Uma metodologia para identificação de perdas não técnicas em unidades consumidoras irrigantes de café do estado de São Paulo

• Ano de defesa: 2022
• Autor(a) principal: Sousa, Natalia Bastos de
• 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: Perdas não técnicas; Consumidores rurais; Irrigantes de café; Aprendizado de máquina; Non-technical losses; Rural consumers; Coffee Irrigators; Machine learning; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://repositorio.ufsm.br/handle/1/27053

Within the rural network, several factors can influence the difficulty on the part of distributors to locate fraudulent or irregular consumer units that may be causing non-technical losses, such as, for example, the difficulty in correlating the consumption of their irrigating customers with occurrences of divergent consumption. than expected, due to the necessary technical knowledge about the characteristics of each type of crop. Thus, this dissertation aims to present a methodology for detecting possible occurrences of non-technical losses for coffee irrigating consumer units that are CPFL Paulista's clients. For this, the study of the phenological characteristics of the coffee tree, management attributes of the irrigation system, and agrometeorological data essential for consumption estimates was carried out. Mapping the expected periods of irrigation system management, the clustering by Kmeans technique was used to group the consumer units, generating classes of consumption profiles based on their energy billing history and the period of billed energy registration (month). These classes and consumption (monthly) relationships are used as input to the implementation of a Random Forest model, used as the predictive model that classifies the units based on the most current consumptions and compares the predicted class with the class stored in the database. . The identification of customers that may be fraudulent occurs if the predicted class is of a lower level than the expected class for the same observed month. Complementing the methodology, since consumer units of lower classes would not change their Consumption Profile in the event of lower energy consumption, the consumption estimation method was used to detect possible occurrences of non-technical losses, comparing the real consumption with the estimated. The classification predictive model was able to detect changes in Consumption Profiles in a satisfactory way, and the estimation model shows that, for consumer units in the coffee growing area of São Paulo, the irrigation method is still used in a complementary way to the farm productivity.