Generating Comprehensive All-day Schedules: Expanding Activity-based Travel Demand Modelling



Generating Comprehensive All-day Schedules: Expanding Activity-based Travel Demand Modelling

Authors

M Feil, M Balmer, K W Axhausen, Institute for Transport Planning and Systems (IVT), CH

Description

The agent-based micro-simulation toolkit MATSim implements an activity-based approach to travel demand generation for large samples. This paper presents our enhancements of the MATSim utility function.

Abstract

Activity-based travel demand micro-simulation generates activity schedules for a certain period of time (e.g. a day) for every member of a population. Travel demand is derived from these activity schedules by the fact that most activities take place at different locations and people need to travel between these. Hence, understanding people's daily activity schedules is fundamental to understand and predict the dynamics of transport.

The agent-based micro-simulation toolkit MATSim implements an activity-based approach to travel demand generation for large samples. The utility of an activity schedule is iteratively improved against the background of overall travel costs which are calculated using a suitable traffic flow simulation. The co-evolutionary learning process stops when none of the agents can further improve their schedule by changing activity sequence, type, number, timings and durations, modes, locations or routes. This paper will present our enhancements of the underlying utility function:

a) Disaggregation of the existing attributes: While the current utility function features three attributes, namely utility of activities performed by type, disutility of travel time and a late arrival penalty, we will further disaggregate these attributes, i.e. more activity types, differentiation of the travel disutility by mode, and differentiation of the delay penalty by activity type.

b) Incorporation of additional attributes: Several further attributes are to be introduced in order to increase the explanatory power of the utility function. Conceivable attributes are the monetary cost of travelling, the quality of activity locations/facilities, technical and behavioural penalties, and socio-economic agent-specific attributes.

c) Definition of the functional form of the utility function: The current utility function features a log form for the duration/performance of activities. This leads to unrealistic results when we allow for changes in the number of activities in the schedule. When the number of activities in a schedule is a dimension of the learning process the log form leads to a lot of very short acitivities due to the decreasing marginal utility of the log-form. In other words, a schedule of two 30 minutes activities of a certain type is always better than a schedule of once 60 minutes of the same activity. We therefore introduce and discuss a modified form of the utility function. It follows some considerations by Joh (2004) and features an S-shaped utility curve for the performance of activities. The S-shape now allows one to cope with a flexible number of activities in a schedule.

d) Empirical estimation of the corresponding parameters: The empirical estimation of the parameters of the utility function will be done through an enhanced Multinomial Logit (MNL) approach. To account for the "Independence from Irrelevant Alternatives" (IIA) property of the MNL model we will introduce a similarity attribute in the systematic part of the utility function reflecting the structural similarity between activity schedules following the logic of the path-size logit or C-logit approach. The estimation will be based on the revealed behaviour data in the Swiss Microcensus (national travel survey) and 2003 Thurgau 6-week travel diary.

The choice set will be generated for each observed person by a heuristic tabu-search around the observed schedule. This algorithm will be briefly described.

Having implemented above 4 steps we will test the utility function and its estimated parameters with a large-scale scenario of the greater area of Zurich, Switzerland, for which detailed validation information is available.


Reference: Joh, C-H. (2004) Measuring and Predicting Adaptation in Multidimensional Activity-Travel Patterns, Dissertation, Technical University of Eindhoven, Eindhoven

Publisher

Association for European Transport