Master Thesis: Learning Air Quality Simulation with Fourier Neural Operators

Background

Fourier Neural Operators are layers of neural networks capable of capturing global information of physical systems which exhibit periodic behavior. This is helpful when dealing with solutions to (partial) differential equations which model physical systems like fluid dynamics. Zongyi et al. (https://arxiv.org/abs/2010.08895) have demonstrated, that a neural network using Fourier Neural Operators can learn the behavior of physical systems modelled via Computation Fluid Dynamics (CFD) with sufficient accuracy while being much faster in their predictions.
The objective of this thesis is to investigate whether Fourier Neural Operators can be used to learn urban wind fields – or even dispersion of pollutants in an urban environment – which were calculated via CFD simulations.

Tasks

1. Obtain an understand of the workings as well as the opportunities and limitations of Fourier Neural Operators (FNOs)
2. Implement a Neural Network based on FNOs and train it using wind field/air quality data
3. Evaluate the results according to scientific principles

Skillset

• Good python skills and experience with numpy (or PyTorch)
• Some knowledge of and preferably experience with machine learning/neural networks and architectures like CNNs

If you are interested in this topic, please contact Paul Tremper (tremper@teco.edu)