Lin, Xuhui; Lu, Qiuchen; Broyd, Tim; Cheng, Tao; Zhang, Xianghui; Erfani, Tohid; Tran, Trung Hieu; (2024) Enhancing Urban Flood Response: Traffic Flow Prediction with Field Theory-Inspired Physics-Informed Graph Neural Network. Presented at: 3rd International Conference: Natural Hazards and Risks in a Changing World Addressing Compound and Multi-Hazard Risk, Amsterdam, Netherlands.

Preview

Text 002_poster.pdf - Submitted Version Download (5MB) | Preview

Abstract

In the context of urban transportation, the industry is facing escalating challenges due to the impact of climate change, particularly the increased severity of flood disasters that disrupt traffic flow and demand advanced predictive models for effective urban planning and disaster response. Current methods are limited in their ability to dynamically capture the complex alterations in transportation networks during floods, lacking real-time adaptability to road closures and emergency evacuation routes, and they often fail to integrate the physical laws governing traffic flow, leading to less reliable predictions. This paper addresses these gaps by introducing a novel traffic flow prediction model, the PINNs-GNN, which combines the strengths of Physics-Informed Neural Networks (PINNs) and Graph Neural Networks (GNNs) to provide a more accurate and robust approach to predicting traffic flow during flood events. The most significant innovation of this study lies in its integration of physical equations and GNNs, abstracting the transportation network as a graph and introducing a diffusion equation within the GNN to enhance physical consistency, while also defining a field effect term to represent the impact of floods, extracted from node and edge features by the GNN, thus improving the model's adaptability to dynamic environments. The proposed model demonstrates superior accuracy, real-time performance, and robustness compared to existing methods through experiments on real flood event datasets, effectively capturing spatiotemporal dynamics and reducing errors. The PINNs-GNN model's ability to embed physical laws into the prediction process results in more reliable traffic flow predictions, especially with limited data, and its generalization to new situations, such as dynamic flood conditions, reduces the dependency on extensive historical data. This method is particularly beneficial for urban planners, disaster response teams, and traffic management authorities, offering them a powerful tool to anticipate and mitigate the impacts of flood events on traffic flow, aiding in informed decision-making regarding resource allocation, route planning, and evacuation strategies, and ultimately enhancing the resilience of urban transportation systems.

Download activity - last month

Export as CSVJSONXML

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item