Optimizing Photovoltaic Soiling Loss Predictions in Louisiana: A Comparative Study of Measured and Modeled Data Using a Novel Approach

Abstract

Previous studies have shown that soiling losses on photovoltaic (PV) modules can lead to reduced power output of up to 80% in PV systems. Therefore, accurate determination of soiling loss plays a crucial role in predicting PV output and ensuring optimized cleaning schedules. The study focused on measuring soiling loss at a 1.1 MW outdoor testing facility in Louisiana, United States, using a DustIQ device, a commercially available soiling sensor. The maximum soiling loss recorded for DustIQ Sensor 1 was 7.5% on August 27, 2023, during the dry season. The measured data was fitted using the well-established Kimber and HSU models (based on PM_2.5 and PM₁₀) by optimizing the least squares error, resulting in observed mean absolute percentage error (MAPE) of approximately 0.82% and 0.78%, respectively. One feature of these models is that it is assumed that the solar panels will be completely cleaned after a rain event that reaches a set threshold limit. However, in-field testing at the site shows that assumption to be flawed, because the soiling ratio did not return to 1 or 100% even after significant rainfall events. To address this, improved versions of the Kimber and HSU models were developed to more accurately represent the recovery of the soiling ratio after rainfall events. The results demonstrated significant improvements, with the modified Kimber models achieving reductions in root mean squared error (RMSE) of 23%, 13%, and 1% compared to the optimized Kimber model, while the modified HSU model exhibited a 12% reduction in RMSE over the optimized HSU model. The overall MAPE was less than 1% for all models.