Translation

library(reneuro)
library(energyRt)

This article documents how a PyPSA network becomes an energyRt model: which component becomes which object, where a slot’s value comes from, and the four places where the two representations do not correspond one to one.

It describes the conversion behind the shipped models. Reading it is not required to use them — vignette("reneuro") covers that — but it is what you need to check a converted number against its source, or to decide whether a question the model answers is one the translation supports.

Scope is the electricity-only network at the base_s_{clusters}_elec.nc cut point: after the network is built, before policy overlays and temporal averaging are applied.

The component table

PyPSA	energyRt
`Bus` carrier	`commodity`	bus carriers only — see below
`Load`	`demand`	series in `@demand`, one row per region and timeslice
`Generator`	`technology` (+ `supply`)	one object per carrier, regions in a `region` column
`Generator` `p_max_pu` series	`weather`	kept out of the technology object
`StorageUnit`	`storage`	one object; `duration` from `max_hours`
`Store` + two `Link`s	`storage`	one object with three part capacities
`Line` (AC)	`trade`	transport, not Kirchhoff — see below
`Link` (DC)	`trade`	transport, faithful
`StorageUnit` `inflow`	`commodity` + `supply` + `storage` + `technology`	four objects — see below

Everything below is a case where that one-line summary is not the whole story.

`carrier` names two different things

PyPSA’s carrier field is used for two purposes, and only one of them is a commodity:

on a bus it names the energy flowing there — AC, H2, battery — and that is a commodity;
on a generator or link it names the technology or the fuel — CCGT, onwind, battery charger — and that is not.

So commodities are taken from bus carriers, and generator carriers become technology names. The carriers component itself is a lookup table of attributes, not a set of objects to create.

Generators carry their fuel explicitly

In an electricity-only network a generator has no fuel bus: marginal_cost already folds fuel price divided by efficiency into one number. Translating that literally gives an output-only technology — a CCGT that consumes no gas. It is arithmetically equivalent, but it cannot carry a fuel price, a resource limit, or an emission factor.

Thermal carriers therefore get an explicit fuel @input with combustion = 1 and the efficiency in @ceff$cinp2use. CO₂ then follows from the fuel’s carbon content on the commodity, rather than from a coefficient on the technology.

The fuel supply is priced at zero and varom keeps the whole marginal_cost. The fuel price is not recoverable from the network file, and charging the supply as well would count it twice. This is the one place in the translation where double counting could enter.

Capacity depends on whether the generator is extendable:

PyPSA	energyRt
not extendable	`@capacity$stock` — exogenous, no capital charged
extendable	`@capacity$cap.lo` / `cap.up` — built and paid for

Storage: two shapes in, one shape out

PyPSA has two storage representations and a network may use both:

StorageUnit — one component carrying a power rating p_nom, a duration max_hours, and a round trip split into efficiency_store and efficiency_dispatch.
Store + two Links — an energy reservoir on its own bus, plus a charger and a discharger that carry the conversion efficiencies and their own ratings.

Both become one energyRt storage, because a storage has three independent capacity variables — charge, discharge and reservoir — so the second shape needs no extra objects:

stg <- getObject(pypsa_eur_5, class = "storage")[["STG_BATTERY"]]
grep("^(inp|out|stg)\\.stock$", names(stg@capacity), value = TRUE)
#> [1] "out.stock" "inp.stock" "stg.stock"
stg@seff[, c("stgeff", "inpeff", "outeff")]
#>   stgeff    inpeff    outeff
#> 1     NA 0.9797959 0.9797959

inp.* is the charger, out.* the discharger, stg.* the reservoir, and @seff carries the two conversion efficiencies plus standing loss. A converted model contains no separate charger or discharger technologies.

Transmission is a transport model

PyPSA is a hybrid: Kirchhoff’s voltage law applies to the AC Lines, which form sub-networks and receive cycle constraints, while DC Links are free transport. energyRt’s trade is transport throughout.

The consequence is asymmetric and worth stating plainly:

DC links translate faithfully — a controllable branch is what a transport model represents natively.
AC lines are a relaxation. The converted model’s AC flows are less constrained than the source, so its objective is a lower bound, and it should not be used to draw transmission-expansion conclusions.

energyRt::newACLine() with kvl = TRUE at interpolation restores the cycle constraints where that matters.

Capacity belongs to the trade object rather than to a route, so each bus pair becomes its own object and parallel lines on a pair are merged by summing s_nom — indistinguishable in a transport model in any case.

Quadratic AC losses are linearised. The default gives each corridor one flat teff evaluated at a chosen loading; tranches = n splits it into n equal capacity tranches with rising loss rates instead, approximating the curve without integer variables. Tranches were the default once, which is why pypsa_eur_5 still carries six of them – at continental scale they cost a generated constraint per corridor, tranche and timeslice, which is why the larger models are flat.

Availability leaves the technology object

PyPSA keeps availability in two places at once. p_max_pu is a static column, and generators_t.p_max_pu is a series covering the weather-driven generators only. For every generator that has a series the static value is exactly 1, a placeholder — so reading the static column alone would make every wind farm permanently available.

energyRt keeps the bulk series out of the technology: a weather object holds region × timeslice values and the technology references it by name.

names(getObject(pypsa_eur_5, class = "weather"))
#> [1] "W_OFFWIND_AC"    "W_OFFWIND_DC"    "W_OFFWIND_FLOAT" "W_ONWIND"       
#> [5] "W_SOLAR"         "W_SOLAR_HSAT"

Hydro inflow becomes four objects

storage_units_t.inflow has no direct equivalent — an energyRt storage can be charged from a commodity, but not from an exogenous series. A hydro reservoir therefore decomposes:

object	role
`commodity` `HYDWAT`	virtual reservoir energy, balanced hourly
`supply` `SUP_HYDWAT`	availability follows the inflow series
`storage` `STG_HYDRO`	the reservoir
`technology` `E_HYDRO`	`HYDWAT` → electricity at the dispatch efficiency

Conventions that differ

Two conventions produce correct-looking numbers that mean different things, and both are places to check before comparing against a source model.

Rating side. A Link’s p_nom is rated on its input bus; an energyRt capacity is rated on activity, i.e. output. Converting one to the other scales by the efficiency: capacity = p_nom * eff, and invcost = capital_cost / eff.

Cost annualisation. Where capital_cost is already annuitised it maps to @invcost$eac, not to invcost — supplying it as invcost would annuitise a second time.

Revised assumptions

Everything above translates PyPSA-Eur faithfully — same data, same assumptions, verified against the source model’s own solutions. Three assumption families are also revised, in model versions built beside the faithful one so the two can be solved side by side and the difference read as the effect of the assumption, never of the translation. The revised versions are not among the eleven models the package ships today: they arrive with the repository update and ship as separate data.

Capacity factors

PyPSA-Eur offers each region’s whole eligible area at one blended, best-cells-weighted profile: all the land at the quality of the good sites. The revised layer replaces the blended VRE generators with resource clusters split into quality tiers — wind calibrated to the Global Wind Atlas along an aggregate park power curve, solar levels carved from the Global Solar Atlas — so capacity buys the tier it occupies, with the area each tier actually has. Fixed-tilt panels use a per-location optimal angle (measured effect vs the flat 35°: about zero in Europe — the change is kept for consistency with the atlas assumptions), and single-axis tracked and two-axis tracked variants compete alongside fixed panels at their own capacity factors, packing densities and costs. The construction is documented in the wind energy and solar energy articles.

Economic retirement

The overnight convention treats the existing fleet asymmetrically: non-extendable capacity survives to the planning year whole and free, while extendable carriers are implicitly retired in full and must be rebuilt at least to today’s size. The revised version replaces both halves with one consistent rule set:

Long-lived, non-replicable assets — nuclear, geothermal, and the hydro chain (reservoir, run-of-river, pumped storage’s sibling dam) — keep their capacity with no investment option, paying explicit fixed O&M.
Everything else in the fleet can retire economically. Fixed O&M is split out of PyPSA’s capital_cost annuity (which includes it), so the solver may retire capacity whose O&M exceeds its value (optimizeRetirement), and new builds pay the pure annuity plus the same O&M without double-counting. The fossil floors are gone: existing gas is free stock that can retire, not a rebuild obligation.
Carbon-free floors stay. Wind, solar and offshore keep cap.lo = today’s capacity — a no-backsliding assumption — and pumped-hydro and battery storage join them, rebuilt at the cost table’s annuities. Reinvested capacity lives its technology lifetime.

Demand

The measured 2025 load is a level, not a forecast: solved against 2050 costs it understates what a 2050 system must build. The revision keeps the measured hourly shapes and levels at 2025 and grows them by the TYNDP 2024 scenarios (ENTSO-E/ENTSOG, CC-BY 4.0): National Trends+ (national plans; extrapolated beyond its 2040 horizon), and the carbon-neutrality pathways Distributed Energy and Global Ambition, which follow the National Trends path to 2030 and then diverge — the scenarios’ own construction. Growth factors are per country and milestone year (tyndp_demand), anchored so 2025 equals the measured load exactly; countries the TYNDP perimeter lacks take the average factor, marked as such. scale_demand() applies a scenario to any shipped model — one year for an overnight solve, all milestones for the multi-year models.

What the translation does not do

The conversion reports rather than repairs. A parameter that cannot be represented is named and its entity excluded; it is never replaced with a plausible value. The register travels with each model:

str(attr(pypsa_eur_5, "reneuro_provenance")[c("describe", "commit", "source")])
#> List of 3
#>  $ describe: chr "v2026.02.0"
#>  $ commit  : chr "d6383ebf602767b1adbb676fe8a16e37a6e9f932"
#>  $ source  : chr "base_s_5_elec.nc"