Hyperparameter optimization of a deep radial basis neural learning approach for wind speed forecasting

With the application of deep learning for different predictions and classifications, it has become essential to employ the most suitable optimized hyperparameters to attain better results. The occurrence of hyperparameters in deep learning models is utilized in the learning rules and in the weight update mechanism. Due to this, in this research study, methods are proposed to evolve optimal hyperparameters for the considered novel deep radial basis neural learning (DRBNL) model, and these attained optimally tuned hyperparameters are used to carry out the wind speed and subsequently, wind power prediction in the renewable energy sector. For obtaining the optimal hyperparameters for the deep learning model, this study develops a hybrid version of the Harris Hawks optimization and differential evolution algorithm resulting in a novel Harris Hawks differential evolution optimization (HHDEO) algorithm and thereby training and testing the deep learning model with optimized hyperparametric values. The developed novel HHDEO-based DRBNL model is employed for its effectiveness over benchmark test functions and on wind farm datasets from varied locations. Results computed during the simulation process prove the efficacy of the developed optimized DRBNL model over the other models from early works of literature. Furthermore, the developed HHDEO–DRBNL model performed time scale predictions—very short-term, short-term, medium-term, and long-term forecasting for the wind farm datasets. The proposed algorithm outperforms the considered benchmark functions and developed a hybrid model to better the prediction in multiple horizons.