Contribution Au Développement De Procédures Monitoring Temps Réels De Diagnostic Et De Détection Des Pannes Dans Les Systèmes Photovoltaïques

Résumé: Faults in the stand-alone photovoltaic systems cause a significant power loss and put the system in critical and hazardous situations. To bring more safety features and increase the system efficiency, reliability, and lifetime of the components, this thesis aims to propose new fault detection and classification methods in the stand-alone photovoltaic systems as well as battery. These methods would overcome the limitations of existing fault analysis as well as the shortcoming of electrical protection devices such as a fuse. Two types of unique faults are treated in this thesis, the one that occurs in PV modules and those that evolve in battery cells, which have not been studied in the literature. The amount of data available from the system as well as the growing interest in machine learning techniques encourages the application of these methods as fault detection in stand-alone photovoltaic systems. Several techniques are adopted for fault detection and classification such as Artificial Neural Network, Anomaly detection, and Support Vector Machine. Their performance has been validated using data from both simulation and experimental platforms. For this purpose, an adequate simulation has been developed for fault investigation in a stand-alone photovoltaic system, and a benchmark system has been set up to experiment the faults in real working conditions. The test results demonstrate the feasibility of the proposed machine learning methods to detect and classify the faults in stand-alone photovoltaic systems which could be directed toward practical application.

Publié dans la revue: