AI predicts rail delays

At the University of Illinois at Urbana-Champaign (UIUC), scientists used real British Railways data and an artificial intelligence model to better predict delays in rail networks. The research results are presented at the IEEE 2020 International Conference on Intelligent Transport Systems.

Over the past 20 years, the number of passengers traveling on the British rail network has nearly doubled to 1.7 billion annually. Obviously, UK residents rely on rail links, and traffic delays could disrupt many.

“We wanted to study this problem using our experience with graph neural networks,” explains Hi Tran, a UIUC Aerospace Engineering Department member. “This is a special class of artificial intelligence models that focus on data represented in graph areas.”

A Graph Neural Network (GNN) is a type of neural network that directly works with a graph’s structure. A typical application for GNN is node classification. A graph neural network concept was first proposed in 2009 in a work that augmented existing neural networks to process data represented in graph domains.

A graph is a data structure made up of two components: vertices and edges. The graph G is described by the set of vertices (nodes) V and edges E.



Using GNN allows you to work with graph data without preprocessing. This approach allows you to preserve the topological relationships between the nodes of the graph.

Scientists have applied a space-time graph convolutional network model to predict delays within one of the British rail network’s busiest parts.

“Compared to other statistical models, this model is superior in predicting delays of up to 60 minutes,” Tran emphasizes.

At the IEEE 2020 International Conference on Intelligent Transport Systems, the study “Predicting Railroad Delays Using Convolutional Networks with Space-Time Graph” written by S.V. Jacob was presented. Heglund, Panukorn Taleongpong, Simon Hu, and Hai T. Tran.

Google News button