Abstract:
"The document outlines the development of ""SolarForecast,"" a novel system for solar power
forecasting using feature-based model redistribution. It aims to address the challenge of creating a
flexible and generalizable solar forecasting system.
This research proposes a system that utilizes a meta-learning model to combine the forecasts of an
ensembling through weighted averaging and redistribution. This works by using the meta-learner
on the extracted time series features of a dataset where the meta-learner will assign weights to the
base models, which are used for forecasting, in the ensemble and redistribute them if below a
threshold.
This approach outperforms previous approaches utilizing model averaging on the tested dataset in
the metrics such as sMAPE, MAE, RMSE and OWA.
"
