Analysis of Truck Travel behaviour using passive GPS data – Case study of Calgary Region, Alberta

Abstract

Freight demand modelling requires detailed information about the volumes and travel patterns of freight trucks and other commercial vehicles. This thesis explores the feasibility of using passive truck GPS data as a complementary source of data collection to traditional truck surveys to obtain information related to the travel behaviour of freight trucks in the Calgary region. GPS data is inexpensive and has a larger sample size compared to traditional surveys. A stop detection model was developed that used a heuristic algorithm to extract information on truck stop activities, including stop locations and stop dwell times. This model was successful in identifying stops with an accuracy of more than 97%. A combination of truck dwell times and the value of entropy, which provided a measure of the homogeneity of observed trucking carriers, at the stop location could be used to classify these stops by purpose. It was found that stop locations with a higher value of entropy tended to have a larger range of dwell times, while locations with a lower entropy had a lower range of dwell times. A separate heuristic model was developed to derive the truck trips and tours from these stops. This model used parameters that were derived from the observed dwell time behaviour of trucks at the stop locations. Additionally, destination choice models were developed for trucks in the Calgary region using trips derived from the GPS data, Info-Canada data, and travel impedances extracted from the existing Calgary Regional Transportation model. An importance based sampling approach was used to identify the choice set for the observed trips. The model estimation results showed that travel distance, a zone-specific explanatory variable related to employment, and a dummy variable indicating trips between the ‘City of Calgary’ and surrounding areas were significant in explaining the destination choices. The models had a satisfactory coincidence ratio of approximately 0.55 between the observed and modelled trips. The model simulation results show that trips shorter than 20 km are consistently underestimated, and therefore, the models require further calibration.