The conversion side decides what the roof can do.

An inverter converts the direct current the modules produce into alternating current at the building’s voltage and frequency, and it does that every second of daylight for twenty-five years. On a commercial array it does three further jobs the headline rating does not describe. It holds each group of strings at its maximum power point as irradiance and cell temperature move through the day. It is the only part of the system that negotiates with the network, so it carries the type-test certification, the protection settings and the export setpoint your Distribution Network Operator agrees. And it is the sole source of data about how the array is performing, at whatever resolution its architecture allows.

Because of that, the conversion side is where the shape of the roof, the electrical infrastructure of the building and the terms of the connection have to be reconciled. We design in that order. The survey establishes what the roof can carry and how it is broken up. The model establishes what the array would generate against your own consumption. The network operator establishes what may be exported, which on a constrained network is frequently well below what the roof could push. Only then is the AC architecture set and the inverter chosen to serve it. Reversing that order, picking a platform first and fitting the array to it, is how a good roof ends up with strings grouped badly and a plant room nobody can work in.

There is no price on this page and there will not be one. Above 50 kWp the figure comes from an on-site survey and a half-hourly PV*SOL model run against your own load, so a published number would be a guess wearing an engineer’s coat. What follows is the engineering instead: how the conversion is split across the site, how strings are grouped, what the equipment does to a plant room and a switchroom, what the grid connection requires, and what happens when the electronics reach the end of their life while the modules are still under warranty.

Module-level power electronics sit on a different axis from that choice rather than forming a fourth option. A DC-optimised design puts an electronic device at each module, which recovers yield where shading from rooftop plant, parapets, flues or a neighbouring building would otherwise drag a whole string down, reports per module, and de-energises the DC at each module for maintenance crews and first responders, the behaviour RC62 fire-safety guidance looks for. It also adds a component per module on the roof, so we specify it where the survey shows shading or complexity that earns it, and not as a habit. The fuller comparison of distributed against central conversion, including redundancy, serviceability and phasing, is in the string versus central inverters guide. Where storage is in the scheme, whether the battery is AC-coupled alongside the PV inverters or handled by a hybrid unit is settled on the battery storage side and then reflected back into this layout.

A maximum power point tracker is an independent input that finds and holds the voltage at which the strings connected to it produce the most power. The number of trackers on a unit therefore decides how finely the roof can be divided. Every string on one tracker is pulled to a shared operating point, so strings that face different ways, sit at different tilts or shade at different times of day should not be sharing one. An eight-tracker machine on a plain south-facing roof buys nothing it can use; a two-tracker machine on a roof with five pitches quietly gives up output every morning and every afternoon for the life of the asset.

String length is fixed by voltage rather than by how the layout looks on a drawing. A module’s open-circuit voltage rises as cell temperature falls, so the longest permitted string is set by a cold, bright winter morning: the series count has to keep the array below the inverter’s maximum DC input, commonly 1,100 V on the three-phase commercial units and 1,500 V on the platforms used for large ground-mount. The shortest string is fixed at the other end of the year, by the warmest afternoon, when the operating voltage still has to sit inside the tracker’s working window or the unit falls off its maximum power point at the moment there is most light to collect. Strings sharing a tracker are kept to the same length and the same orientation for exactly the same reason.

The ratio between array size and inverter rating is then a deliberate decision rather than an accident of what was in stock. A commercial array is normally oversized against its inverters, because a UK roof spends very little of the year anywhere near its rated output, and the additional modules earn through the whole of a dull morning while the loss at the peak of a clear June day is small and brief. Where the network operator has capped export, the arithmetic changes again, since the ceiling is then the connection rather than the equipment, and the surplus is worth more used inside the building than pushed at a limit that will not take it. We model that clipping in PV*SOL against your half-hourly consumption, so the ratio is chosen with the loss quantified rather than by rule of thumb.

Inverters reject heat, and they derate when they cannot, so position is chosen for airflow rather than for tidiness: outdoors on a shaded wall, or inside a room with genuine ventilation, never in a sealed cupboard that becomes an oven in July. Enclosures rated to IP65 or IP66 will sit outside without a separate cabinet, though direct sun on the case still costs output over a long summer, and fan-cooled units make a noise that matters next to an office window or a residential boundary. On a ground-mount scheme the same questions become a housing, a hardstanding and a track that a vehicle can still use in February.

The AC side is usually where the surprises are. The distance from the array to the intake drives cable size, because volt drop has to stay inside the limits BS 7671 sets, and on a long industrial building that pushes the cable larger than the current alone would suggest. Several inverters have to be combined, which means an AC combiner or a dedicated distribution board, and then a way in to the existing switchgear. Spare ways in the main panel, the prospective fault level at that board, the rating of the incoming supply and the physical room to land a new cable are all checked at survey, because a switchroom with nothing spare in it is a programme item rather than a detail. On a larger scheme it can be the item that decides where the inverters go.

Access is designed in from the start. A commercial inverter is a two-person lift or a mechanical one, and it has to be carried in when it is installed and carried back out when it is replaced, past whatever the building has become by then. We keep clear working space in front of every unit, label the DC and AC isolation so it reads correctly to somebody who has never seen the system, and record the stringing on the as-built drawings. Design, inspection and verification sit under BS 7671 and IEC 62446 rather than MCS, because a system this size is outside the domestic scheme, and the works run under CDM 2015 with a principal designer and a real construction phase plan. Our quality without MCS guide sets out what that assurance stack covers and what it replaces.

Every inverter on a commercial connection needs G99 compliance evidence for the exact model and firmware. A full type test on the ENA register carries that case, and the current edition of G99 grants that status only to units of 50 kW or less; a larger unit usually carries a partial one, which covers part of the case, with the rest supplied with the application and shown at commissioning. We check which applies against the register rather than take it from a brochure. The G99 application runs in parallel with the design, because the offer that comes back sets the ceiling everything else is built to. The network operator may grant full export, may cap it, or may offer a connection on the condition that nothing is exported at all, and on a constrained part of the network that answer arrives late enough to be the longest single item in the programme. We start it early for that reason and size the array to the position the operator confirms.

Where export is limited, the limit has to be enforced rather than promised. That means a metering point at the intake, current transformers on the incoming supply and a controller that instructs the inverters to reduce output inside the response time the operator specifies, which is the arrangement our export limitation and G100 guide describes. The inverter has to be able to accept that setpoint and to fail to a safe state if the controller stops talking to it, so the capability is confirmed at specification and then demonstrated at commissioning in front of the witness the scheme requires. A design that relies on everyone remembering not to export is not a design.

Monitoring resolution follows the architecture. A string inverter reports at unit and string level, which is enough to show that one group has dropped away; a DC-optimised design reports per module, so a single failed or shaded panel is visible instead of being averaged into a total that still looks broadly right. Either way the portal account is created in your name with the data exportable, so the performance record does not sit inside a contractor’s login and leave with them. That feed is what our operations and maintenance work runs on, and it is the difference between catching a fault in a week and catching it in an annual bill. The symptoms worth watching for are set out in our performance audit.

The corporate story, the ownership, the manufacturing footprint, the warranty terms and the honest caveats for every maker we specify are written once, on one page per manufacturer, covering everything that maker supplies to us instead of repeating itself across product categories. The full roster of modules, inverters and storage sits at the manufacturer directory, alongside the module and storage sides of the same design.

Two things stay fixed wherever a scheme lands. An inverter is never described here as BloombergNEF Tier 1, because that is a module financeability scheme and attaching it to conversion equipment would be borrowing a credential that does not apply; there is a separate inverter bankability survey, and it is cited only on the page of a maker that genuinely leads it. And where a manufacturer carries a real financial or track-record caveat, it is stated on that maker’s page rather than quietly left out. Our reasoning on the brands we decline to specify is set out in the why we do not specify Huawei guide.