A Data-Driven Solar Irradiance Forecasting Model with Minimum Data

An emerging new challenge introduced to solar generation forecasting is the accumulation and effective processing of raw weather data. This paper aims to address this challenge by presenting a hybrid approach to forecasting the solar irradiance, incorporating both clustering and feature extraction techniques. The developed method aims to significantly reduce the amount of data required for forecasting, and at the same time increase the accuracy of the forecast. A clustering and data selection strategy is developed that yields a reduced dataset for prediction. The performance of the forecasting approach is evaluated with real solar irradiance data collected throughout the year. Case studies demonstrate that solar irradiance can be accurately forecasted using only 20% of the full-scale training data, while also improving the forecast error compared to using the entire dataset.