A practical framework for expected vs actual PV performance

An expected-versus-actual comparison is useful only when both sides of the comparison describe the same operating interval and system boundary. The expected series is a counterfactual estimate: what the plant could reasonably have produced under the observed conditions and known configuration. The actual series is measured output, subject to meter accuracy, communications health, and curtailment or outage states.

1. Define the comparison boundary

Start by identifying the electrical boundary represented by each signal. A revenue meter, inverter AC channel, and plant controller can all report power, but they do not necessarily represent the same losses or equipment. Record whether the comparison is site-level or component-level, whether values are instantaneous power or interval energy, and which timezone defines the operating day.

• Use timezone-aware timestamps and one explicit interval convention.
• Normalize physical units before combining sources.
• Do not add inverter channels to a site meter unless the measurement boundaries are understood.
• Keep curtailment, export limits, and planned maintenance visible as operating context.

2. Establish data quality before modeling

Duplicate timestamps, missing intervals, frozen values, negative measurements, and implausible capacity excursions should be flagged before performance is calculated. A missing interval is uncertainty, not automatically zero production. Reporting data coverage alongside performance prevents a clean-looking percentage from hiding an incomplete day.

3. Construct a defensible expected series

Expected production may come from a validated physical model, a calibrated reference model, or another documented baseline. The model should use the plant configuration and weather inputs available for the interval. Calibration should use a known-good period and should not absorb persistent faults into the baseline. Model version, weather source, and calibration period belong with every analysis result.

4. Calculate comparable performance measures

Integrate valid power intervals to obtain actual and expected energy. Energy deficit is the non-negative difference between expected and actual energy for intervals that are valid and comparable. Performance percentage is actual energy divided by expected energy when expected energy is positive. Divide-by-zero conditions should return an unavailable result rather than an invented zero or perfect score.

1. Review data coverage and availability first.
2. Compare the timing and shape of the two curves, not only daily totals.
3. Look for persistence, recurrence, and component-relative evidence.
4. Attach a probable cause and confidence only when the evidence supports it.
5. Translate validated deficits into estimated business impact with explicit assumptions.

5. Move from variance to action

A variance becomes operationally useful when it is connected to an inspection decision. Rank persistent, high-impact deviations above short low-energy events, while keeping low-coverage sites visible as data-quality work. The outcome should answer what changed, how much energy may be affected, how confident the system is, and what a technician or analyst should inspect next.