Power Systems · AI · Research
Practical AI-assisted analysis for power-system models and data.
Modern power-systems analysis is changing faster than the methods used to study it. System behaviour has got harder to study, and AI has got useful enough to do real work. This site is about how those two meet, in practice - what the methods actually do, what they catch, and where they fall short.
Calm, reasoned, nuanced - not the breathless “AI will transform everything” version. The practical version, where the work gets more thorough, more consistent, and faster, without losing engineering judgement. Most of the analysis uses publicly available test networks - the IEEE 14-bus, IEEE 39-bus, and the NESO Reduced GB model - so the work can be reproduced and pulled apart.
Where to start
If you are looking at this for the first time, the case studies are the most substantial reads - longer technical pieces showing the methods in action. The blog is shorter, opinion-led, and updated more often. The interactive map is the visual centrepiece, and the NESO data tools are the public-data side of the work.
Open to discussion and collaboration. Reach me at stevesommerville_website@proton.me.
When working in research it is often beneficial to use publicly available test networks. This helps other researchers replicate findings and avoids IP and NDA concerns. The work uses publicly available models and data wherever possible. For this project I am currently focusing on three main standard models:
- IEEE 14-bus model - a simple, well-established model used for testing concepts and ideas. It is a simple test network, intended to have enough detail in to let a simulation engine think and solve problems, but it is nowhere near complicated enough for a real network.
- IEEE 39-bus model (aka New England 10-machine) - a more detailed model representing a larger system with multiple voltage levels and large synchronous machines. Big enough to show interesting transmission-level behaviour, small enough to navigate easily and debug when things don’t work.
- NESO Reduced GB model - produced by the UK system operator; a reduced / simplified model of the whole GB system, split into 28 zones. A flexible generation mix, complex transmission, HVDC interconnectors and realistic transmission characteristics.
With the public models, the main areas of interest and research cover the existing and new challenges emerging with power systems analysis, and explores how AI can help us investigate them.
- Loadflow and network analysis
- System strength and short-circuit analysis
- Reduced network models and screening studies
- Synchronous inertia and frequency response
- Grid Forming Inverters
- RMS / EMT dynamic studies
- Reactive power and voltage stability
- HVDC and interconnector behaviour
- AI tooling for power-system workflows
Latest from the blog
Notes on AI and power systems.
Working notes and longer-form thoughts on AI-assisted power systems analysis. Written in plain English, with the engineering rigour preserved.
Using AI to Introspect an EPRI PSCAD Grid Forming Inverter Model
Using AI to Introspect a PSCAD Grid Forming Inverter Model. An exploration of the EPRI GFM publicly available model.
About
I am an experienced power systems engineer with 25+ years in HV and EHV power system design and analysis. I am currently completing a PhD in Power Systems Stability at Brunel University London looking at voltage stability, sub-synchronous oscillation, damping controls for IBRs, and the modelling of system strength in inverter-dominated networks. Alongside this, I have developed a keen interest in AI and how it will transform how we do power systems analysis. I am also the Managing Director of Aurora Power Consulting.