Abstract Manager

In comparison to passenger vehicles, large trucks have a greater impact on safety, traffic congestion, pollution, and pavement wear. In order to reduce these impacts, it is essential that truck movements are planned and operated appropriately. Heavy truck movements have traditionally been measured through surveys. However, these remain challenging and time-consuming due to their cost and complexity. The purpose of this study is to examine the spatial and temporal patterns of truck movement in a traffic network by analyzing large streams of GPS data. It has always been challenging to use large streams of GPS data because there are typically no descriptors for key events that occurred during a trip unless the GPS data is accompanied by a travel diary for the trip. It is therefore of primary interest in this paper to develop a systematic framework for identifying truck movement patterns through the clustering of truck trajectory data. In our framework, no prior knowledge is required and trajectory data is directly applied without the need for information about the underlying road networks. More specifically, the proposed framework consists of three steps: identification of trip stop points, measurement of similarity between trips, and truck trajectory clustering.

The first step in our process is to develop an objectively defined speed threshold for identifying truck stops, as well as a multilevel time threshold for identifying temporary stops and freight trip ends. In the next step, as a measure of similarity between two truck trajectories, we propose using the Longest Common Subsequence (LCS), which assumes that the extent to which the routes of trucks overlap determines the degree of closeness and relatedness of these vehicles, as well as potential interactions between them. Next, to address our trajectory clustering problem, we extend a density-based clustering algorithm (DBSCAN), to incorporate LCS-based similarity measurements to the clustering problem. In summary, a few spatially distinct clusters of network traffic streams are produced as a result of the proposed clustering approach. Together, these clusters provide a concise and informative picture of major network traffic streams.

An actual truck trajectory dataset collected in Calgary, Alberta, Canada, is used to demonstrate the proposed framework. The results of our method incorporated the impact of city freight context on truck trajectory characteristics and can give an indication of the spatial distribution and chain patterns of intercity heavy truck freight trips, which have a wide range of practical applications.