Analyzing Distributions of Free-Flow Speed on Urban and Rural Roads

The development of faster, improved, and less costly sensor technologies has made it easier to obtain large and much more disaggregate (“BIG” data) streams of traffic related data along freeways and arterials. As such, the raw data that are collected have great impact on providing more credible insight with regard to information such as speed and travel time distribution(s), which can subsequently provide reliable travel information. The average speeds on a segment of roadway can be impacted by different factors that can exist for a specific period. However, there are limited studies that have focused specifically on investigating speed distribution fitting under free-flow and recurrent congestion based on day-to-day real traffic observations. Whereas it is common to assume that speeds are normally distributed, it is essential to take into consideration all possible alternative models for the specification of a distribution that can most appropriately characterize daily vehicles’ speed variability. The aim of this study is to investigate the characteristics of speed distribution patterns under free-flow and recurrent congestion by fitting different models based on day-to-day real traffic observations. Conceptually, vehicles’ speed can experience different ranges of uncertainty because of stochastic traffic flow, random delay due to weather events, traffic direction, vehicle classification, and the number of lanes. Another goal of this study is to determine the speed distribution patterns based on different temporal aggregation levels on two different types of roadways. Traffic data, along with vehicle classifications, have been collected from two study sites located on Interstates I-71 and I-75 in the state of Ohio. In this study, different statistical probability distribution models were explored. The probability distribution models are analyzed based on different scenarios, such as temporal aggregations, spatial scales, lane types, and vehicle classifications. A goodness of fit test has been conducted for a comprehensive evaluation of the performance of distribution models. Results showed that the Burr, log-logistic, and t-location distributions would provide a relatively promising fitting to vehicle speed in the urban segment considering all temporal and lanes aggregations. Standard normal distribution, Burr, and t-location models are evaluated as superior fitting distribution to vehicle speed compared to their alternatives in the rural segment. Maximum likelihood has been used to estimate the probability distributions parameters. The results also exhibited that heavy vehicles, particularly class 9 trucks, have high acceptance rate distribution fitting compared to other classes. Also, the distribution of class 9 vehicles traveling on the outer lane shows a significant difference in the probability distribution parameters compared to inner lanes. This approach offered details on speed distribution and provided critical insights into the dynamics of congestion and the variability of travel times. Other applications that can benefit from these data could be enhancing the reliability of travel time prediction and estimation, mode choice, and travel demand models.