Classificação Multiestágio Aplicada ao Monitoramento Não-Intrusivo de String Fotovoltaica
Ano de defesa: | 2022 |
---|---|
Autor(a) principal: | |
Orientador(a): | |
Banca de defesa: | |
Tipo de documento: | Dissertação |
Tipo de acesso: | Acesso aberto |
Idioma: | por |
Instituição de defesa: |
Universidade Federal do Espírito Santo
BR Mestrado em Energia Centro Universitário Norte do Espírito Santo UFES Programa de Pós-Graduação em Energia |
Programa de Pós-Graduação: |
Não Informado pela instituição
|
Departamento: |
Não Informado pela instituição
|
País: |
Não Informado pela instituição
|
Palavras-chave em Português: | |
Link de acesso: | http://repositorio.ufes.br/handle/10/15719 |
Resumo: | Brazil has a high rate of solar incidence in all its regions solar energy is today the fourth most important source of renewable energy in the country, with a great chance of expansion and growth, a room for the need to search for solutions to operating problems of such systems. Regardless of the cause of the atypical string condition, its effect is somehow registered in the electrical generation itself, an electrical signature. Thus, the objective of this dissertation is to identify the operating condition of a photovoltaic string among twenty possible, one being normal and the others atypical. In the case of different atypical conditions. For this, a non-intrusive monitoring method is used based on electrical voltage and current samples generated by itself, different techniques based on Artificial Intelligence (AI) Classifiers. A methodology is adopted in order to divide a complex problem into smaller and less sub-problems. Therefore, two classification stages are considered: the first one whose objective is to identify one of five PV operating conditions, that is, normal, full panel shading, partial panel shading, short circuit and line break (electric arc); and the second stage, whose objective is to identify the PV panel causing an atypical condition. The classifiers used in both stages are the K-Nearest Neighbors (kNN), the Support Vector Machine (SVM) and the Multi-Layer Perceptron (MLP). The results achieved led to an average accuracy of 93.9% when using the classifier with the best performance in each subproblem treated. |