Accurate day-ahead forecasting of photovoltaic (PV) power requires both strong temporal modeling and adherence to physical PV system constraints, according to a new arXiv paper. The abstract emphasizes that these two elements are essential for reliable predictions.

The paper notes that purely data-driven models often ignore the underlying PV conversion process. This omission can lead to forecasts that are not physically consistent, potentially reducing their practical value for grid operators and energy traders.

To address this, the authors propose physics-informed state space models that integrate physical knowledge into the temporal modeling framework. The abstract suggests this approach aims to improve forecast accuracy by respecting the physical constraints of PV systems, though further details are not provided in the available excerpt.