Fault Detection in Photovoltaic Systems Using a Machine Learning Approach

Project Code :TCMAPY2193

Objective

This project focuses on detecting and classifying faults in photovoltaic (PV) systems using machine learning techniques such as XGBoost, Random Forest, Decision Tree, SVM applied to real-world PV operational data. The application allows users to upload photovoltaic system data in CSV format and provides accurate predictions of different fault conditions, including normal operation, short circuit faults, open circuit faults, and partial shading. Model performance is evaluated using metrics such as accuracy, precision, recall, F1-score, and confusion matrix to ensure reliable fault detection. A real-time prediction interface is provided for instant fault identification. Built using Flask for the web framework and MySQL for data management, the system aims to improve the reliability, efficiency, and maintenance of solar power plants by enabling early fault detection and data-driven decision-making through machine learning.

Abstract

This project focuses on detecting faults in photovoltaic systems through a machine learning approach. The system uses a dataset with parameters such as current, voltage, temperature, and ranges to classify conditions into normal or fault categories. Machine learning models analyze these features to identify issues like variations in current or voltage that indicate problems. The proposed approach combines multi-source data, including operational parameters such as I1, I2, maximum and minimum currents, variances, voltage means, power, temperature, and range values, to build a robust predictive model. A dual-layer architecture integrates supervised learning classifiers including Decision Tree, SVM, Random Forest, and XGBoost with a feature encoding mechanism that adapts to unseen data. The system analyzes real-time parameters like current variances, voltage indicators, and anomaly patterns to classify threats into two categories: Normal and Fault. By employing ensemble learning strategies, the detection accuracy is significantly enhanced, achieving over 90% classification performance across multiple models. Furthermore, the framework includes actionable recommendations tailored to the identified fault level, enabling faster mitigation and proactive risk management. The proposed methodology provides a scalable and adaptable solution for securing photovoltaic systems, reducing downtime, and minimizing the economic impact of faults.

Keywords :

Photovoltaic Systems, Fault Detection, Machine Learning, Intelligent Classification, Ensemble Learning, Decision Tree, SVM, Random Forest, XGBoost, Predictive Maintenance, Anomaly Detection, Solar Infrastructure Protection.

NOTE: Without the concern of our team, please don't submit to the college. This Abstract varies based on student requirements.