Localized Online Weather Predictions with Overnight Adaption

Abstract

This paper extends the results presented in a conference paper showcasing an approach for online forecasting of ambient temperature and solar irradiation. The proposed method creates a localized prediction with an improvement over the available weather predictions ranging from 52 to 92% in ambient temperature forecast and 8-42% for solar irradiation forecast. This localized forecast can be used for improved predictions in smart homes or PV power plants for a more efficient operation. A new method for adapting the parameters of the autoregressive model with external input (ARX) for the solar irradiation over the night is proposed. This allows the model to be tuned to changing weather conditions without relying on external inputs.