Activity travel repertoires objectified as Bayesian inference networks (AT-ROBIN)

Abstract

To assess the impact of TDM aiming at reducing negative effects of motorized traffic, policy makers rely on travel demand models of various kinds. Over the past decades a paradigm shift in travel demand modelling could be observed from trip and tour based modelling towards activity-based modelling. The latter consider travel to be a derivative of people’s activity participation. At the same time modelling methods evolved from aggregate approaches towards disaggregate and agent-based principles, an evolution supported by the increase in computational capacity. In this transition the need for behavioural realism in travel demand models is growing. Indeed, the ultimate goal is to improve predictive capacity and accuracy of travel demand models by mimicking individual’s activity travel (AT) choices and their behavioural response to various TDM. Based on a qualitative, contextual and descriptive exploration of daily AT behaviour using semi-structured interviews with open-ended questions, travel diaries and GPS registration, a generic descriptive model of daily AT choices on a strategic level (activity choice, destination choice, transport mode choice) was developed (Hannes, et al., 2007). Based on further analysis of the in-dept interviews, in this research we objectify the individual descriptions of fixed routines in daily AT and the circumstances in which they appear, in a Decision Network shaped as a Bayesian Inference Network. Both the structure of the network and the parameters are learned from experts’ qualitative specifications. Results show that this approach constitutes a valuable modelling framework for the integration of individual AT repertoires in an agent-based model of travel demand.