Document Type

Thesis

Degree Name

Master of Applied Computing

Department

Physics and Computer Science

Program Name/Specialization

Applied Computing

Faculty/School

Faculty of Science

First Advisor

Dr. Sukhjit Singh Sehra

Advisor Role

Thesis Supervisor

Second Advisor

Dr. Emad A. Mohammed

Advisor Role

Thesis Co-Supervisor

Abstract

Digital twin technology has emerged as a transformative paradigm for intelligent transportation systems, driven by the growing need to analyze, simulate and optimize increasingly complex urban transportation networks. Digital twin systems help fulfill this requirement through the exchange between physical transportation infrastructure and its virtual counterpart. Urban mobility, characterized by independent transport modes, volatile traffic patterns and city-specific road network characteristics, tends to benefit significantly from digital twins. However, current digital twin systems often face significant challenges due to limited adoption of advanced artificial intelligence-driven prediction, natural language interaction capabilities, expensive computation at city-scale, and difficulty integrating multi-modal transportation services. Therefore, this thesis advances the digital twin system for urban mobility through link-level travel time estimation (LinkETA), meta-learning based generalization (MetaETA), multi-modal travel time estimation and a LangGraph-driven natural language simulation interface. LinkETA, a multi-relational graph transformer represented the road network with eight relation graphs and employed physics-guided calibration to estimate link-level and route-level travel time. MetaETA extended this by jointly training LinkETA with MetaTraffic, a graph attention network that predicts the traffic profile for each road segment within a Model-Agnostic Meta-Learning paradigm, where morphology-based clustering and city-level structural similarity weighting helped achieve robust predictions with minimal data available. For multi-modal travel time estimation, LinkETA is extended for bus-stop dwell time and stop-to-stop travel time prediction and integrated with a multi-modal routing engine. This replaced bus schedule-based estimates with context-aware predictions for complete door-to-door itineraries in a multi-modal journey. An accessible natural language interface was provided through a Large Language Model (LLM) with LangGraph that enabled users to configure and execute simulations through raw text without technical expertise. The proposed algorithms were empirically validated using trajectory data from multiple cities, exhibiting state-of-the-art results and deployed within the Digital Twin Framework for Urban Mobility Operating Systems (DTUMOS) under an open-source license, hence advancing the knowledge in intelligent digital twin simulation for urban mobility and providing reproducibility to further research in this domain.

Convocation Year

2026

Convocation Season

Spring

Download

Available for download on Thursday, October 21, 2027

Share

COinS