Examining Associations, Identifying Chokepoints, and Modeling Truck Travel Time Performance Measures
Analytics
30 views ◎12 downloads ⇓
Abstract
Trucks transport a significant amount of freight tonnage and are more susceptible to complex interactions with other vehicles in a traffic stream. While traffic congestion continues to be an important ‘highway’ problem, delays in truck travel result in loss of revenue to the freight trucking companies. There is significant research on traffic congestion mitigation, but not many studies focused on data exclusive to trucks. This research is aimed at a link-level truck travel time data analysis to identify roads for improving truck traffic mobility and reducing congestion. The objectives of this dissertation research are: 1) compute and evaluate the truck travel time performance measures (by time of the day and day of the week), 2) use selected truck travel time performance measures to examine their correlation with on-network and off-network characteristics, 3) generate geospatial maps to visualize truck travel time performance measures, and, 4) develop truck travel time estimation models using on-network and off-network variables as independent variables. Truck travel time data for the year 2019 were obtained and processed at the link level for Mecklenburg County, Buncombe County, and Wake County in North Carolina. Various truck travel time performance measures were computed by time of the day and day of the week. Pearson correlation coefficient analysis was performed to select the average travel time (ATT), planning time index (PTI), travel time index (TTI), and buffer time index (BTI) for further analysis. On-network characteristics such as the speed limit, reference speed, annual average daily traffic (AADT), and the number of through lanes were extracted for each link. Similarly, off-network characteristics such as land use and demographic characteristics in the near vicinity of each selected link were captured using 0.25 miles, 0.50 miles and 1-mile buffer widths. The relationships between the selected truck travel time performance measures and on-network and off-network characteristics were analyzed using Pearson correlation coefficient analysis. The results indicate that urban areas, high-volume roads, and roads with through lanes <= 6 are positively correlated with the truck travel time performance measures. Further, the presence of agriculture, light commercial, heavy commercial, light industrial, single-family residential, multi-family residential, office, transportation, and medical land uses increase the truck travel time performance measures (decrease the operational performance). Using the selected four performance measures, geospatial mapping was performed to visualize variations and identify potential chokepoints. The maps were generated across the study area and visualized for various times of the day and days of the week. Selected maps were used to interpret and identify "truck-exclusive" chokepoints. Truck travel time estimation models were developed using generalized estimating equations (GEE) with the average truck travel time per mile (ATTPM) as the dependent variable and the on-network and off-network characteristics as independent variables. The influence of the off-network characteristics was incorporated based on spatial proximity (buffers) and spatial weights (distance decay function, 1/d, 1/d2, 1/d3) to check the best approach for modeling.A longitudinal dataset was created using time of the day and day of the week as variables. Modeling was performed using data for 75% of the links while data for the remaining 25% of the links was used for validation. Linear and gamma log link models were developed to estimate the ATTPM using all variables from multiple buffer widths, all variables with specific buffer widths (individually), and all variables from the spatial weights.All the model results indicated significant influence of time of the day, day of the week, and on-network characteristics like AADT, speed limit and number of through lanes on truck travel time. The model results indicate that office, transportation, heavy commercial, and light industrial land uses, population and employment density of the surrounding areas have a significant increasing influence on the ATTPM (i.e., an increase in these land use areas/demographic estimates result in an increase in the ATTPM). Contrarily, some of the variables like government land use and number of household units have a decreasing influence on the ATTPM (i.e., an increase in these land use areas/demographic estimates result in a decrease in the ATTPM). The model results including the goodness of fit and validation indicated that using data from individual buffer widths performs better compared to the spatial weights.The data was segregated based on the speed limit and county to estimate the ATTPM. The data was classified into three categories based on the speed limit; (1) <= 50 mph; (2) > 50 mph & <= 60 mph; and (3) > 60 mph. In all the models developed (the three datasets), variables like time of the day, day of the week, and on-network characteristics like AADT and number of through lanes were found to be significant. The model results indicate that the ATTPM for the evening peak is the highest followed by the morning peak and afternoon peak hours. Similarly, the results indicate that the ATTPM decreases as the number of through lanes increase. The model results also indicate a significant increasing influence of institutional, light commercial, and light industrial land uses and population on the ATTPM. Contrarily, the heavy industrial land use and population have a significant decreasing influence on the ATTPM.The data was classified into three categories based on the county. The results from the models indicate the dependency of local spatial parameters on the ATTPM. For example, heavy commercial land use and employment density had a significant decreasing influence on the ATTPM in the case of Mecklenburg County, whereas model results indicate a significant increasing influence on the ATTPM in the case of Buncombe and Wake Counties. This can be attributed to the county development patterns and demographic characteristics. Overall, the model results suggest a buffer width of 0.25-mile and gamma log link models being ideal to capture the off-network characteristics to estimate truck travel times. The research demonstrated the potential influence of the on-network and off-network characteristics on truck travel time performance and related models. The ATTPM was considered as the dependent variable. The research can be extended to estimate the 95th percentile travel time (or PT) accounting for factors that results in recurring and non-recurring congestion. Additionally, supervised machine learning techniques could be explored to model travel time performance measures.