Uma abordagem de engenharia de dados para a avaliação de desempenho em plantas fotovoltaicas de geração de energia elétrica

• Ano de defesa: 2025
• Autor(a) principal: Nonato, Nicolle
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Pato Branco; Brasil; Programa de Pós-Graduação em Engenharia Elétrica e de Computação; UTFPR
• Palavras-chave em Português: Sistemas de energia fotovoltaica; Energia - Fontes alternativas; Energia solar; Mineração de dados (Computação); Photovoltaic power systems; Renewable energy sources; Solar energy; Data mining; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA; Engenharia/Tecnologia/Gestão
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/38973

The generation of electrical energy from photovoltaic systems (PV, from the English *photovoltaic*) has established itself as one of the main alternatives for diversifying the energy matrix and reducing greenhouse gas emissions. Due to its abundance, modularity, and progressive cost reduction, solar energy plays a strategic role in the energy transition and in the security of the electrical grid. However, the prolonged operation of photovoltaic plants is affected by multiple degradation factors, such as material aging, dust accumulation, variations in solar irradiance, and increases in cell temperature. These effects reduce the predictability of generation and introduce uncertainties regarding the expected economic return, posing a central challenge for managers and energy policymakers. This dissertation proposes a dynamic and adaptable approach for evaluating the performance of PV plants. It consists of two stages: the first corresponds to data engineering, which aims to adapt the operational records of any plant through data extraction, preprocessing, and discretization, in order to enable performance analyses and their comparability; the second stage corresponds to the performance analysis itself, structured as a cycle that involves discovering the process model from the data, evaluating that model, capturing performance metrics, proposing improvements for the real process, and continuous reevaluation with new data that cyclically describe new scenarios. The practical operationalization of this cycle, especially the model’s feedback with real-time data, can be achieved through integrated monitoring systems, such as Internet of Things (IoT)-based platforms and supervisory control and data acquisition (SCADA) systems, which allow for the continuous collection of operational and environmental variables. Although the reevaluation stage was not empirically executed in this study due to the lack of direct access to the plant for post-analysis interventions, the approach proposes an iterative implementation in real operational environments, with automatic model updates at defined intervals (for example, daily or weekly), depending on the size of the plant (scalable for small to large-scale plants, with data volumes ranging from megabytes to gigabytes) and the available infrastructure. The methodology was applied in a real case study, using data from an operating photovoltaic plant. The results indicated that solar radiation and cell temperature were, in that case, the most influential factors on performance, in addition to evidencing a slow but consistent degradation rate, on the order of 0.02% per year, linked to these factors. This rate is substantially lower than the average reported in the literature (approximately 0.5% per year), possibly attributed to regular maintenance practices that mitigate cumulative effects, such as aging and dust accumulation. Considering the typical lifespan of solar panels, estimated at 25 to 30 years, a cumulative degradation of about 0.6% over 30 years represents a minimal and non-significant loss in economic terms for well-maintained systems, although this rate may vary according to panel quality—high-quality panels (for example, with materials resistant to UV and humidity) exhibit lower degradation, while low-quality options may exceed 1% per year, impacting long-term viability. This approach can naturally be extended to other plants, contributing to the advancement of performance evaluation methodologies by offering an integrated and reproducible solution that integrates data science into the practice of energy generation.