Under/Over Performance

The Under/Over Performance page (available at /kpi/underOverPerformance) compares, device by device, the energy each machine actually produced with the energy it should have produced given the resource it measured: hub wind speed for wind, irradiation for photovoltaic. The comparison is limited to the periods when the machine was free to produce.

The difference between the two is the deviation:

deviation = (actual − expected) / expected × 100

Negative means under performance (the machine yields less than the available resource would allow), positive means over performance.

Toolbar

On the left:

• Date range — The period the analysis runs on.
• Devices — Multi-select tree: pick turbines and inverters, or whole plants and sections (the leaves are collected automatically). A badge next to the tree shows the locked technology.
• Threshold ±% — The band within which a device counts as "on line": above it is over, below it is under. Adjust it with the − and + buttons.

On the right:

• Trend — Opens the Trend page with the current selection (or just the focused device) already set, to inspect the raw signals. Enabled only once an analysis has been generated.
• Export CSV — Downloads the per-device table of the current analysis.
• Generate — Starts the computation. Results arrive in groups of devices, with a progress banner. If the selection, period or threshold change afterwards, an amber dot appears on the button: what is on screen is no longer up to date.

KPI strip

After the computation, the summary of the selection sits on top of the results.

The yield, availability and coverage cards carry an icon that opens the formula in use and the filters applied.

Device ranking

The left column lists the devices from worst to best. For each one: percentage deviation, energy lost or gained, data coverage and a sparkline with the monthly trend. The Whole selection row on top carries the aggregate values.

Clicking a device filters every analysis on the right on it, and a chip at the top recalls the active filter; clicking it again (or closing the chip) goes back to the whole selection. The column can be collapsed to give the charts more room.

Analysis views

The seven views work on the same result and follow the focused device.

Fleet

Distribution of the devices per deviation band, with the fleet median highlighted. The bands adapt to the data.

Power

Measured power curve against the reference one (for photovoltaic: measured specific power against reference). The samples behind the curves are 10 minutes for wind and 15 minutes for photovoltaic.

Density cloud. The real distribution is drawn behind the curves: one cell per resource / power level pair, the darker the more often the machine ran in that condition. It tells a drop made of a few scattered samples from one the machine spent most of the period in. With a device focused the cloud is its own; on the whole selection it is the one of all the devices together, and it shows how much the fleet spreads around the mean curve: two machines at full power and three at a third, at the same resource, average out to a curve that would not say so. For a device whose nominal power is not configured the distribution is not available and the curves are all that is left. The cloud uses the same samples as the rest of the page and does not capture the mouse: the tooltip stays the one of the resource band.

Wind speed / Irradiation

Two charts per resource band:

• Expected vs actual energy per band — The thin bar under the axis is the sample count of the band.
• Deviation % per band — Tells at which resource level the energy is lost. Bands under 60 samples stay as amber outlines and do not enter the totals.

Sectors / Hour profile

For wind, the rose of direction sectors (radius = resource hours, colour = energy-weighted deviation); for photovoltaic, the same cut by hour. Next to it, the device × sector/hour matrix.

For photovoltaic the cut is by hours from the plant solar noon (−3 h, 0 h, +2 h) instead of clock hours: dropping time zone, daylight saving and longitude keeps different plants comparable within the same analysis and stops the edge bands from absorbing all the light outside the window. Solar noon is the midpoint between sunrise and sunset of the day; for a plant without coordinates local noon is used instead (12:00 on the plant clock), an approximation that shifts the bands by a few tens of minutes. Empty bands are informative: they tell how long the day was.

Benchmark

Z-score of each device against the median of its peer group, that is the devices of the same model (turbine or inverter), with the ±1σ band highlighted: it separates the outliers from a drop that affects the whole fleet.

Trend

Monthly trend of the deviation. With a focused device it is compared against the fleet mean; with no focus, a grid of sparklines with one cell per device.

Detail

Table with every value per device (resource, availability, coverage, capacity factor, expected and actual energy, deviation, status). Clicking a row focuses that device.

In every chart, hovering shows a tooltip with the values of the point under the cursor.

Data coverage and applied filters

In the Fleet view a bar shows how the sample set everything else is computed on is composed (it applies to every view): used samples, machine downtime, grid limitation, above full scale (or above cut-out), below the resource threshold (or below cut-in), invalid or missing.

Excluded samples enter no calculation on the page: outside the working range the reference is zero, so any produced energy would read as over performance. The mask is the same for all the charts and is always declared.

CSV export

The downloaded file holds one row per device with the same values as the Detail view, plus a summary section with the fleet KPIs. Numbers are formatted in the user locale.