The GB network model

Great Britain, as a solvable object. Not a diagram — an operating point.

gb-full is the national network as a case file the solver actually runs: every distribution licence area in Britain plus the transmission spine, distilled from 38,109 real network nodes, geolocated bus by bus, and shipped already converged. Load it, and you are holding the country at a residual of 3.6×10−11 pu. This page is the datasheet.

0
buses
0
branches
0 GW
demand carried
0/14
licence areas
The reduction pipeline

Thirty-eight thousand raw nodes walked in. Three and a half thousand buses walked out.

The DNOs publish their networks the way accountants publish ledgers — exhaustively, and in a form you cannot solve. Long-term development statements model every busbar section, every zero-impedance jumper, every star point as its own node. The pipeline's job is to collapse the bookkeeping without touching the electricity.

Ingest the raw publications 38,109 nodes

Every node the DNO and NESO publications describe, taken in whole from the lake — long-term development statements, asset registers, the TEC register, GSP geography. No sampling, no "representative subset". The full ledger, exactly as Britain filed it.

Zero-impedance collapse → 4,240 buses

Nodes joined by zero-impedance links are the same electrical point wearing different name badges. The collapse merges 2,910 busbar ties and dissolves 10 star nodes — pure bookkeeping removed, admittance matrix untouched in meaning.

Busbar-section merge → 3,539 buses

The subtle one. UKPN's three areas (EPN, LPN, SPN) and SP Manweb publish busbar sections as separate nodes without the couplers that join them — so a single substation arrives as two or three disconnected fragments. The merge rejoins 702 split section nodes, gated on matching name and matching voltage, so two distinct substations can never be combined by accident.

Re-validate the reduction machine ε

After every reduction stage the model is re-solved and re-checked by the independent residual instrument. The reduced model re-validates to the same machine-epsilon residual — the merge is electrically faithful, and I can prove it rather than assert it.

The point of the exercise is not a smaller number. It is a model where every bus is a real electrical node with a real name, a real voltage and a real place on the map — and nothing else.

The inventory

What is actually in the case file. Line by line.

A network model is a list of promises. Here is the full set, so you can audit mine before you rely on them.

Quantity Value Notes
Branches
Branches, total 5,918 every one carrying an impedance and a rating
— series branches 2,719 lines and cables
— transformers 1,413 modelled with ratio and impedance
— of which OLTCs 1,383 on-load tap changers under live tap control in the solve loop, not frozen at nominal
Demand
Demand carried 48,870 MW real LTDS demand, not a uniform smear
Load points 1,141 each attached where the DNO says the load lives
Generation
Embedded generation, capacity 127,616 MW from the embedded capacity registers
Embedded generation, dispatched 49,867 MW the operating point actually solved
Transmission generation (TEC-attached) 92,997 MW 572 units, placed from the TEC register
Topology & reference
Islands 2 largest island 3,536 buses
Slack bus bus 194 — Melksham the spine node with the largest incident branch rating: 21,000 MVA. Chosen by the network, not by taste.
Voltage filter ≥132 kV lower voltages appear only as attachment points
Series branches
0
lines and cables
Transformers
0
1,383 OLTC-controlled
Load points
0
48,870 MW carried
TEC generation units
0
92,997 MW attached
The voltage ladder

Seven voltage tiers. From the supergrid down to the substation gate.

The model is filtered to 132 kV and above as a network; the sub-132 kV buses that remain are attachment points — the places where demand and embedded generation reach up and grab the grid.

Nominal voltage Buses Role
400 kV 113 the supergrid — the country's spine
275 kV 203 transmission, the older half of the spine
132 kV 2,281 the distribution backbone — where most of Britain actually hangs
66 kV 54 attachment points
33 kV 843 attachment points
25 kV 34 attachment points (railway supplies, mostly)
22 kV 11 attachment points
Total 3,539

Note where the mass is: 2,281 of the 3,539 buses sit at 132 kV. Great Britain is not 400 kV with some detail bolted on — it is a distribution country with a transmission spine, and the model respects that ratio.

Coverage

All fourteen licence areas. The distribution half of the story most tools ignore.

Most GB studies stop at the transmission spine and hand-wave the rest. This model was assembled the long way — from NGED, UKPN, Northern Powergrid, SSEN, SPEN, ENW and NGET published data, area by area, with each area's buses counted, named and placed.

Area Operator Buses
EMIDNGED — East Midlands432
ENWLElectricity North West168
EPNUKPN — Eastern321
LPNUKPN — London126
NGETNational Grid — the transmission spine316
NPGNNorthern Powergrid — Northeast166
NPGYNorthern Powergrid — Yorkshire372
SEPDSSEN — Southern369
SPMSP Energy Networks — Manweb224
SPNUKPN — South Eastern185
SWALESNGED — South Wales336
SWESTNGED — South West227
WMIDNGED — West Midlands297
TOTAL 14/14 licence areas represented 3,539

The two Scottish licence areas are carried on the spine: their demand and generation attach at NGET GSP buses rather than as separately-modelled distribution networks — see the full disclosure below.

Thirteen distribution and transmission networks, seven publishing organisations, one coherent admittance matrix. If your study touches a GSP, a DNO boundary or an embedded fleet, the network it lives on is already in the file.

Geography

Every bus knows where it is. Zero unplaced.

A bus without coordinates is a spreadsheet row. A bus with coordinates is a substation. Geolocation was done the honest way — direct matching first, GSP-region hinting second, and the stragglers run to ground rather than quietly deleted.

Direct-matched
0
named-site coordinate match
GSP-hinted
0
placed within their GSP region
Initially unlocated
0
every one resolved
Unplaced in final model
0
the number that matters

The GeoJSON layers it ships with

The model is not just solvable, it is drawable. Three ready-made layers drop straight onto any web map or GIS tool, in plain GeoJSON with no proprietary wrapper.

Layer Size Contents
assets.geojson 5.2 MB every bus and branch, placed
dno_areas.geojson 1.3 MB the fourteen licence-area polygons
gsp_regions.geojson 7.6 MB Grid Supply Point region boundaries

The boundaries NESO actually manages

The ETYS boundary corridors are modelled with their real MW limits — not decorative lines on a map, but constraints the solver can bind against. Run a constraint study and it runs against the same boundaries the control room watches.

B4 — Scotland internal B6 — Anglo-Scottish B7 — upper North of England B8 — North Midlands SEIMP — South East import

Each corridor carries its published transfer limit, so a boundary-flow question gets a number with a constraint attached — and a verdict when the constraint binds.

The solved state

The model ships warm. A converged country, out of the box.

Most tools hand you a topology and wish you luck with initialisation. gb-full ships with a converged operating point baked in — solved at 0.70 × LTDS maximum demand, taps set, voltages banded, losses accounted. Your first solve is a warm start on a national grid.

Newton iterations
3
final pass
OLTC passes
8
to settle 1,383 taps
Max mismatch
3.6×10−11
pu
Voltage band
0.9802–1.0457
pu, network-wide
Losses
267
MW at the operating point
Taps off-nominal
953
doing real voltage work

Read the convergence, don't take my word

Three Newton iterations on the final pass is what a well-conditioned national model looks like. Eight OLTC passes means the tap-control loop genuinely hunted — 953 of 1,383 taps ended off-nominal, which is the model doing the voltage regulation the real network does, rather than pinning everything at 1.0 pu and calling it a day. And the voltage band sits comfortably inside statutory limits without a single cosmetic clamp.

Behind the solved case sits the full asset graph it was distilled from — 734,913 assets and 1,016,912 edges — so every bus can be traced back through the reduction to the published records that put it there.

gb-full — solve summary CONVERGED
case:                gb-full
operating_point:     0.70 × LTDS maximum demand
newton_iterations:   3          # final pass
oltc_passes:         8          # 1,383 controlled taps
max_mismatch_pu:     3.6e-11
voltage_band_pu:     [0.9802, 1.0457]
losses_mw:           267
taps_off_nominal:    953
slack:               bus 194 "Melksham" (21,000 MVA)
asset_graph:         734,913 assets / 1,016,912 edges
status:              CONVERGED — ships in this state
Full disclosure

One part of this model is reconstructed. Here is exactly which part.

The synthetic-spine disclosure

The transmission spine's wiring is geometric reconstruction — because NGET publishes everything except how the circuits connect.

NGET does not publish circuit-by-circuit connectivity. Nobody outside the control room has the real wiring diagram, and any vendor implying otherwise is selling you a guess without the label. My guess wears the label: the spine topology is a Delaunay/Gabriel geometric reconstruction over the real GSP sites — triangulate the actual substation locations, keep the edges physics and geography would keep, dress them with real ratings.

Everything the spine connects is real: real buses, real demand, real generation, real transformers, all from published DNO and NESO data. The reconstruction is the wiring between them, nothing more. And it has artefacts I will name rather than bury — the Alverdiscott–Pembroke corridor spans 96 km over the Bristol Channel, a line no sane planner would build, kept because the reconstruction needs it for connectivity. Scottish demand and generation attach at GSP spine buses rather than through separately-modelled Scottish distribution networks.

None of this moves the physics: the model re-validates to machine epsilon either way. But you deserved to know before you licensed it, not after.

This is the only asterisk on the page. Everything above it — the bus counts, the demand, the taps, the residual — is exactly as stated.

The model is one case file. The lake behind it is the other half of the licence.

gb-full ships with GridSim Core; the GDA data lake keeps it pinned to the public record, settlement period by settlement period.