My first try of defining and describing the Autonomous Grid
Disclaimer: This article was originally published on the AI-native Power Grid Channel on Substack, but adapted and refreshed for the Energy Central Community.
Last week I was at Enlit Europe in Vienna. I had three great days. Great discussions, great talks, great reunions with old and new friends. And, of course, some great Schnitzels. If there is one thing that stuck with me, one word I heard or saw most, then it was “autonomous”. Autonomous grid, autonomous grid operations, autonomous control room. All the big booths used it. And many of the smaller players, too.
But I did not hear a clear definition. Everybody uses the term autonomous grid. And once again, as with many other buzzwords, not everybody seems to mean the same thing. So let me try to give my definition. My understanding. My vision. This is part one of two. I focus on how I see the autonomous grid and where control should sit. In part 2, I’ll look at what it means for the grid, for grid operations and the control room.
The Grid is already autonomous today
Large parts of the grid already act autonomously today. And they have done so for mant years. Think about protection tripping a line in milliseconds. Without asking anyone. Think about frequency response, automatic generation control, tap changers or fault isolation on feeders: they all work without an operator having to press a button.
Interestingly, do we call it an autonomous grid? No, we call it grid automation. And we trust it. Because it is simple. It does one clear job. It is tested. And we understand it.
So what is the new thing? Machines now start to take decisions that formerly needed the judgments of an operator or planner. Think about switching plans. Think about which flexibility to dispatch. Think about updating line ratings or operational limits. Those kind of decisions. Decisions that depend on the real and correct picture of the whole grid. Decisions that have many possible answers and outcomes. And where the right one is not always obvious. Now we talk about autonomy.
The Autonomous Grid
Let me try a definition. My definition.
An autonomous grid is a grid where defined decisions are taken and executed by machines, inside boundaries and limits that human operators set and own. The machines act on a trusted picture of the grid state, and they fall back to a safe mode when the state or the limits cannot be trusted.
Let’s dive into it. Especially the highlighted or bold parts.
Defined decisions: Autonomy is given per type of decision (and most probably depending on additional conditions), not to the grid as a whole. A DSO can define that a machine can manage voltage on LV feeders. But, maybe, changes to the topology the DSO wants to keep with the operator. So the question “is our grid autonomous?” makes little sense. The question should be: which decisions do we let machines take,. And why exactly these.
Boundaries and limits that human operators set and own. The machine acts inside a defined operational envelope. The human operator defines the boundariess, policies or limits in this envelope. And (s)he is responsible for it, too. It is about accountability and where accountability sits. Without it, nobody can say who is accountible when something goes wrong.
A trusted picture of the grid state. A machine’s decisions can only be that good as the data the machine sees. If the topology is wrong, if the LV grid is not observable, or if the model is old, then even the best AI model will reason and decide wrongly. Fast. With full confidence. But wrong. This is why the DSO4DT report I mentioned last week matters so much. Most DSOs are still working with building this picture.
A safe fallback. Every autonomous function must know what to do when its input is bad, when the data quality is bad, when communication is lost, or when it reaches the edge of its boundaries and limits. This has to be there from day one. Not day two or three.
Levels of Autonomy
For me, it helps to think in levels. Like we did for the Journey of Trust.
Recommendation: The machine recommends. The operator decides and acts.
Operator approved: The machine decides how to act. The operator approves.
Automatic inside limits: The machine decides and acts on its own inside the defined boundaries, policies and limits. The operator supervises.
Automatic with exception handling: The machine decides and acts, even in the most unusual situations. The machine delegates to the operator only when it reaches the edge of its limits.
To be honest, most of the demos I saw in Vienna were level 1. Some few demos 2. I don’t think this is a bad thing. We all learn while we walk. And these are the right first steps to build the trust we will need later. But maybe we should not call level 1 for autonomous and keep that for level 3 and 4?
Where does Control sit
My initial take: Decisions should be taken as close to the grid as possible. And the layer above should set the limits for it and orchestrate the policies.
System level, the TSO, from reserves to day-ahead planning. Balancing, and the coordination or integration between TSO and DSO.
Regional level, the DSO control centre, minutes to hours. Topology optimisation, congestion management, dispatch of flexibility. Sets the limits for the edge.
Edge level, the secondary substation and the feeder, seconds to minutes. LV voltage control, local flexibility, operating limits for customers, fault isolation. Sits local because of the amount of data, and because communication is not always reliable. Acts inside the defined limits.
Device level, milliseconds to seconds. Protection, inverter behaviour, droop. This is already autonomous. Stays local, because it has to work also without communication.
Looking at it, we see two paths. Limits are pushed down. Information about the grid state are pushed up. When the edge loses the connection to the control center, it keeps working with the last valid limits, or with a safe default. Why? The grid stays resilient when communication fails. And a cyber attack on the operations center can do much less damage in the field.
So we move control closer to the edge. But this does not mean the central system disappears.
TSO: the EMS stays mostly centralized. A meshed transmission grid cannot be split into independent zones. For now.
DSO: the ADMS changes its role. It does less of the real-time switching. Less of the grid control. It holds the trusted state of the grid. It calculates the limits and publishes these policies to each zone. It coordinates the borders between zones. It orchestrates the distributed edge intelligence at the zones. And it keeps the record of what was decided and why.
What comes next
In part two I look at what this changes in the control room. And how the autonomous grid impacts grid operations. Stay tuned!