The SmartgridOne Controller uses machine learning to predict the grid power of the near future (12h to 36h ahead of time, based on the available weather and price data). The prediction is based on historical data, for a given time, weekday and solar irradiation from weather data.
Why predict the grid power, and not the PV power and consumption power? Firstly, most control objectives care about what the final power at the grid is - because you get billed based on the energy you get from the grid - so this is also one of the most important parameters in the control algorithm. Secondly, the SmartgridOne Controller almost always has real, measured values from the grid energy meter. This is not always true for PV (not all inverters might be read out) and very often not for the base load consumption (this is usually a calculated value from the grid power, after subtracting all the values measured from the devices the SmartgridOne Controller communicates with). Predictions based on directly measured values are often more accurate.
Typically, within one to two days, the EMS can already detect patterns in your grid power. After one to two weeks, the predictions should be reasonably accurate.
Yes, seasonal variations are accounted for.
Yes. Because the predicted grid power is affected by how much energy your PV installation actually produces (which is a function of both the solar irradiation and the orientation of the panels), the effect of the orientation of the panels is inherently included in the data the prediction is based on. As such, the orientation is accounted for.
The Prediction data reset action in the advanced settings erases the historical energy data used by the forecasting model. Use it only when the stored history is no longer representative, for example after a substantial change to the installation or its measurement topology.
After a reset, the controller must collect new measurements before its forecasts become reliable again. Initial patterns normally become visible after one or two days; allow one to two weeks for a more representative history.