Abstract

Trucks contribute disproportionally to traffic congestion, emissions, road safety issues, and infrastructure and maintenance costs. In addition, truck flow patterns are known to vary by season and time-of-day as trucks serve different industries and facilities. Therefore, truck flow data are critical for transportation planning, freight modeling, and highway infrastructure design and operations. However, the current data sources only provide partial truck flow or point observations. This dissertation developed a framework for estimating path flows of trucks by tracking individual vehicles as they traverse detector stations over long distances. Truck physical attributes and inductive waveform signatures were collected from advanced point detector systems and used to match vehicles between detector locations by a Selective Weighted Bayesian Model (SWBM). The key feature variables that were the most influential in distinguishing vehicles were identified and emphasized in the SWBM to efficiently and successfully track vehicles across road networks.

The initial results showed that the Bayesian approach with the full integration of two complementary detector data types – advanced inductive loop detectors and Weigh-in-Motion (WIM) sensors – could successfully track trucks over long distances (i.e., 26 miles) by minimizing the impacts of measurement variations and errors from the detection systems. The network implementation of the model demonstrated high coverage and accuracy, which affirmed the capability of the tracking approach to provide comprehensive truck travel patterns in a complex network. Specifically, the model was able to successfully match 90 percent of multi-unit trucks where only 67 percent of trucks observed at a downstream site passed an upstream detection site.

A strategic plan to identify optimal sensor locations to maximize benefits from the truck tracking model was also proposed. A decision model that optimally locates sensors to capture the maximum truck OD and route flow was investigated using a goal programming approach. This approach suggested optimal locations for tracking implementation in a large truck network considering a limited budget. Results showed that sensor locations from a maximum-flow-capturing approach were more advantageous to observe truck flow than a conventional sensor location approach that focuses on OD and route identifiability.