An underperforming afternoon on a solar plant is usually nothing — a cloud, a hazy sky, the weather being the weather. Which is exactly why real faults hide there. A drooping inverter and an overcast hour can look identical on a generation chart, and by the time a dip is obviously not weather, it has been quietly costing money for days.
The operator here didn’t need more alarms. They needed the one alarm that meant something, told apart from the dozens that didn’t.
Telling weather from failure
Vastvic’s detector didn’t just notice that output was low. Using robust statistics that a normal midday peak or a passing cloud wouldn’t trip, it flagged a persistent shortfall — and then the diagnostic layer did the part that mattered: it weighed the shortfall against irradiance, performance ratio and capacity and concluded this wasn’t a cloud. The generation was low while the sky said it shouldn’t be.
The chart showed a dip. The detector explained a fault. Those are very different tickets.
Caught early, fixed cheap
Because the anomaly came with a cause rather than just a red mark, it went to the field team as a probable inverter issue, not a “please investigate.” The fix happened while it was still a small fix — before a marginal inverter became a dead string and days of lost generation became weeks.
The value wasn’t in detecting that something was off; a spreadsheet can do that eventually. It was in separating the signal from the weather early, and attaching a reason — so a human spent their time repairing a fault, not debating whether there was one.
Key takeaways
- Robust detection flagged a persistent shortfall without tripping on normal peaks or clouds.
- Rule-based diagnostics used irradiance, performance ratio and capacity to attribute the cause.
- The alarm arrived as a probable inverter fault, not a vague “investigate” ticket.
- Early, explained detection turned a small repair into a small repair — before it compounded.