Evaluating soiling effects to optimize solar photovoltaic performance using machine learning algorithms

This is the authors' abstract. We don't add key points for this paper.

Proposed methodology overview for soiling effects and estimation using various machine learning algorithms with Bayesian Optimization. • Machine learning algorithms are employed to predict power reduction due to soiling. • Soiling losses are also estimated by machine learning models. • Algorithm performances are improved by Bayesian optimization. • Predictive performance decreases with increasing soiling. • Gaussian process regression and artificial neural network excel other algorithms. Fossil fuel environmental issues and escalating costs have prompted a global shift towards renewable energy sources like solar photovoltaic. However, optimizing the performance of photovoltaic systems requires a comprehensive investigation of the various factors that reduce their power generation. Dust accumulation is prevalent in arid regions like the United Arab Emirates, posing a significant challenge to solar photovoltaic performance. Therefore, this study investigates the effect of soiling (from 1% to 5%) on electrical parameters (open circuit voltage and short circuit current), photovoltaic panel characteristics (cell temperature and module efficiency), and environmental variables (wind speed and irradiance) in the United Arab Emirates based Noor Abu Dhabi Solar Project. Additionally, machine learning algorithms such as artificial neural networks, support vector machines, regression trees, ensemble of regression trees, Gaussian process regression, efficient linear regression, and kernel methods are employed to predict power reduction due to soiling and soiling losses across various soiling percentages. Hyperparameter optimization using Bayesian methods enhances predictive performance. Results show Gaussian process regression and artificial neural networks excel in accuracy, though all models’ performance declines with increased soiling. Economic analysis via system advisor model highlights significant revenue drops in power purchase agreements with higher soiling, emphasizing need for proactive cleaning and maintenance.