Soiling type classification and prediction of power loss of a PV panel using CNN

Abstract

With the rapid advancement of technology, greener and more efficient means for energy sources are always sought after. Harvesting solar energy is an effective way to generate electricity. Unfortunately, PV panel surface soiling is a major disruption in energy harvesting since it massively lowers the ability of the solar panel to be exposed to sunlight. Given how dire the air pollution situation is in Bangladesh, this is undoubtedly one of the major problems which have to be addressed when it comes to solar panels setup. When thousands of solar panels are setup in a remote location in which sunlight is abundantly available, the PV panel site has to be monitored to check if there are any issues, one of the issues being soiling. Manually checking thousands of PV panel images for soiling is laborious and time-intensive. We intend to automate that process using a lightweight deep learning model that can be incorporated into any system with fairly average computational power. More specifically, our deep learning model can determine if a particular PV panel is clean or soiled and classify the type of soiling. It can also make an approximate power loss prediction through image classification. This process will massively optimize the process of monitoring and negate the need for manually checking all the PV panels for soiling. In this paper, we propose the aforementioned deep learning model and discuss in detail how it has been developed from scratch and how feasible it is