Sistemas IoT e processamento de dados baseados em inteligência artificial para monitoramento de pressões e vazões na rede de distribuição de água

• Ano de defesa: 2024
• Autor(a) principal: Silva, Jonatha Bizerra
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Engenharia Elétrica; Programa de Pós-Graduação em Engenharia Elétrica; UFPB
• Palavras-chave em Português: Abastecimento de água monitoramento; Rede Neural Artificial (RNA); Outliers; TEDA; Artificial Neural Network.; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/34745

Water is a natural resource essential for sustaining life on Earth. Although most of our planet is composed of water, only 25 % is freshwater, and just 1 % is available for consumption. Brazil holds approximately 12 % of this percentage. Water supply systems accompany the development of cities, population growth, and geographical distribution, resulting in pipeline systems with heterogeneous characteristics, such as varying diameters and installation ages, which present various challenges in operation and maintenance, as well as compromising the guarantee of minimum technical consumption requirements. Losses in water supply systems, inherent to any network, have received increasing attention in recent years due to water scarcity and the risk of contamination of treated water. Efficient management of water resources is crucial for survival, highlighting the need for research addressing losses associated with water distribution to the population. In this context, this work aims to develop a methodology based on Typicality and Eccentricity Data Analytics (TEDA) for detecting outliers in data collected from an experimental water supply monitoring bench at the Laboratory of Energy Efficiency and Hydraulic Engineering in Sanitation at the Federal University of Paraíba (LENHS/UFPB), as well as the application of an Artificial Neural Network (ANN) for predicting flow in the monitored system. The results indicate that, when applying the ANN without the presence of outliers, the model achieved a Mean Absolute Percentage Error (MAPE) of 3.1 %. In contrast, when outliers were introduced only in the test data, the model resulted in a MAPE of 5.2 %, highlighting the system’s vulnerability to atypical data. Following the application of TEDA for detecting anomalies and replacing detected data with previous readings, the ANN yielded a MAPE of 3.0 %, reflecting a significant improvement in prediction accuracy.These advancements have the potential to positively impact the management of water resources by minimizing losses and promoting a more sustainable use of water, especially in a scenario where losses in distribution systems are significant. The research contributes to a better understanding of the behavior of flow and pressure in distribution networks, enabling more effective management of natural resources.