A Novel Machine Learning based Computing Algorithm in Modeling of Soiled Photovoltaic Module

Abstract

Soiling of Photovoltaic (PV) Panels is the accumulation of these dust particles on the panel surfaces. Soiling absorbs, scatters, and reflects a portion of incoming sunlight, lowering the intensity that reaches the solar cell's active part. Predicting expected power under soiling conditions is a difficult problem, and a variety of models have been proposed, each with a different set of inputs. Here, we report on a novel machine learning based computing algorithm in modeling of soiled photovoltaic module. The daily measurements (current, voltage, temperature, and wind speed), as well as a dependent variable (power expected) of PV performance loss due to soiling are available for 90 days, of which 80 days were considered for modeling. Two methods of cross validation were used to ensure that there was no overfitting or underfitting in the training and testing data. The data was split into two groups with the hold-out cross validation approach used. In this study, 75\% of the data was used to train the models and 25\% was used to test them. The efficacy of the proposed model was evaluated against the multilinear Regression (MLR) model during calibration, and validation periods based on Mean Absolute Error (MAE) and Nash–Sutcliffe efficiency (NSE). According to the results of appraisal, the models developed in this study performed good in estimating the predicted power of a soiled PV module with minimal error in Gaussian Process Regression Matern 5/2 GPR-M (MAE = 0.0784 and 0.0784 and high efficiency in NSE = 0.9745 and 0.8604) in training and testing respectively. Furthermore, the results indicate the better performance and suitability of the model with five input parameters in predicting the expected power.