Congestion Probability and Traffic Volatility

Abstract

The primary purpose of this study is to understand better the roadway performance and the conditions that trigger congestion. Incidents of recurrent congestion were encountered at a location on New Hampshire Interstate 93 northbound, where radar measurements of speed were continuously recorded from April through November 2010. The root cause for the onset of congestion, both recurrent and nonrecurrent, is impossible to discern and explain by using exploratory data analyses alone; the data are too noisy. A time series modeling approach suggests that the magnitude of traffic flow and volatility, measured as the second (variance) and third (skewness) moments of flow residuals, can explain the triggering of congestion events for a highly variable environment that changes time by the time and day. The approach includes two types of mathematical models: generalized additive binomial probability models for roadway congestion that use functions of traffic flow and volatility as explanatory variables and functional data models that decompose and smooth traffic data to add insight to the roles that flow and volatility play in the congestion process. Most notably, the probability of congestion is shown to be a function of the short-term history of flow and volatility. The changes in these values—the first derivatives of flow and second and third moments of flow residuals derived from functional data model—are shown as well. Model selection, parameter estimation, and checking are presented.