Commissioning Errors in Solar: The Setup Mistakes That Make Wrong Data Look Right
Some of the most expensive errors in a solar plant are not faults at all. They are setup mistakes made on commissioning day — a current transformer ratio entered wrong, two devices mapped to the same data register, a sensor range set incorrectly, a logger clock on the wrong time zone. None of them raises an alarm, because the data they produce looks perfectly healthy. They simply make every number downstream wrong, often for years. Ellume Vector checks for these errors when a site is first connected and keeps checking every day after, so the figures you report rest on readings you can trust.
Most monitoring is built to catch things that change: an inverter that trips, a string that drops out, output that falls away on a sunny afternoon. Commissioning errors do not change. They are present from the first reading, they behave consistently, and they produce trends that look entirely normal. That is exactly why they survive. The result is a plant whose data is internally consistent and wrong. Performance ratios are reported to lenders, underperformance is investigated on site, warranty claims are built or dropped — all on top of a number that was off by a fixed multiple from day one.
A fault makes the data look wrong. A commissioning error makes wrong data look right.
Who this matters to
Anyone who signs off on plant figures: asset managers reporting performance to owners and lenders, O&M teams deciding where to send technicians, and buyers running due diligence on an operating portfolio. If your numbers have never been checked against the physics of the plant itself, this is the error you cannot rule out.
Why a constant error is the hardest one to see
Alarm logic looks for departures from normal. A commissioning error defines normal. If a meter has read half the true power since the plant was energised, its history has no step change, no anomaly and no event to flag. Year-on-year comparisons look stable. Month-to-month trends look stable. Even a degradation analysis looks reasonable, because the error is the same in every period. Teams usually find these errors by accident — a new analyst reconciles SCADA against the settlement meter, a technician on site reads a display that does not match the dashboard, or a buyer's engineer asks why a site has always underperformed its neighbours. By then, the wrong figure has often been in board packs and lender reports for years.
Four commissioning errors that hide in plain sight
- •CT/PT ratio misconfiguration. Meters do not measure plant current and voltage directly. They measure the scaled-down output of current transformers (CTs) and potential transformers (PTs), and multiply it back up by a configured ratio. If an 800:5 CT is entered as 400:5, every power reading is exactly half the truth. If it is entered the other way round, every reading is double. The error is a fixed multiple, so it never looks like a fault.
- •Two channels wired to the same register. Inverters and meters report over communication protocols such as Modbus, where each value lives at a numbered register. A copy-paste slip in the register map can point two channels at the same register, so one device's data appears twice and the other device's real data is never read. The worst case is a failed inverter hidden behind its neighbour's healthy readings.
- •A misconfigured sensor range. Irradiance and temperature transmitters are scaled to a configured range. Set the range wrong and the readings are scaled wrong; set it badly wrong and the signal sits at the noise floor, moving but meaningless. Because irradiance drives the expected-output calculation, a mis-scaled sensor makes the whole plant look like it is over- or underperforming.
- •Clocks on the wrong time. A logger set to the wrong time zone, or one that ignores daylight saving, shifts production against the sun. Production then appears to peak before solar noon, irradiance and output no longer line up, and every time-based figure — daily totals, availability, losses by hour — is quietly corrupted.
What each error does to your numbers
| Commissioning error | What the data looks like | What it costs you |
|---|---|---|
| CT/PT ratio set wrong | Smooth, stable output at a fixed multiple of the truth | Wrong performance ratio, wrong revenue reporting, site visits to chase losses that are not real — or real losses hidden |
| Two channels on one register | Two devices with identical readings | A failed device stays invisible; fleet comparisons are distorted |
| Sensor range set wrong | Irradiance or temperature too high, too low, or flat at the noise floor | Expected output is wrong, so every performance judgement on the site is wrong |
| Clock or time zone wrong | Production peaking before solar noon; output out of step with irradiance | Daily totals, availability and hourly loss figures quietly corrupted |
One site, one wrong ratio: an illustrative case
Numbers illustrative. A 10 MW site with three feeders has run about 16% below its expected performance ratio since it was commissioned. Over two years the O&M team has cleaned the array twice, tested strings and opened an inverter warranty discussion. Nothing explains the gap. When the site is connected to Ellume Vector, the data checks find that one feeder's meter reads exactly half of what the inverters behind it are producing — on cloudy days and clear days, morning and afternoon. Real faults vary with conditions; a fixed ratio does not. The CT on that feeder was entered at half its true ratio. One configuration change closes the gap, the cleaning budget goes back to being spent on soiling, and the asset manager has a corrected history to explain to the lender. The same error in the other direction is worse. A meter reading high can hide a real loss for years, because the site appears to be performing on budget.
How Ellume Vector catches it
- •A design audit at connection. When a site is first connected, Vector runs a design audit that checks configuration and layout — array-to-inverter sizing and string layout among them — for problems that no amount of monitoring can fix afterwards.
- •Named checks, running continuously. Vector looks for the signatures of setup errors directly: CT/PT ratio misconfiguration, two channels wired to the same register, sensor ranges that leave a signal at the noise floor, timestamp errors, and corrupted or partial meter reads. These run every day, not only at onboarding, so an error introduced by a later equipment swap or firmware change is caught too.
- •Physics as the referee. Readings outside what solar physics allows are rejected rather than averaged into your figures, and when the data is too poor to judge, Vector says so instead of producing a confident number from bad inputs.
- •Every figure traceable. Every number on every screen can be opened up to the raw readings behind it, and every figure is labelled as measured, derived or modelled. When a ratio error is corrected, you can see exactly which figures moved and why.
A check you can run this week
Take one month of data for each site and compare three numbers: the sum of inverter output, the plant meter, and the settlement meter. They should agree to within normal cable and transformer losses. If one sits at a steady, round-number ratio to the others — half, double, two-thirds — you are probably looking at a ratio error, not a performance problem. Then plot a clear day's output against irradiance: if production peaks an hour away from solar noon, check the logger clock before you check the plant.
Frequently asked questions
- What is a CT/PT ratio error?
- Meters measure current and voltage through instrument transformers that scale them down, then multiply the reading back up by a configured ratio. If that ratio is entered wrong, every reading is off by a fixed multiple — usually without any alarm.
- If the settlement meter is correct, does a SCADA ratio error matter?
- Yes. Revenue may be billed correctly, but performance ratio, availability, loss analysis and the decisions built on them come from SCADA. A wrong ratio sends technicians after losses that do not exist, or hides ones that do.
- How can an error last for years without anyone noticing?
- Because it is constant. Alarms and trend analysis look for change, and a commissioning error is present from the first reading. It only shows up when the data is checked against an independent reference, such as the inverters behind the meter or the physics of the plant.
- Does Ellume Vector only check new sites?
- No. An operating site is brought online with its full history, the design audit runs when it is first connected, and the data checks then run continuously, so errors made years ago are found as well as new ones.
