Seminar speaker: Naghmeh Shafiee

Abstract

    The rapid growth of urbanization and the development of smart city technologies have generated vast amounts of spatiotemporal data, presenting both challenges and opportunities for urban planning and resource management. Accurate forecasting of various urban phenomena, such as traffic congestion, meteorological variables, hydrometric data, and energy consumption, is crucial for efficient and sustainable smart city operations. By modeling urban data as a graph, where nodes represent locations and edges capture spatial dependencies, Graph Neural Networks (GNNs) can effectively capture the intricate relationships between different urban elements. In this research, we propose novel spatiotemporal forecasting approaches in smart cities using GNNs. The proposed framework leverages GNNs to learn the dynamic spatiotemporal patterns from historical data and predict future values. To demonstrate the effectiveness of the proposed approaches, we apply them to smart city applications such as traffic speed prediction and water flow forecasting. We utilize real-world datasets collected from various sensors and urban monitoring systems. The graph representation captures the spatial layout and interdependencies between different traffic or water system nodes responsible for capturing sequential dependencies. The proposed models also include other deep learning modules responsible for capturing sequential dependencies, enabling accurate predictions even in complex urban environments. Experimental results show that our approach outperforms traditional forecasting methods, such as time-series models and conventional deep-learning architectures, in terms of predictive performance metrics.

Examining Committee

• Dr. Yann-Gael Gueheneuc (Chair) 
• Dr. Charalambos Poullis, Ursula Eicker, Zachary Patterson (Supervisor)
  
• Dr. Adam Krzyzak (Examiner)
• Dr. Yann-Gael Gueheneuc (Examiner)
• Dr. Nizar Bouguila (External Examiner)