Programa de Pós-Graduação em Engenharia Elétrica - PPGEE/ITEC

URI Permanente desta comunidadehttps://repositorio.ufpa.br/handle/2011/2314

O Programa de Pós-Graduação em Engenharia Elétrica (PPGEE) do Instituto de Tecnologia (ITEC) da Universidade Federal do Pará (UFPA) foi o primeiro e é considerado o melhor programa de pós-graduação em Engenharia Elétrica da Região Amazônica. As atividades acadêmicas regulares dos cursos de mestrado e doutorado são desenvolvidas principalmente nas Faculdades de Engenharia Elétrica e Engenharia de Computação, supervisionadas pela Coordenação do Programa de Pós-Graduação em Engenharia Elétrica (CPPGEE).

Navegar

Agora exibindo 1 - 15 de 15

Acesso aberto (Open Access)
Análise de desempenho de algoritmos para classificação de sequências representando faltas do tipo curto-circuito em linhas de transmissão de energia elétrica
(Universidade Federal do Pará, 2019-12-05) FREIRE, Jean Carlos Arouche; MORAIS, Jefferson Magalhães de; http://lattes.cnpq.br/5219735119295290; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
Maintaining power quality in electrical power systems depends on addressing the major disturbances that may arise in their generation, transmission and distribution. Within this context, many studies have been developed aiming to detect and classify short circuit faults in electrical systems through the analysis of the electrical signal behavior. Transmission line fault classification systems can be divided into two types: online and post fault classification systems. In the post-missing scenario the signal sequences to be evaluated for classification have variable length (duration). In sequence classification it is possible to use conventional classifiers such as Artificial Neural Networks, Support Vector Machine, K-nearest neighboors and Random forest. In these cases, the classification process usually requires a sequence preprocessing or a front end stage that converts the raw data into sensitive parameters to feed the classifier, which may increase the computational cost of the classification system. An alternative to this problem is the FBSC-FrameBased-Sequence Classification (FBSC) architecture. The problem with FBSC architecture is that it has many degrees of freedom in designing the model (front end plus classifier) and it should be evaluated using a complete dataset and rigorous methodology to avoid biased conclusions. Considering the importance of using efficient short-circuit fault classification methodologies and mainly with low computational cost, this paper presents the results of the KNN-DTW (K-Nearest Neighbor) algorithm analysis study associated with Dynamic similarity measurement. Time Warping (DTW) and HMM (Hidden Markov Model) algorithm for fault classification task. These two techniques allow the direct use of data without the need for front ends for signal pre-processing, as well as being able to handle multivariate and variable time series, such as signal sequences for the post-miss case. To develop the two proposed systems for classification, simulated data of short-circuit faults from the UFPAFaults public database were used. To compare results with methodologies already presented in the literature for the problem, the FBSC architecture was also evaluated for the same database. In the case of FBSC architecture, different front ends and classifiers were used. The comparative assessment was performed from the measurement of error rate, computational cost and statistical tests. The results showed that the HMM-based classifier was more suitable for the problem of classification of short circuits on transmission lines.
Acesso aberto (Open Access)
Classificação de arritmias cardíacas através de uma estrutura competitiva de redes neurais convolucionais autoassociativas
(Universidade Federal do Pará, 2023-05-11) CORRÊA FILHO, Sérgio Teixeira; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
This work proposes a system for classifying cardiac arrhythmias based on a competitive structure of Autoassociative Convolutional Neural Networks. Three neural networks were trained to reconstruct Electrocardiogram (ECG) signals for cases of patients with supraventricular, ventricular and normal beats. After training, the networks were allocated in a competitive parallel structure for classification of arrhythmias. The MIT-BIH arrhythmia public database of ECG signals was used for training and testing the networks, and for each ECG signal, from each patient, the QRS complexes of the heartbeats were extracted, which were the characteristics used as input. for the system, and these signals, which were in the form of temporal signals (1D), were transformed into digital images (2D) in order to use the capacity of convolutional neural networks for pattern recognition and feature extraction in images. For the development and performance analysis of the proposed structure, two paradigms that have been used in works already presented in the literature were used: interpatient paradigm and intrapatient paradigm, and the system obtained an accuracy of 96.97%, sensitivity of 96.30% and precision of 93.59% for the intrapatient case and accuracy of 94.05%, sensitivity of 70.43% and precision of 65.74% for the interpatient case. A comparative analysis with results from arrhythmia classification systems already presented in the literature shows that the proposed system presented similar results or, in some cases, better results than those already obtained, thus showing the applicability of the proposed structure to the problem
Acesso aberto (Open Access)
Classificação de tumores cerebrais: um estudo comparativo entre rede neural convolucional e rede neural convolucional com mecanismo de atenção
(Universidade Federal do Pará, 2024-09-30) SILVA, Ulrich Kauê Mendes Alencar da; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
Brain tumors are neurological diseases with a high potential impact on the lives of affected individuals, requiring a rapid and accurate diagnosis through complementary imaging tests, such as magnetic resonance imaging, which is considered the gold standard in this process. Considering the need for faster diagnosis, classification systems based on Machine Learning have been developed and within this context, this dissertation aims to present a comparative study between a Convolutional Neural Network (CNN) and a CNN with an attention mechanism, developed for the classification of brain tumors from magnetic resonance images. The comparative study aims to identify the impact of the attention mechanism on the performance of the CNN for tumor classification. For the development and evaluation of the proposed models, a public database was used, collected from the Kaggle website and made available by Masoud Nickparvar, which is composed of 7023 brain magnetic resonance images, segmented into four classes: glioma, meningioma, no tumor and pituitary. As a result, from the performance metrics obtained, considering the image base used for testing in both CNNs, an improvement in the CNN performance was observed after the introduction of the attention mechanism, where the network with this mechanism presented an increase of 1.98% in the accuracy metric, 2.07% in the precision metric, 2.18% in the sensitivity metric and 1.72% in the F1-score metric in relation to the CNN without the attention mechanism. It is also possible to highlight the results obtained in particular for the meningioma tumor class, since the CNN without the attention mechanism presented difficulties in classifying this class and, after the integration of the attention mechanism, the model obtained an accuracy increase of 6.54% for this class.
Acesso aberto (Open Access)
Estrutura competitiva de redes neurais autoassociativas para classificação de fadiga mental através de sinais de eletroencefalografia
(Universidade Federal do Pará, 2018-12-21) FERREIRA, Mylena Nazaré Medeiros dos Reis; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
The complexity of mental fatigue signals in healthy people is due to the absence of specific perturbations in the electroencephalographic activity, and by the singularity and variability of the cognitive profile of each individual. Identifying this mental state requires the analysis of several factors that involve the brain behavior in its regions in various frequency bands. In concern to the industry, mental fatigue compromises the efficiency of the production chain by affecting the perception (concentration and attention) of people, which increases the risk of accidents and production costs. Thus, monitoring the cognitive condition is necessary for the maintenance of the productive and cognitive performance of the evaluated subject. This work proposes the classification of fatigue using a competitive structure of Associative Neural Networks. This type of neural network allows to find the association between the input data and the reconstructed data from a compact architecture, being indicated for real-time applications. The characteristics vector used for classification is composed of the normalized information of three frequency bands (theta, beta and alpha) and four metrics that, according to the literature, differentiate mental states from electroencephalographic data in terms of Power Spectral Density. The results show the capacity and usability of autoassociative neural networks in patterns classification.
Acesso aberto (Open Access)
Estrutura competitiva de redes neurais convolucionais auto-associativas para classificação de arritmias
(Universidade Federal do Pará, 2019-04-17) BAIA, Alexandre Farias; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
This work presents the proposal of two automatic systems to aid in the detection of anomalies in heart beats and medical decision support. The systems were developed for the identification of rhythmic arrhythmia and morphological arrhythmias from signals obtained from an electrocardiogram (ECG). Both systems are based on a competitive structure of Convolutional Autoencoders (CAE), and each network was trained to reconstruct the signals presented at its entrance. For the case of the rhythmic classifier, the system was developed from the use of the ECG signals, without undergoing a feature extraction process, and for the case of the morphological classifier, the system was based on the QRS complex extracted from the ECG signal. For the development and testing of the systems, the database MIT-BIH Arrhythmia of ECG signals was used. An accuracy of 88.9% was achieved for the Rhythmic Classifier and 81.73% for the Morphological Classifier, in the case in which the evaluation basis is considered. The results obtained demonstrate the applicability of the proposed competitive structures to the arrhythmia classification problem.
Acesso aberto (Open Access)
Estrutura de redes neurais auto-associativas aplicadas ao processo de identificação de equipamentos elétricos em sistemas de monitoramento não intrusivo de cargas
(Universidade Federal do Pará, 2019-10-23) MORAIS, Lorena dos Reis; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
The pursuit of reducing and rationalizing electricity consumption is increasingly becoming a priority for all consumers worldwide. Residential environments are responsible for a large part of electricity consumption. Non-intrusive load monitoring systems were created with the aim of helping consumers, providing the possibility of obtaining information about the individual consumption of equipment and thus allowing a monitored consumption and the consequent increase in energy efficiency. In a Non-Intrusive Load Monitoring System, four steps are critical: acquiring aggregate data through a single sensor, detecting equipment on / off events from the aggregate load, extracting disaggregated signal characteristics and the identification of equipment based on the characteristics extracted from the disaggregated signal. In this context, this work proposes a new methodology for identification of electrical equipment in a residential environment employing a competitive structure of Auto-Associative Neural Networks. The proposed system is based on power signal measurements obtained from equipment on / off events. To test the proposed methodology 3 scenarios were developed using 3 different public databases. Due to the good results achieved, analyzed using statistical metrics, it is evaluated that the proposed methodology is able to efficiently perform the task of identifying electrical equipment, thus contributing to the development of future non-intrusive monitoring systems. meet market demands.
Acesso aberto (Open Access)
Extração de conhecimento em forma de regras difusas a partir de mapas auto-organizáveis de Kohonen: aplicação em diagnóstico de faltas incipientes em transformadores
(Universidade Federal do Pará, 2013-03-11) SILVA, Ana Carla Macedo da; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
Despite the many advantages offered by the artificial neural networks, some limitations still prevent their widespread use, especially in applications that require making decisions essential to ensure safety in environments such as in Power Systems. A major limitation of artificial neural networks with respect to the inability of these networks is to explain how to arrive at certain decisions. This explanation must be humanly understandable. Thus, this paper proposes a method for extracting fuzzy rules from Kohonen self-organizing map, designing a fuzzy inference system capable of explaining the decisions taken by the map. To verify its effectiveness, the method is applied to solve the problem of classification for the diagnosis of incipient faults in power transformers used.
Acesso aberto (Open Access)
Inteligência computacional aplicada à detecção e correção de outliers em séries temporais: estudo de caso em consumo de energia elétrica
(Universidade Federal do Pará, 2015-09-04) MELO, Diemisom Carlos Romano de; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
The electric load prediction is a task that requires accurate models, as should properly influence the decision making in hydroelectric plants and power stations. These computer models are implemented from a data set that must faithfully represent the behavior of the variables. However, these data sets are quite common the presence of outliers, which arise due to sensor reading errors, errors in the actual processing system / storage of data or faults in the distribution system or power station. This paper proposes a new methodology based on Computational Intelligence for detection and treatment of outliers in time series of electric power load. An auto associative artificial neural network is used for outlier detection. Subsequently, it is reused together with a genetic algorithm to correct detected outliers. This approach was applied to a time series of electrical power load in the State of Pará. The computational experiments were performed using the MATLAB tool and the results demonstrate the efficiency of the proposal, which identified and corrected all virtual outliers introduced during the evaluation phase of the methodology.
Acesso aberto (Open Access)
Metaheurísticas populacionais: estudo comparativo na sintonia de parâmetros de controladores clássicos
(Universidade Federal do Pará, 2016-12-02) VIDAL, Juan Ferreira; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
Population metaheuristics are techniques belonging to the field of Computational Intelligence and are based on natural models, have emerged as alternatives to solve optimization problems where the traditional techniques cannot be applied, or even where a solution model for the problem is not available with which the solution is found through empirical means. Given these capabilities to provide acceptable solutions in a timely manner for most of the complex problems encountered, metaheuristics has been applied successfully in most of the control system problems found in the literature. This work presents in general how the metaheuristics are being applied in the solution of control problems and performs a comparative study of performance among four algorithms bioinspirados in the tuning of the PID parameters. The following algorithms were used: Genetic Algorithm (AG), Genetic Algorithm in the Islands Model (AGMI), Bacterial Foraging Optimization (BFO) and Particle Swarm Optimization (PSO). The results demonstrate that the algorithms present an excellent performance in the tuning of the PID producing response that met the project requirements. Different systems with different characteristics were used to evaluate the algorithms. The PSO was shown as the best algorithm among the four used, producing response in a faster time and presented lower deviated standard in the trials.
Acesso aberto (Open Access)
Modelo de previsão hidrológica utilizando redes neurais artificiais: um estudo de caso na bacia do Rio Xingu- Altamira-Pa
(Universidade Federal do Pará, 2019-10-10) SILVA, Arilson Galdino da; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
Knowledge about the extent of riverbed overflow is extremely necessary for the determination of areas at risk. The City of Altamira-PA, located on the banks of the Xingu River, historically suffers from extreme events of floods that provoke floods, causing great damages to the population. Considering the problem, this paper presents a monthly level prediction system of the Xingu River based on neural networks perceptron of multiple layers. For the development of the system, rainfall data were used in the basin and sub-basins of the Xingu River, and SST information (Sea Surface Temperature) from 1979 to 2016. The Satisfactory results demonstrate the great applicability of Artificial Neural Networks to the flood prediction problem, as compared to other methodologies have greater precision in finding solutions for nonlinear problems. For the treatment and selection of the input variables, the correlation approach was used, with the objective of improving the accuracy of the results, thus selecting the best information with their respective lags, in which they are inserted in three prediction scenarios: model with rainfall data, model with sea surface temperature information and application using the SST junction with rainfall. To measure the prediction capacity of the proposed methods, the Mean Squared Error (MSE) and coefficient of determination (R²) values were obtained for the best strategy, using only oceanic variables, SST, being the values 2,99x104 and 0,9991 considering, mainly, the treatment of input values of the Neural Network.
Acesso aberto (Open Access)
Modelos para previsão de carga a curto prazo através de redes neurais artificiais com treinamento baseado na teoria da informação
(Universidade Federal do Pará, 2011-11-04) ALVES, Wesin Ribeiro; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
The previous knowledge of the load value is almighty important to the electric power system planning and operation. This paper presents results of an investigative study of application of Artificial Neural Networks as a Multilayer Perceptron with the training based on Information Theory to the problem of short term load forecasting. The learning based on Information Theory focuses on the use of the amount of information (Entropy) for the training of neural network. Two forecaster models are presented, and that they was developed using real data from an energy utility. To compare and verify the efficiency of the proposed systems, it was also developed a forecasting system using neural network trained based on the traditional criterion of mean square error (MSE). The results has showed the efficiency of proposed systems, which had better results when compared with the forecasting system based on neural network trained by criterion of MSE and with forecasting system already was presented in the literature.
Acesso aberto (Open Access)
Previsão de geração de energia fotovoltaica utilizando transformação de séries temporais em imagens e redes neurais convolucionais bidimensionais
(Universidade Federal do Pará, 2023-10-26) MONTEIRO, Diego Ramiro Melo; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
This research presents a novel approach based on a Bidimensional Convolutional Neural Network (CNN) and techniques for transforming time series data into images, such as Gramian Angular Field (GAF) and Recurrence Plot (RP), for short-term forecast of electricity generation from a photovoltaic microgrid connected to the electrical grid, located at the Center of Excellence in Energy Efficiency of the Amazon (Centro de Excelência em Eficiência Energética da Amazônia –CEAMAZON) at the Federal University of Pará (Universidade Federal do Pará –UFPA). The GAF and RP techniques were employed to transform the time series data into images, which were used as input for the CNN. More accurate electricity generation forecasts enable users to better estimate the potential costs for grid implementation and the payback periods, as well as assess the available load capacity that can be connected to the system with higher precision. The prediction results using GAF and RP with a 2D CNN were compared with results obtained using other established neural network architectures in the field, such as Multilayer Perceptron and 1D CNNs, yielding satisfactory Root Mean Square Error (RMSE) values. This demonstrates the applicability of using images generated from the transformation of photovoltaic time series data in a 2D CNN for this problem.
Acesso aberto (Open Access)
Reconhecimento de atividades humanas utilizando redes neurais auto-associativas e dados de smartphone
(Universidade Federal do Pará, 2016-12-16) SIQUEIRA, André Luis Carvalho; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
Human Activity Recognition (HAR) is an important challenging research area with many applications in intelligence ambient, healthcare and homeland security systems. HAR is the process whereby a person is monitored through sensors and analyzed to infer the undergoing activities during a period of time. This work presents the development of two systems for the HAR using auto associative neural networks. The activity recognition systems are based on public dataset that has signal from three static postures (standing, sitting, lying) and three dynamic activities (walking, walking downstairs and walking upstairs).The dataset was captured by using accelerometer and gyroscopic sensor of a Smartphone. The features extracted from the time and the acceleration due to body motion were used to the development of the proposed systems. Our experimental results illustrates the effectiveness of the proposed system.
Acesso aberto (Open Access)
Rede neural convolucional aplicada à identificação de equipamentos residenciais para sistemas de monitoramento não-intrusivo de carga
(Universidade Federal do Pará, 2018-04-03) PENHA, Deyvison de Paiva; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
This research presents the proposal of a new methodology for the identification of residential equipment in non-intrusive load monitoring systems. The system is based on a Convolutional Neural Network to classify residential equipment, which uses directly as inputs to the system, the transient power signal data of 7 equipment obtained at the moment they are connected in a residence. The methodology was developed using data from a public database (REED) that presents data collected at a low frequency (1 Hz). The results obtained in the test database show an accuracy of more than 90%, indicating that the proposed system is capable of performing the task of identification. In addition, the results presented are considered satisfactory when compared with the results already presented in the literature for the problem in question.
Acesso aberto (Open Access)
Redes neurais profundas aplicadas ao diagnóstico de faltas incipientes em transformadores imersos em óleo isolante.
(Universidade Federal do Pará, 2019-09-11) MORAES, Hugo Riviere Silva; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
Diagnosing incipient faults in transformers is a major challenge because it is very difficult to define the source and type of fault, due to the variability in the conditions under which faults occur. Conventional methods based on the analysis of dissolved gases in oil have been used by companies to diagnose faults, however, these methods still need to be applied together to reach a satisfactory result, as well as relying heavily on the knowledge of a specialist. In order to solve the difficulties related to conventional methods, some systems based on Computational Intelligence have been proposed in the literature and have presented promising results. This paper presents the results of the study developed of the application of deep neural networks to fault diagnosis, considering then the importance of fault diagnosis in transformers. Two models are proposed using Convolutional Neural Networks and Stacked Autoencoding Neural Networks. For the development of the systems we used the TC 10 database with faulty transformer situations. This base was used to develop the IEC 60599 method, which is one of the main methods used by power utilities for transformer diagnostics through the analysis of dissolved gases in oil. The promising results achieved with the two proposed models (100% accuracy in the test base) show the great applicability of deep neural networks to the problem of incipient transformer fault diagnosis, however showing a great alternative to the conventional methods commonly used.

Navegar

Navegando Programa de Pós-Graduação em Engenharia Elétrica - PPGEE/ITEC por Orientadores "CASTRO, Adriana Rosa Garcez"