Application of Machine Learning and Convolutional Neural Networks for the Fault Detection and Classification Monitoring System in PV Plants

To ensure high reliability of the larger photovoltaic (PV) systems, fault detection is an essential tool to keep the system under optimal functioning. In this paper, a fault detection model is proposed using Machine Learning Techniques (MLT) for the detection and classification offaults occurring in PVsystems. For the experimental verification, normal state and various fault state and datasets are couected using a circuit simulated in MatLablSimulink. From the couected datasets, relevant features were extracted and then fed into a Machine Learning (ML) subsystem containing Decision Tree and Random Forest models. To predict and classiff faults with new data, the ML models were trained by means of the historical data. The trained Random Forest model showed a classification accuracy of 90%, exceeding the performance of a baseline Decision Tree method which showed 76% accuracy. The condition of partial shading in PV arrays is also addressed, and a progressive, evolutionary Maximum Power Point Tracking (MPPT) method is working to combat it; the results of which proved its effectiveness and superiority compared to a baseline MPPT model. Fouowing this, a Convolutional Neural Network (CNN) model for PV fault detection by predefined thermographic images is constructed and optimized to yield a model that can be deployed on ultra-edge devices in the form of a web application. This application proves the effectiveness of the proposed system for quick and accurate response with low computational capacity.