Plant irradiation

Plant irradiation is not the raw reading of a single sensor: it is a value computed by the system, obtained by combining the readings of the irradiation sensors configured as technical. Its purpose is to produce a single reliable, representative value that is robust against dirty, faulty, or noisy sensors.

It is the value used, for example, by Performance Ratio as the reference irradiation, and shown in the realtime views.

Where it starts: the technical sensors

Each photovoltaic plant can have several irradiation sensors. In Wizard Plants → Sensors some of them can be marked as Technical: these are the reference sensors, considered trustworthy for the calculation.

The plant irradiation calculation uses only the irradiation sensors marked as technical. Non-technical sensors are ignored.

How the readings are combined

At each instant, the system takes the readings of the technical sensors (on the same time base as the computed value) and picks the most representative value through a series of checks, in order:

1. Validity — readings that are missing or flagged as invalid by the system are discarded (a data-quality flag, independent of the value).
2. Physical plausibility — among the valid readings, only those with a value in 0–1500 W/m² are kept: this excludes off-scale numbers produced by faulty or miscalibrated sensors.
3. Daylight — only readings between sunrise and sunset are considered, to prevent night-time noise from polluting the value. Sunrise and sunset are computed with the NOAA solar algorithm from the plant coordinates; without valid coordinates a fixed 06:00–18:00 UTC window is used and, in polar conditions (sun always above or below the horizon), nothing is excluded.
4. Outlier rejection (Chauvenet's criterion) — applied only with more than 2 valid readings: with μ the mean and σ the population standard deviation (dividing by N), a reading x is discarded if erfc( |x − μ| / (σ·√2) ) < 1 / (2N), where erfc is the complementary error function. If σ is zero, or if the criterion would discard all readings, nothing is discarded; with 2 or fewer readings the step does not apply.
5. Final value — the median of the surviving readings (with an even number of readings, the actual reading closest to the average of the two central ones, so as to preserve its timestamp and metadata): more stable than the average and not skewed by a single off-scale sensor.

The computed value is tagged with the VireoXcube (Computed) agent and carries a short diagnostic (which sensor was selected, how many readings were valid, how many survived, and which sensors were rejected), useful to understand how it was obtained.

Time series

For charts and analyses over a time range the logic is the same, applied point by point after a temporal alignment: the technical sensors' readings are grouped into 1-second windows, keeping for each sensor the latest reading of each window; the rules described above are then applied to each window. The grouping absorbs the small misalignments between sensor timestamps that would otherwise produce a "zigzag" signal.

Plant irradiation vs inverter irradiation

• Plant irradiation — the single, filtered value described on this page, computed from the technical sensors.
• Inverter irradiation — each inverter can be linked to a specific irradiation sensor; if none is linked, the inverter falls back to the irradiation of the plant's control device. It does not apply the statistical filter: its purpose is to have the reference closest to the individual inverter.