Dissertation Defence: Towards a Next Generation of Integrated Urban Models: Advanced Machine Learning and Microsimulation Modelling for Transportation and Urban Systems
July 31 at 9:00 am - 1:00 pm
Mohamad Ali Khalil, supervised by Dr. Mahmudur Fatmi, will defend their dissertation titled “Towards a Next Generation of Integrated Urban Models: Advanced Machine Learning and Microsimulation Modelling for Transportation and Urban Systems” in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Civil Engineering.
An abstract for Mohamad Ali Khalil’s dissertation is included below.
Examinations are open to all members of the campus community as well as the general public. Please email mahmudur.fatmi@ubc.ca to receive the Zoom link for this exam.
Abstract
Transportation challenges have imposed substantial economic, societal, and environmental burdens on urban areas over the past fifty years. Addressing these challenges requires recognizing the complex interdependencies between transportation and urban systems. In response, Integrated Urban Models (IUMs) have emerged as comprehensive computational frameworks designed to capture these interconnected systems.
Recent global shifts, including remote working, evolving shopping habits, the adoption of electric vehicles, and advancements in computational power and data availability, underscore the need for a new generation of IUMs.
This dissertation contributes towards developing a next-generation IUM framework by focusing on three primary pillars: advanced modelling techniques, innovative simulation methods, and expanded modules and applications.
Most existing IUMs rely on econometric models grounded in random utility theory. To advance beyond these traditional modes, this dissertation employs machine learning (ML) techniques to develop two critical modules: demographic dynamics (DYx) and ActivX (modelling in-home activity participation and duration). Comparative analyses between ML and econometric models assess predictive performance, interpretability via explainable AI (xAI), and computational efficiency. The research also explores multi-task learning, joint ML models that predict multiple interrelated outcomes, demonstrating that shared input features across related tasks significantly enhance accuracy and efficiency.
For simulation methods, a vectorized (batch-based) approach was introduced to replace the traditional iterative loops to improve simulation speed and efficiency.
The dissertation further expands IUM applications by introducing detailed in-home activity modelling, a component historically underrepresented in IUMs. This module simulates activities like sleeping, leisure, mandatory activities, and personal maintenance. Additionally, the output of this module is linked to EnergyPlus, a residential energy simulation tool developed by the U.S. Department of Energy, enabling detailed simulations of residential energy consumption based on evolving household routines and thermostat behaviours. That being said, a case study simulating the effect of changes in in-home activity participation and duration during COVID-19 is presented.
This dissertation represents a step towards realizing next-generation IUM frameworks; however, numerous other aspects defining such frameworks remain open for exploration.
