Tree-Enhanced Deep Neural Network Framework for Solar Power Forecasting

This research article presents a study on forecasting solar power generation using a Tree-Enhanced Deep Neural Network (TEDNN) framework. The document outlines a three-phase approach that begins with optimizing hyperparameters of regression models through algorithms such as Grid Search, Bayesian Optimization, and Particle Swarm Optimization. The second phase involves applying feature engineering techniques, specifically Squared Irradiation and Temperature Differential, to enhance the extracted data. In the final phase, various tree-based regression algorithms, including Decision Tree Regression, Random Forest Regression, Gradient Boosting Regression, and Xgboost Regression, are combined with Deep Neural Networks to improve forecasting accuracy. The proposed model is evaluated using real-world data from a photovoltaic solar power plant in India, demonstrating that the combination of Gradient Boosting Regression and Deep Neural Networks with feature engineering achieves the highest performance metrics. The findings indicate that the optimization techniques significantly enhance the predictive capabilities of the models employed in solar power forecasting.