Modelling short-term dynamics in activity-travel patterns: the feathers model

Abstract

The main contribution of current activity-based models is to offer an alternative to the four-step models of travel demand, better focusing on the consistency of the submodels and proving increased sensitivity to a wider range of policy issues. In terms of short terms dynamics in activity-travel patterns, these activity-based models at their current state of development have much less to offer. For example, route choice and the aggregate impact of individual-level route choice decision on activity generation and rescheduling behaviour is not included in these models. Short term dynamics are really not addressed at all, while issues such as uncertainty, learning and non-stationary environments are also not considered. Especially in the context of day-to-day management of traffic flows, such activity-based models of short-term dynamics in activity-travel patterns would serve their purpose. A prototype, activity-based model of transport demand will be developed for Flanders, Belgium. The basis of this model, which has been given the acronym Feathers (Forecasting Evolutionary Activity-Travel of Households and their Environmental RepercussionS), will be the extended version of Aurora, complemented with some additional concepts. Feathers is an agent-based micro-simulation system in which each individual of the population is represented as an agent. It is also an activity-based model in the sense that the model simulates the full pattern of activity and travel episodes of each agent and each day of the simulated time period. At the start of the day, the agent generates a schedule from scratch and during the day he executes the schedule in space and time. It is also dynamic in that (i) perceived utilities of scheduling options depend on the state of the agent, and implementing aschedule changes this state; (ii) each time after having implemented a schedule, an agent updates his knowledge about the transportation and land use system and develops habits for implementing activities (i.e. the agent learns), and (iii) at each time an agent arrives at a node of the network or has completed an activity during execution of a schedule, he may reconsider scheduling decisions for the remaining time of the day. This may happen because an agent’s expectations may differ from reality. This may be the result from imperfect knowledge, but it may also be due to the non-stationarity of the environment. As a result of the decisions of all other agents, congestion may cause an increase in travel times on links or transaction times at activity locations. Furthermore, random events may cause a discrepancy between schedule and reality. This paper reports the current development of this agent-based micro-simulator that allows one to simulate activity-travel scheduling decisions, within day re-scheduling and learning processes in high resolution of space and time. It summaries some concepts and discusses a series of projects and activities that will be conducted to further operationalize the models for Flanders.