Converting a Conventional Transport Model into an Agent Based Transport Model with As Much About Activities and Lifestyle Choices As Possible
Peter Davidson, Peter Davidson Consultancy, Shadi Sadeghian, Institut Vedecom, Helen Porter, Peter Davidson Consultancy
A conventional transport model has been developed of a suburban area of Lyon. This model has been used to predict initial demand for a new vehicle, but has been developed into an agent based model incorporating activities and some lifestyle choices.
A conventional transport model has been developed of a suburban area of about 40 km², located within the Greater Lyon area (France) about 12 km east of the city centre of Lyon. The model has 52 sectors, each representing the catchment for a new car sharing service. Trip data has been provided from the 2006 Lyon Travel Survey. Public transport options are included with particular emphasis on combining the new car sharing service with existing modes. The model uses a choice model element combined with a network to predict mode share of the new vehicle, and a sub mode choice is included for public transport. This conventional model has been used to predict initial responses to the car sharing service, explore a number of scenarios relating to the availability of the new service (which was designed to complement the existing public transport services), forecast the potential impacts of its introduction on the transport demand and also on the actual transport supply of the area.
What we then required was a much more detailed model of people’s lifestyle, activity and travel behaviour, and initially, only the conventional Lyon model was available. We needed to learn as much as possible about travel in Lyon from the conventional model, if necessary augmented by data from elsewhere. The results of this learning process was to be used to help define a more detailed household survey to underpin development of a more detailed lifestyle model.
To do this, we decided to develop Lyon’s conventional transport model into an agent based model capturing as much about activities and lifestyle as possible so as to learn as much as possible from what already existed. The advantage of using an agent based approach is that individuals can be followed through the model and their choices analysed directly rather than always considering fractional probabilities. The choices that the individual make are directed by the underlying probabilities, through Monte Carlo techniques, but the individual decisions that contribute to overall behaviours can be analysed. Individual agents can carry additional descriptive attributes that can be included within the choice nests so as to include such important variables as income for example.
A population synthesiser was developed to generate the agents and their individual characteristics for the population of the study to match the conventional model and other data about the population. The population was applied to a nested choice model comprising mode and destination choice estimated so as to match the conventional model. In the second stage model, this choice nest was expanded to model tours with choices of time period, tour secondary (tertiary etc) destination, purpose and mode. In the third stage model, the second stage tour generation model was replaced by a simplified activity generator/ scheduler. Where necessary we used coefficients from other studies in order to develop the model. A fourth stage model was developed to address the long term ‘lifestyle’ household decisions using data from other sources. Instead of focussing on the disutility of travel, the overall positive utility of undertaking an activity was also considered in the new model.
In this way the conventional Lyon model was successively converted into an agent based model with as much about activities and lifestyle choices as possible, as described in the paper. The benefits of the agent based model are also discussed, and some comparisons are drawn about the conventional and agent based approaches.
Association for European Transport