Inverter-Level Performance

A plant reporting 78% PR tells you almost nothing actionable. The figure is an average, and averages are where problems go to hide — one badly degraded block and forty healthy ones produce the same number as forty-one mediocre ones.

Without a level below the plant, every investigation starts with a site visit and a guess about where to look.

Each inverter's output is normalised for its own DC capacity and the measured irradiance, then ranked against the rest of the fleet. Deviation bands — 0%, −3%, −5%, −10% — turn a continuous spread into a triage list.

The discriminator is persistence. A block sitting in the bottom band every day across rolling 7 and 30-day windows is a physical problem: degradation, mismatch, a string outage or a missed cleaning. A block that moves in and out of the band is a data problem — a sensor, a comms dropout, a mapping error.

• Normalise each inverter for DC capacity and measured irradiance
• Rank against the fleet and assign deviation bands
• Test persistence across rolling 7 and 30-day windows
• Separate physical faults from data faults by their pattern
• Attach a cause hypothesis to each laggard

A ranked laggard list with a cause hypothesis attached, plus an inverter-by-day heatmap that makes recurring patterns obvious at a glance.

• The laggard is named, not averaged away
• One persistent laggard is worth more than fifty scattered red cells
• Crews are dispatched to a specific block, not a plant
• Physical and data problems separated before anyone travels