A hybrid microsimulation model of urban freight travel demand Rick Donnelly, Marcus Wigan & Russell Thompson * July 15, 2010 The distribution of freight within large urban areas is a highly complex process involving a vast mul- titude of interactions between an equally large number of agents. Shippers, carriers, intermediaries, warehousers, consumers (businesses as well as households), and the government all dynamically exchange goods and information in order for this system to work. Planners and policy-makers have long sought to understand and predict these interactions and their impact on the urban transportation system. They have perhaps naturally turned to modeling the interactions, most often using sketch planning techniques or adapting the sequential modeling approach used for person travel model- ing. Not surprisingly, such simplifications have not fared well, as they fail to capture some of the key dynamics of urban freight. In particular, they miss the high prevalence of trip chaining and the widespread use of distribution centers in the supply chains that generate freight movements. As these and other patterns have come to define urban freight existing models have been seen as increasingly deficient (Taylor & Button 1999, Hunt & Steffan 2007). Our dissatisfaction with the status quo in freight modeling inspired the idea for an agent-based modeling approach capable of representing these complex interactions, their diversity, and their inherent variability. The chance to do so presented itself in Oregon, and the initial framework was originally developed as part of their Transportation and Land Use Model Integration Program (TLU- MIP) almost a decade ago. The goal was to translate economic flows between firms, as well as firms and households, into freight flows by mode of transport. While the approach chosen was well founded in theory and current thinking, the data required to comprehensively estimate such a model were unavailable. Each known source of data only provided a narrow glimpse of the overall freight picture, and a means for fusing them into a holistic dataset suitable for model estimation seemed forlorn. Thus, it was decided to use a microsimulation approach that separately modeled each of the important decisions that characterize the demand for freight transport. The initial model was known simply as CT. Another model in the TLUMIP suite estimated economic growth and allocated it to production, exchange, and consumption between firms and households. Many of these flows were mapped to commodities that were modeled in CT, which translated them into discrete daily shipments carried by specific vehicles. The shipments, which in- cludes attributes such as departure and dwell times as well as origin and destination, were organized into truck tours. The stops on each tour were optimized using the well-known traveling salesman problem and then assigned to the network. The land use and regional economic component has since evolved into the PECAS modeling framework, which is being deployed both in Oregon (replacing its predecessor in the TLUMIP framework) as well as in other cities and states. * Rick (donnellyr@pbworld.com) is a principal consultant at Parsons Brinckerhoff and Senior Fellow at the University of Mel- bourne. Marcus is a Professorial Fellow at the University of Melbourne, and Russell Thompson is a Research Fellow at Monash University. 1