Eigenvalue analysis of the Reduced GB Network model
Introduction
In this post I present a more detailed Case Study of carrying out an Eigenvalue analysis using the NESO published GB Reduced grid model. The aim of the exercise is to show how we can use an AI blended approach to what is otherwise quite a slow tedious set of studies and analysis patterns.
Eigenvalue Analysis
Eigenvalue analysis is one of those areas, that many engineers do not interact with, as it is historically aimed at whole system users, rather than for smaller network analysis. Today it has fallen slightly out of favour, and is more often replaced with EMT based time domain analysis and impedance frequency scans. There are multiple reasons for this, which I will cover in a separate post. The short version is though, that the eigenvalue approach is based on a positive sequence assessment of control systems that are ‘open’ (whiteboxed), and a positive sequence representation does not capture all of the dynamics. That said, eigenvalue is a very useful screening tool to identify general problem areas.
One of the slightly tedious parts of Eigenvalue analysis (actually any SSO study) is that you need to linearise the system around a specific operating point. So on a big networks with lots of scenarios, this gets a difficult to do by hand. Traditionally we would often write a script to do this, but writing and tweaking a script continually is time consuming. I was able to just setup an MCP interface to Powerfactory and just instruct the AI in natural language about the cases I wanted to see and analyse, it could also use numpy directly to carry out more complex aggregated analysis and look for trends, rather than manually export lots of plots and inspect the individually.
GB Reduced Model
In this study I used the reduced GB Grid model available from NESO, as an interesting case to play with. It is a large enough to be interesting, and has a well-known Scotland-England mode at about 0.6 Hz, and the model has a nice mix of generators and potential loading and generation patterns. Obviously the model is very simplified over the actual UK system, so I was hoping some of the general trends would be identified.
The things I asked it to explore were:
1) Identify any unstable or poorly damped operating modes in normal operating model and light load (summer) conditions.
2) Network outages cases that significantly impact the stability.
3) Different renewable penetration levels, reducing classical system strength and inertia.
4) Future loading cases for AI data centres in Scotland for high and low wind cases.
5) Visibility of controllers for IBR generation - to represent whitebox vs blackbox performance.
Summary
The overall report is attached, all generated by the AI. Which provides the detailed narrative, scenarios, study cases and results. As with all AI analysis, the actual model is largely irrelevant, what is impressive is the ability of the AI to carry out a pretty detailed analysis that is fairly niche and specialist area. I had to do a reasonable amount of directing, and dig it out of a couple of rabbit holes, but overall it was a comprehensive set of results.