Comparative Study of Customizable Neural Network Models for Solar Irradiance Forecasting

The solar irradiance is one of the key parameters used for the generation of electrical power from solar energy. This study discusses about one of the scopes of improving the solar irradiance forecasting models. The solar irradiance forecasting models presented in the study has been analyzed for two types of feature engineering techniques, i.e., PCA (Principal Component Analysis) and kPCA (kernel Principal Component Analysis). Simultaneously, the feature engineered dataset has been trained with different activation functions and different combinations of hidden layer neurons-based Cas-NN (Cascaded Neural Network) forecasting models. The training of different test models suggests that the original dataset engineered with kPCA and combined with Tan sigmoid activation function including higher hidden layer neurons is performing better than other considered test models.