Power Reference

---
name: power-reference
description: "Use when checking a power-engineering anti-pattern, looking up a voltage, rating or system value, asking what actually moved (the inverter-dominated grid and datacenter load growth, verified August 2026), finding the books, or needing a picker and the orientation questions to ask before starting grid software work. Companion to the other power-engineering skills."
---

# Power Engineering: Anti-Patterns, Numbers, What Moved, and Canon

> **Part 5 of 5** of the *Power Engineering* reference (plugin `power-engineering`), covering §13–§18. Sibling skills: `power-ac-fundamentals-generation-and-grid` (§0–§3), `power-system-analysis-and-protection` (§4–§5), `power-scada-ems-and-protocols` (§6–§7), `power-inverters-storage-markets-and-datacenters` (§8–§12). Section numbers are shared across the set; a reference written as §N → `skill` points into that sibling skill.
>
> **Currency:** Circuit theory and power system analysis are settled; the inverter-dominated grid and datacenter load growth moved fast. See §15 below for both, dated.

> **Scope.** Complements an electrical-engineering reference, which is circuit-level —
> components, op-amps, PCB layout, signal integrity. ⚠️ **This is grid scale: kilovolts
> and megawatts, and the software that runs it.**
>
> **⚠️ GOTCHA** boxes mark what kills people, destroys equipment, or causes blackouts.
>
> **The three ideas that organize the whole field:**
> 1. **⚠️ Generation must equal load, continuously, everywhere.** The grid stores almost
>    nothing. **Frequency IS the balance signal** — it falls when load exceeds generation
>    and rises when it doesn't, in real time, across a continent (§1.4 → `power-ac-fundamentals-generation-and-grid`).
> 2. **⚠️ Reactive power is not "wasted" power — it's a separate commodity that must also
>    balance, and it's local.** Real power flows across a system; reactive power doesn't
>    travel well, so voltage is a local problem and frequency is a global one (§1.2 → `power-ac-fundamentals-generation-and-grid`).
> 3. **⚠️ Protection is the fastest software in the system and it acts on trust.** A relay
>    decides in milliseconds to disconnect equipment, and it must be right — a false trip
>    causes an outage, a missed trip destroys equipment or kills someone (§5 → `power-system-analysis-and-protection`).

---

## §13. Anti-Patterns

| Anti-pattern | Why |
|---|---|
| Treating reactive power as waste | ⚠️ **It holds voltage up. The question is where it's sourced** (§1.2 → `power-ac-fundamentals-generation-and-grid`) |
| Assuming voltage is a system-wide quantity | ⚠️ **Voltage is local; frequency is global** (§1.2 → `power-ac-fundamentals-generation-and-grid`) |
| Undersized neutral in a nonlinear-load facility | ⚠️ **Triplen harmonics add rather than cancel** (§1.3 → `power-ac-fundamentals-generation-and-grid`) |
| Ignoring per-unit conventions | Every number will confuse you (§1.5 → `power-ac-fundamentals-generation-and-grid`) |
| Relaxing tolerance when power flow won't converge | ⚠️ **Non-convergence may mean genuine infeasibility** (§4.1 → `power-system-analysis-and-protection`) |
| Acting on raw telemetry rather than the state estimate | That's what state estimation is for (§4.2 → `power-system-analysis-and-protection`) |
| Overcurrent protection settings assuming synchronous fault current | ⚠️ **Inverters supply ~1.1–2× rated. The relay may not see it** (§5 → `power-system-analysis-and-protection`, §8.1 → `power-inverters-storage-markets-and-datacenters`) |
| Treating GFM inverters as a solved replacement for inertia | ⚠️ **~12% of UK contracted inertia by 2026; the rest is rotating mass** (§8.2 → `power-inverters-storage-markets-and-datacenters`) |
| Assuming an air gap protects OT | ⚠️ **It's mostly mythical. Segment properly** (§11.2 → `power-inverters-storage-markets-and-datacenters`) |
| Applying IT patch cadence to OT | ⚠️ **Availability outranks confidentiality here** (§11.2 → `power-inverters-storage-markets-and-datacenters`) |
| Assuming a protocol authenticates the sender | ⚠️ **Most in §7 → `power-scada-ems-and-protocols` don't** (§7 → `power-scada-ems-and-protocols`) |
| Phasor-domain simulation for inverter control studies | ⚠️ **Misses the fast dynamics. Use EMT** (§11.3 → `power-inverters-storage-markets-and-datacenters`) |
| Sloppy time synchronization | ⚠️ **Sequence-of-events analysis becomes worthless** (§11.3 → `power-inverters-storage-markets-and-datacenters`) |
| Alarm design that floods during real events | Documented blackout escalation factor (§6 → `power-scada-ems-and-protocols`) |
| Greedy battery dispatch ignoring state of charge | ⚠️ **Empty when the obligation lands** (§9 → `power-inverters-storage-markets-and-datacenters`) |
| Storage optimization without degradation cost | Destroys the asset profitably (§9 → `power-inverters-storage-markets-and-datacenters`) |
| Optimizing PUE while ignoring workload efficiency | ⚠️ **PUE cannot see software at all** (§12.1 → `power-inverters-storage-markets-and-datacenters`) |
| Air cooling at modern AI rack density | ⚠️ **It doesn't work above ~40 kW/rack** (§12.2 → `power-inverters-storage-markets-and-datacenters`) |
| Assuming grid capacity is available on your timeline | ⚠️ **4–10 year interconnection queues** (§15.2) |

---

## §14. Numbers

```
FREQUENCY   60 Hz (NA, parts of Asia/LatAm) · 50 Hz (most of the world)
⚠️ Normal band ±0.05 Hz typical; under-frequency load shedding begins ~59.3 Hz (60 Hz sys)
Governor droop 4–5%

VOLTAGE LEVELS
Transmission 115 / 138 / 230 / 345 / 500 / 765 kV
Distribution 4.16 / 12.47 / 13.8 / 34.5 kV
Utilization 120/240 V (NA residential) · 208Y/120 · 480Y/277 V (NA commercial)
230 V / 400Y/230 V (most of the world)

THREE-PHASE
Wye: V_line = √3 V_phase · Delta: I_line = √3 I_phase
P_3φ = √3 · V_L · I_L · cos θ

POWER FACTOR
PF 0.7 draws ~43% more current than PF 1.0 for the same real power

PROTECTION (ANSI)
50/51 overcurrent · 87 differential · 21 distance · 27/59 volt · 81 freq · 79 reclose
⚠️ IEC 61850 GOOSE delivery requirement ~4 ms
⚠️ Inverter fault current ~1.1–2× rated (vs many× for synchronous machines)

DATACENTER
PUE: 1.0 ideal · ~1.1 hyperscale · 1.5–2.0 older enterprise
⚠️ AI rack density to ~140 kW (vs 5–15 kW traditional)
AI site draw 100–750 MW  ·  ⚠️ a 500 MW site at 90% util ≈ 3.9 TWh/yr
```

---

## §15. What Actually Moved — verified August 2026

### 15.1 The inverter-dominated grid
**⚠️ This has moved from academic concern to operational reality.** One trade source puts
it directly: **in 2026 the conversation moved from IEEE papers to utility boardrooms
because high-profile outages showed how fast frequency collapses when inverter-dominated
regions lose a transmission tie.** ⚠️ **The Iberian Peninsula event of 28 April 2025 has an
official expert panel report (October 2025) and is now a standard reference in the
protection literature — worth reading directly rather than through commentary.**

**Instantaneous non-synchronous generation penetration has reached 60–80% in many small
power systems**, which makes this a present-tense engineering problem in those systems,
not a future one.

**⚠️ The honest state of GFM deployment is the §8.2 → `power-inverters-storage-markets-and-datacenters` gotcha and I'd weight it heavily**:
the UK's procurement outcome — **~12% of contracted inertia from GFM by 2026, the rest from
synchronous machines and mainly synchronous condensers** — is a market revealing that
**adding back rotating mass has so far beaten synthetic inertia on cost and confidence.**
**Germany's 2026 technology-differentiated procurement is the test to watch.**

**Standards**: **IEEE 2800** and NERC guideline revisions now explicitly address IBR fault
performance, and ⚠️ **grid-forming capability is increasingly asked about in 2026
interconnection requests.**

### 15.2 ⚠️ Datacenter load growth — the biggest grid story for this audience
**The structural picture, which is consistent across sources:**
- **Data centers used roughly 415 TWh in 2024, ~1.5% of global demand**, with **IEA
  projections around 945 TWh by 2030 (~3%)** — ⚠️ **roughly 15% annual growth, far above
  overall electricity demand growth.**
- **⚠️ Concentration, not aggregate growth, is the actual problem.** A single training
  facility can draw **several hundred MW to over 1 GW**, and **clustering in Northern
  Virginia, Texas, Ireland and parts of East Asia** is what stresses specific systems.
- **⚠️ The timescale mismatch is the core tension**, and it's well put in one source:
  **datacenter demand moves at the speed of capital markets, while grid infrastructure
  moves at the speed of permitting, procurement, and construction.**
- **Interconnection queues**: ⚠️ **reported delays of 4–10 years**, with median time from
  request to commercial operation **over five years.** ⚠️ **And a sobering base rate — of
  capacity that submitted interconnection requests 2000–2019, only 13% had reached
  commercial operation by end-2024; 77% was withdrawn.**
- **⚠️ Physical bottlenecks are transformers, substations, switchgear and transmission
  capacity** — long-lead-time hardware, not software.

**⚠️ The responses, and the one that matters most to software people:**
- **"Bring your own power"** — on-site generation, behind-the-meter, microgrids, fuel
  cells; ⚠️ **mandated in some markets, and operators are moving from PPAs to directly
  funding generation.**
- **⚠️ Flexibility as an interconnection strategy.** **PJM's board announced a
  connect-and-manage framework in January 2026** under which **incremental large load that
  doesn't bring its own generation may be subject to curtailment** ahead of
  pre-emergency demand response. ⚠️ **This is the important one: a datacenter that can
  modulate its consumption can connect sooner than one that can't — which turns workload
  flexibility from an efficiency nicety into a capital-deployment lever.** **If you write
  schedulers, that is your problem now.**

> **⚠️ GOTCHA — the numbers in this subsection are forecasts from interested parties and
> should be read as such.** ⚠️ **Sources include consultancies, real estate firms,
> equipment vendors, grid-analytics companies, and investment banks — all of whom benefit
> from the growth narrative.** **Projections varied noticeably across the sources I saw**
> (2030 datacenter consumption at ~945 TWh from IEA-derived figures, with other estimates
> running higher; US demand forecasts differ by tens of GW). **The physical constraints —
> queue lengths, transformer lead times, PJM's actual policy — are better attested than
> the demand forecasts.** ⚠️ **Treat direction as solid and magnitude as contested.**

---

## §16. Books

| Author | Work | Why |
|---|---|---|
| **Kundur** | ***Power System Stability and Control*** | ⚠️ **The canonical stability reference. §4.4 → `power-system-analysis-and-protection` is this book** |
| **Glover, Sarma & Overbye** | ***Power System Analysis and Design*** | ⚠️ **The best all-round textbook; start here** |
| **Wood, Wollenberg & Sheblé** | ***Power Generation, Operation, and Control*** | §10 → `power-inverters-storage-markets-and-datacenters` — unit commitment and dispatch |
| **Blackburn & Domin** | ***Protective Relaying: Principles and Applications*** | ⚠️ **§5 → `power-system-analysis-and-protection`, definitively** |
| **Grainger & Stevenson** | *Power System Analysis* | Classic fundamentals |
| **Kersting** | *Distribution System Modeling and Analysis* | ⚠️ **Distribution is genuinely different; this is the reference** |
| **Denholm et al. (NREL)** | ***"Inertia and the Power Grid: A Guide Without the Spin"*** | ⚠️ **Free, and exactly the §8 → `power-inverters-storage-markets-and-datacenters` explainer you want** |
| **Expert Panel** | *Grid Incident in Spain and Portugal, 28 April 2025* (Oct 2025) | ⚠️ **Read the primary report** |
| **Bergen & Vittal** | *Power Systems Analysis* | Rigorous |

**Practical**: **NERC standards and reliability guidelines** (⚠️ **public and readable**),
**IEEE 1547 / 2800**, **IEC 61850** documentation, **EPRI reports**, **NREL publications**
(⚠️ **consistently excellent and free**), **MATPOWER and pandapower** docs as working
introductions, and **FERC/RTO filings** for how markets actually operate.

---

## §17. Quick Reference

### 17.1 Picker
| Need | Use |
|---|---|
| Bus voltages and line flows | **Power flow, Newton-Raphson** (§4.1 → `power-system-analysis-and-protection`) |
| Fast approximate flows for markets | ⚠️ **DC power flow** (§4.1 → `power-system-analysis-and-protection`) |
| Best estimate of current system state | **State estimation** (§4.2 → `power-system-analysis-and-protection`) |
| Breaker sizing, relay settings | **Fault analysis, symmetrical components** (§4.3 → `power-system-analysis-and-protection`) |
| Inverter control dynamics | ⚠️ **EMT simulation (PSCAD/EMTP), not phasor domain** (§11.3 → `power-inverters-storage-markets-and-datacenters`) |
| Distribution feeder study | **OpenDSS / GridLAB-D** (§11.3 → `power-inverters-storage-markets-and-datacenters`) |
| Open research / scripting | **pandapower, MATPOWER, PowerModels.jl** (§11.3 → `power-inverters-storage-markets-and-datacenters`) |
| Fast protection signalling | ⚠️ **IEC 61850 GOOSE (~4 ms)** (§7 → `power-scada-ems-and-protocols`) |
| Network model exchange | **CIM** (§7 → `power-scada-ems-and-protocols`) |
| Wide-area dynamics visibility | **PMUs / synchrophasors** (§4.2 → `power-system-analysis-and-protection`) |
| Inertia in a low-inertia system | ⚠️ **Synchronous condenser or GFM inverter — and see §8.2 → `power-inverters-storage-markets-and-datacenters`** |
| Fast frequency response | **BESS** (§9 → `power-inverters-storage-markets-and-datacenters`) |
| Connect a large load sooner | ⚠️ **Demonstrable demand flexibility** (§15.2) |

### 17.2 Orientation questions for grid software work
- [ ] Which timescale am I in — protection, control, dispatch, or planning? (§11.1 → `power-inverters-storage-markets-and-datacenters`)
- [ ] Per-unit or physical units, and on what base? (§1.5 → `power-ac-fundamentals-generation-and-grid`)
- [ ] Is this phasor-domain or does it need EMT? (§11.3 → `power-inverters-storage-markets-and-datacenters`)
- [ ] Does the network model match reality, and when was it last validated? (§11.3 → `power-inverters-storage-markets-and-datacenters`)
- [ ] Are timestamps traceable to a synchronized source? (§11.3 → `power-inverters-storage-markets-and-datacenters`)
- [ ] What happens to this system during an alarm flood? (§6 → `power-scada-ems-and-protocols`)
- [ ] Is this in NERC CIP scope? (§11.2 → `power-inverters-storage-markets-and-datacenters`)
- [ ] What does it do when it can't converge / can't reach a device / is late? (§4.1 → `power-system-analysis-and-protection`, §10 → `power-inverters-storage-markets-and-datacenters`)

---

## §18. Method

**§1–§14 → `power-ac-fundamentals-generation-and-grid`, `power-system-analysis-and-protection`, `power-scada-ems-and-protocols`, `power-inverters-storage-markets-and-datacenters` rest on settled material** — **Fortescue's symmetrical components (1918)**,
per-unit, Newton-Raphson power flow, protection principles, and market optimization
formulations — sourced from the references in §16, chiefly **Kundur**, **Glover/Sarma/
Overbye**, **Blackburn & Domin**, and **Wood/Wollenberg**. ⚠️ **None of that needed
verification; it's been stable for decades and the textbooks are the authority.**

**Scoped to complement**: component-level circuit design, op-amps, PCB layout and signal
integrity sit in an electrical-engineering reference. **This picks up at the kilovolt
scale.** ⚠️ **§11.1 → `power-inverters-storage-markets-and-datacenters` and §11.2 → `power-inverters-storage-markets-and-datacenters` deliberately parallel the flight-software and automotive
references — safety-critical embedded practice converges across industries, and the
reasoning transfers.**

**Two searches were run in August 2026**, on the two areas that genuinely moved: **the
inverter-dominated grid** and **datacenter load growth.**

**Confidence.** **High** in §1–§11 → `power-ac-fundamentals-generation-and-grid`, `power-system-analysis-and-protection`, `power-scada-ems-and-protocols`, `power-inverters-storage-markets-and-datacenters` and §14 — textbook material stated with its
assumptions, and the assumptions are the useful part. **High** in §8 → `power-inverters-storage-markets-and-datacenters`'s technical content:
the GFL/GFM distinction, the inertia and short-circuit-ratio mechanism, and the
protection consequences of limited inverter fault current are **consistent across
peer-reviewed sources** (IET, ScienceDirect, MDPI, arXiv reviews) **and NREL/NERC
material.**

⚠️ **Two hedges, weighted differently.**

**§8.2 → `power-inverters-storage-markets-and-datacenters`'s UK procurement figure I'd treat as solid and important** — it comes from a
consultancy analysis but is a **concrete, checkable procurement outcome** rather than a
forecast, and ⚠️ **it cuts directly against the prevailing enthusiasm, which is part of
why I've given it prominence.** **Verify the current figure if you're making a
technology decision on it.**

**⚠️ §15.2 is the weak section and is flagged in place.** The sourcing is **consultancies,
real estate firms, equipment vendors, grid-analytics companies and investment banks, all
of whom benefit from the growth narrative**, and ⚠️ **the demand projections varied
noticeably between them.** **I have deliberately separated the better-attested physical
facts — interconnection queue lengths, the 13%/77% historical completion base rate,
transformer and switchgear lead times, PJM's January 2026 connect-and-manage
announcement — from the forecasts, and I'd trust the former considerably more than the
latter.** ⚠️ **Direction is solid; magnitude is contested.**

**One thing I'd flag as the actionable insight rather than trivia**: ⚠️ **PJM's
connect-and-manage framework makes workload flexibility an interconnection asset.** **If
that model spreads, scheduler design becomes a determinant of how fast compute capacity
can be built** — which is an unusually direct line from software architecture to physical
infrastructure.