Utilização de geolocalização para detectar perdas não técnicas em ligações clandestinas em sistemas de distribuição de energia elétrica

• Ano de defesa: 2024
• Autor(a) principal: Stromm, Kamila Cassol
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Engenharia Elétrica; UFSM; Programa de Pós-Graduação em Engenharia Elétrica; Centro de Tecnologia
• Palavras-chave em Português: Isolation Forest; Geolocation; Energy theft; Non-technical losses; Built-up areas; Geolocalização; Furto de energia; Perdas não técnicas; Áreas construídas; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://repositorio.ufsm.br/handle/1/33027

The reduction of non-technical losses is a significant challenge for electric utility companies in Brazil, directly impacting both the rates paid by consumers and the quality of service provided. These losses occur mainly due to energy theft or fraud. This study aims to use geolocation tools to enhance the identification of theft and fraud. Various data sources were employed, such as the National Address Database for Statistical Purposes (CNEFE), municipal geoportal data, and the Geospatial Database of the Utility (BDGD), along with specific methods to achieve these objectives. The analysis focused on areas with high levels of irregularities and urban clusters, based on the principle that properties with buildings should have a formal connection to the utility, as electricity is an essential service. Properties with buildings but lacking an official connection were analyzed, as well as those without registered construction in municipal data but with address registration, suggesting that the National Institute of Geography and Statistics (IBGE) identified these areas as residential. These suspicious points were processed and classified using the Isolation Forest machine learning algorithm, which identifies anomalous patterns by grouping electrical regions based on the available information. The case study identified 409 suspicious properties within an area of 4.342 analyzed properties, of which 50 lacked registered construction with the municipality. The analysis also revealed that some of these properties had shown irregularities in previous inspections, indicating possible recidivism. The results allow the utility to more accurately and efficiently target its inspection efforts.