Graph Neural Networks for Wind Power Modelling

This project aims to develop new machine learning models that can help wind energy companies design wind farms more efficiently and at lower cost. Today, planning a wind farm requires running large, high-resolution computer simulations to understand how wind flows around turbines and how much power a proposed layout can produce. These simulations are accurate but extremely slow and financially expensive, especially when companies need to test a large number of different layouts or new locations to optimize the power production and transmission. Our research will create advanced prediction models that learn directly from existing wind flow simulations and real atmospheric data. By using graph-based neural networks and advanced generative methods, our models can adapt to different turbine layouts and geographic domains, while also estimating uncertainty in the predicted wind and power output. This makes them more flexible and scalable than current tools. For Veer Renewables, the partner organization, this project will produce a new commercial product that can deliver fast, early-stage assessments for new wind farm layouts and new geographic domains, reducing reliance on costly physical simulations.

Faculty Supervisor:

Adam Monahan;Slim Ibrahim

Student:

Partner:

Veer Renewables

Discipline:

Mathematics

Sector:

Professional, scientific and technical services

University:

University of Victoria

Program: