycle is found to better replicate the frequencies as well as the average durations of the modal events observed in the sample data. Such differences result from the different construction approaches and can be expected to impact emissions inventory estimation. Finally, we compare the chase car driving data among the five data sets, the EPA's combined 3-city driving data, the California combined 3-region driving data, and the three individual California regional driving data. We conclude that there exists regional driving variability. Further comparison among the California three regional driving cycles using the new method confirms that regional driving differences are sufficiently large enough to result in important driving cycle differences, which implies important regional variability in vehicle emissions estimation.