Solar Irradiance Forecasting using Bayesian Optimization based Machine Learning Algorithm to Determine the Optimal Size of a Residential PV System

This work presents a new method for determining the optimum size of a Photovoltaic (PV) system for meeting a residential load in an unelectrified area. To evaluate the optimum power size of the PV system, the proposed work takes into consideration of the forecasted global horizontal irradiance (GHI) values determined using a Bayesian optimization based regression tree (BORT) machine learning algorithm. At each iteration, the prediction error of the regression tree is reduced by tuning the hyperparameter (leaf size) using the Bayesian optimization algorithm which results in higher prediction accuracy. Based on the forecasted GHI, the optimal power generation of the PV system is estimated. Furthermore, the effectiveness of Bayesian optimization has been compared with grid search and random search methods. The proposed forecasting process is implemented and verified using Matlab -Regression learner toolbox. The statistical results revealed that the BORT algorithm estimates GHI with a mean square error of 2% and 5% lesser than that obtained in grid search and the random search respectively.