MV collection system
The MV collection system is the plant between the unit transformers and the substation: the medium-voltage feeders that gather every PCS block onto one bus, together with the cable, taps and terminations that make up those runs.
It is built radial — a breaker at the collection bus, blocks tapped along the cable as it runs out across the site, no normal parallel path — so each feeder carries the summed output of everything behind it and the segment nearest the bus is the most heavily loaded metre of the plant.
Two limits govern the design and they are not the same limit: ampacity, which asks what the conductor survives, and the loss budget, which asks what the aluminium costs in delivered megawatt-hours across twenty years of cycling. It is also the last electrical plant between the converters and the revenue meter, which is one reason the megawatts recorded at the point of interconnection are never the megawatts leaving the PCS units.
Reviewed August 2026 by Sergey Syrvachev
New to BESS? Start free with the 7-email fundamentals course — no cost, no account.
What it is (precise)
The collection system begins at the medium-voltage terminals of each unit transformer and ends at the substation MV bus, where the main power transformer takes over. Upstream of it sits the block — a PCS at low voltage, typically in the 400-800 V range, and the step-up transformer that lifts it to a collection level commonly between 13.8 and 34.5 kV. Downstream sits the substation. Between them is cable: commonly single-core XLPE with a metallic screen, one circuit of three cores per feeder, direct-buried or pulled into a duct bank, tapped or teed at each block position.
The circuit is three-wire — there is no neutral to tap on an MV feeder, which is why station service comes from its own auxiliary transformer rather than off the collection run. The equipment at either end belongs to other entries: the feeder breakers, the bus and their ratings to MV switchgear, the step-up itself to the transformer entries, and the neutral arrangement that decides how a ground fault behaves to the grounding system.
The topology of a site falls out of one piece of arithmetic. Apparent power on a three-phase feeder is S = √3 × V(L-L) × I, so at 34.5 kV one megawatt is about 17 A and a 4.4 MVA block draws about 74 A. A feeder held to 400 A therefore carries about 23.9 MVA — five of those blocks load it to roughly 368 A — and a 600 A feeder carries about 35.9 MVA.
Run the same division over a whole plant: 100 MW at 34.5 kV is roughly 1,700 A in total, which at a few hundred amperes per cable is four or five feeders, each with a string of blocks tapped along it. That is why a plant single-line draws one feeder in detail and ghosts the rest — the repeated unit is the feeder, and the design work is deciding how much load goes on each one and how far it has to travel.
The distinction worth holding onto is that a feeder is not one current. Each segment carries only what is behind it, so the cable between the bus and the first tap sees the sum of every block on the run, while the last section sees one block.
Ampacity is judged on the worst segment, voltage drop accumulates along the whole length, and loss is the sum over segments of 3 × I² × R with a different I in every term. Sizing one conductor for the whole feeder is a common simplification, not a physical fact, and the choice to keep or break that simplification is where most of the money on this page changes hands.
Feeder loading: how many blocks on one cable
Ampacity for a buried MV feeder is a computed number, not a catalogue lookup. Cable makers publish tables derived by the method of IEC 60287, and IEEE Std 835 serves the same purpose in North American practice; either way the published figure carries an assumed installation, and the design number is what survives correction for soil thermal resistivity, burial depth, ambient ground temperature, duct versus direct burial and — usually the largest single correction — the mutual heating of several circuits sharing a trench.
Screen bonding matters too: bonding the metallic screens at both ends allows circulating currents that add loss and cut ampacity, while single-point bonding avoids that at the cost of a standing screen voltage that has to be limited and managed. A feeder loaded to its free-air catalogue rating is loaded above its trench rating, and the trench is the installation the project actually built.
With a derated ampacity in hand, the block count is division, but the answer is deliberately short of the limit. Reactive dispatch consumes the same amperes as real export. The charge direction loads the feeder just as hard as discharge. Soil dries out and ambient ground temperature rises in the years after the survey.
And the feeder breaker and its relay impose their own ceilings — the equipment ratings belong to MV switchgear, the settings to the protection relay, and the conductor has to sit inside both. Five 4.4 MVA blocks at 368 A on a feeder derated to 400 A is a working example of the shape of that answer: enough room for the reactive obligation and a hot summer, not enough to add a sixth block later.
Radial topology sets the exposure. There is no normal parallel path, so a fault anywhere on a feeder removes every block behind it until the cable is repaired — on a five-feeder plant that is about a fifth of the export, and a buried cable fault is located and jointed in the field, a different class of outage from resetting a converter.
Some sites answer that with a normally open tie between the ends of two feeders, which allows part of a faulted run to be picked up from the other side after switching. That is a decision made in the switchgear and the protection scheme, and it has to be made before the trench is closed, because the tie cable is part of the same civil works.
S = √3 × V(L-L) × I: at 34.5 kV a 400 A feeder carries about 23.9 MVA and a 600 A feeder about 35.9 MVA, and the currents drawn here are that arithmetic at 34.5 kV, and collection systems are commonly 13.8–34.5 kV — the same megawatts on a 13.8 kV feeder draw two and a half times the amps. Ampacity is computed rather than looked up: IEC 60287, or IEEE Std 835 in North American practice, corrected for soil thermal resistivity, burial depth, ambient, duct versus direct burial and trench grouping. Where the blocks sit changes the loss as much as the conductor does: 400 A through 1.5 km at an assumed 0.10 Ω/km is 3I²R = 72 kW, or 0.30% of 23.9 MVA, if the load is lumped at the far end, and about 32 kW or 0.13% with the blocks tapped along the run. Five equal taps at equal spacing give (n+1)(2n+1)/6n² = 0.44 of the all-at-the-end loss, tending toward one third.
- What it spans
- MV terminals of each unit transformer to the substation MV bus — feeders, taps, terminations and the collection bus; commonly 13.8-34.5 kV, three-wire, no neutral
- Feeder arithmetic
- S = √3 × V(L-L) × I: at 34.5 kV a 400 A feeder carries ~23.9 MVA and a 600 A feeder ~35.9 MVA; a 4.4 MVA block draws ~74 A, so five blocks load a 400 A feeder to ~368 A
- Ampacity is computed, not looked up
- Tables per IEC 60287 (or IEEE Std 835 in North American practice), corrected for soil thermal resistivity, burial depth, ambient, duct vs direct burial and trench grouping; both-ends screen bonding adds circulating-current loss and lowers it further
- Loss identity
- Loss fraction = √3 I R / (V(L-L) cos φ); resistive percent drop = √3 I R cos φ / V(L-L) — the same number at unity power factor, and loss % = resistive drop % ÷ cos²φ in general
- Worked feeder
- 400 A through 1.5 km at an assumed 0.10 Ω/km: 3I²R = 72 kW, 0.30% of 23.9 MVA — with the blocks tapped along the run instead of lumped at the end, ~32 kW or 0.13%
- Distributed-tap factor
- n equal taps at equal spacing give (n+1)(2n+1)/6n² of the all-at-the-far-end loss — 0.44 at five taps, tending toward one third
- Ledger position
- A fraction-of-a-percent line beside ~0.5-1% per transformer stage — load-shaped and quadratic, subtracting on discharge and adding on charge; at 0.95 power factor the same MW costs ~5% more current and ~11% more loss
- Not the same as
- MV switchgear (the breakers and bus at either end), the unit transformer (the step-up into it), the substation (the main transformer out of it)
Sizing against the loss budget, not only ampacity
The loss on a feeder has a clean closed form worth memorising. Total conductor loss is 3 × I² × R against a delivered √3 × V(L-L) × I × cos φ, so the loss as a fraction of delivered power is √3 × I × R / (V(L-L) × cos φ). The resistive part of percent voltage drop on the same run is √3 × I × R × cos φ / V(L-L). At unity power factor those two percentages are the same number — a feeder at 1% resistive drop is a 1% loss line — and in general the loss percentage is the resistive drop percentage divided by cos²φ.
Both scale linearly with length and with current, which is what makes a feeder loss budget something you can do on one line per run before any software is opened. Use the hot resistance: at roughly 0.4% per degree C, a conductor at 90 degrees C carries about 28% more resistance than its 20 degrees C catalogue value, and the ampacity check assumes the conductor is at its rated temperature.
Worked through on the 400 A feeder: take 1.5 km at an assumed 0.10 Ω/km — a large aluminium conductor at operating temperature — for 0.15 Ω per phase. Lump all the load at the far end and the loss is 3 × 400² × 0.15 = 72 kW, which is 0.30% of 23.9 MVA and matches the identity above exactly. Real feeders do not lump; the blocks are spread along the run.
For n equal taps at equal spacing the feeder loss is (n+1)(2n+1)/6n² of the lumped case — 0.44 at five taps, tending toward one third as the taps get more numerous — so the same run lands nearer 32 kW, about 0.13%. That is how a collection network comes out at a fraction of a percent while every individual segment looks unremarkable, and it is also why the distribution of blocks along a feeder, not just their number, is a design variable.
Three levers move the budget and each has a cost. Cross-section buys resistance down for the life of the asset, and since the loss is quadratic in current it is largest at exactly the contracted full-export condition the guarantees are tested at. Conductor tapering — a heavier section near the bus, lighter toward the end — tracks the actual segment currents, at the price of more splice kits, more spare types in the store, and less freedom to re-tap the run later.
And the collection voltage itself is the biggest lever of all: at 13.8 kV the same power is 2.5 times the current and 6.25 times the loss through the same conductor, which is the real content of a 13.8-versus-34.5 kV trade study once switchgear and transformer costs are on the other side of the ledger. Whichever way it resolves, the trench is opened once.
Why POI megawatts are not PCS megawatts
Between the converter terminals and the revenue meter sit the unit transformer's load and no-load losses, the collection cables, the main transformer and the auxiliary load — roughly 0.5-1% per transformer stage plus a fraction of a percent across the cable network. The collection line has a distinct shape among those: it is load-shaped and quadratic in current, so it disappears at rest and peaks at full export, where transformer no-load loss and standby auxiliaries do the opposite.
It also charges in both directions. Discharging, the POI reads less than the sum of the converters; charging, the grid must supply the converters plus the same losses, so the metered draw exceeds what the batteries take in. A round trip pays the collection line twice, which is why it appears on both halves of the efficiency ledger that round-trip efficiency assembles and why the boundary discipline of the measurement-boundary entry has to be applied to any collection loss figure before it can be compared with another.
Feeders are sized in amperes, and amperes carry apparent power. A reactive obligation at the point of interconnection travels the collection system as current like any other: holding the same megawatts at 0.95 power factor takes about 5% more current and produces about 11% more conductor loss, with no additional real power delivered for it.
That makes a POI power-factor commitment a feeder-loading question as much as a converter question, and it is the reason a load-flow study has to visit the corners of the capability envelope rather than the unity-power-factor centre. The converse also holds: a feeder sized only on nameplate real export has no separate allowance left for the reactive dispatch the interconnection agreement obliges.
Two positional effects come with the same cable. Blocks at the far end of a feeder sit at a different terminal voltage from those at the bus — higher on export, lower on charge — so identical converters given an identical setpoint are not operating at identical points, and the volts themselves belong to the voltage-drop entry.
And a large MV cable network is a distributed shunt capacitance: it contributes to ground-fault current, which the grounding system entry owns, and it shifts network resonances, sometimes into the lower harmonic orders where converter output has content. That is why the harmonic study is run against the as-designed cable schedule rather than a generic model; in North American practice the resulting limit is checked at a stated bus under IEEE 519.
How it shows up in specs, studies and contracts
The documents to ask for are the single-line and the MV cable schedule, and the schedule is only useful if it states conductor size and material, route length, installation method, screen bonding arrangement and the ampacity basis — which standard, which soil thermal resistivity, which ground ambient, how many circuits per trench.
Ask for the derated ampacity next to the catalogue figure, and for the segment currents at full export in both directions, not the feeder total. Ask which feeder each block lands on and how far along it. Those three answers determine the loss budget, the fault duty at the far end and how much of the plant one cable fault takes out, and none of them is recoverable from a nameplate.
In studies the collection system appears everywhere the plant meets the grid. Load flow runs the full chain from converter terminals through unit transformers and feeders to the POI, at the corners of the capability envelope, because that is where feeder current and reactive flow are largest. The short-circuit study uses feeder impedance twice: it sets the fault level a tap sees and it sets how far the feeder relay can see, so cable data errors propagate straight into protection settings.
A thermal or ampacity study covers the trench with its real grouping, and a harmonic study covers the cable capacitance. At commissioning, the loss line becomes directly observable for once: at a steady known export, the difference between the sum of the block meters and the revenue meter is the collection and transformer loss as built, and an as-built loop resistance measurement on a run confirms the conductor schedule including its joints.
Contractually the asymmetry is simple: guarantees are written at the POI, and everything between the converters and that meter is the EPC's design choice. A plant contracted for a given MW at the POI needs converter capacity and feeder ampacity sized for that number plus the whole loss chain at the worst ambient, in both directions. Two scope items go missing often enough to name.
One is the ampacity basis — who owns the soil thermal resistivity survey, and what happens if the measured value comes back worse than the design assumption after the cable is in the ground. The other is spare capacity: whether the feeders and the collection bus were sized for the phase being built or for the expansion someone has drawn on the site plan, because adding a sixth block to a five-block feeder is a civil works project, not an electrical one.
The collection system is cable between the skids and the substation. Size each run so it passes ampacity, and the losses are small enough to leave out of the model.
In reality: Ampacity answers what the conductor survives; it says nothing about what the conductor costs. A run at its thermal limit is at its maximum continuous loss by definition, and that loss is quadratic in current, so it peaks at the contracted full-export condition the guarantees are tested at and rises another 11% when a 0.95 power-factor obligation is added without a single extra megawatt delivered. The ampacity figure itself is not the catalogue number either — soil thermal resistivity, burial depth, ambient and trench grouping all take a bite, and both-ends screen bonding takes another. Meanwhile the radial topology means one cable fault removes every block behind it, and the whole design is committed the day the trench closes: a fraction of a percent is a real stream of megawatt-hours bought at the charging price for twenty years, and a feeder sized with no reactive or expansion allowance is a civil works project to fix.
- I²R losses Glossary
- MV switchgear Glossary
- The BESS Single-Line Diagram, Explained Article
- Interactive: BESS Site Component Map Interactive visual · bess.engineer
MV collection system, in context.
The Grid-Scale BESS course covers mv collection system — and the rest of the system — from the ground up, the way it actually gets deployed.