Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
DISCENTE: JOSE CARLOS CORREIA LIMA DA SILVA FILHO
DATA: 10/07/2025
HORA: 14:30
LOCAL: Sala de Aula do Núcleo de Computação de Alto Desempenho
TÍTULO: Localizacao de Falta em Redes de Distribuicao de Energia Eletrica atraves de Redes Neurais Convolucionais
PALAVRAS-CHAVES: Energy Quality, Fault Location, Distribution Networks, Convolutional Neural Networks, SmartGrid, Digital Image.
PÁGINAS: 101
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
RESUMO:

The increasing complexity of electrical distribution networks, driven by infrastructure modernization and the incorporation of distributed generation, makes it essential to develop efficient methods for fault location. This work proposes a methodology based on the combination of Genetic Algorithms and Convolutional Neural Networks (CNN) to enhance the identification of the bus affected by faults in electrical distribution networks. The employed method segments the system into strategic regions, using key buses to reduce computational complexity and optimize data analysis. A fundamental aspect of the proposed methodology is the transformation of data into visual representations, which serve as input to the CNN. To this end, the characteristics of voltage signals—such as root mean square (RMS) value, peak value, peak factor, average value, and distance to the substation—are extracted and converted into pixel matrices. These matrices are then normalized and combined into a five-channel image, where each channel represents a specific characteristic of the signal. This approach allows the CNN to process the information efficiently by exploiting complex patterns in the data, thereby improving the accuracy of fault location. The generation of these images is essential to the effectiveness of the method, as it transforms temporal data into a visual representation that facilitates the identification of relevant patterns for fault detection. The modeling and simulation of the IEEE 34-bus system using ATP software were employed to generate a dataset, enabling the analysis of different fault scenarios. The results demonstrated that the proposed approach outperformed conventional methods by exhibiting high accuracy in detecting the affected bus, achieving an accuracy rate of 99.16% and surpassing techniques such as AdaBoost with Artificial Neural Networks (98.30%) and standalone Artificial Neural Networks (97.60%). The comparison with approaches based on Graph Neural Networks (GNN) across various fault types evidenced that the proposed method attains superior performance in fault location while simultaneously optimizing data processing. Moreover, representing the data as panoramic images enhanced the extraction of relevant patterns for fault detection, leveraging the ability of CNNs to recognize complex structures. In simulation tests, the model’s accuracy reached up to 98.50%, depending on the fault type and the segmented group, further demonstrating the robustness of the methodology.

MEMBROS DA BANCA:
Externo ao Programa - 000.***.***-19 - FABBIO ANDERSON SILVA BORGES - UFPI
Interno - 2025885 - FLÁVIO HENRIQUE DUARTE DE ARAÚJO
Presidente - 2061294 - RICARDO DE ANDRADE LIRA RABELO
Externo à Instituição - 279.***.***-50 - ROGÉRIO ANDRADE FLAUZINO - USP
Externo à Instituição - 299.***.***-18 - SILVIO GIUSEPPE DI SANTO - USP
Interno - 1446435 - VINICIUS PONTE MACHADO

Cadastrada em: 12/06/2025