Augmenting deep learning with hybrid AHA-CO optimization technique for solar radiation prediction

Precise and reliable Solar Radiation (SR) prediction is an integral part of the thermal systems for renewable energy production. A lightweight convolutional neural network (CNN) is designed to generate complex features and a hybrid optimization algorithm combining the artificial hummingbird algorithm (AHA) and Cheetah optimizer (CO), named as AHA-CO, is introduced for Feature selection (FS). The CNN model comprising three convolution layers interlaced by max pooling layers is used to transform inputs to complex features and the AHA-CO is employed to retain only the most informative features without compromising prediction performance and avoid getting stuck in local minima. The developed method is evaluated against 12 individual and three hybrid metaheuristic methods, demonstrating the superior performance of the AHA-CO. Subsequently, the CNN features selected by the AHA-CO are used as input to five regression models. The effectiveness of the SR prediction is assessed on a publicly available dataset. The support vector regression (SVR) model performed the best, forming the CNN-AHA-CO-SVR framework. The results show that the developed framework attained superior performance with 0.0365 mean absolute error, 0.0074 mean squared error, and 0.9251 coefficient of determination, highlighting its robustness and suitability for accurate SR prediction.