Journal of the Korean Solar Energy Society ISSN:1598-6411(Print) 2508-3562(Online) 한국태양에너지학회 논문집

To achieve the greatest performance, photovoltaic (PV) systems need to be continuously monitored for fault detection and diagnosis, which could affect their performance under operational conditions. Various fault detection and diagnosis methods have been developed to improve these technologies. For example, the “Voltage and Current Measurement Method” based on the One-Diode model has been widely adopted in many case studies to predict a maximum power point (i.e., power, current, and voltage) and then detect any issues that occur under harsh outdoor conditions. Although the One-Diode model has demonstrated accurate performance for fault detection and diagnosis, the accuracy has typically been limited to crystalline PV modules. To overcome this limitation, many fault diagnosis methods have been proposed, and machine-learning fault detection and diagnosis methods are effective alternatives because of the nonlinear output features and varying operating conditions of PV arrays. Therefore, this study presents the prediction model of maximum power point (MPP) using artificial neural network (ANN) algorithms. The prediction model of MPP was generated through an optimization analysis. The MPP was validated by using measured data obtained from a building attached to a PV (BAPV) system in Korea. The results of the optimization study showed the highest predictive accuracy when the structure of the ANN model was trained and learned with one hidden layer, 20 neurons per hidden layer, a sigmoid function, and an 0.01 learning rate. The coefficient of variation of RMS error (CvRMSE) between the predicted value and the labelled value was 3.14%. The prediction model for MPP was verified by using the measured data. The analysis of the measured and predicted values resulted in CvRMSE values of 11.66%, 11.70%, and 18.01% for power, current, and voltage, respectively.