Recognizing personalized flexible activity patterns

Abstract

In people’s daily plans, mandatory activities like rest, work or study, are already prearranged or fixed. Time planning or scheduling is therefore primarily to organize flexible activities and trips during the remaining time windows. Models of scheduling become powerful when they can be used to obtain macroscopic insights from microscopic changes. This paper presents a method to extract and model flexible activity patterns from real data, and to use them for activity chain generation. It proposes a multi-activity scheduling method which (i) doesn’t prioritize or fix any scheduling dimension, (ii) uses commonly available data (i.e. travel surveys and land use datasets) as its input, (iii) generates personalized solutions, and (iv) can be configured to be computed in tractable time for realistic time windows. Based on the concepts of known places and activity agenda, this method uses socio-demographic characteristics and matched behavioural parameters of the decision maker to schedule his/her flexible activities. To show the applicability and efficiency of this approach, flexible activity patterns from Singapore were extracted from a travel survey carried out in 2012. This survey contains reported motorized trips of 1% of the population during one day. A dataset with information of more than 100.000 activity locations, and estimated dynamic travel times from a large- scale agent-based transport model were employed for the calculations. The method was tested with a fraction of the survey observations which were not used for the estimation of the models. In order to assess the relevance of the personalized mechanisms of the model, the utility maximization algorithm was applied several times, comparing random locations versus systematically selected locations and randomly generated agendas versus systematically constructed agendas. Although in both cases optimal activity schedules are successfully generated, results show much better fit when personalized models are applied.