Fault Prediction in Photovoltaic Systems with Extreme Learning Machine Modeling

Modern electric power systems are highly complex and dynamic interconnected systems. The integration of new and renewable energy plants (EBT) such as solar power plants (PLTS) will affect operations. In photovoltaic (PV) systems, the PV output characteristics are nonlinear. This happens due to certain conditions, such as different weather conditions. Many errors may go undetected due to the nonlinear nature of the PV output characteristics. Many researchers and engineers have been working to find the best solution for detecting various early faults in PV with high accuracy and low computational load. In this study, the PV errors discussed are prediction errors in photovoltaic arrays, inverter faults, network anomalies, feedback sensor errors, and MPPT controller errors with varying severity. This study aims to predict errors in the photovoltaic system automatically without having to identify and perform measurements manually. The method used in this research is Extreme Learning Machine (ELM). In this study, 13 types of parameters were used with a total of 1070068 data. This research succeeded in predicting photovoltaic errors automatically using K-Fold Cross Validation with an average accuracy of 77.83%. The number of neurons in the hidden layer is three neurons and the data used is data for the last two years.