Overview
Key functions
- Prepare and solve:
newModel(),setGeoscale(),solve_base_year(),interpolate_model(),solve_scenario(). - Analyse:
getData(),autoplot(),levcost(). - Publish:
report().
This vignette takes the packaged
[topia]$modules model – the one built object
by object in TOPIA I: building the model – assembles it and
runs scenarios on it.
library(energyRt)
library(dplyr)
library(ggplot2)
have_sf <- requireNamespace("sf", quietly = TRUE) # maps need sf
have_pal <- requireNamespace("energypal", quietly = TRUE) # palette + order
set_scenarios_path(file.path(tempdir(), "topia"))
set_default_solver(solver_options$glpk) # set once, used by every solve
set_default_solver(solver_options$julia_highs)
# get_default_solver()Two calls run every scenario in this vignette:
interpolate_model() expands the model over regions, years
and timeslices; solve_scenario() writes the solver files,
runs the default solver (set once above) and reads the solution.
verbose = FALSE keeps the solver log out of the page — drop
it to watch the solver work.
The horizon marks 2025 as its base year, so each scenario is
two solves and not one: the base year on its own, then
the projection that starts from it. solve_base_year() does
the first, and base_year_scenario = hands its result to the
second.
The TOPIA model
topia$modules$electricity carries the model: a base
repository $repo, the regions it covers, its geoscale, and
the scenario levers used below. Calendars, horizons and map layouts sit
alongside it. (For a different layout or calendar, build the blocks
yourself following TOPIA I: building the model.)
topia_elc <- topia$modules$electricity # the model
# A SAMPLED calendar: one day per month, every hour of it, drawn from the
# model's own `d365_h24`. The kit's weather is measured on that calendar, so a
# sample of its labels resolves against the same series while the LP shrinks
# about thirty-fold -- which is what makes this vignette runnable. `s4_h24`
# would not: its labels (`WIN_h00`) are not `d365_h24`'s (`d001_h00`), so
# every weather row would be dropped and the renewables would run unbounded.
calendar_sample <- calendars[["d365_h24_1dpm"]]
horizon_BY2025_50_by10 <- topia$modules$horizons$BY2025_50_by10
topia_elc$regionsAssembling and solving the base model
set_progress_bar()
mod <- newModel("TOPIA", data = topia_elc$repo,
calendar = calendar_sample,
region = topia_elc$regions,
horizon = horizon_BY2025_50_by10, discount = 0.05)
mod <- setGeoscale(mod, topia_elc$geoscale)The base year
2025 is given, not chosen, so it is solved by itself.
solve_base_year() slices the base year out of the horizon,
switches endogenous retirement off – an observed retirement is data –
and returns the solved scenario together with the ledger of what it
decided.
A well-stated base year builds nothing: every generator standing in 2025 is already in the fleet. New capacity here means the input data understates the system, and capacity left idle means it overstates it.
by <- solve_base_year(mod, name = "BY2025", verbose = FALSE)
getData(by, "vObjective", merge = TRUE) # the base year's own cost, MUSD
# nothing should appear here
new_by <- getData(by, "vTechNewCap", merge = TRUE)
new_by[new_by$value > 1e-6, ]The projection
base_year_scenario = carries the base year forward: what
it decided becomes exogenous stock, and the solved years are 2030, 2040
and 2050.
scen_BASE <- interpolate_model(mod, name = "BASE", base_year_scenario = by,
verbose = TRUE, check = TRUE)
scen_BASE@modInp@sets$year # 2030, 2040, 2050
scen_BASE <- solve_scenario(scen_BASE, verbose = TRUE)
getData(scen_BASE, "vObjective", merge = TRUE) # projection cost, MUSDThe regions are wired west to east with bi-directional transmission
links (TBD_ELC_*), and the fuels move zone to zone.
gg_trade_map() draws that network over a map layout — pass
any of topia$geoscales, here honeycomb:
gg_trade_map(topia_elc$repo, map = topia$geoscales$honeycomb)Reading the solution
The electricity balance
Generation is only one side of the account. The bands below are drawn
from the carriers palette in energypal,
which supplies the colours and the order:
merit_order is a sequence the palette declares, so neither
is invented here. TOPIA’s own codes are mapped onto that vocabulary
explicitly – a fuzzy lookup mis-reads them, matching NUC to
nothing and EGAS to enhanced geothermal.
pal <- energypal::energypal("carriers", order = "merit_order",
include_groups = FALSE, include_subtypes = FALSE)
carrier_of <- c(ESOL = "Solar", EWIN = "Wind", EHYD = "Hydro",
ENUC = "Nuclear", EBIO = "Bioenergy",
ECOA = "FossilCoal", EGAS = "FossilGas")
stem <- function(x) sub("_(VIN|CL).*$", "", x) # drop vintage/cluster suffixesSources stack above zero and consumers below it. Electricity leaves
the system through three doors: final demand (vDemInp), the
electrolyser drawing power to make hydrogen (vTechInp), and
storage charging (vStorageInp).
gen <- getData(scen_BASE, "vTechOut", comm = "ELC", merge = TRUE)
elc_use <- getData(scen_BASE, "vTechInp", comm = "ELC", merge = TRUE)
elc_dem <- getData(scen_BASE, "vDemInp", comm = "ELC", merge = TRUE)
elc_sto <- getData(scen_BASE, "vStorageInp", comm = "ELC", merge = TRUE)
sources <- gen |>
mutate(band = unname(carrier_of[stem(tech)])) |>
group_by(year, band) |>
summarise(GWh = sum(value), .groups = "drop")
consumers <- bind_rows(
elc_dem |> mutate(band = "Electricity"),
elc_use |> mutate(band = "Hydrogen"),
elc_sto |> mutate(band = "Storage")) |>
group_by(year, band) |>
summarise(GWh = -sum(value), .groups = "drop")
balance <- bind_rows(sources, consumers)
balance$band <- factor(balance$band, levels = names(pal))
ggplot(balance, aes(factor(year), GWh, fill = band)) +
geom_col() +
geom_hline(yintercept = 0, linewidth = 0.3) +
scale_fill_manual(values = pal, breaks = names(pal), drop = TRUE) +
labs(x = "year", y = "GWh", fill = "carrier",
title = "The electricity balance (BASE)",
subtitle = "sources above zero, consumers below") +
theme_bw()Generation runs well above final demand, and the chart says why: the electrolyser is a load, and a large one. Three further readings of the same figure are worth making explicit.
- The storage band is empty – the battery is never built. At a sampled calendar the multi-day calm spells that make storage valuable are not present to be found, and the model has cheaper ways to meet a peak.
-
Trade does not appear as a consumer.
vTradeIrmoves electricity between regions rather than out of the system, and no electricity is exported to the rest of the world, so at system level trade shows up only as the small residual of line losses. Per region it is a genuine source and sink. - The gap between the two sides is exactly that residual: 3% is lost at each transmission hop.
balance |>
mutate(GWh = round(GWh)) |>
arrange(band) |>
tidyr::pivot_wider(id_cols = band, names_from = year, values_from = GWh,
values_fill = 0) |>
knitr::kable(caption = "The electricity balance, GWh (consumers negative)")The colour of the hydrogen
Hydrogen is conventionally described by the electricity that made it: green from renewables, pink from nuclear, grey from unabated fossil fuel. The model itself supports no such label. Electricity is a pooled commodity, so the solution records how much each plant generated in a region-hour and how much the electrolyser drew in the same region-hour, and nothing about which flowed into which. A colour is therefore a convention laid over the solution, and the one used here is the hourly average mix: the draw in each hour is divided among the plants generating in that hour, in proportion to their output.
The hourly resolution carries the argument. An electrolyser exists
partly to absorb surplus, so an annual average would credit it with the
year’s mix and conceal whether it actually ran when the wind blew.
timeframe = "highest" keeps the timeslice detail that
getData() otherwise aggregates away.
gen_h <- getData(scen_BASE, "vTechOut", comm = "ELC", merge = TRUE,
timeframe = "highest")
use_h <- getData(scen_BASE, "vTechInp", comm = "ELC", merge = TRUE,
timeframe = "highest")
colour_of <- c(Solar = "GreenHydrogen", Wind = "GreenHydrogen",
Hydro = "GreenHydrogen", Nuclear = "PinkHydrogen",
Bioenergy = "OtherHydrogen", FossilCoal = "GreyHydrogen",
FossilGas = "GreyHydrogen")
draw <- use_h |>
filter(stem(tech) == "H2ELC") |>
group_by(region, year, timeslice) |>
summarise(drawn = sum(value), .groups = "drop")
h2_colour <- gen_h |>
group_by(region, year, timeslice) |>
mutate(share = value / sum(value)) |>
ungroup() |>
inner_join(draw, by = c("region", "year", "timeslice")) |>
mutate(colour = unname(colour_of[carrier_of[stem(tech)]]),
GWh = share * drawn) |>
group_by(year, colour) |>
summarise(GWh = sum(GWh), .groups = "drop")
ggplot(h2_colour, aes(factor(year), GWh, fill = colour)) +
geom_col(position = "fill") +
scale_fill_manual(values = energypal::energypal_colors(
unique(h2_colour$colour), palette = "carriers")) +
labs(x = "year", y = "share of the electricity consumed", fill = "colour",
title = "The colour of the hydrogen (BASE)",
subtitle = paste("the electrolyser's draw attributed to the",
"hourly generation mix")) +
theme_bw()Most of the hydrogen is green, a third of it grey, and a small pink share comes from the nuclear fleet. The attributed total matches the electrolyser’s recorded draw, which is the check that the attribution neither invents nor loses energy:
The national average hides the more instructive result. Attribution by region shows the colour following whatever that region happens to generate: hydrogen made beside the hydro reservoirs is almost entirely green, hydrogen made in the coal regions is predominantly grey, and the region holding the nuclear fleet produces hydrogen that is substantially pink.
h2_region <- gen_h |>
group_by(region, year, timeslice) |>
mutate(share = value / sum(value)) |>
ungroup() |>
inner_join(draw, by = c("region", "year", "timeslice")) |>
mutate(colour = unname(colour_of[carrier_of[stem(tech)]]),
GWh = share * drawn) |>
filter(year == max(year)) |>
group_by(region, colour) |>
summarise(GWh = sum(GWh), .groups = "drop_last") |>
mutate(share = round(100 * GWh / sum(GWh), 1)) |>
ungroup()
h2_region |>
tidyr::pivot_wider(id_cols = region, names_from = colour,
values_from = share, values_fill = 0) |>
knitr::kable(caption = "The colour of the hydrogen by region, final year (%)")The convention is one of several defensible ones, and the choice changes the answer. An average attribution of this kind credits the electrolyser with the mix it ran alongside. A marginal attribution would instead ask which plant changed its output because the electrolyser ran, which requires either the duals or a counterfactual solve. A stricter hourly-matching rule would count only renewable generation in the same hour and leave the remainder uncoloured. None is more correct than the others; what matters is that the rule is stated alongside the number.
The dispatch over a representative summer day – generation stacked by
technology, storage charging below zero, demand as the line
(autoplot(scenario) reads the solution via
getMix(); note that getData() aggregates
sub-annual values to ANNUAL by default – pass
timeframe = "highest" for timeslice-level data):
autoplot(scen_BASE, "generation", timeslice = "^SUM_")Scenarios: layering the levers
Each scenario adds a pre-built lever to the base model. The pattern
is always
interpolate_model(mod, name, LEVER, base_year_scenario = by)
then solve_scenario(). Every scenario starts from the same
base year: 2025 is an observation, so a lever cannot change it.
# emissions cap
scen_CO2CAP <- interpolate_model(mod, "CO2CAP", topia_elc$CO2_CAP,
base_year_scenario = by) |>
solve_scenario(verbose = FALSE)
# carbon tax
scen_CTAX <- interpolate_model(mod, "CARBONTAX", topia_elc$CT_CO2,
base_year_scenario = by) |>
solve_scenario(verbose = FALSE)
# early fossil retirement
scen_ERET <- interpolate_model(mod, "EARLY_RET", topia_elc$EARLY_RET,
base_year_scenario = by) |>
solve_scenario(verbose = FALSE)
# renewable-backed hydrogen
scen_GRN <- interpolate_model(mod, "GREEN_H2", topia_elc$GREEN_H2,
base_year_scenario = by) |>
solve_scenario(verbose = FALSE)The third lever, EARLY_RET, puts a declining per-region
ceiling on installed coal and gas capacity – the plants carry
optimizeRetirement = TRUE, so the model chooses
which units to close first while the constraint sets the pace.
The ceiling bites hardest on gas; the generation chart above shows what
fills the gap.
Whether the battery is built at all is worth reading as a result
rather than assumed. The model carries several ways to meet a peak – a
legacy gas fleet alive to 2050, unlimited LNG above the domestic cap,
transmission to a neighbour, and an electrolyser that can soak up
surplus and sell the hydrogen – so shifting electricity in time has to
beat all of them before it earns its capital. Check
vStorageOutCap across the scenarios before concluding
either way, and note that on a sampled calendar the multi-day calm
spells that make storage valuable are not present to be found: the
sample takes one day per month, so nothing in it lasts longer than a
night.
Compare the scenarios with compare_scenarios(): the
overview carries every run’s objective (with differences against the
baseline), the value tables flag which solved variables moved, and
autoplot() draws the comparison charts. The full
comparative document is one report(cmp) call away.
scl <- list(BASE = scen_BASE, CO2CAP = scen_CO2CAP,
CARBONTAX = scen_CTAX, EARLY_RET = scen_ERET,
GREEN_H2 = scen_GRN)
cmp <- compare_scenarios(scl)
cmp
autoplot(cmp, "emissions")Total system cost (objective) by scenario, straight from the overview:
(The same charts can always be hand-rolled from the tidy layer –
getData(scl, ...) and getMix(scl, ...) accept
named scenario lists and add a scenario column.)
A carbon tax and a CO2 cap both cut emissions but through different mechanisms – a price the model responds to, against a quantity it must meet – and the two reach a given level of abatement at different cost. That contrast is the point of running them side by side.
Requiring green hydrogen, and what that does not buy
The GREEN_H2 lever makes the matching rule of the
previous section binding rather than descriptive: in every region and
timeslice, the electrolyser’s draw may not exceed renewable generation
at that moment. It is the test used to certify renewable hydrogen, and
it is costly – hydrogen output falls and total system cost rises, both
visible in the comparison above.
What it does not do is turn the attribution wholly green.
colour_share <- function(scen) {
g <- getData(scen, "vTechOut", comm = "ELC", merge = TRUE,
timeframe = "highest")
u <- getData(scen, "vTechInp", comm = "ELC", merge = TRUE,
timeframe = "highest")
d <- u |>
filter(stem(tech) == "H2ELC") |>
group_by(region, year, timeslice) |>
summarise(drawn = sum(value), .groups = "drop")
g |>
group_by(region, year, timeslice) |>
mutate(share = value / sum(value)) |>
ungroup() |>
inner_join(d, by = c("region", "year", "timeslice")) |>
mutate(colour = unname(colour_of[carrier_of[stem(tech)]]),
GWh = share * drawn) |>
group_by(colour) |>
summarise(GWh = sum(GWh), .groups = "drop") |>
mutate(pct = round(100 * GWh / sum(GWh), 1))
}
base_mix <- colour_share(scen_BASE)
green_mix <- colour_share(scen_GRN)
base_mix |>
select(colour, BASE = pct) |>
full_join(select(green_mix, colour, GREEN_H2 = pct), by = "colour") |>
knitr::kable(caption = "Attributed colour, whole horizon (%)")The share attributed to renewables rises but does not reach 100%, and the reason is that the constraint and the attribution answer different questions. The constraint compares two quantities in the same hour: electrolysis against renewable output. The attribution divides the electrolyser’s draw among everything generating in that hour, and the coal and gas plants serving the rest of the load are still running. A single pooled commodity cannot say which generator supplied which consumer, so one solution supports both statements at once – the hydrogen is fully backed by renewable generation, and roughly a quarter of the electricity mix it drew from was fossil.
Reconciling the two would require either a stricter rule – no fossil generation anywhere in the region while the electrolyser runs, which needs integer variables – or a physically separate green commodity that only renewable capacity feeds. Both are heavier than the teaching point they would buy, and neither is more honest than stating the convention.
Levelized cost of a process
levcost() on a solved scenario returns
the ex-post levelized cost of a process: the discounted sum of
its own costs (annualised investment vTechEac, fixed and
variable O&M, and attributed fuel) divided by its output in that
solution. Because it reads the actual results – not a mini-model – it is
exact at the model’s time resolution, with no calendar caveat.
lc <- levcost(scen_BASE, name = "ENUC") # the existing nuclear fleet
lc$levcost_npv # NPV levelized cost, MUSD/GWh
autoplot(lc, type = "components") # eac / fixom / varom / supply, by yearENUC runs on base-year capacity, so its capital
is already sunk – the annualised-investment term
eac is ~0 and the realized cost is fixed O&M
plus fuel; a newly built plant would add its
eac here. The cost per GWh also depends on how hard a plant
is run: more output spreads the fixed costs, lowering the levelized
cost. Priced across scenarios, ENUC’s LCOE tracks how much
the existing fleet is dispatched as the policy mix changes:
sapply(scl, function(s) {
lc <- suppressMessages(levcost(s, name = "ENUC"))
if (inherits(lc, "levcost")) round(as.numeric(lc$levcost_npv), 2) else NA_real_
})(levcost(scen, name) attributes fuel from
vTechInp, so it works for the single-fuel technologies
here; a plant with a grouped fuel input would need its group
activity instead.) A full technology datasheet with the ex-post cost
embedded is one call away (report() renders HTML/PDF; needs
pandoc):
