A priority based approach to dynamic multi-day activity generation: illustrated with PEV users’ charging behavior

The aim of this paper is to develop a conceptual framework for activity scheduling in the latest version of Albatross (Rasouli, et al., 2018), which can simulate activity-travel patterns across multi-day or multi-week time periods. We suggest a heuristic to make activity generation process dynamic, allowing time-varying, individual-specific, activity priority for scheduling the flexible part of an activity agenda. The proposed framework is illustrated using users’ charging activity for plug-in electric vehicle (PEV) as an example. The model is derived from multi-week activity diary data collected through a smartphone-based prompted recall survey (Geurs et al., 2015). The current version of Albatross consists of two major components that together define an activity schedule for a single day. The first component generates an activity skeleton consisting of fixed activities (i.e., work, school, business and bring/get) and their exact start time and duration. The second component determines the part of the schedule related to flexible activities (e.g., daily shopping, service, non-daily shopping, social, leisure, touring) to be conducted that day, their travel party, duration, time-of-day and travel characteristics. It should be noted that for both components the activities are generated and scheduled according to a predefined priority list. Although such a predefined priority list assumes that all respondents feel the same level of urgency for all activities which cannot be the case, the approach can still be defendable for 1-day activity pattern prediction where any interrelation between the conduct of a certain activity in one day with the need/desire to conduct it in any time in the future is ignored. In this case a predefined priority associated to activities can be attributed to the sequence with the highest probability derived from the data. However, if the one day activity agenda is to be extended to multi-day/multi-week activity patterns, such an approximation may cause substantial error as the priority of each activity in any subsequent days may depend on the urgency of conducting the activity in that specific day which may in turn depend on the time elapsed since the last previous conduct of that activity. In other words and in order to make the model dynamic, it is required to incorporate a memory of when the activity was conducted the last time, and to make scheduling decision sensitive to this history information. Based on the dynamic activity priority deduced in this paper, in the new version of Albatross, the activities will be generated and scheduled in the order of the priority.