PhD Oral Exam - Shayan Nejadshamsi, Information and Systems Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Understanding urban-scale building occupancy and commuting flow patterns is critical for sustainable city planning, energy efficiency, and transportation optimization. However, existing Urban Building Energy Modeling (UBEM) frameworks rely on standardized occupancy schedules that fail to capture spatial and temporal variations, leading to significant inaccuracies in energy consumption estimates.

This thesis introduces a data-driven approach to improve urban-scale occupancy estimation and commuting flow prediction by integrating transportation and mobility data with machine learning techniques. First, a Transportation-Informed Building Occupancy (TIBO) framework is developed to estimate dynamic building occupancy profiles using large-scale transportation datasets, including metro, bus, bike-sharing, and traffic flow data. The results demonstrate that TIBO-based occupancy profiles significantly enhance the accuracy of UBEM simulations compared to conventional ASHRAE schedules, reducing errors in energy demand predictions.

Since transportation-based building occupancy estimation frameworks use transportation data (point-based/origin-destination-based data types) as their input, enhancing the prediction of transportation demand and flow leads to better prediction of building occupancy as well. Therefore, this thesis presents a multi-task spatiotemporal deep learning framework is introduced for short-term bike-sharing demand (as one type of point-based transportation data) prediction at the station level. By incorporating historical demand patterns, meteorological data, and a dynamically evolving semantic adjacency graph, the model jointly predicts bike pick-ups and drop-offs, addressing imbalances in bike-sharing systems.

Additionally, a novel geographic-semantic graph-based model for commuting flow prediction (as an origin-destination-based transportation data) was introduced. By leveraging Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs), the model captures both geographic adjacency and semantic connectivity—such as metro network linkages—providing a more precise estimation of urban mobility patterns. The proposed approach outperforms existing models, improving accuracy in forecasting commuting demand across city zones.

By integrating different data sources and learning frameworks, this research provides a holistic methodology for modeling urban dynamics. The findings contribute to more accurate energy simulations, better-informed transportation planning, and the development of smarter, more sustainable cities.