Using Linear Regression and Back Propagation Neural Networks to Predict Performance of Soiled PV Modules

Abstract

This paper presents a study on neural network-based modeling techniques and sensor data to estimate the power output of photovoltaic systems under soiling conditions. Predicting maximum power output under soiling conditions is considered an important and difficult problem and a variety of models using a host of factors including temperature and weather profiles have been proposed. This study used linear regression models and artificial neural networks and used only solar irradiation and ambient temperature, as well and the maximum power point (MPP) characteristic variables of photovoltaic (PV) modules obtained from online current-voltage (IV) tracers in the site of a PV installation. The two models were trained and validated using actual monitoring data of two 100-Watt PV modules installed in the UAE. One reference panel was cleaned on a weekly basis and the second panel was left to accumulate dust over the entire period between July 1, 2018 and 17 September, 2018. The results show that it is possible to predict maximum power output of soiled PV modules at about 97% accuracy. The proposed models perform at an accuracy comparable to more complex models in literature.