Solar Radiation Forecasting Under Variable Atmospheric Conditions Using Signal Processing and Knowledge-Driven Clustering Approach

Spatial–temporal forecastability challenges solar radiation forecasting, making it difficult to predict the energy generation from solar photovoltaic-based smart buildings. Accurate forecasting is essential to align the energy generation, storage, and consumption in terms of synchrony and helps achieve efficiency, reliability, and robustness in the grid. The novelty of the research is the formation of a hybrid forecasting model that combines signal processing, neural forecasting, and knowledge-induced clustering techniques to resolve such forecasting complexities under varied atmospheric conditions. The model combines Discrete Orthogonal S-Transform (DOST) for signal decomposition, Nonlinear Auto-Regressive Exogenous Neural Network (NARXNN) for forecasting the decomposed signals, and Knowledge-induced Multiple Kernel Fuzzy Clustering combined with the Green Chameleon Algorithm (KMKFC-GCA) for clustering atmospheric conditions. MATLAB-based simulation and comparison reveal that the hybrid model offers improved performance with a root mean square error of 0.15, a mean absolute error of 0.12, and an R2 value of 0.99 compared to standard models like the conventional model. The model provides prediction errors within ± 5%, offering a stable and precise solution for optimizing energy-efficient building systems and supporting sustainable energy management.