Using Machine Learning for Performance Classification and Early Fault Detection in Solar Systems

Abstract

The steady increase in the world's population has directly influenced global climate change, resulting in catastrophic environmental consequences. This has created an immediate need for scientists from interdisciplinary domains like clean technology innovation in solar energy and computer science to join in the effort to save the world for future generations. As such, the United Nations has set a goal to ensure global access to affordable, sustainable, and clean energy. As a leading influential G20 economy, Saudi Arabia has recently established the Green Saudi initiative to align with the UN goal for enhancing the use of green energy. However, research in this area is sparse and greater effort is still required. This work is among the first to address the issue of enhancing and expanding the use of clean energy by means of studying the data collected from solar plants around Saudi. We used machine learning-based methods to assess the energy output performance of solar plants and employed the collected data to train the models to make early detection of faults. Our models achieved the highest performance at an accuracy score of 98.85% and 0.98 weighted F-score using the J48 model trained on a publicly available dataset of 874 instances collected from 26 different sites across Saudi. We anticipate that the findings of this work to serve as testbed to facilitate further research in this area and enhance the early fault detection in solar energy stations.