Search

This paper presents an integrated design of a fault classifier which uses a hybrid Wavelet-Artificial Neural Network (ANN) based approach. The data for the fault classifier is produced by PSCAD/EMTDC simulation program for 34.5 kV Sagmalcilar-Maltepe distribution system in Istanbul, Turkey. It is aimed to design a classifier capable of recognizing ten classes of three-phase distribution system faults. A database of line currents and line-to-ground voltages is built up including system faults at different fault inception angles and fault locations. The characteristic information over six-channel of current and voltage samples is extracted by the wavelet multi-resolution analysis technique. Then, an ANN-based tool was employed for classification task. The main idea of this approach is to solve the complex fault (three-phase short-circuit) classification problem under various system and fault conditions. A self-organizing map, with Kohonen's learning algorithm and type-one learning vector quantization technique is implemented into this study. The performance of the wavelet-neural fault classifier is presented and the results are analyzed in the paper. It is shown that the technique correctly recognizes and discriminates the fault types and faulted phases with a high degree of accuracy in the simulated model distribution system. © 2004 IEEE. IEEE Power Engineering Society, PES;Chinese Society of Electrical Engineering, CSEE 2004 International Conference on Power System Technology, POWERCON 2004 -- 21 November 2004 through 24 November 2004 -- -- 65033

This paper presents an integrated design of a fault classifier for distribution systems using a hybrid Waveletartificial neural network (ANN) based approach. Data for the fault classifier is produced by PSCAD/EMTDC simulation program for 34.5 kV Sagmalcilar-Maltepe distribution system in Istanbul, Turkey. It is aimed to design a classifier capable of recognizing ten classes of three-phase distribution system faults. A database of line currents and line-to-ground voltages is built up including system faults at different fault inception angles and fault locations. The characteristic information over six-channel of current and voltage samples is extracted by the wavelet multi-resolution analysis technique. Afterwards, an ANN-based tool was employed for classification task. The main idea in this approach is solving the complex fault (three-phase short-circuit) classification problem under various system and fault conditions. A self-organizing map, with Kohonen's learning algorithm and typeone learning vector quantization technique is implemented into the fault classification study. The performance of the wavelet-neural fault classifier is presented and the results are analyzed in the paper. It is shown that the technique presented correctly recognizes and discriminates the fault type and faulted phases(s) with a high degree of accuracy for different location and time of occurrence in the simulated model distribution system.

Tez (Yüksek Lisans) -- İstanbul Teknik Üniversitesi, Fen Bilimleri Enstitüsü, 2002, Thesis (M.Sc.) -- İstanbul Technical University, Institute of Science and Technology, 2002, Bu çalışmada, dağıtım sistemlerinde hibrid “Dalgacık-Yapay Sinir ağı (YSA) tabanlı” bir yaklaşımla arıza sınıflama işlemi gerçeklenmiştir. 34.5 kV “Sağmalcılar-Maltepe” dağıtım sistemi PSCAD/EMTDC yazılımı kullanılarak arıza sınıflayıcı için gereken veri üretilmiştir. Tezin amacı, on farklı kısa-devre sistem arızalarını tanımlayabilecek bir sınıflayıcı tasarlamaktır. Sistemde kullanılan arıza işaretleri 5 kHZ lik örnekleme frekansı ile üretilmiştir. Farklı arıza noktaları ve farklı arıza oluşum açılarındaki hat-akımları ve hat-toprak gerilimlerini içeren sistem arızaları ile bir veritabanı oluşturulmuştur. “Çoklu-çözünürlük işaret ayrıştırma” tekniği kullanılarak altı-kanal akım ve gerilim örneklerinden karakteristik bigi çıkarılmıştır. PSCAD/EMTDC ile üretilen veri bu şekilde bir ön islemden geçirildikten sonra YSA-tabanlı bir yapı ile sınıflama islemi gerçekleştirilmiştir. Bu yapının görevi çeşitli sistem ve arıza koşullarını kapsayan karmaşık arıza sınıflama problemini çözebilmektir. Bu çalışmada, Kohonen’in öğrenme algoritmasını kullanan bir “Kendine-Organize harita” ile “eğitilebilen vektör kuantalama” teknikleri kullanılmıştır. Bu “dalgacık-sinir ağı” tabanlı arıza sınıflayıcı ile eğitim kümesi için % 99-100 arasında ve sınıflayıcıya daha önce hiç verilmemiş test kümesi ile de %85-92 arasında sınıflama oranları elde edilmiştir. Elde edilen başarım oranları literatürdeki sonuçlara yakındır. Geliştirilen birleşik “dalgacık-sinir ağı” tabanlı sınıflayıcı elektrik dağıtım sistemlerindeki arızaların belirlenmesinde iyi sonuçlar vermiş ve iyi bir performans sağlamıştır., In this study an integrated design of fault classifier in a distribution system by using a hybrid “Wavelet- Artificial neural network (ANN) based” approach is implemented. Data for the fault classifier is produced by using PSCAD/EMTDC simulation program on 34.5 kV “Sagmalcılar-Maltepe” distribution system in Istanbul. The objective is to design a classifier capable of recognizing ten classes of three-phase system faults. The signals are generated at an equivalent sampling rate of 5 KHz per channel. A database of line currents and line-to-ground voltages is built up including system faults at different fault inception angles and fault locations. The characteristic information over six-channel of current and voltage samples is extracted by the “wavelet multi-resolution analysis” technique, which is a preprocessing unit to obtain a small size of interpretable features from the raw data. After preprocessing the raw data, an ANN-based tool was employed for classification task. The main idea in this approach is solving the complex fault (three-phase short-circuit) classification problem under various system and fault conditions. In this project, a self-organizing map, with Kohonen’s learning algorithm and type-one learning vector quantization technique is implemented into the fault classification study. The performance of the wavelet-neural fault classification scheme is found to be around “99-100%” for the training data and around “85-92%” for the test data, which the classifier has not been trained on. This result is comparable to the studied fault classifiers in the literature. Combined wavelet-neural classifier showed a promising future to identify the faults in electric distribution systems, Yüksek Lisans, M.Sc.

Directing at the problems that it is hard to determine fault type if the vibration of aero-engine's rotor system is over standard during the maintenance testing, a fault diagnosis approach based on kernel principal component analysis (KPCA) feature extraction and multi-class support vector machines (SVM) is proposed. This method first takes the use of nonlinear feature extraction of KPCA to extract the feature of testing cell standard fault samples. By computing the inner product kernel functions of rotor vibration signal's original feature space, the nonlinear map of rotor vibration signal transformed from low dimensional feature space to high dimensional feature space is achieved. The nonlinear principal components of original feature are obtained by performing PCA on the high dimensional feature data. Then, the nonlinear principal components are taken as eigenvectors of multi-class SVM to perform training and test. During the training period, optimize the relatively parameter by adopting cross optimization algorithm to find out the best penalty parameter and kernel function parameter. A high classification accuracy of training set and test set is sustained, overfitting and underfitting are avoided. Experiment results indicated that this method had good performance in distinguishing different fault modes for high-speed rotor, and was suitable for machinery's state recognition.

The reliability requirements of wind turbines have increased significantly in recent years inthe search for a lower impact on the cost of energy. In addition, the trend towards larger wind turbinesinstalled in remote locations has significantly increased the cost of repair or replacement of the compo-nent. In the wind industry, therefore, condition monitoring is crucial for maximum availability [1]. Thiscontribution makes a review of supervised machine learning classification techniques for wind turbinecondition monitoring using only SCADA data already available. That is, without installing extra sensorsor costly purpose-built data sensing equipment. Although there has been extensive research into the useof machine learning techniques for wind turbine monitoring, the more recent trend in this type of litera-ture is to focus on a specific WT sub-assembly: the bearings and planetary gearbox [2], the generator andpower converter [3], the blades [4], etc. Oil debris systems can detect pitting failures but cannot detectcracking faults. Vibration based systems can detect both pitting and cracking, but most cannot determinethe health of components in the planetary section. This work approaches condition monitoring of variouswind turbine components (torque actuator, pitch actuator, pitch sensor, and generator speed sensor) witha unique strategy. In particular, for this purpose, a review of supervised machine learning classificationtechniques is performed and analyzed.

In recent years, power quality has become the main concern in power system engineering. Classification of power system faults is the first stage for improving power quality and ensuring the system protection. For this purpose a robust classifier is necessary. In this paper, classification of power system faults using Empirical Mode Decomposition (EMD) and Support Vector Machines (SVMs) is proposed. EMD is used for decomposing voltages of transmission line into Intrinsic Mode Functions (IMFs). Hilbert Huang Transform (HHT) is used for extracting characteristic features from IMFs. A multiple SVM model is introduced for classifying the fault condition among ten power system faults. Algorithm is validated using MATLAB/SIMULINK environment. Results demonstrate that the combination of EMD and SVM can be an efficient classifier with acceptable levels of accuracy.

Accurate Fault location in an Electric Power Distribution System (EPDS) is important in maintaining system reliability. Several methods have been proposed in the past. However, the performances of these methods either show to be inefficient or are a function of the fault type (Fault Classification), because they require the use of an appropriate algorithm for each fault type. In contrast to traditional approaches, an accurate impedance-based Fault Location (FL) method is presented in this paper. It is based on the voltage-sag calculation between two measurement points chosen carefully from the available strategic measurement points of the line, network topology and current measurements at substation. The effectiveness and the accuracy of the proposed technique are demonstrated for different fault types using a radial power flow system. The test results are achieved from the numerical simulation using the data of a distribution line recognized in the literature.

Computational grids are established with the intention of providing shared access to hardware and software based resources with special reference to increased computational capabilities. Fault tolerance is one of the most important issues faced by the computational grids. The main contribution of this survey is the creation of an extended classification of problems that incur in the computational grid environments. The proposed classification will help researchers, developers, and maintainers of grids to understand the types of issues to be anticipated. Moreover, different types of problems, such as omission, interaction, and timing related have been identified that need to be handled on various layers of the computational grid. In this survey, an analysis and examination is also performed pertaining to the fault tolerance and fault detection mechanisms. Our conclusion is that a dependable and reliable grid can only be established when more emphasis is on fault identification. Moreover, our survey reveals that adaptive and intelligent fault identification, and tolerance techniques can improve the dependability of grid working environments.

A new approach is attempted for the differential protection of power transformer using advanced signal processing such as time to time transform along with Fuzzy clustering technique. The time-time representation is derived from inverse Fourier transform of S-transform, a method of representation of real time series as a set of complex, time local spectra. Here, TT-transform used to generate time-time TT contour) from the samples of differential current. Parseval’s theorem extracts the features like energy and standard deviations. Later these features are used as inputs to the Fuzzy clustering to distinguish inrush current, external fault current and internal fault current. Simulation of the fault (with and without noise) was done using MATLAB/SIMULINK software taking 2 cycles of data from each 400 samples. The advantage of the proposed algorithm provides accurate results even in the presence of noise inputs (energy and standard deviations).