Global horizontal irradiance forecasting using online sparse Gaussian process regression based on quasiperiodic kernels
Shab Gbémou, Hanany Tolba, Stéphane Thil, Stéphane Grieu · 2019
The application of Gaussian process models is intractable for large datasets, because of time complexity and storage. To overcome this limitation, online sparse Gaussian process regression (OSGPR) based on quasiperiodic kernels is used to model and forecast global horizontal irradiance (GHI), at three forecast horizons (30 min, 4 h and 24 h). Using two years of GHI data, OSGPR models are trained. Forecasting accuracy is evaluated for various levels of sparsity in training data.