The goal of this project is to develop a new travel demand model based on dynamic discrete choice theory and activity based modelling. This includes multiple steps, and in ongoing projects a sub-model for daily travel and activity planning is being estimated using travel surveys. The current data set includes 3400 observations, and estimating the model currently requires approximately 720 core-hours (taking approximately 3 days on our local server exclusively), thereby severely restricting what is possible to accomplish. This is an embarrassingly parallel problem, and it could probably scale well up to the number of observations. With more resources, it would be possible to use data from more sources to obtain better estimates.
There is also ongoing work to: extend the model to cover long-term planning of activities, where available computer resources will determine the details that will be achievable; introduce more advanced model specifications; and use the model to analyze how accessibility measures influence decisions of, e.g., car ownership and work and residence location. The latter could benefit from Swedish register data potentially including millions of observations. Finally the goal is to use the model for large scale simulation (with millions of agents). With the current solution method, this would imply approximately 100 times the time of the above mentioned estimation and thus 72000 core hours. All of these sub-projects would greatly benefit from, and might be dependent on, the availability of super computers.