A fault can cut daily output by 8 to 12%, while the monthly PR change may be only 0.3 to 0.4%. That contrast explains both the value and the limits of Performance Ratio. The metric appears on O&M dashboards, investor reports, and quarterly reviews, but the losses it misses are where plant performance often slips.
What PR Actually Measures
Performance Ratio compares actual energy yield with the energy a plant would theoretically produce at its nameplate efficiency under the incident irradiance. IEC 61724-1 defines the standard formula as:
PR = E_actual / (G_POA × A × η_STC)
With known peak power, the equivalent form is:
PR = (E_actual / P_nom) / (H_POA / G_ref)
H_POA is plane-of-array irradiation in kWh/m2, while G_ref is the standard 1 kW/m2. A PR of 0.80 means the plant delivered 80% of its expected output under STC with the measured irradiance. In Brazil, well-maintained plants typically fall between 0.78 to 0.84, depending on technology, climate zone, and age.
PR combines every major loss source: wiring, inverters, soiling, shading, module temperature, mismatch, and availability. That broad view gives it value, but also limits what the number can explain.
The Averaging Problem
Monthly PR is an average across time. An inverter trip lasting 3 hours on a high-irradiance afternoon can cut daily output by 8 to 12%, depending on timing. Across 30 days, the effect may equal only 0.3 to 0.4% of monthly generation. Monthly PR shifts from 0.812 to 0.809, still within ordinary measurement uncertainty and seasonal variation. No alarm fires.
This happens often. Brief inverter trips, partial resets, communication gaps that put inverters into fallback modes, and short grid voltage events can all disappear inside monthly PR. O&M teams that use only this metric will repeatedly undercount fault frequency.
Daily PR narrows the averaging period, but a fault that starts and clears within one day can still vanish into the average. Reliable intraday detection calls for sub-daily analysis, ideally at 15-minute intervals.
Temperature: The Hidden Seasonal Bias
For standard silicon, module output usually falls by 0.35 to 0.45% per degree Celsius above 25 degrees C. In Minas Gerais in January, cell temperatures can routinely reach 60 to 65 degrees C, or 35 to 40 degrees C above STC. That temperature difference alone cuts peak output by 12 to 18% versus nameplate.
Without correction, PR includes this temperature penalty. A plant can report 0.76 in January and 0.84 in July while performing the same after temperature is considered. Seasonal movement can therefore look like degradation.
Temperature-corrected PRc adjusts for this:
PRc = PR × (1 + γ × (T_cell - 25))
Gamma is the module temperature coefficient, around -0.004 per degree C for most silicon modules. Applying PRc across months supports year-over-year trending. Without that adjustment, each seasonal comparison mixes different operating conditions.
What PR Does Not Decompose
Temperature correction still leaves PR as one undivided figure. A 2 percentage points drop in PRc from one month to the next could reflect more soiling, a persistent string fault, changed inverter clipping after a firmware update, or module degradation. Each cause produces the same headline number.
The distinction affects the response. Soiling calls for a cleaning crew, a string fault calls for a technician with electrical equipment, and clipping calls for an inverter configuration check. A PR decline alone cannot identify the right action.
Breaking PR into loss categories requires more data: string-level current for mismatch and fault detection, irradiance from a calibrated reference sensor or modeled source, and module or ambient temperature paired with a thermal model. These inputs are standard in modern monitoring systems, but PR itself does not contain them.
Using PR Alongside Fault Scoring
A practical setup uses monthly PR as the health indicator and short-window anomaly scores for fault detection. Each layer operates on a different time scale and reveals different issues.
Monthly PR trend: a steady PRc decline over 6 to 12 months points to a likely degradation process. First-year module LID (light-induced degradation), PID (potential-induced degradation) in high-humidity environments, and encapsulant discoloration over time can all create gradual PR drift. This is what PR is designed to catch.
Short-window fault scoring: anomaly detection on 15-minute string and inverter data can expose events that monthly PR hides. Comparing each string with cluster peers under the same irradiance conditions can identify an open-circuit string within hours rather than at month end.
Soiling index: PR may show soiling when cleaning frequency changes, but a dedicated metric is clearer. Comparing irradiance-adjusted yield from a reference string with the rest of the array can track soiling accumulation directly between cleaning events.
The combination requires no extra sensing. All three rely on telemetry that a plant with a working SCADA already captures. The work is in the analysis layer, not the sensing layer.
A Realistic Expectation for PR-Based Monitoring
PR remains a sound KPI. It is useful for long-term asset health, PPA compliance verification, and portfolio benchmarking. With temperature correction and consistent irradiance normalization, it condenses a complex system into a comparable figure and shows whether the plant is meeting design expectations over time.
The problem begins when PR becomes the only alarm layer or replaces time-resolved analysis. A plant can meet its monthly PR target and show no inverter trip alarms while still losing 2 to 4% of generation to soiling and string-level faults visible only in sub-daily data. At 30 MWp, 3% represents roughly 1,400 MWh per year.
The better question is not "is our PR good?" It is "what does our PR show, and what does it miss?" Answering the second question requires an analytical layer operating at the resolution used for real O&M decisions.
For more on data patterns that complement PR monitoring, see our article on weather normalization for solar generation data and how continuous string-level monitoring reveals what daily averages hide.