A different kind of power station
According to Ember's Global Electricity Review 2026, published in April 2026, renewable sources generated 33.8% of the world's electricity in 2025, overtaking coal (33.0%) for the first time in a century of electricity. Solar generation grew by 30% in one year, to 2,778 TWh, overtaking wind for the first time, and solar alone met three-quarters of the growth in the world's electricity demand. Whatever happens next, the engineering of the grid is being rebuilt around sources that are unlike the synchronous generators of the previous lessons in three ways: they are variable, producing when the sun shines and the wind blows rather than when demand calls; they connect through inverters, not spinning machines; and many of them are small and distributed, on rooftops and in fields rather than in a few large stations.
This lesson takes each piece from first principles to practice: the solar cell's I–V curve and the algorithm that finds its best operating point; the wind turbine and the 16/27 that bounds it; the storage that moves energy in time and its hazards; the inverters that must take over the stabilising role of synchronous machines; and the electric vehicles that are becoming both a large new load and a potential resource.
Note
Safety. Lithium-ion batteries store a great deal of energy, and a damaged, overcharged, short-circuited or overheated cell can catch fire or release toxic gases in a process that is hard to stop. Never open, puncture, crush or modify a battery pack, never charge a swollen or damaged battery, and use only the charger made for it. Solar panels make a dangerous DC voltage whenever light falls on them, and cannot simply be switched off; rooftop systems, home batteries and EV chargers must be installed and serviced by qualified people. Safe ways to explore are simulation, a single small solar cell with a multimeter, and small low-voltage kits with protected cells.
Solar cells and the I–V curve
A solar cell is a large silicon diode, the p–n junction of Semiconductors and diodes, with a thin front layer that lets light in. Each absorbed photon with more energy than silicon's bandgap of 1.12 eV frees an electron–hole pair, and the junction's built-in field sweeps the electrons to one side and the holes to the other. In the dark the cell is just a diode; in the light it is a diode with a current source in parallel, pumping current backwards through it.
Definition— Single-diode model
A cell, or a module of cells in series, delivering current at voltage obeys where is the light-generated current (proportional to the irradiance), the diode's saturation current (which rises steeply with temperature), an ideality factor between 1 and 2, the thermal voltage (25.7 mV at 25 °C), the series resistance of the contacts and fingers, and a shunt resistance from leakage paths.
The I–V curve runs from the short-circuit current at to the open-circuit voltage at . Neither end delivers any power, because one of and is zero. In between, the product has a single peak, the maximum power point (MPP), at the knee of the curve. Panels are rated at standard test conditions (STC): 1000 W/m² of sunlight at a cell temperature of 25 °C.
Light and heat move the curve in different ways. The current is proportional to irradiance, so halving the light halves and roughly halves the power, while the voltage falls only logarithmically. Heat is worse than it looks: grows so fast with temperature that falls, by roughly 0.3% per kelvin for silicon, and on a sunny roof cells easily run 25 to 35 °C above the air. The program computes the curve from the model, solving for as a function of (explicit when is neglected), and then runs the simplest tracking algorithm.
import numpy as np
# Single-diode model of an illustrative 60-cell silicon module (shunt resistance neglected).
k, q = 1.380649e-23, 1.602176634e-19
N_s, n, R_s = 60, 1.2, 0.30 # cells in series, diode ideality factor, series resistance (ohm)
I0_ref, T_ref, E_g = 1e-8, 298.15, 1.12
def iv_curve(G=1000.0, T_c=25.0, points=400):
"""Current from 0 to short circuit, and the module voltage at each: V = n N_s V_T ln((I_L - I)/I_0 + 1) - I R_s."""
T = T_c + 273.15
V_T = k * T / q
I_L = 10.0 * G / 1000 * (1 + 0.0005 * (T - T_ref)) # light current, A
I_0 = I0_ref * (T / T_ref) ** 3 * np.exp(E_g / (n * k / q) * (1 / T_ref - 1 / T))
I = np.linspace(0, I_L, points, endpoint=False)
V = n * N_s * V_T * np.log((I_L - I) / I_0 + 1) - I * R_s
keep = V >= 0
return V[keep][::-1], I[keep][::-1] # sorted by rising voltage
def mpp(G=1000.0, T_c=25.0):
V, I = iv_curve(G, T_c, 20000)
j = np.argmax(V * I)
return V[j], I[j], V[j] * I[j], V[-1], I[0]
print(" G (W/m2) T (C) Voc (V) Isc (A) Vmp (V) Imp (A) Pmax (W)")
for G, T_c in [(1000, 25), (1000, 60), (500, 25), (200, 25)]:
Vmp, Imp, Pmp, Voc, Isc = mpp(G, T_c)
print(f"{G:10.0f} {T_c:6.0f} {Voc:9.2f} {Isc:8.2f} {Vmp:9.2f} {Imp:8.2f} {Pmp:10.1f}")
p25, p60 = mpp(1000, 25)[2], mpp(1000, 60)[2]
print(f"temperature coefficient of power: {100 * (p60 / p25 - 1) / 35:.2f} %/K")
# Perturb and observe: nudge the operating voltage, keep going while the power rises, reverse when it falls.
def power_at(v, G):
V, I = iv_curve(G, 25.0, 4000)
return v * np.interp(v, V, I)
v, step, p_old = 28.0, 0.5, 0.0
print("\nstep G (W/m2) V (V) P (W)")
for k_step in range(40):
G = 1000.0 if k_step < 20 else 600.0 # a cloud arrives at step 20
p = power_at(v, G)
if p < p_old:
step = -step # power fell: turn round
if k_step % 4 == 0 or k_step in (19, 20, 21, 39):
print(f"{k_step:4d} {G:10.0f} {v:7.1f} {p:7.1f}")
p_old = p
v += step
print(f"true maximum at 600 W/m2: {mpp(600, 25)[2]:.1f} W at {mpp(600, 25)[0]:.1f} V")The module gives 284.9 W at STC, at 30.41 V and 9.37 A. At 60 °C it gives only 245.9 W, a temperature coefficient of per kelvin, typical of silicon; at 200 W/m², a cloudy day, it still holds a voltage close to its full-sun value while its current falls fivefold.
Maximum power point tracking (MPPT) is the job of the DC-DC converter between the panels and the inverter (Power electronics): it adjusts its duty cycle to hold the panels at whatever voltage gives the most power, which changes with every cloud and every degree. Perturb and observe, the method above, nudges the voltage, keeps going in the same direction while the power rises and turns round when it falls. It climbs from 28 V to the peak in a few steps and then dithers around it, between 30.0 and 30.5 V, a small, permanent cost of the method. When the cloud arrives at step 20 it finds the new peak, 172.0 W at 30.5 V, at once, because the MPP voltage barely moves with irradiance.
Wind power and the Betz limit
The wind carries kinetic energy through a rotor of swept area at a rate with air density kg/m³. The cube is the most important fact about wind power: double the wind speed and the power rises eightfold, so siting, tower height and rotor size matter enormously. But a turbine cannot capture all of it, because the air must keep moving to get out of the way of the air behind it.
Theorem— Betz limit
An ideal rotor, modelled as a thin disc that extracts energy uniformly from a steady, incompressible stream, can extract at most of the power of the wind passing through its area.
Proof
Let the wind speed be far upstream, at the disc and far downstream. Mass flows through the disc at . The force on the air is the rate of change of its momentum, , and the power the disc takes is . That must equal the kinetic energy the air loses per second, . Comparing the two gives : the speed at the disc is the average of the upstream and downstream speeds. Write , where is the axial induction factor; then and Then , which is zero at (for ), where has its maximum, .
∎
Real rotors reach power coefficients of about 0.45 to 0.5 at their best, losing the rest to the wake's rotation, drag on the blades and losses at the tips. A turbine's power curve has three regions: below the cut-in speed (about 3 m/s) it does not turn usefully; between cut-in and rated speed it follows , with its control system adjusting the rotor speed to hold near its best; above rated speed it pitches its blades to spill the excess and hold rated power; and at the cut-out speed (often around 25 m/s) it feathers its blades and stops to protect itself.
Definition— Capacity factor
The capacity factor of a generator over a period is the energy it actually produced divided by the energy it would have produced running at its rated power the whole time. Over a year, .
The capacity factor of a wind turbine depends on the distribution of wind speeds at its site. A common model is the Rayleigh distribution, set by the mean speed alone.
import numpy as np
import sympy as sp
# 1. The Betz limit: maximise the power coefficient Cp(a) = 4 a (1 - a)^2 over the induction factor a.
a = sp.symbols("a", positive=True)
Cp = 4 * a * (1 - a) ** 2
best = [r for r in sp.solve(sp.diff(Cp, a), a) if r < 1]
print("dCp/da = 0 at a =", best, " Cp max =", Cp.subs(a, best[0]), "=", float(Cp.subs(a, best[0])))
# 2. An illustrative 5 MW turbine with a 150 m rotor.
rho, D = 1.225, 150.0
A = np.pi * (D / 2) ** 2
P_rated, cp, v_in, v_out = 5e6, 0.45, 3.0, 25.0
for v in (6.0, 10.0):
P_wind = 0.5 * rho * A * v**3
print(f"wind at {v:4.1f} m/s carries {P_wind / 1e6:5.2f} MW through the rotor; Betz allows "
f"{P_wind * 16 / 27 / 1e6:5.2f} MW; at Cp = {cp} the turbine makes {min(P_wind * cp, P_rated) / 1e6:4.2f} MW")
v_rated = (P_rated / (0.5 * rho * A * cp)) ** (1 / 3)
print(f"rated power is reached at {v_rated:.1f} m/s")
def power_curve(v):
p = np.minimum(0.5 * rho * A * cp * v**3, P_rated)
return np.where((v >= v_in) & (v < v_out), p, 0.0)
# 3. Capacity factor: average the power curve over a Rayleigh distribution of wind speeds.
v = np.linspace(0, 35, 3501)
dv = v[1] - v[0]
print("\nmean wind (m/s) capacity factor energy per year (GWh)")
for v_mean in (6.0, 7.5, 9.0):
pdf = (np.pi / 2) * v / v_mean**2 * np.exp(-np.pi / 4 * (v / v_mean) ** 2)
P_avg = np.sum(power_curve(v) * pdf) * dv
print(f"{v_mean:15.1f} {P_avg / P_rated:15.1%} {P_avg * 8760 / 1e9:21.1f}")SymPy confirms the maximum at , . At 10 m/s, 10.82 MW of wind passes through the 150 m rotor; Betz allows 6.41 MW and the turbine, at , makes 4.87 MW, reaching its 5 MW rating at 10.1 m/s. Over a year, the same turbine has an ideal capacity factor of 31.3% at a mean wind of 6 m/s and 56.9% at 9 m/s: the cube law at work. These figures are upper bounds; real turbines lose further energy to downtime, the wakes of neighbouring turbines, electrical losses and curtailment. For measured capacity factors by technology and country, see IRENA's annual reports on renewable generation and national statistics such as those of the US Energy Information Administration.
Variability and forecasting
Solar output follows the sun through the day and the seasons, with clouds on top; wind follows the weather, often for days at a time, with seasonal patterns that differ by region. Neither can be dispatched like a gas turbine, but both can be forecast: numerical weather prediction gives the shape of the next days, and statistical and machine-learning models trained on each plant's history refine it to the hour and the minute. Errors shrink as the forecast horizon shortens and as more plants are added over a wider area, because the weather is not the same everywhere at once (geographic smoothing).
What operators must balance is net load, demand minus wind and solar. In sunny grids with much solar, net load sags at midday and ramps up steeply in the evening as the sun sets and people come home (California's operator popularised the picture of this as a "duck curve"). Grids meet the challenge with every kind of flexibility: storage, demand that can shift (heating, cooling, industrial processes, and EV charging), interconnections that pool larger areas, flexible hydro and gas plants, and, when there is more than can be used, curtailment, deliberately reducing wind or solar output. The markets of The power grid signal all of this through prices that fall at times of plenty and rise at times of scarcity.
Storage: batteries, pumped hydro and more
Lithium-ion batteries
A lithium-ion cell shuttles lithium ions between two electrodes through an electrolyte: into the graphite anode as it charges, back into the cathode's crystal lattice as it discharges, while electrons take the external circuit. The cathode chemistry largely sets the cell's character:
- LFP (lithium iron phosphate) cells store less energy per kilogram but last for thousands of cycles, are more tolerant of abuse, enter thermal runaway at a higher temperature and less violently, and use no nickel or cobalt. They dominate grid storage and a large and growing share of electric cars, especially in China.
- NMC (lithium nickel manganese cobalt oxide), and the related NCA, store more energy per kilogram, which is why they power long-range cars and devices, at the cost of the more expensive and supply-constrained metals and a lower thermal-runaway threshold.
- Sodium-ion cells, which replace lithium with abundant sodium, began entering mass production around the mid-2020s for cheaper vehicles and stationary storage, trading energy density for cost and better cold-weather performance.
Definition— Battery degradation
A cell loses capacity and gains internal resistance in two ways. Calendar ageing happens even at rest, mainly by slow growth of the solid–electrolyte interphase (SEI) film on the anode, which consumes lithium; it is faster at high temperature and high state of charge. Cycle ageing happens with use: deep cycles, high currents, and especially fast charging at low temperature, which can deposit metallic lithium on the anode (lithium plating), both waste capacity and stress the electrodes.
This is why a battery management system monitors every cell's voltage and temperature, balances cells, limits charge and discharge currents with temperature, and why both cars and grid batteries usually avoid sitting at 100% charge for long.
Safety is the other half. In thermal runaway, a cell that is overheated, overcharged, crushed or internally shorted heats itself through chemical reactions faster than it can shed heat, venting flammable gases and possibly setting its neighbours off in turn. Large battery installations are therefore engineered against propagation, with spacing, fire barriers, gas detection and ventilation, and tested against national and international fire-safety standards before they are allowed on the grid.
Pumped hydro and other storage
Pumped-storage hydropower pumps water uphill when energy is plentiful and lets it run back through turbines when it is needed. Its stored energy is for a volume of water raised a head : an upper reservoir of 5 million m³, 400 m above the lower one, stores J, about 5,450 MWh, of which a 90%-efficient turbine returns about 4,900 MWh. Round-trip efficiencies are typically 70 to 80%, plants last for many decades, and in the mid-2020s pumped hydro still held most of the world's stored energy (see the International Hydropower Association), even as batteries, with typically one to four hours of storage, were being added far faster.
Other technologies fill other niches: compressed-air storage in underground caverns; flow batteries, whose energy sits in tanks of liquid electrolyte, so energy and power can be sized separately; thermal storage in molten salt, hot rock or sand; and hydrogen made by electrolysis, which can store energy for months but returns only a minority of it as electricity.
The last program follows a sunny day on a small island grid with a 120 MW solar farm and gas generators, with batteries of different sizes. The battery charges whenever solar exceeds demand and discharges whenever it falls short.
import numpy as np
# One sunny day on a small island grid (illustrative): hourly demand and the output of a 120 MW solar farm.
hours = np.arange(24)
demand = 60 + 15 * np.exp(-((hours - 8) / 2.5) ** 2) + 30 * np.exp(-((hours - 19.5) / 2.5) ** 2) # MW
solar = 120 * np.clip(np.sin(np.pi * (hours - 6) / 13), 0, None) # MW, 06:00 to 19:00
def run_day(power_mw, energy_mwh, efficiency=0.9):
"""Charge from surplus solar; discharge whenever demand exceeds solar. Gas covers the rest."""
soc = 0.2 * energy_mwh # start the day 20% full
gas, curtailed, socs = [], [], []
for d, s in zip(demand, solar):
net = d - s # positive: a shortfall; negative: a surplus
if net < 0: # store what we can (charging losses counted on the way in)
charge = min(-net, power_mw, (energy_mwh - soc) / np.sqrt(efficiency))
soc += charge * np.sqrt(efficiency)
curtailed.append(-net - charge)
gas.append(0.0)
else:
discharge = min(net, power_mw, soc * np.sqrt(efficiency))
soc -= discharge / np.sqrt(efficiency)
gas.append(net - discharge)
curtailed.append(0.0)
socs.append(soc)
return np.array(gas), np.array(curtailed), np.array(socs)
print(f"demand {demand.sum():.0f} MWh, solar available {solar.sum():.0f} MWh, evening peak {demand.max():.0f} MW")
print("battery (MW / MWh) gas used (MWh) solar curtailed (MWh) highest gas output (MW)")
for p_mw, e_mwh in [(0, 0), (25, 50), (50, 200), (80, 400)]:
gas, cut, _ = run_day(p_mw, e_mwh)
print(f"{p_mw:8d} / {e_mwh:<8d} {gas.sum():16.0f} {cut.sum():23.0f} {gas.max():25.0f}")
gas, cut, soc = run_day(50, 200)
print("\nhour demand solar gas battery state (MWh), with 50 MW / 200 MWh")
for h in range(0, 24, 2):
print(f"{h:4d} {demand[h]:7.0f} {solar[h]:6.0f} {gas[h]:4.0f} {soc[h]:6.0f} " + "#" * int(round(soc[h] / 10)))Without storage, 308 MWh of the day's 988 MWh of solar is curtailed and gas supplies 958 MWh, with an evening peak of 89 MW. A 50 MW, 200 MWh battery (four hours) saves 228 MWh of gas and cuts curtailment to 97 MWh; with 80 MW and 400 MWh nothing is curtailed. The small 25 MW, 50 MWh battery saves energy but not the peak: this simple rule spends it the moment solar falls short, at 18:00, and it is empty by the 19:30 peak.
Grid-following and grid-forming inverters
Almost every solar farm, wind turbine and battery connects to the grid through an inverter. The conventional kind is grid-following.
Definition— Grid-following and grid-forming inverters
A grid-following inverter measures the grid's voltage and phase with a phase-locked loop and injects a controlled current in step with it: it behaves as a current source that follows the voltage set by others. A grid-forming inverter behaves instead as a voltage source behind an impedance, setting its own voltage magnitude and frequency and letting its current be whatever the network draws, adjusting its frequency and voltage with power output through droop laws or by emulating the swing equation of a synchronous machine (a virtual synchronous machine).
The difference matters in exactly the situations of The power grid. A grid-following inverter needs a strong, stable voltage to follow; on a weak grid, or during a disturbance, its phase-locked loop can lose track. A grid-forming inverter responds instantly and naturally to a change in load or frequency, as a synchronous machine does, so it provides inertia-like response, helps stabilise voltage, can run an island and can restart a dead network. Its limits are those of its transistors: it cannot supply large fault currents for long, so the headroom for that response must be designed in, usually in batteries.
As inverter-based resources reach high shares, operators have begun to require or buy grid-forming capability. Australia's AEMO, Great Britain's system operator NESO and North America's NERC have published specifications and guidance, and large grid-forming batteries have been operating in Australia and elsewhere since the early 2020s. How such inverters should behave, be specified and be tested is an active area of standards work in 2026.
Electric vehicles and smart charging
An electric car carries a battery of typically 40 to 100 kWh. Charged at home from a 7.4 kW single-phase charger, a 60 kWh battery fills from empty in about eight hours; a public DC fast charger can add most of a charge in well under an hour, using the converters of Power electronics. The IEA's Global EV Outlook 2026, published in May 2026, tracks sales, charging and battery supply worldwide, and expected close to 30% of new cars sold in 2026 to be electric.
Millions of cars charging at once add a large load, but a flexible one: a car is typically parked for most of the day and needs only a few hours of charging. Smart charging shifts it to when the grid has spare capacity or cheap, clean energy, overnight or at the solar peak, by tariffs, by signals from the network operator, or by an aggregator that manages many cars as one resource. Vehicle-to-grid (V2G) goes further, letting cars discharge into the grid or a house at peak times; it needs bidirectional chargers and the communication of standards such as ISO 15118-20, and in 2026 it is moving from pilots to early commercial services in several countries. The engineering questions are the familiar ones: distribution transformers and feeders sized for diversity of demand, local voltage limits, and battery degradation from the extra cycling.
The energy transition, country by country
The transition looks different everywhere, and the data change quickly; the figures below are from Ember's Global Electricity Review 2026 (April 2026), which has country-by-country data, and you should check its latest edition.
- China led the world's solar build-out in 2025, accounting for more than half of the global increase in solar capacity and generation; solar and wind reached 22% of its generation, above the OECD average of 20%, and its fossil generation fell by 0.9%, its first fall since 2015. China is also the largest manufacturer of solar modules, batteries and electric cars.
- India's fossil generation fell by 3.3% in 2025, and it installed more new solar capacity than the United States for the first time. Its single national synchronous grid makes moving solar power from the sunny west and wind from the south across the country a central engineering task.
- The European Union has some of the highest shares of wind and solar of any large region, with large differences between countries; its challenges are interconnection, flexibility and the lessons of the Iberian blackout about voltage control.
- The United States has very different grids, from Texas, where wind, solar and batteries have grown rapidly in a market with little connection to its neighbours, to California, with large battery fleets that shift midday solar into the evening.
- Brazil has a grid dominated by hydropower, with wind and solar growing rapidly in its sunny, windy north-east, linked to the south by long transmission lines including HVDC.
- Kenya generates most of its electricity from renewable sources, with geothermal power from the Rift Valley as a steady base; South Africa, long dependent on coal, endured years of rolling load shedding in the early 2020s, during which households and businesses installed rooftop solar and batteries on a large scale.
- Australia has one of the highest shares of rooftop solar in the world; at times its grids, especially in South Australia, run with wind and solar supplying more than local demand, which is why its operator has been an early specifier of grid-forming inverters.
Common misconceptions
Misconception
"A solar panel's rated power is what it produces." The rating is at standard test conditions: 1000 W/m² and a cell temperature of 25 °C. Real panels rarely see both at once; on a hot sunny roof they produce noticeably less, and annual energy depends on the climate, orientation, shading and the capacity factor of the site.
Misconception
"A better wind turbine could capture nearly all the wind's energy." No rotor in an open stream can exceed the Betz limit of 16/27: stopping the air completely would stop the flow through the rotor. Real progress comes from larger rotors, taller towers and better sites, not from beating Betz.
Misconception
"Storage makes energy." Every store returns less than it takes: round-trip efficiencies are about 85 to 95% for lithium-ion batteries, 70 to 80% for pumped hydro, and far lower for hydrogen. Storage is valuable because it moves energy to when it is worth more, and because it can respond in milliseconds, not because it adds energy.
Misconception
"Electric cars will overload the grid." Unmanaged, simultaneous evening charging could strain local distribution networks, and some reinforcement is needed. But most charging can be shifted, and managed charging can even help a grid absorb midday solar or overnight wind. The outcome depends on tariffs, controls and planning, not on the cars themselves.
Exercises
1. Renewables, storage and electric vehicles
Warm-upHumanA wind turbine has a rotor 100 m in diameter, and the wind blows at 8 m/s with air density 1.225 kg/m³. What is the most power any rotor of that size could extract, according to the Betz limit? Give the answer in kilowatts, to the nearest 10 kW.
2. Renewables, storage and electric vehicles
Warm-upHumanA 3 MW wind turbine generated 7,900 MWh in a (non-leap) year. What was its capacity factor, in per cent, to the nearest whole number?
3. Renewables, storage and electric vehicles
CoreHumanA pumped-storage plant's upper reservoir holds 3 million m³ of water, 250 m above its turbines. If the turbines and generators are 90% efficient, how much electrical energy can it deliver from a full reservoir, in MWh (nearest whole number)? Use m/s² and 1000 kg/m³ for water.
4. Renewables, storage and electric vehicles
CoreHumanWhich statements about lithium iron phosphate (LFP) cells, compared with nickel-manganese-cobalt (NMC) cells, are true? Select all that apply.
5. Renewables, storage and electric vehicles
CoreHumanWhat is the essential difference between a grid-forming inverter and a grid-following inverter?
6. Renewables, storage and electric vehicles
CoreComputationalWrite perturb_and_observe(power, v_start, step, iterations), the maximum power point tracker of the lesson, for any function power(v). Start at v_start with a previous power of 0; at each iteration measure power(v), reverse the step if the power fell, then move by the step. Return the list of voltages at which power was measured. The tests use a smooth hill and a solar-module-like curve, starting from both sides of the peak.
7. Renewables, storage and electric vehicles
StretchComputationalIn the lesson's island grid, a 25 MW, 50 MWh battery dispatched by the simple rule was empty by the evening peak. Write peak_shaving(net_load, power_mw, energy_mwh, threshold, efficiency=0.9, soc0=0.0), which instead charges from any surplus solar, discharges only the part of net load above threshold, and otherwise holds its charge. Return (gas, soc), lists of the gas output (MW, zero when there is surplus) and the stored energy (MWh) after each hour. Use a one-way efficiency of sqrt(efficiency) in each direction. The tests check that the same battery now holds the gas peak to 75 MW and 72 MW, while respecting its limits.
Further reading and resources
- Global Electricity Review 2026 · Ember · 2026Electricity generation by source for the world and for individual countries in 2025, published April 2026.
- Global EV Outlook 2026 · International Energy Agency · 2026The IEA's annual report on electric vehicle sales, charging and batteries, published May 2026.
- IRENA (International Renewable Energy Agency)Statistics on renewable capacity and generation, and annual reports on generation costs and capacity factors by technology.
- International Hydropower AssociationData and reports on hydropower and pumped storage worldwide.
- Australian Energy Market Operator (AEMO)Operator of a grid with some of the world's highest shares of rooftop solar, and an early specifier of grid-forming inverters.
- Das Maximum der theoretisch möglichen Ausnützung des Windes durch Windmotoren · Albert Betz · 1920The paper that derived the limit on the power a wind turbine can extract.