A hybrid clustering and online gaussian process regression model for solar power forecasting in Hanoi, Vietnam

The integration of photovoltaic (PV) power into the electrical grid has increased significantly in recent years, both on the generation and consumption sides. However, the stochastic and intermittent nature of solar energy has posed considerable operational challenges for modern grids. Therefore, accurate short-term PV output forecasting is crucial to ensuring grid stability and efficient energy management. This study proposes a hybrid forecasting framework integrating Gaussian Mixed Model (GMM) with Online Gaussian Process Regression (OGPR), using a synthetic kernel that combines periodic and rational quadratic components. GMM is used to probabilistically cluster historical PV output data, enabling the OGPR model to perform adaptive, computationally efficient real-time predictions. Experimental results under highly variable summer weather conditions in Hanoi show that the GMM-OGPR framework significantly outperformed the standalone OGPR model in terms of prediction accuracy (reducing nRMSE by over 30%) and computational cost (reducing training time by nearly 60%). Furthermore, the model exhibited high stability in both clear and cloudy conditions, demonstrating its applicability in real-world scenarios where meteorological data might be unavailable.