Machine Learning Approach to Detect Dust on Solar Panels in UAE. A Contribution toward Optimizing Cleaning Plan

United Arab Emirates is one of the leading countries in the production of concentrated solar power (CSP) having the world’s largest single-site solar project located in Abu Dhabi (Sweihan Independent Power Project (938 MW) – “Noor Abu Dhabi”, led by Marubeni and JinkoSolar). As well as the world’s largest CSP plant (Shams 1 in Abu Dhabi) with 2.5 square kilometer plant that has the capacity to feed 100 MW of electricity into the national grid, and many more promising projects. In addition to that, UAE has set up initiatives to start deploying solar photovoltaic (PV) panels on numerous residential properties, working toward achieving zero carbon emissions by end of 2050. Since solar power became widely used and developed, so focusing on the reliability and performance of solar panels became a main concern for government, producers, and owners. One of the main factors, which causes a drop in the efficiency of solar panels, is the accumulation of dust, which requires an innovative solution to monitor and control soiling measurement, especially in those regions with high soiling amounts such as the Middle East where energy loss due to soiling could reach 50%. UAE and Gulf region in general extended their efforts to measure and monitor the impact of dust and air pollution on solar energy production, especially since the UAE is within a zone with the highest dust intensity.

The accumulation of dust on the surface of solar panels reduces the efficiency of the solar modules and hence the amount of produced energy. In a country like UAE where output power loss due to dust accumulation is considered among the highest rate; monitoring and cleaning solar panels is a crucial task, hence developing an optimal procedure to monitor and clean these panels is very important in order to increase modules efficiency, reduce maintenance cost and reducing the use of resources. This project investigates different machine learning approaches to detect dust and bird droppings on solar panel surfaces to report the highest possible accurate approach..

The objective of this work is to investigate the ability of different machine learning classifiers to detect dust on solar panel surfaces with the highest possible accuracy and to develop a dashboard/open approach to contribute toward optimizing the solar panel cleaning process.

With the duration of 4-weeks, this project aims to: - Collect the maximum possible dataset of solar panel images in UAE (dusty, clean) and level of dust. - Research, data pre-processing, and augmentation - Investigate convolutional neural networks (CNN) for deep learning classification, and (if possible) explore the existing approach in literature (feature extraction-GLCM, SVM, etc,...) - Classify images using the trained models - Fine-tuning of the pre-trained model - Reporting the machine learning approach with the highest accuracy, and developing a simple dashboard for cleaning schedule

During this project, participants will be mainly able to

– Perform data collection and pre-processing for solar panel images

– Explore and examine the existing pre-trained convolutional neural network and investigating the Deep Learning Toolbox and transfer learning

– Implementing different Machine learning models to classify solar panel images

– Building a dashboard to visualize and present the finding

– Provide a recommendation report on the best approach for detecting dust on solar panel surfaces