ANN-ANFIS Comparison for Power Prediction of PV and PV-PCM Panels with Different Fin Geometries

In this study, artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS)-based modeling approaches are employed to predict the electrical power output of uncooled photovoltaic (PV) panels and photovoltaic panels integrated with phase change materials (PV-PCM). The dataset used in the study consists of power output measurements and environmental parameters obtained from an uncooled PV and PV-PCM systems with different fin geometries. Solar irradiance, ambient temperature, and panel surface temperature are used as input variables to train and test both ANN and ANFIS models. The predictive performance of the developed models is quantitatively evaluated using the root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination (R²). To ensure the statistical reliability of the models, a 5-fold cross-validation scheme was implemented due to the limited dataset size. The results indicate that the ANFIS model significantly outperformed the ANN across all configurations. For the uncooled PV, ANFIS achieved a mean R² of 0,7390 ± 0,0795 and an RMSE of 0,4652 ± 0,2070, whereas the ANN model's performance was more sensitive to data variability, yielding a mean R² of 0,2161 ± 0,1776. Notably, among the various fin geometries, the configuration with vertical perforated separated fins (Model-2) achieved the highest predictive accuracy with a mean R² of 0,8762, demonstrating the model's ability to effectively capture complex heat transfer dynamics in advanced cooling designs. Across the different fin geometries, ANFIS demonstrated superior robustness and predictive stability, identifying it as a more reliable modeling approach for PV-PCM systems under restricted experimental conditions.