Introduction
TenneT is a Transmission System Operator (TSO) that operates, designs and manages the high-voltage electricity grid in the Netherlands and large parts of Germany with a fundamental purpose of reliably and efficiently transporting electric power at all times. To achieve this goal, several processes such as grid security analysis, capacity calculations, congestion management, grid expansion studies, just to name a few, are performed on a regular basis at TenneT and in almost all these technical processes, power flow simulations play an integral role. Furthermore, with an increasing demand to optimize the electricity grid which is being pushed to its limits due to the energy transition and rapid expansion of the energy sector, faster power flow calculations are crucial to ensure grid reliability and efficiency.
AC Power Flow
Due to the fact that the electricity grid is predominantly based on Alternating Current (AC), the underlying physics of the grid is inherently non-linear. The AC power flow problem is hence a set of simultaneous non- linear equations solving which determines the state of the grid. The conventional and perhaps the most popular method used to solve the AC power flow problem is the Newton-Raphson method. Even though this method works well and leads to the most accurate solution, it is computationally expensive and hence not suitable when the problem size is very large. To circumvent this difficulty, the AC power flow problem is often linearized based on a set of strong assumptions to form what is called the DC power flow problem which can be solved much faster, although with a compromise in accuracy. In recent times however, machine learning has emerged as a promising tool to solve non-linear problems and literature claims that machine learning models approximate the AC power flow problem better than the DC power flow method. As the electricity network is inherently a graph, there is currently a lot of interest in the community to solve the AC power flow problem using Graph Neural Networks (GNNs). As GNNs are still a nascent field of research, this is an exciting time to use them for solving power grid problems.
Thesis Project
At TenneT, we are looking for an interested master student to work on solving the AC power flow problem using graph neural networks for real grids. The master thesis will be part of the GridOptions project which aims to develop a decision support tool that recommends topological remedial actions to the operators in the TenneT control room to alleviate congestion in the grid. The GNN algorithms to be developed for solving AC power flow problems for the onshore Dutch TenneT grid will be used to train a reinforcement learning agent that finds remedial actions to reduce grid congestion.
The thesis project can be done as an internship at TenneT with daily supervision by TenneT employees and the opportunity to get to know the business. But if the student prefers to work purely at the University with a focus more on research, that can be arranged as well. We are open to discussion.
Approach
1. Literature review on graph neural networks in the context of power flow analysis.
2. Solving AC power flow for small test networks using GNNs.
3. Solving AC power flow for the onshore Dutch TenneT grid using GNNs.
4. Benchmarking a few promising GNN architectures for solving power flow problems.
Requirements
1. Background in computer science, mathematics, physics, or engineering.
2. Good programming skills in Python, knowledge of libraries for deep learning is a plus.
3. Knowledge on power flow analysis or network calculations is helpful, but not required.
4. Interest in the energy transition is a big plus.
The thesis project can be done as an internship at TenneT with daily supervision by TenneT employees and the opportunity to get to know the business. But if the student prefers to work purely at the University with a focus more on research, that can be arranged as well. We are open to discussion.
Contact: Yuliya Shapovalova