Estimating and Modeling Soak Time Distributions with the 2009 National Household Travel Survey Data

摘要

Vehicle soak time is defined as the duration of time a vehicle’s engine is at rest prior to beingstarted. The distribution of soak time is a key input for mobile-source emission models, such asthe EPA MOVES. This paper estimates various soak time distributions, and develops statisticalmodels of those distributions.The data source is the National Household Travel Surveys (NHTS) in 2009, which containsinformation about person and vehicle trips in a 24-hour period for all sampled households. Whenthe weights are introduced, the total vehicle trips for the national level are 467,505,568. We firstdevelop a comprehensive methodology for extracting vehicle soak time distribution informationfrom NHTS data files including the day trip file, person file, vehicle file and household file. Theobtained soak time information is then employed in the development of a series of statisticalmodels that can directly provide inputs to mobile-source emission models.Vehicle emission rates are heavily influenced by soak time distributions due to their impact onvehicle start emissions and evaporative emissions. Since the distribution and duration of soakperiods preceding the first vehicle start of day is quite different from those of soak periodspreceding non-first starts, we analyze these two types of soak period with separate models.Results show that time of day, day of week, trip purpose, vehicle type, gas price, metropolitanstatistical area size and several interaction variables have significant impacts on soak timedurations.A model analyzing the start mode fraction is built with logistic regression methods. The modelrho-squared is 0.88 based on more than 0.41 million observations. Again, time of day, trippurpose, day of week and their interactions are found to be the main factors explaining thedifferences between soak periods prior to first start and those of non-first starts. Following thestart mode fraction model, a statistical model on non-first start soak time durations is alsoestablished. After fitting the data with several parametric distributions, the generalized Gammamodel is chosen for its superior goodness of fit. This model enables emission modelers andanalysts to predict soak time distributions based on several demographic, socioeconomic andtravel behavior characteristics. The impact of fuel price on soak time is also considered in themodel.