Development of local speed-flow models accounting for road condition to facilitate reliable traffic assignment procedures

Abstract

Traffic assignment models, integral to the four-step urban travel demand modelling process, utilize speed-flow functions to depict the interaction between speed and traffic volume. This, in turn, informs the assignment of traffic across network links. The commonly used speed-flow functions such as the Bureau of Public Roads (BPR) function, were developed in the 1960s. With changes in traffic flow conditions, there is a pressing need to update these models, particularly to accommodate the characteristics of unpaved roads prevalent in developing countries like Uganda. This study sought to address the above limitation by developing local speed-flow functions that take into account road conditions i.e. differentiating between paved and unpaved roads. The research utilized data gathered from a selection of roadways in Uganda, where traffic data was recorded using high-definition CCTV cameras and analysed with the aid of Python programming language. The developed local models have calibrated shape parameters 𝛼 and 𝛽, which are crucial for accurately representing traffic dynamics. The 𝛼 coefficient indicates the ratio of travel time per unit distance at practical capacity relative to the free-flow speed. This means that it reflects how travel time increases as roads approach their maximum capacity. On the other hand, the 𝛽 parameter determines the rate at which the estimated average speed decreases from the free-flow speed as the volume-to-capacity ratio (v/c) increases. In simpler terms, 𝛽 shows how quickly congestion impacts travel speeds. For paved roads, the local models had 𝛼 values of 0.98 and 𝛽 values of 5.71. For gravel roads, the models had 𝛼 values of 0.71 and 𝛽 values of 6.25. These parameters provide a more accurate representation of traffic flow and road pavement conditions in Uganda. When compared with the BPR function, the developed local models displayed superior fit with lower root mean square error and lower average relative error values. Additionally, the study provides roadway capacity ranges for gravel roads between 830 to 1,192 passenger car units (pcu) per hour per lane, and for paved roads, from 964 to 1,348 pcu per hour per lane. These findings contribute valuable insights into the planning and management of road networks in Uganda, demonstrating the importance of context-specific traffic models