Decision Theory in Activity-Based Models

Introduction

AB demand models generate a sequential list of activities and trips connecting these activities for every person in the study area. Demand generation is embedded in a concept of daily activity demand from which the need for transport is derived (Kitamura, 1988). A major advantage of AB models is that the spatial and temporal consistency of travel behaviour can be ensured. This is superior to traditional demand generation where aggregate traffic quantities represent isolated trips. Generally speaking, an AB model of travel demand consists of various model components to streamline the process from population input to travel output. The main components of such a system are a population synthesizer, which is required to generate population input data sensitive to demographic evolutions. The scheduler uses this input and additional data such as transportation and land use characteristics from other model components to generate detailed activity and travel plans. Subsequently, various components to model route choice, to assign traffic to the road network and to account for interaction between demand and supply can be added to establish a full AB micro-simulation model.

Clearly, the core of an AB model of travel demand is the scheduler since this model component produces a detailed calendar of activities and travel for each individual, indicating what to do, when, for how long, where and how to travel to that location. AB schedulers generally generate such a detailed activity calendar from scratch for each individual in the population based on their socio-demographic characteristics, although some schedulers use a predefined frame of activities depending on person characteristics.

Roughly speaking, building the scheduler of an AB model involves a three phase process. Firstly, the actual model is learned and its parameters or rules are estimated based on observed activity and travel behaviour in diary data (the training set). In the next step, the model performance is tested. Therefore, current population characteristics are used to generate individual activity and travel calendars, and this output is compared to actual travel behaviour as observed in travel surveys (the test set). Once validated, the model can be applied to forecast the impact of policy measures, such as road pricing (e.g. Arentze & Timmermans, 2008), and general socio-demographic changes in so-called policy scenarios.