Pumped-Hydro Storage

How pumped-hydro storage works

A pumped-hydro plant is two lakes at different heights joined by a pipe and a machine. When the grid has surplus power — at night, or on a sunny windy afternoon when solar and wind are flooding the lines — the plant runs its machine as a pump and lifts water from the lower lake to the upper one. The electricity has not vanished; it is now stored as the gravitational potential energy of millions of tonnes of water sitting hundreds of metres in the air. When demand spikes — the evening cooking peak, or a generator tripping offline — the valves open, the water rushes back down, and the same machine spins the other way as a turbine, regenerating electricity.

That is the whole idea: a battery whose charge is altitude. The cleverness is in the engineering that makes the round trip efficient. The water falls through a steel-lined shaft called a penstock, is throttled by guide vanes, and drives a reversible Francis pump-turbine coupled to a motor-generator. One machine does both jobs, which roughly halves the cost of the powerhouse compared to a separate pump and turbine set.

Crucially, the energy stored per litre is small, so the reservoirs must be huge. A litre of water dropped 500 metres yields only about 1.2 watt-hours at the shaft — barely enough to light a small LED for a couple of hours. To store a useful amount you need a reservoir the size of a small lake — which is exactly why pumped-hydro plants are landscape-scale civil works, not factory products.

The governing physics

The energy a plant can store is just the gravitational potential of the usable water volume, derated by the generating efficiency:

Stored (generating) energy:

  E = ρ · g · V · H · η_gen

where
  ρ      = water density       ≈ 1000 kg/m³
  g      = gravity             = 9.81 m/s²
  V      = usable volume       (m³)
  H      = head (vertical drop) (m)
  η_gen  = generating efficiency ≈ 0.88 to 0.92

Power delivered at flow rate Q:

  P = ρ · g · Q · H · η_gen           (Q in m³/s)

Round-trip efficiency:

  η_rt = η_pump · η_gen ≈ 0.85 × 0.88 ≈ 0.75

Note the same equation governs every hydro turbine — the only difference here is that the plant also runs the reaction in reverse to charge. Putting numbers in: a plant with V = 7 × 10⁶ m³ of usable water and H = 500 m of head, at η_gen = 0.90, holds

E = 1000 × 9.81 × 7e6 × 500 × 0.90
  = 3.09 × 10¹³ J
  = 8.6 GWh  (divide by 3.6e6 J/kWh, then 1e6)

To deliver that energy as power, you trade flow against time. The same plant pushing Q = 400 m³/s of water through 500 m of head generates

P = 1000 × 9.81 × 400 × 500 × 0.90
  = 1.77 × 10⁹ W
  = 1.77 GW   →   can sustain 8.6 GWh / 1.77 GW ≈ 4.9 hours

This is the defining property of pumped-hydro: power and energy are decoupled. Want more hours of storage? Make the reservoir deeper or wider — V goes up, the expensive powerhouse is untouched. Want more peak power? Widen the penstock and add machine capacity — the reservoir is untouched. Batteries cannot do this; their power and energy are both set by the number of cells.

Worked example: sizing Dinorwig

Dinorwig in north Wales, the "Electric Mountain", is the textbook fast-response plant. It was built inside a slate mountain specifically to provide near-instant backup for the UK grid. Run its headline numbers:

Upper reservoir (Marchlyn Mawr) usable volume:  V ≈ 6.7 × 10⁶ m³
Average head:                                    H ≈ 500 m
Rated output:                                    P = 1,728 MW (6 × 288 MW)
Generating efficiency:                           η_gen ≈ 0.90

Stored energy:
  E = 1000 × 9.81 × 6.7e6 × 500 × 0.90
    ≈ 2.96 × 10¹³ J ≈ 8.2 GWh

Discharge time at full output:
  t = 8.2 GWh / 1.728 GW ≈ 4.7 hours
  (the published figure is ~5 to 6 hours at typical dispatch)

The famous part is the response time. Dinorwig keeps machines spinning in air (turning but unloaded) so it can reach full 1,728 MW in about 12 seconds — fast enough to catch the "TV pickup" surge when millions of British kettles switch on at the end of a popular broadcast. From a complete standstill it still reaches full power in about 75 seconds. That speed, not the energy, is what the grid pays for.

Real-world plants

Plant	Country	Power	Head	Notable for
Fengning	China	3,600 MW	~425 m	World's largest by capacity (2024)
Bath County	USA	3,003 MW	~380 m	~24 GWh — "world's largest battery"
Guangdong	China	2,400 MW	~500 m	Backs the Daya Bay nuclear plant
Dinorwig	UK	1,728 MW	~500 m	12 s spinning-reserve response
Goldisthal	Germany	1,060 MW	~300 m	First large variable-speed units in Europe
Tianhuangping	China	1,836 MW	~590 m	High-head, large-volume design
Snowy 2.0	Australia	2,200 MW	~700 m	~350 GWh; closed-loop tunnel link (under construction)

For scale: Bath County's ~24 GWh dwarfs the largest lithium battery installations (a few GWh), and its reservoirs hold enough water to power roughly 3 million homes for about 11 hours. Globally, pumped-hydro accounts for over 90% of utility-scale storage capacity — roughly 180 GW of power and well over 8,000 GWh of energy.

Pumped-hydro vs other grid storage

	Pumped-hydro	Lithium battery	Compressed air (CAES)	Flywheel	Green hydrogen
Round-trip efficiency	70 to 80%	85 to 95%	40 to 70%	85 to 90%	30 to 45%
Response time	Seconds to minutes	Milliseconds	Minutes	Milliseconds	Minutes
Discharge duration	6 to 20+ hours	1 to 4 hours	4 to 24 hours	Seconds to minutes	Days to seasons
Energy scaling cost	Cheap (bigger lake)	Linear (more cells)	Cheap (bigger cavern)	Very expensive	Cheap (bigger tank)
Lifetime	50 to 100 yr	10 to 15 yr	30 to 40 yr	20+ yr	10 to 20 yr
Capacity fade	None	Significant	Minimal	None	Minimal
Siting constraint	Needs head + geology	Almost anywhere	Needs cavern/aquifer	Anywhere	Anywhere
Self-discharge	Negligible (evaporation)	Low	Low	High	Low (boil-off)

The pattern is clear: batteries win on efficiency, speed, and siting; pumped-hydro wins on energy scale, longevity, and cost-per-stored-kilowatt-hour over decades. They are complements, not rivals — a modern grid uses batteries for sub-second frequency response and short peaks, and pumped-hydro for the long evening discharge and the multi-day lulls in wind and sun.

Design choices and machine types

• Reversible Francis pump-turbine. The workhorse for heads of roughly 100 to 700 m. A single runner pumps and generates; cheap, compact, but a hydraulic compromise (the runner cannot be optimal for both directions) and slow to switch (1 to several minutes to reverse).
• Ternary set. A separate Francis or Pelton turbine and a dedicated pump on one shaft with a clutch. Costs more, but switches between pumping and generating in seconds and can run in "hydraulic short-circuit" — pumping and generating at once to provide continuous fine power regulation.
• Variable-speed (doubly-fed) machines. The motor-generator runs over a speed range instead of a fixed grid frequency. This lets the plant vary its pumping power (fixed-speed pumps are all-or-nothing), regulate frequency while pumping, and squeeze 1 to 3 points more efficiency. Goldisthal and many new Chinese plants use them.
• Open-loop vs closed-loop. Open-loop ties one reservoir to a natural river; closed-loop cycles the same water between two built ponds and only fills once. Closed-loop avoids most river-ecology damage and can reuse old mines and quarries — the fastest-growing design.
• Surge chamber. A vertical shaft or tank between the tunnel and penstock that absorbs the pressure wave when valves slam shut, protecting the conduit from water hammer.

When pumped-hydro is the right answer

• Long-duration storage (6+ hours). Riding through the evening peak, or storing a windy night for a calm day. This is where battery cost explodes and pumped-hydro shines.
• Energy arbitrage. Buy cheap off-peak power, sell expensive peak power. The 25% round-trip loss is worth it when the peak/off-peak price spread exceeds about 1.33× (1 ÷ 0.75) — and in practice you need a wider spread than that to clear costs and earn a margin.
• Grid stability services. Spinning reserve, fast frequency response, black-start capability, and reactive power. A spinning pump-turbine adds real inertia to the grid — something inverter-based batteries must synthesise.
• Firming variable renewables. Soaking up midday solar overproduction (the "duck curve" belly) and releasing it after sunset is the modern flagship use case.

Reach for something else when the geography lacks head, when you need millisecond response (battery or flywheel), when storage must last days or seasons (hydrogen), or when the project must be built in two years rather than ten — pumped-hydro's permitting and civil-works timeline is its biggest practical drawback.

Common misconceptions and pitfalls

• "It generates net power." No — pumped-hydro is a net consumer of energy. You always get back less than you put in (the 25% round-trip loss). It is a storage and timing device, not a source. Its value is moving cheap energy to an expensive moment, plus grid services.
• "It needs a dam on a river." Closed-loop plants need no river at all — just two ponds and a one-time fill. Disused mines, quarries, and even seawater coastal sites (Okinawa ran a seawater pilot) all work.
• "Bigger head is always better." Higher head means more energy per litre and a smaller reservoir, but also higher penstock pressure (thicker, costlier steel) and a runner that must handle large pressure swings. Real designs optimise head against civil cost and cavitation limits, not maximise it.
• Cavitation in pump mode. If the runner inlet pressure drops below the water's vapour pressure, bubbles form and collapse violently, pitting the steel. Pump-turbines are set deep below the lower reservoir surface to keep suction pressure high enough — the submergence depth is a hard design constraint, not an afterthought. (See cavitation.)
• Water hammer on shutdown. Closing a valve on a column of fast-moving water in a long penstock creates a pressure spike that can burst the conduit. Surge chambers and timed, staged valve closure tame it; ignoring it has destroyed penstocks.
• "Efficiency is the only number that matters." A battery's 90% beats hydro's 75%, but if the battery costs ten times as much per stored kWh and lasts a fifth as long, the 15-point efficiency gap is irrelevant for long-duration jobs. Match the technology to the duration, not the spec sheet.