Graph Neural Networks for Travel Distance Estimation and Route Recommendation Under Probabilistic Hazards

TL;DR

The paper addresses the need for fast, scalable estimation of travel distances and routes under hazard conditions in large road networks. It introduces GNN-SDE, a graph neural network framework that learns to approximate single-source shortest distances and then derive routes via a predecessor-based path-finding mechanism. Through experiments on synthetic graphs with up to 100k nodes and a coastal urban case study (Manhattan and Hillsborough County) for flood evacuation planning, the method achieves high accuracy (MAPE near or below 2%) and substantial speedups over Dijkstra as graph size grows. The results demonstrate the framework's potential for emergency planning and disaster response in large, dynamic networks and highlight its applicability to flood-risk analysis and shelter routing. The study also suggests avenues for extending to multi-modal networks and incorporating uncertainty quantification.

Abstract

Estimating the shortest travel time and providing route recommendation between different locations in a city or region can quantitatively measure the conditions of the transportation network during or after extreme events. One common approach is to use Dijkstra's Algorithm, which produces the shortest path as well as the shortest distance. However, this option is computationally expensive when applied to large-scale networks. This paper proposes a novel fast framework based on graph neural networks (GNNs) which approximate the single-source shortest distance between pairs of locations, and predict the single-source shortest path subsequently. We conduct multiple experiments on synthetic graphs of different size to demonstrate the feasibility and computational efficiency of the proposed model. In real-world case studies, we also applied the proposed method of flood risk analysis of coastal urban areas to calculate delays in evacuation to public shelters during hurricanes. The results indicate the accuracy and computational efficiency of the GNN model, and its potential for effective implementation in emergency planning and management.

Figures (11)

• Figure 1: Message passing with both node and edge features
• Figure 2: The original node and edge features are initially transformed into feature embeddings. These embeddings are then propagated through the GNN layers, where they are concatenated using a projection vector. Subsequently, the features of the source node and destination node are concatenated and passed through a MLP to make the final prediction.
• Figure 3: Shortest distance prediction under different graph sizes. The prediction from GNN-SDE is compared with Dijkstra's Algorithm under three different graph size, including 1k, 4k, and 50k.
• Figure 4: Relationship between error distribution versus different graph sizes. The blue distributions are error histograms tilted by 90 degrees. The red line connects the mean absolute percentage errors (MAPE) and mean absolute errors (MAE), respectively, between graph sizes. The taller blue distribution refer to wider range of errors. The peak of error histograms are mostly at the lowest-error bin.
• Figure 5: Computational time comparison between graph neural network and Dijkstra