Tree-anchored ensembles reduce photovoltaic forecast error by 3.21%

A tree-anchored expert combination method reduces mean absolute error (MAE) for multi-horizon photovoltaic forecasting. By weighting LightGBM with neural temporal predictors, the approach achieves a 3.21% reduction in scaled MAE on PVDAQ datasets.