Novel Incremental Machine Learning Approach for Predicting Pv Power in Systems with and Without Hydrophobic Coating

Abstract

The increasing demand for power, diminishing natural gas and oil resources, and climate change necessitate the integration of renewables like photovoltaics (PV) into the energy matrix. PV performance is significantly affected by environmental factors such as dust and temperature, especially in desert environments like Qatar. This study investigates the use of hydrophobic coatings on PV panels to mitigate the adverse effects of soiling and high temperatures. To accurately predict energy output under Maximum Power Point Tracking (MPPT) conditions, we developed a specialized PV monitoring system integrating environmental and climatic sensors. Over two months, data from PV panels with and without hydrophobic coatings were collected and made accessible via an IoT-Cloud system using LabVIEW™. Our findings indicate that hydrophobic-coated panels exhibit a 4-5% improvement in power output compared to non-coated panels. We employed machine learning (ML) techniques, including regressors, incremental learning, and Shapley Additive Explanations (SHAP), to rank features, train models, and predict maximum power output based on climatic and environmental parameters such as solar irradiance, temperature, relative humidity, dust density, and wind speed. The best performance was achieved using XGB Regressor-based stacking of fine-tuned Decision Tree, Random Forest, and Gradient Boosting models, with a mean-square error (MSE) of 0.09799 and an R² of 0.9769. Incremental learning further improved these metrics, achieving an MSE of 0.08732 and an R² of 0.9794. These results highlight the effectiveness of hydrophobic coatings in enhancing PV performance and demonstrate the potential of ML techniques in accurately predicting PV power output.

Keywords: Photovoltaic cell performance, Hydrophobic coating, PV power prediction, wireless monitoring, Machine learning