Simulação e classificação de falhas no lado CC de sistemas fotovoltaicos

• Ano de defesa: 2025
• Autor(a) principal: Silva, José Leandro da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Engenharia Elétrica; Programa de Pós-Graduação em Engenharia Elétrica; UFPB
• Palavras-chave em Português: Energia fotovoltaica; Classificação de falhas; Classificação multiclasse; Comissionamento de usinas fotovoltaica; Curvas I-V e P-V; Aprendizado de máquina; Fotovoltaic energy; Fault classification; Multiclass classification; Photovoltaic power plant commissioning; I–V and P–V Curves; Machine learning; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/36897

In this work, a method is presented for classifying faults in the direct current side of photovoltaic power plants, using a simulation model developed in Matlab/Simulink 2023b and a classification algorithm implemented in Python. The proposed methodology is based on the extraction of features from current–voltage and power–voltage curves to classify faults that directly affect the energy efficiency and safety of photovoltaic installations. For method validation, two photovoltaic plant models were developed: one with a single string and another with three strings, simulating different environmental and operational conditions. The generated data totaled 3,712 samples for the single-string system and 4,640 for the three-string model, which were divided into three subsets and subsequently split randomly into 70% for training and 30% for testing. Faults were labeled in both configurations. In the single-string system, the assigned labels were: 0 (normal condition), 1 (shading), 2 (short circuit), and 3 (connector failure). In the three-string model, label 4 (open circuit) was added. The proposed method employs multiclass classification techniques, applying the One Versus Rest and One Versus One approaches. In the tests performed, the method achieved 100% accuracy in fault classification for both techniques. In the single-string system, this performance was obtained with 704 samples, while in the three-string model, the same accuracy was reached with 2,480 samples. These results demonstrate the effectiveness of the developed approach, which significantly contributes to fault diagnosis in photovoltaic power plants, promoting greater reliability and efficiency in operation and maintenance activities, in addition to providing support to professionals and companies in the sector.