Shading anomaly detection framework for bi-facial photovoltaic modules

Abstract

ABSTRACT Solar photovoltaics have been widely used as solar energy harvesting systems for many years throughout the world. Lately, bi-facial modules have been gaining a lot of popularity due to utilizing solar irradiance on both sides of the module. Therefore, the presence of partial static shading on such modules may lead to some ambiguity regarding their output powers and efficiencies. In this work, a shading anomaly detection framework comprised three stages: An autoencoder-convolutional neural networks CNN model, a mean absolute error MAE threshold, and data filters. The framework was developed to detect the occurrence and location of partial shading on bi-facial modules. Several experiments were carried out using two bi-facial modules under different shading settings. The modules were connected to solar chargers and batteries to analyze their performances. The experimental results showed the modules’ generated current and the batteries’ state of charge SOC in all shading settings. The results also showed that anomalies or shading can be detected with an accuracy of more than 99% merely from the second stage of the framework. However, the location of shading can be classified and predicted with an accuracy of 91% by utilizing all three stages of the framework.