The Monitoring Numbers Worth Watching, and the Ones That Waste Your Time

Performance & Operations   6 min read

Modern monitoring platforms present dozens of metrics and a dashboard that updates every few seconds. Very little of it changes any decision. Here is the subset I would actually watch on a commercial plant.

Watch weekly

String-level current comparison. The single most valuable signal available. Strings of identical configuration and orientation should track each other closely. When one drifts a few percent below its neighbours, something is developing – a connector, a cracked cell, partial shading, a module fault.

Plant-level output will not reveal this. A 4% drop on one of sixteen strings is a 0.25% change at the meter, comfortably inside daily noise.

Inverter availability. Not just faults – also how long recovery takes. An inverter that trips and self-restarts within a minute is fine. One that trips and waits for a manual reset while nobody is watching costs a day of production each time.

Watch monthly

PR against the same month last year. Seasonal variation makes month-on-month comparison meaningless. Year-over-year for the same month is the real trend line.

Specific yield (kWh/kWp). Simple, comparable, easy to sanity-check against regional benchmarks or a neighbouring installation.

Peak output on clear days. Compare against the same period in prior years. A declining clear-day peak with unchanged total output suggests derating or a capacity loss being masked by favourable weather.

Watch quarterly

Inverter internal temperature during peak hours. Rising trend means blocked filters, failing fans or degraded ventilation. Catching this before it derates is straightforward; noticing after is a lost summer.

The soiling gap. Production immediately after heavy rain versus production a week later. The size of that swing is your soiling rate, free of charge.

What to mostly ignore

Real-time instantaneous power. It is compelling to look at and tells you nothing about plant health. A cloud passing produces exactly the same dip as a fault, and you cannot distinguish them in the moment.

Cumulative lifetime energy. Nice on a lobby display. Contains no diagnostic information.

CO₂ avoided. Useful for reporting, useless operationally.

Daily production compared against yesterday. Weather noise dominates completely. People act on this constantly and it generates false alarms.

Alerts worth configuring

Most platforms ship with alerting that either floods you or never fires. A workable set:

  • Any inverter offline for more than 60 minutes during daylight.
  • Any string more than 8% below the median of its comparison group, sustained across three consecutive days.
  • Plant output below 40% of expectation for the given irradiance, sustained two hours.
  • Inverter internal temperature above manufacturer threshold.
  • Communication loss for more than 24 hours – because a plant you cannot see is a plant you are not monitoring.

The three-day persistence on string alerts matters. Without it, every partly cloudy day with uneven cloud cover across a large roof generates alerts, and within a fortnight everyone stops reading them.

Data retention

Check how long your platform keeps interval data, and at what resolution. Many compress to daily totals after a year, which destroys exactly the data you need for a warranty claim or a year-three diagnosis.

Export raw interval data periodically and keep it yourself. It costs nothing and it is the evidence base for every performance conversation you will have over the plant's life.