Danalet, A., and Bierlaire, M.

Activity pattern modeling: A path choice approach

Speaker: Danalet Antonin

14th Swiss Transport Research Conference (STRC)

May 15, 2014

In Switzerland, the largest railway stations are called "RailCities" by the Swiss Railways. It emphasizes their transformation into place to live and perform several different activities, similar to a small-scale city. Similar concerns than in urban areas are increasing: escalating costs of new infrastructure, increasing concerns regarding traffic congestion, scarcity of land use. Pedestrian demand management strategies would allow to modify individual travel behavior. The activity-based approach models the interactions shaping the activity participation patterns. Traveling is seen as a derived demand from the need to pursue activities. Several models have been proposed. Activity scheduling model in an entire-day framework is a mix of rule-based algorithm, duration models and discrete choice structures. The biggest drawback of most of these models is the postulated rules: they are structured on home and tours from home, with models applied sequentially according to priorities of activity types. Very often, the large dimensionality of the problem (activity types, continuous time, number of episodes in the day) implies aggregation (broad periods of time, mandatory vs non mandatory, primary vs secondary) or hierarchy of dimensions. We are developing a modeling framework based on path choice. The activity-episode sequence is seen as a path in an activity network. The sequence is not home-based nor tour-based, so that the model can be applied in different contexts, both urban and pedestrian. The large dimensionality is managed through an importance sampling based on Metropolis-Hastings algorithm for the generation of the choice set. The time is discretized in regular intervals. The utility of an activity-episode sequence is the sum of individual trips and activities, including the time-of-day preferences and the satiation effects. First estimation results are presented based on data from WiFi traces.