Most of the activity and travel behavior theories and models focus on long-term, average behaviors, to predict the overall travel demand and to guide transportation-related policy decisions. Nowadays, with the development of technologies, especially the Global Position System (GPS) technology, we can track each individual for a long time at a low cost and monitor the network in real time. The emrging GPS technologies provides a great opportunity for researchers to examine how each take activities and travels on different days in real life. This study targets to explore the how different people arrange their activities and travels, and to predict how people dynamically make their activity mobility schedules in short-term.

First, a theoretical framework is proposed to explain how each individual makes activity travel plan dynamically. The theoretical framework not only describes how an individual makes daily activity schedule at the beginning of a day but also shows how the individual adjust the activity schedule if any unexpected disturbance occurs during the day. At any given time, the individual is assumed to make activity schedule through maximizing the total expected utility based on the current knowledge, preference and expected utilities of different activities. People will learn from experience and update their knowledge as new information and experience become available.

Different model components are built based on the proposed theoretical framework. The individual’s short-term activity travel schedule problem, either before-day or within-day, is formulated as the Short-term Activity Mobility Schedule Problem (SAMSP), where different decision variables are used to represent people’s choices in different dimensions with the objective function to maximize the total expected utility. Three different parts of utilities are considered: utility generated by conducting an activity for a certain period, penalty or disutility caused by early or late arrival and disutility of travel on the road. The nonlinear terms in the formulation are linearized, which converts the problem to a convex mixed-integer quadratic programming problem.

To apply the proposed theoretical framework and methods to predict people’s short-term activity and travel behaviors, a trip purpose imputation method is first proposed to impute trip purpose using passively collected location. Then the inverse optimization theory is introduced for SAMSP model estimation. The multi-day cutting plane algorithm is proposed to solve the multi-day inverse mixed integer linear programming (Multi-InvMILP) problem.

A smartphone app named as Travel Helper Pro is developed for Android and iOS platforms which can passively track user’s location for a long time and automatically upload data to the cloud server without any user intervention. The collected data is used to estimate the SAMSP models for different individuals and exhibit the capability of the proposed approach.