--- title: "Twins on the grid: fast physics, slow assets" date: "2023-05-01T05:00:00.000Z" categories: "Digital twins" canonical: "https://xraph.com/writing/twins-in-the-energy-grid" --- # Abstract Power systems combine millisecond electrical dynamics with equipment measured in decades. A twin that picks one timescale is useless for the other, and reconciling them is a modelling decision. ## Two clocks Electrical behaviour on a transmission network happens in milliseconds. Protection operates faster than a person can react, and phasor measurement units sample tens of times per second. The assets carrying that behaviour age over decades. A transformer's insulation degrades on a curve measured in years, and the useful question about it is whether it will survive the next summer peak. Both are legitimate uses of the word twin and they need different models. A real-time dynamic model is a solver over a network topology. An asset health model is a statistical model over slow degradation with sparse ground truth, because failures are rare by design. ## Where the coupling is They are not independent, which is what makes this interesting. Thermal loading drives insulation ageing, so the fast model produces the input the slow model integrates. A transformer that ran hot for three days in August has aged more than the calendar says. The clean way to express that is a hierarchy of models exchanging aggregates rather than raw states, which is the standard framing in the simulation literature [2]. The fast model emits a thermal duty summary per interval, the slow model consumes summaries. Nothing tries to run a decades-long simulation at millisecond resolution. ## The data problem is different from oil and gas Grid data is comparatively well structured, since the equipment is instrumented for protection and settlement, and both have regulatory requirements behind them. Identity is less of a mess because assets have registered identifiers. The volume is the issue instead. Sub-second measurements from thousands of points produce a data rate where storage strategy becomes a design constraint rather than a detail. What I have seen work is keeping full resolution for a short window, aggregating aggressively after that, and retaining full resolution indefinitely only around events. That is a lossy decision and it needs to be made deliberately, because the analysis somebody wants in two years is the one whose data you downsampled. ## Validation is the honest difficulty A model that predicts failure has very little to learn from. Well-maintained equipment fails rarely, the failures that do occur are heterogeneous, and the interventions that prevented failures are not recorded as counterfactuals. This is the scaling problem the computational science community has named directly: moving from artisanal, individually calibrated twins to industrial fleets requires uncertainty quantification and validation methodology that mostly does not exist yet [3] [1]. In practice that means a model output without an uncertainty estimate should not be shown to an operator, because a point prediction invites a decision that the evidence does not support. ## The line you do not cross Protection and control are safety functions with their own assurance regime [5]. A twin informs planning, maintenance and operator awareness. It does not sit in a protection path, and anyone proposing otherwise should be asked which safety integrity level the model has been assessed against [4]. Keeping that boundary explicit in the architecture is worth doing early, because it constrains where the twin can be deployed and who has to approve it. # References 1. Adil Rasheed, Omer San, Trond Kvamsdal, "Digital Twin: Values, Challenges and Enablers From a Modeling Perspective", IEEE Access, vol. 8, pp. 21980-22012, 2020 2. Stefan Boschert, Roland Rosen, "Digital Twin: The Simulation Aspect", Mechatronic Futures, Springer, 2016 3. Steven A. Niederer, Michael S. Sacks, Mark Girolami, Karen Willcox, "Scaling Digital Twins from the Artisanal to the Industrial", Nature Computational Science, vol. 1, pp. 313-320, 2021 4. Edward A. Lee, "Cyber Physical Systems: Design Challenges", IEEE International Symposium on Object and Component-Oriented Real-Time Distributed Computing (ISORC), 2008 5. International Electrotechnical Commission, "Functional Safety of Electrical, Electronic and Programmable Electronic Safety-Related Systems", IEC 61508 --- *Unpublished draft. Not peer reviewed and not submitted to a venue.*