Classificação de falhas em plantas fotovoltaicas usando aprendizado de máquina

• Ano de defesa: 2020
• Autor(a) principal: Costa, Clayton Hilgemberg da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Brasil; Programa de Pós-Graduação em Engenharia Elétrica e Informática Industrial; UTFPR
• Palavras-chave em Português: Geração de energia fotovoltaica - Aspectos Ambientais; Sistemas de energia fotovoltaica - Monitorização; Localização de falhas (Engenharia); Aprendizado do computador; Redes neurais (Computação); Modelos matemáticos; Simulação (Computadores); Photovoltaic power generation - Environmental aspects; Photovoltaic power systems - Monitoring; Fault location (Engineering); Machine learning; Neural networks (Computer science); Mathematical models; Computer simulation; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA::SISTEMAS ELETRICOS DE POTENCIA; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/5158

With the progressive increase in interest in solar photovoltaic energy, there is also an growing interest in monitoring plants, as well as the development of practical methods to detect and even classify faults that may occur in any part of photovoltaic systems. Among some of the motivators for this growth, it can be mentioned the fact that most of these systems are in an open and unprotected environment, being susceptible to the effect of meteorological events, such as strong winds and storms, which leads to a greater tendency to failures. Another factor is that the failure of a single system element, such as a photovoltaic module for instance, can impair the functioning of the others, in addition to the fact that a system operating with failures for a long time can reduce the lifespan of the equipment and even generate risks as the occurrence of fires. Thus, this work presents a classification system for short-circuit, degradation, open circuit and shading failures in on grid photovoltaic plants. This system was built using machine learning models, using measurements of local irradiance and temperature, as well as the voltages and currents generated by the photovoltaic modules. Also, the classification model was trained with data gathered from simulation and was also implemented in a monitoring system, which was used to collect the electrical and weather variables around the photovoltaic plant. The work was validated in a real 5kWp plant and took into account two approaches, one with and other without fault detection. The final system presented a competitive result with those found in the literature, reaching over 90% accuracy, while still presenting relevant advantages such as: non-intrusive classification, classifier embedded in the monitoring system, training with simulation data only and validation with data including the natural shading on the photovoltaic plant.