Solar Irradiance Prediction with Ensemble Learning Method as Input for Battery Operation Optimization

The current use of fossil fuels as a major source of energy on a global scale has had a negative impact on the environment in terms of pollution and global warming. There is currently a paradigm shift to the utilization of renewable energy sources such as solar and wind power. However, the challenge is that these options are inherently intermittent and expensive to install. Data from these sources can be used as input to algorithms to optimize the operation of microgrids to provide energy to both residential and commercial buildings. A key component to enhance the performance of a microgrid is a battery energy storage system (BESS). Our goal is to design an algorithm to optimize battery operation. The objective of this paper is to predict the solar irradiance to provide input to battery operation optimization. We investigate the comparative performances of different strategies. The accuracy of five popular machine learning algorithms are estimated, namely Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), Kernel ridge regression (KRR), and Linear Regression. Voting, stacking, and bagging ensemble methods are utilized to further enhance the accuracy. Our results show stacking to outperform the other methods. Our experiment produced better predictions than other work using the same dataset.