The Electrical Grid: From Generation to Consumption

5.1 — The grid cannot store electricity

Start with the constraint that shapes everything else: the wires do not hold energy. Generation and demand must balance continuously, second by second, across the entire synchronous system.

At 100 GW of generation against 100 GW of demand the system is balanced. If demand jumps to 105 GW while generation stays at 100 GW, frequency begins to fall immediately — not in minutes, not after a delay, but in the same instant — and operators must respond within seconds. Unlike almost every other commodity supply chain, you cannot manufacture electricity today and leave it sitting in the network for tomorrow.

This is precisely why storage is not simply another generation technology competing on cost. It is the first thing in the history of the electricity system that gives the network a buffer — and why the arithmetic in Chapter 4 treats a battery as an asset that earns from several different buyers at once.

5.2 — Generation: six conversion paths, one AC output

Every generator on the network is doing the same job — delivering alternating current at the system frequency — but the conversion chains that get them there look nothing alike.

Four of those six are heat engines with a turbine in the middle, which is why they inherit the Carnot ceiling discussed in Chapter 3. Two of them — solar and storage — produce or store direct current and reach the grid only through power electronics. That distinction turns out to matter far more than it first appears.

5.3 — Synchronous machines versus inverter-based resources

Traditional plants use synchronous generators: large rotating machines physically locked in step with grid frequency. Because they are spinning masses coupled to the network, they supply several things at once, most of them for free as a by-product of the physics: real power, reactive power, voltage support, rotational inertia and short-circuit strength.

Solar and battery systems produce or store DC and reach the grid through power electronics.

These are inverter-based resources, or IBRs, and their rising share of total generation is quietly rewriting how engineers keep the system stable. An inverter does not naturally do any of the things a spinning mass does by accident. Everything it contributes to stability, it contributes because a control engineer deliberately programmed it to.

5.4 — Why transmission exists

Demand concentrates around cities, industrial clusters, data centres and transport corridors. The best energy resources — solar-rich deserts, wind corridors, hydro basins, coastal nuclear and coal sites — are frequently hundreds of kilometres away. Transmission is the high-voltage network that closes that gap.

It is built on the principle established in Chapter 2: for a fixed quantity of power, higher voltage means proportionally lower current, and because resistive loss follows I ²R, cutting current sharply cuts loss. Halve the current and you quarter the loss over the same conductor. Over hundreds of kilometres that difference is the whole argument.

5.5 — HVAC versus HVDC

High-voltage alternating current dominates most grids for historical and practical reasons: transformers make voltage changes trivial, and AC circuit breakers can interrupt current at the natural zero crossing fifty times a second.

High-voltage direct current becomes attractive in four specific situations: very long overhead distances, submarine cables, interconnecting grids that are not synchronised with each other, and high-capacity point-to-point transfers where controllability matters. HVDC requires converter stations at each end.

Those converter stations are expensive, so HVDC carries a fixed cost penalty that HVAC does not. But HVDC has no reactive charging current, so its per-kilometre cost is lower. Beyond a break-even distance — conventionally a few hundred kilometres overhead, and much shorter for submarine cable, where AC charging current becomes prohibitive — DC wins. Chapter 16 works through that trade-off properly.

5.6 — Substations, busbars and transformers

A substation is where the network changes state. It changes voltage, switches circuits, protects equipment, measures parameters and isolates faults. The canonical path through one is short:

5.7 — Protection systems and circuit breakers

A fault is an unintended current path — phase to ground, phase to phase — and it can drive tens of thousands of amperes through equipment designed for a few hundred. Unchecked, a fault destroys plant, starts fires and can cascade into a wider collapse.

Protection systems exist to detect that condition and disconnect the smallest possible section of the network. The chain is current transformers and voltage transformers to measure, relays to decide, and circuit breakers to act. The governing principle is selective protection, sometimes called discrimination: the device closest to the fault should operate first, and everything upstream should hold.

1. 1Detect. The relay measures current, voltage, impedance or differential current and determines that the condition is a fault rather than an inrush or an overload.
2. 2Decide and command. The relay issues a trip signal, often within a single power-frequency cycle.
3. 3Open the contacts. The breaker mechanism physically separates the contacts.
4. 4Interrupt the current and extinguish the arc. As the contacts part, an arc forms and must be quenched — in SF6, vacuum or oil — typically at the natural current zero.

The entire sequence completes in a fraction of a second, and the difference between a well-coordinated protection scheme and a badly coordinated one is the difference between a feeder outage and a regional blackout.

5.8 — Distribution and the rise of the prosumer

Distribution is the last mile: the network that takes power from the transmission system down to the voltage at which it can be safely used in a home or a workshop.

Distribution was designed on an assumption that has quietly stopped being true: that power flows one way, from grid to consumer. Rooftop solar, behind-the-meter batteries and eventually bidirectional EVs turn many consumers into prosumers who both consume and produce, and who at times push excess power back upstream.

Collectively these are distributed energy resources, and millions of small DERs can represent grid capacity comparable to a large power plant — a fact section 5.22 builds on directly.

5.9 — Load, peak demand and the worst hour of the year

Demand is called load, and it follows a predictable daily shape: low overnight, rising through the morning, holding through the working day and often spiking sharply in the evening.

The fleet that serves this curve is built in tiers. Baseload plants sit at the bottom and run continuously. Intermediate or mid-merit plants follow the daily swing. Peaking plants cover only the narrow spike, which is why a peaker may run a few hundred hours a year and still be indispensable.

And here is the structural fact: infrastructure must be sized for the peak, not the average. A system averaging 20 GW but peaking at 30 GW must still build generation, transmission and distribution capacity for 30 GW, even though a third of that capacity sits idle for most of the year.

5.10 — Peak (GW) versus energy (TWh)

Annual demand of 175 TWh and peak demand of 30 GW describe entirely different properties of the same system. One is the total quantity of energy consumed over a year. The other is the instantaneous rate of demand at the single worst moment.

Both matter independently, and for different pieces of the plan. Energy sizes fuel contracts, generation build-out and emissions accounting. Peak sizes transmission, transformers, switchgear and reserve margin. A system can have ample annual energy and still be short of peak capacity; it can also have abundant peak capacity and be short of energy through a dry hydro year.

5.11 — Ramp rate: how fast can output change?

Serving the peak is one problem. Getting there fast enough is another, and it is increasingly the harder one.

A plant that moves from 100 MW to 200 MW in ten minutes has a ramp rate of 10 MW per minute. That is a plant characteristic set by thermal and mechanical limits: a large steam unit heats and cools slowly and cannot be pushed faster without damaging it.

As solar output falls through the late afternoon at precisely the moment demand climbs toward its evening peak, the net load that the rest of the fleet must serve rises steeply. This is the neck of the duck curve introduced in Chapter 4, viewed now as an engineering requirement rather than a price signal.

5.12 — Frequency regulation and inertia

When generation and demand mismatch, frequency moves. Too much generation and it rises; too much demand and it falls. Resources respond by raising or lowering output, or by charging and discharging storage. Frequency is therefore the system’s continuous, real-time measure of whether supply and demand are in balance — the single number that tells an operator the state of the whole machine.

Traditional synchronous generators contain large rotating masses storing kinetic energy. During a sudden disturbance — a large unit tripping, say — that stored energy is released or absorbed automatically by the physics, briefly resisting rapid frequency change. This property is inertia, and it buys the system the few seconds it needs for controlled responses to act.

Solar and battery systems connected through power electronics have no rotating mass and provide none of this naturally. As their share rises, the system’s inertia falls, and the rate of change of frequency after a disturbance rises — which shrinks the window in which anything can be done about it.

5.13 — Grid strength and the weak-grid problem

Grid strength describes a system’s ability to hold voltage steady when something disturbs it. A strong grid has substantial synchronous generation and low-impedance connections nearby, so voltage barely moves when a large plant switches. A weak grid — typically a thin network far from major generation, which is exactly where the best wind and solar resources tend to be — sees voltage swing significantly.

High penetration of inverter-based resources in weak-grid conditions creates real engineering problems: control interactions between neighbouring inverters, voltage instability, and unpredictable behaviour during faults. Before connecting a large new solar or wind project to a thin part of the network, engineers analyse short-circuit ratio, voltage stability, control interactions and fault behaviour.

5.14 — Congestion and curtailment

When generation in a region exceeds what the network can carry out of it, or what the system can absorb, the operator instructs plants to reduce output. That instruction is curtailment, and the energy it prevents is revenue that never existed.

5.15 — Interconnection: the hidden bottleneck

A new power plant cannot simply plug into the grid. The system operator must study where it can connect, how much power can be injected at that node, whether existing equipment can carry it, and whether stability problems would arise under fault or contingency conditions.

This grid interconnection process routinely becomes the longest lead-time item in a project — frequently exceeding construction itself. A solar or storage project can have technology, capital, land and offtake all secured and still sit stalled for years waiting for a connection study and a queue position.

5.16 — Interconnected grids, islanding and microgrids

Regional grids are typically interconnected through transmission, which enables power sharing, reserve sharing, economic dispatch across a wider area and mutual reliability support. A larger synchronous area is generally a more stable one: the same disturbance produces a smaller frequency excursion when spread across more inertia.

A section of the grid can also become electrically separated from the rest, either deliberately or as a consequence of a fault. That condition is islanding. Unintentional islanding is dangerous — a supposedly de-energised line being fed by an unnoticed local generator is a hazard to line workers, which is why anti-islanding protection is mandatory on grid-connected inverters.

Deliberate islanding is a design feature. A well-designed microgrid — solar plus storage plus backup generation plus local loads plus an energy management system — can detect a wider outage, disconnect cleanly, and continue serving critical load until the grid returns. That capability is increasingly valuable for hospitals, data centres, and remote or defence facilities, and Chapter 19 develops it fully.

5.17 — Data centres as grid-scale loads

A large AI data centre can require hundreds of megawatts of continuous power. At that scale it stops resembling a commercial building and starts resembling an industrial power project — one that happens to consume rather than produce.

Serving it means dedicated substations, multiple transmission connections, redundant feeders, on-site storage, backup generation and active power-quality management, all negotiated years before the first server rack is energised. It also means a load with a very high capacity factor and very little tolerance for interruption, which is a materially different proposition for a system operator than a load that varies with the working day.

The consequence is that data-centre siting has become an electricity-infrastructure decision. Series 8 covers this in depth; for now the point is simply that the largest new loads on many networks are being planned on the same timescales, and against the same constraints, as the largest new generators.

5.18 — Storage operates at every layer of the grid

Most technologies belong to one layer of the architecture. Storage is unusual in belonging to all of them.

Energy, power, flexibility and resilience from a single asset class, deployable at five different points in the system, is what makes storage strategically unusual rather than merely useful. It is also the physical reason the revenue stack in Chapter 4 exists at all: the same box is genuinely providing several distinct services to several distinct buyers.

5.19 — Smart grids: SCADA, BMS, PCS and EMS

Modern grids layer sensors, smart meters, communication networks, forecasting and increasingly machine learning on top of the physical infrastructure. The objective shifts from simply moving electricity to continuously measuring, predicting, optimising and controlling the whole system.

Four acronyms do most of the work in a storage-heavy system, and they are constantly confused because all four sit, in some sense, inside the same installation.

5.20 — N-1 and the economics of redundancy

The core planning principle of transmission is N-1: the system must remain operational following the loss of any single critical component. If one line trips, another path carries the load. If one transformer fails, another continues serving demand. If one generator disconnects, reserves cover it.

Redundancy is expensive. It means building and maintaining equipment that, on any ordinary day, is not needed. But the cost of widespread outages is typically far higher, which makes grid reliability an economic optimisation problem rather than a pursuit of theoretical perfection.

That framing matters. Nobody builds N-2 everywhere, because the marginal cost of the second layer of redundancy usually exceeds the marginal value of the outages it prevents. Planners choose the standard by weighing capital cost against the value of lost load, and the answer is legitimately different for a hospital feeder and a rural distribution spur.

5.21 — Reliability versus resilience

These two words are used interchangeably in casual conversation and mean quite different things in system planning.

A system can be highly reliable and poorly resilient: excellent on ordinary days, and catastrophic on the rare day when several things go wrong at once. As the frequency of extreme weather events rises, the second column is receiving planning attention it did not historically get.

5.22 — Virtual power plants

A virtual power plant aggregates many small distributed resources — EVs, home batteries, rooftop solar, flexible industrial load — and coordinates them through software so that they behave, from the market’s point of view, like a single dispatchable power plant. Every physical asset remains independently owned and physically distributed.

The engineering challenge is not the individual assets, which are simple. It is telemetry, latency, forecasting and the contractual question of what happens when a participant declines to respond. The commercial challenge is that market rules were mostly written for large single assets, and a resource made of ten thousand pieces has to be allowed to participate before it can earn anything.

Where those obstacles have been cleared, VPPs have delivered capacity at costs that conventional peaking plants cannot match, because the assets were already going to be purchased for other reasons. The marginal cost of the capacity is the cost of coordinating it.

5.23 — The four flows of the modern grid

A useful way to hold the whole system in mind is to notice that four different things flow through it simultaneously, in different units, on different timescales.

The future grid must coordinate all four at once, which is why grid infrastructure increasingly means as much software and data as it does copper and steel. A transmission upgrade and a market-rule change can be substitutes for one another; so can a battery and a better forecast.

5.24 — Why the grid may be the binding constraint

Solar, wind, EVs, storage, data centres, industrial electrification and hydrogen production are all growing at once, and every one of them increases demand for grid infrastructure.

Meanwhile the timescales do not match. A generation project can be developed and built in two to four years. Major transmission infrastructure frequently requires far longer once planning, consenting, land acquisition, equipment lead times and construction are all included. When the thing everything depends on takes two to three times as long to build as the things depending on it, that mismatch becomes structural rather than temporary.

The practical consequence is that the energy transition is, in large part, a grid infrastructure investment cycle: transmission lines, transformers, switchgear, substations, protection equipment and grid software. That is a less photogenic story than falling module prices, and it is where a great deal of the capital and a great deal of the delay actually sit.

5.25 — The complete journey of one megawatt-hour

Putting the whole chapter into a single path, from resource to useful work:

Every stage carries both a cost and a loss. That is the physical reason a megawatt-hour generated and a megawatt-hour usefully consumed are never the same number, and why efficiency arguments made at one layer are so often quietly undone at another.

5.26 — The core grid mental model

Eight questions cover the architecture. If you can answer all eight about a given system, you understand it well enough to reason about almost any proposal made for it.

• •Generation — where is electricity produced, and by what conversion path?
• •Transmission — how does it move over long distances, and what is the evacuation limit?
• •Distribution — how does it reach the end user, and does power flow both ways?
• •Storage — how can energy be shifted through time, and at which layer?
• •Control — how is the system coordinated, measured and optimised?
• •Protection — what happens when something fails, and how small is the disconnected section?
• •Economics — who pays, who earns, and who bears which risk?
• •Reliability — what happens when demand spikes or equipment fails, and what standard was chosen?

Chapter 6 goes back to the start of that chain and looks at the largest new source feeding into it: solar power, from photons to megawatts, and what actually determines how much of the sunlight falling on a module reaches the grid.