An agent based dynamic road freight transport demand generation

Abstract

Competition between freight and passenger transport for the use of the road infrastructure is an increasingly important problem. This research is carried out in the framework of the DIDAM (Disaggregated demand and assignment models for combined passengers and freight transport) research project which aims at proposing some advances in fundamental research in transportation modelling and analysis, and this paper presents some preliminary results obtained on the Belgian network. The DIDAM project's methodology is organized around two axes: on one hand, a disaggregated dynamic demand model for freight transport is proposed. In this model, the freight transport actors are represented by agents. These agents are extracted from existing databases and their behavior is then generated by means of a simulation that tries to represent the interactions between shippers and carriers who both try to minimize their costs. The carriers try to fill their trucks and to combine several trips to maximize their benefits (by minimizing the empty-trips for example). This step of the model faces a well-known optimization problem called Traveling Salesman Problem (TSP). In the meanwhile the shippers try to find the best transport opportunity by putting the carriers into competition. On the other hand, the Belgian road network is represented in a GIS based transport model software developed at FUCaM and called NODUS. Each arc of the network is associated with a dynamic cost function built using standard OD matrices for passenger transport combined with global traffic density data. By doing this, one takes into account the passenger flows everywhere and at every moment of the day. That way, the rush-hours are also taken into account. These dynamic cost functions are then used by the transporter-agents of the simulation that tries to minimize their costs. A time-dependent OD matrix for freight transport is built as result from this process. Finally, a prototype of a joint traffic assignment model is being developed. This new assignment procedure assigns the flow for each time-slice sequentially, keeping the still running flows from previous time slices into account. Still under development, this algorithm should provide the agents a feedback on their traveling time, which could be retained to allow them to change their habits if needed during a next trip. This paper will first present the time-dependent freight origin-destination matrices generation process, discussing the agent's generation, their characteristics and the implementation of their behavior in a simulation. Finally, the first time-dependent freight transport OD matrices built on the basis of this methodology will be assigned on the Belgian network using the experimental assignment model.