Understand your PV forecast
Why does the solar forecast differ from your inverter?
A missed forecast can come from changing clouds, a different roof or two numbers that measure different things. Start with the comparison before changing a battery schedule or assuming a hardware fault.
Make these six things match.
- Date and time zone. Compare the same local day, including daylight-saving changes. A UTC day can cut across two local dates.
- Time covered. At 14:00, compare production so far with forecast production up to 14:00, not with the full-day total.
- Units. kW is power at a moment or averaged over an interval. kWh is energy accumulated over time. A 4 kW average over 15 minutes is 1 kWh.
- Measurement boundary. Use PV generation rather than grid export. Check whether the forecast and meter refer to DC panel energy or AC inverter output, especially with a hybrid battery.
- System settings. Match installed panel capacity in kWp, roof direction, tilt and the arrays included. A city’s reference system is not a measurement of your roof.
- Forecast version. Save the previous evening’s curve when assessing day-ahead performance. Comparing yesterday’s promise with a forecast refreshed after the sun has already arrived gives a different result.
Use the solar forecast calculator for your roof settings. City pages also contain a PVGIS seasonal profile; that multi-year benchmark is not the dated weather forecast.
Worked example · illustrative numbers
20 kWh forecast, 16 kWh generated: what was the miss?
Assume complete, comparable AC-generation data for one local day, without a change in the system configuration.
| Measure | Calculation | Interpretation |
|---|---|---|
| Forecast generation | 20 kWh | The saved forecast for the full day. |
| Measured generation | 16 kWh | Comparable production recorded by the inverter or generation meter. |
| Signed error: forecast − measured | 20 − 16 = +4 kWh | The forecast overestimated production by 4 kWh. |
| Absolute percentage error | 4 ÷ 16 × 100 = 25% | This definition uses measured generation as its denominator. |
Suppose the home used 10 kWh of that solar directly and exported 6 kWh, with no battery flows in this simplified example. Comparing the 20 kWh forecast with 6 kWh of export would report a 14 kWh gap, even though the generation miss was only 4 kWh.
Near zero, percentages become misleading. A forecast of 0.2 kWh against 0.1 kWh measured has a 100% absolute percentage error, but the difference is just 0.1 kWh. When measured energy is zero, this percentage is undefined. Keep the kWh difference visible and do not average together incompatible definitions of “accuracy”.
Use the shape of the difference to choose the next check.
These patterns suggest where to look; none is a diagnosis on its own.
| Pattern | Possible explanation | Useful next check |
|---|---|---|
| The whole curve appears shifted in time. | Time-zone, daylight-saving, interval-label or roof-direction mismatch. | Align timestamps and check whether each point marks an interval’s start or end. |
| Repeated underproduction in the same morning or afternoon hours. | Local shading or an array setup that differs from the model. | Compare several clear days with roof direction and visible obstructions. |
| Actual output has a flat top below the model. | Inverter clipping, an export constraint or other active control limit. | Read inverter status and configured limits; inspect the generation channel, not only export. |
| A short, deep dip while the forecast is smooth. | Local cloud, a temporary stop or a missing measurement. | Check the weather and event log, and distinguish a true zero from missing data. |
| Several days change abruptly. | Changed settings, unavailable equipment, snow or another site condition. | Check app alerts and recent changes. Follow the installer’s support process for a persistent fault indication. |
| The daily total is close but useful hours are wrong. | Cloud timing or curve-shape error. | Compare hourly energy during the battery or EV charging window. |
Weather and physical limits can both matter. Tesla’s performance guidance describes the effects of shading, snow, clipping and curtailment. Check the equivalent documentation for your own inverter.
Keep a small comparison log before recalibrating.
For an initial review, keep two to four weeks of complete days and include a mix of weather. This is a practical starting point, not proof of annual accuracy. Record:
- The local target date, forecast issue time and declared roof settings.
- Forecast and measured daily PV generation, with units and measurement boundary.
- Forecast and actual energy during the hours that mattered to a battery, EV or hot-water decision.
- Missing telemetry, clipping, outages, snow, export controls and unusual household events.
Exclude incomplete measurements from an accuracy score and retain the reason. Look for consistent bias across comparable days before applying a correction. A single cloudy miss should not automatically change the roof capacity used by the model.
Translate uncertainty into a plan with room to change.
The useful question is how much a forecast miss could affect the next decision. Preserve a battery margin, recheck the overnight charge target, and leave an EV charging fallback before departure. Do not treat an uncertainty band as a guaranteed minimum output.
For measured evidence, read our PV forecast case study. It separates the public day-ahead forecast from updated forecasts at one connected home and states the comparison period. It does not establish an accuracy promise for every installation.
FAQ
Forecast accuracy questions.
Why is my solar forecast different from actual production?
First check that both values cover the same local date, time interval and measurement boundary. Then check roof settings, weather changes, shade, snow, inverter limits and missing data. A difference does not on its own prove a faulty inverter or forecast.
Should I compare a PV forecast with grid export?
Usually not. A production forecast estimates generated energy. Grid export excludes solar used in the home or stored in a battery, and can include battery exports. Compare with a matching PV-generation measurement instead.
What is a good solar forecast accuracy percentage?
There is no single useful threshold for every system and purpose. State the forecast horizon, time resolution, measurement boundary and error formula, then examine a period of complete days. A daily energy total can be close while the hourly timing is wrong for an EV or battery.
How do I calculate a solar forecast error?
For daily energy, subtract measured generation from forecast generation to get a signed error in kWh. If measured energy is greater than zero, divide the absolute difference by measured energy and multiply by 100 for absolute percentage error. Avoid this percentage for near-zero production.
Is PVGIS monthly output a forecast for tomorrow?
No. PVGIS monthly profiles are climate-based averages. Use them for seasonal context, and use a dated weather-based PV forecast to plan a particular day.
Sources
Methods and system behaviour.
- European Commission JRC: PVGIS user manual
Defines installed peak power, orientation, slope, system losses and the meaning of monthly and annual model outputs.
- Tesla: understanding solar system performance
Manufacturer explanation of power versus energy, weather, shading, inverter clipping and curtailment.
- GridPassport: measured PV forecast comparison
A separate one-home case study with its own measurement period and limitations; its results are not a guarantee for other homes.
Make it personal
Forecast your own roof, then get tomorrow by email.
Use your location and system size instead of a generic benchmark. The result includes expected output, uncertainty and the strongest solar hours.