HBA-LightGBM: Honey Badger Algorithm With LightGBM Model For Solar Irradiance Forecasting

Abstract

The precise forecasting of solar energy holds significant importance for photovoltaic power plants, facilitating early engagement in energy auctions and cost-efficient resource planning. Due to the intermittent nature of solar irradiance, statistical models often fall short, giving an edge to machine learning models for more accurate forecasting. This article introduces an ensemble learning approach, the Light Gradient Boosting Machine model (LightGBM), that not only takes less computational time but also improves the forecasting performance. A two-stage dimensionality reduction approach is adopted that involves Mutual Information as a filter-based feature selection method and Autoencoder as a deep neural-based feature extraction approach to set the optimal features in order to avoid the problem of overfitting. Furthermore, the outliers present in the optimal features obtained are treated by their forecasted value by the Random Forest method. The performance of LightGBM models directly depends on their architectures and hyperparameters. To handle this problem, a swarm evolutionary optimization algorithm called the Honey Badger Algorithm, is applied to optimize the network architecture. The performance of the proposed method is investigated using two datasets collected from the National Renewable Energy Laboratory. The experimental results validate the superiority of the proposed method in comparison to other benchmark models.