Off/Grid
Method

Four equations decide whether a site can power itself.

Every site is simulated hour by hour over a full ERA5 meteorological year. No capacity factors, no annual averages — the battery either holds through the January calm or it does not.

Every number in this tool comes from one deterministic simulation: 8 760 hourly points of ERA5 reanalysis per site, converted into power, integrated through a battery, and reduced to a single probability. No stochastics, no fitted models.

01

Wind

Ten-metre reanalysis wind is extrapolated to the 100 m hub with a power-law boundary layer, exponent 0.14 for open terrain.

v(zhub)=v10(zhubzref)α,α=0.14v(z_{hub}) = v_{10}\left(\frac{z_{hub}}{z_{ref}}\right)^{\alpha},\quad \alpha = 0.14

Air density follows the ideal gas law, so cold sites extract more power from the same wind speed.

ρ=patmRair T,patm=101 325 Pa, Rair=287\rho = \frac{p_{atm}}{R_{air}\,T},\quad p_{atm}=101\,325\ \text{Pa},\ R_{air}=287

Rotor power is the kinetic flux through a 150 m disc with a real power coefficient of 0.45 — below the Betz limit of 0.593 — clipped at the 6 MW rating and zeroed outside cut-in 3 m/s and cut-out 25 m/s.

Pwind=min⁡ ⁣(12ρAv3Cp, Pnom),Cp=0.45P_{wind} = \min\!\left(\tfrac{1}{2}\rho A v^{3} C_p,\ P_{nom}\right),\quad C_p = 0.45
02

Solar

Global horizontal irradiance scales linearly against 1 000 W/m², derated by cell temperature: the module runs 0.03 K hotter per W/m² of irradiance and loses 0.4 % of its yield per kelvin above 25 °C.

Ppv=G1000 Pc[1−γ (Tcell−25)],Tcell=T+0.03 GP_{pv} = \frac{G}{1000}\,P_{c}\left[1-\gamma\,(T_{cell}-25)\right],\quad T_{cell}=T+0.03\,G
03

Cooling and PUE

All IT power becomes heat. The chiller is modelled as a Carnot machine with 50 % isentropic efficiency, evaporating at 15 °C and condensing 10 K above ambient — so the PUE is a function of local temperature, not a constant.

COP=η TevapTcond−Tevap,PUE=1+1COPCOP = \eta\,\frac{T_{evap}}{T_{cond}-T_{evap}},\qquad PUE = 1 + \frac{1}{COP}

Below 15 °C ambient the site switches to free cooling and the PUE is fixed at 1.10. This is why the Nordic sites carry a structural advantage.

Pload=PIT⋅PUE(T)P_{load} = P_{IT}\cdot PUE(T)
04

Battery

The power balance is integrated with explicit Euler over 8 760 one-hour steps, starting at half charge, with a 90 % round-trip efficiency and hard bounds at zero and rated capacity. Surplus beyond full charge is curtailed.

Et+1=clip⁡(Et+(Pprod,t−Pload,t) Δt η, 0, C),Δt=1 hE_{t+1} = \operatorname{clip}\big(E_t + (P_{prod,t}-P_{load,t})\,\Delta t\,\eta,\ 0,\ C\big),\quad \Delta t = 1\ \text{h}
05

Loss of load probability

An hour counts as a deficit whenever demand exceeds production plus the energy the battery can actually deliver. LOLP is simply the share of such hours in the year — the one number the whole product turns on.

LOLP=#{t : deficit}8760LOLP = \frac{\#\{t\ :\ \text{deficit}\}}{8760}
06

Sizing search

For each site and IT load the grid is swept exhaustively, and the cheapest configuration meeting the LOLP threshold is retained, at €8 M per 6 MW turbine, €0.7 M per MWp of solar and €0.25 M per MWh of storage.

min⁡ (8 nturb+0.7 Ppv+0.25 Ebatt) M€s.t. LOLP≤LOLPmax\min\ \big(8\,n_{turb} + 0.7\,P_{pv} + 0.25\,E_{batt}\big)\ \text{M€}\quad \text{s.t.}\ LOLP \le LOLP_{max}

Data source

Open-Meteo / ERA5 reanalysis, calendar year 2023, 8760 hourly points per site: 100 m wind speed, global irradiance and 2 m temperature.

Search space

A precomputed grid over turbine count, PV capacity and battery energy, evaluated for three IT loads — 10, 25 and 50 MW.

Output

Loss-of-load probability, deficit hours and CAPEX for every combination, plus the cheapest sizing that meets each LOLP threshold.