Bataillard, L., Pougala, J., Haering, T., and Bierlaire, M. (2022)

A comparative analysis of optimization algorithms for activity-based applications

In order to forecast travel demand, activity-based models generate schedules for every individual in a population sample. Most models generate these schedules using a sequential rule-based approach. Recent activity-based models are based on econometric utility-maximisation, expressed as a Mixed-Integer Linear Program. In order to improve computational efficiency and model expressiveness, we explore adaptations of the same scheduling problem in constraint programming. In a first effort, we investigate a direct translation of the existing MILP formulation. We then adapt the direct translation in two different manners to model the problem in ways that exploit the advantages of constraint programming. In these new programs, we use idiomatic constraint programming expressions: implications as constraints, global constraints such as ELEMENT, and interval variables. We find that constraint programming not only makes the model more intuitive, but also improves solver performance by at least one order or magnitude.