Graph Neural Networks for Traffic Prediction

How can we predict traffic in connected systems such as road networks, and how can we use these predictions to place charging stations for electrical vehicles ? Graph neural networks are currently the most promising approach for studying these question.
Master

Graph neural networks represent an important development in deep learning as they allow predictions on unstructured input data such as road networks. While they tend to be very performance hungry, advances in the last three years have made it possible to use them for predictions on large scale networks. The goal of this thesis is to use them in order to model Norwegian road traffic, specifically for electric vehicles. This information can be used to predict optimum placement of charging stations.

Goal

The goal is to build a graph neural network using the PyG library and use Norwegian public data for realistic travel predictions of electrical vehicles in Norway.

Learning outcome

Proficiency with PyG
Understanding of graph neural networks
Famillarity with traffic modeling

Qualifications

Familiarity with deep learning and.
Knowledge of Python and PyTorch.
Interest in the road traffic

Supervisors

  • Johannes Langguth
  • Konstantin Pogorelov

Contact person