Conference Papers 2025

Seminar
Day 2 (18 Sep 2025), Session 7, ABM, 15:30 - 17:30

Status
Accepted, documents submitted

Submitted by / Abstract owner
Mohammed Hajouj

Authors
Mohammed HAJOUJ, Hasselt University (presenter)
Farhan JAMIL, Hasselt University (presenter)
Luk Kanpen, Hasselt University
Muhammad ADNAN, Hasselt University
Tom BELLEMANS, Hasselt University

Short abstract
This research enhances city-scale travel behavior simulation by applying population disaggregation techniques and refining TAZs into smaller units.

Abstract
Cities are complex systems in which daily travel decisions and activity choices are shaped by sociodemographic characteristics and urban spatial structure (such as zoning and land use). Activity-based models (ABMs), such as the Framework for Environmental Analysis of Transport Household Energy Recreational Schedules (FEATHERS), simulate individual and household travel behavior based on daily activity patterns. To ensure accuracy, these models require detailed sociodemographic and travel habit data for the entire population, typically generated through synthetic population methods.
This research adapts the FEATHERS model to Flanders, Belgium, by shrinking the traffic analysis zones (TAZs) into smaller spatial units. This refinement increases spatial resolution and improves the granularity of the demand model. However, due to the greater number of zones, this refinement significantly increases computational complexity. To address this challenge, we will introduce a hybrid zoning layer by prioritizing high-resolution zones in urban centers, balancing the need for computational feasibility with model accuracy.
Our three-stage methodology begins with synthetic population generation using an Iterative Proportional Updating (IPU) algorithm. Microdata from the Onderzoek Verplaatsingsgedrag Vlaanderen (OVG5) survey (8,821 individuals) and 2021 census data are used to generate a population considering age, gender, education, employment, and household attributes. The second stage is creating a hybrid zoning layer with high spatial resolution in city centers by incorporating building points extracted from the OpenStreetMap (OSM). Finally, the synthetic population is spatially disaggregated using area interpolation, dasymetric mapping (building points and residential addresses-based) approaches.
We preprocessed the addresses for Flanders (Belgium) for the residential addresses-based disaggregation. The BOSA-BeST dataset provides addresses and a single coordinate pair but no building functions. To tackle this, we extract building contours along with building and (Point of Interest) POI tags from OpenStreetMap. If an OSM building polygon covers the coordinate pair of a BOSA-BeST address, the address inherits all tags attached to both the polygon and all POIs covered by the polygon. The OSM 'Map Features' wiki creates a general relation between OSM tags and building functions (activity types) used in our models. This finally leads to a set of building function labels for over 90% of the addresses. The labeling algorithm keeps track of the occurrence frequency of each observed label combination. This is a probability distribution to label addresses for buildings not labeled in OSM.
Then, a disaggregation model disaggregates daily travel plans at TAZ level resolution to schedules at street address resolution. The disaggregator aims to transform a predicted TAZ-based travel plan to an address-based schedule by generating a choice set of 'schedule variants' and selecting one. The selector is based on the total daily travel duration. Hence, positions are not sampled independently but in the context of the travel plan. Stable addresses (work, school, ...) are determined first. The 'variant generator' then randomly assigns addresses to flexible activities and typically produces 150 variants.
A household travel survey is used to determine an Artificial Neural Network (ANN)-based 'total daily travel duration' classifier based on the properties of individuals. This classifier has a 'softmax' output layer. The ANN is used in prediction mode for each TAZ-based schedule, and the coefficients of the softmax layer are assigned to the schedule as a probability weight function (PWF) for the total travel duration. Hence, each TAZ-based schedule has its own PWF. This PWF is used by the 'variant selector' to choose from the choice set.
Disaggregating individual characteristics and travel plans from large TAZs to address levels enhances the accuracy of activity-based modeling by integrating high-resolution spatial data. This approach provides a deeper understanding of individual travel behavior, offering better insights into how various personal attributes and daily activity chains interact with different transport modes. This enables more accurate modeling of the multimodal transport modes, helping to assess the impact of MaaS on the daily travel demand.

Programme committee
Transport Models

Topic
Expanding the 21st century planners toolkit with fit for purpose models and decision support systems