Data analysis methodology utilizing the statistical metrics weight of evidence (WoE) and information value (IV) to assist in asset management of power transformers

• Ano de defesa: 2024
• Autor(a) principal: REMA, Gabriela Sampaio
• Orientador(a): BONATTO, Benedito Donizeti
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Itajubá
• Programa de Pós-Graduação: Programa de Pós-Graduação: Doutorado - Engenharia Elétrica
• Departamento: IESTI - Instituto de Engenharia de Sistemas e Tecnologia da Informação
• País: Brasil
• Palavras-chave em Português: Power transformers; Asset management; Preventive actions; Insulating oil; Moisture in the oil; Variables prioritization; Information value; Weight of evidence
• Área do conhecimento CNPq: CNPQ::ENGENHARIAS::ENGENHARIA ELÉTRICA::SISTEMAS ELÉTRICOS DE POTÊNCIA
• Link de acesso: https://repositorio.unifei.edu.br/jspui/handle/123456789/4218

The Brazilian electric power transmission sector faces a significant challenge involving the end of the regulatory useful life of several of equipment. Given the technical and economic infeasibility of renewing all depreciated assets from a regulatory standpoint, the need for an assertive risk management analysis and a reliable assessment of the physical useful life of assets is emphasized, especially power transformers, the main asset in the electrical energy transmission sector. In view of this scenario, the objective of the proposed thesis is to add value to asset management through the development of a data analysis methodology to assist in decision-making regarding the direction of maintenance investment in power transformers. Due to their similarity, reactors are also evaluated. To this end, data on moisture in the insulating oil of the equipment were used and the following categorical variables: voltage class, installation region (Regional), criticality, type, and age of the equipment. It is noteworthy that these variables are technical registry data of the assets, and the water content is an essential characteristic for determining the operational condition of the insulating oil, being one of the properties measured in the physicochemical tests. The original contribution of the thesis is the selection of categories with greater weight and categorical variables with higher predictive power using the statistical metrics Weight of Evidence (WoE) and Information Value (IV). Analyzing the predictive importance of a variable before developing a predictor can lead to better performing models. Furthermore, data based decisions lead to more assertive and proactive actions, and the prioritization of variables for evaluation is an important contribution, especially considering large equipment parks. The methodology was applied to a dataset of almost 10 thousand oil samples from 795 power transformers and reactors from the ISA CTEEP, electrical energy transmission company in Brazil, responsible for approximately 95% of the energy transmitted in the state of São Paulo and about 30% of all energy in Brazil.