Please use this identifier to cite or link to this item:
Authors: Upendar, Jalla
Issue Date: 2010
Abstract: Overall development of modern society heavily relies /depends on electricity. With deregulation, electricity has become a commodity as well as a means for competition. Power quality, as a consequence, is coming into focus to an extent hitherto unseen. Industries as well as commercial and domestic users simply putting pressure to supply quality power. Transmission lines are vital links that achieve the continuity of service from the power generating plants to the end consumers. Protection systems for transmission lines are one of the most important parts in power systems. Most of the power system is frequently affected by the fault, which causes disruption to the power flow. The increasing complexities of the modern power transmission systems have greatly raised the importance of the research studies on classification of type of fault. The restoration or reconfiguration can be accelerated, if the type of a fault is accurately known, resulting in improved economy and reliability of the power system. Methods of fault classification can be divided into two subgroups: one which uses single-end information and /or other which uses information from both ends of the faulted line. Algorithms that use measurements of the signals at both terminals of the line are generally more accurate than the ones using data only from the local terminal. However, in many transmission lines, a communication channel between the local and remote terminals is not available, thus making it necessary to use data from the local terminal only. Fault classification algorithms based on only local terminal current signal data needs some simplifying assumptions to achieve improved performance. However, the techniques using single-ended data could be more attractive for researchers. So in this work, the fault currant signals collected at the source side only are considered for the analysis. At the time of a fault instant, the sudden changes in steady state current signals produce high frequency transients, which cause undesirable effects to normal distance protection algorithms based on power frequency components. The practical setting for a general transmission line protection causes mal-operation of the relay due to the effect various operating conditions. When the fault position is closer to relay, the high frequency component in the transients signals are more prominent. A traveling wave scheme will find it difficult to distinguish the arrival of consecutive wave-fronts when collecting wave-front information with the use of high fidelity transducers. After reviewing the various techniques and their limitations, new protective relying schemes are proposed in this work using timefrequency analysis and pattern recognition approach. The fault transient waveform contains both fundamental and harmonic component during the occurrence of fault that degrade the quality of the current and waveforms sensed by the relay. Therefore, the serious drawback of Fourier transform method is that in transforming to the frequency domain, time information is lost, i.e. it is impossible to tell when a particular event took place. So to detect transients, abrupt changes etc, Fourier analysis is not suitable. Thus, the usage of a wavelet transform (WT) is more adequate to perform frequency-based fault detection, since the WT analyzes a wide frequency bandwidth and not only one frequency component as the Fourier transforms. The protective relays are connected through current & voltage transformers specially, cause of current transformer (CT) saturation and large DC offset currents in the CT primary signal, the distortion in CT secondary current waveform will occur. The second harmonics content is also present in the CT secondary current signal at the time of fault. When higher secondary emf exists during steady state condition, the non-linear transformer exciting impedance only causes odd harmonic distortion. But, during saturation under fault transient condition, it produces many harmonics, with dominance of second and third harmonic contents. Typically, the fault-induced transients are inside a limited frequency bandwidth, that corresponds to frequencies from 0.1 Hz to 1 kHz. So in the proposed work, mainly the characteristics of both second and third harmonic components present in current waveform during fault is used with the help of wavelet transform. There are many different kinds of functions that satisfy the wavelet property. Daubechie's wavelet is most well known. They represent the foundations of wavelet signal processing, and are used in numerous applications. Thus, in this work Daubechie's wavelet is used for the implementation. The current signals Ia, lb, Ic (each of two full cycles) at a frequency of 50 Hz measured at the sending end are being decomposed in 9-levels using multi resolution analysis (MRA) algorithm. The Discrete Wavelet transform (DWT) acts as extractor of distinctive features present in the input current signal, which are collected at source end. The information is then used to obtain the threshold feature values to classify the different type of faults. In li this thesis, a new algorithm is proposed to obtain the threshold feature values with the use of decomposed wavelet coefficients minimum-maximum boundary values. For selecting the various threshold values for the proposed algorithm, training samples were created by simulating all types of fault at different locations along the line. Each fault was initiated at various inception angles considering different values of fault resistance (between 0 to 200 ohms), thereby providing a total of 49140 training fault cases randomly. However, total 1209600 fault cases were selected as the test cases for checking the accuracy. The overall performance of 99.72% is obtained by determining the threshold values with the help of Minimum-Maximum values of the wavelet features, and only 3416 fault cases are not classified accurately. Nowadays, Genetic Algorithms (GA) is used as an effective tool for analyzing data and for pattern recognition which are very difficult or even impossible with traditional methods. To improve the fault classification accuracy, genetic algorithm is used to generate the threshold values of fault classification algorithm by defining the fault classification objective function as an optimizing function. The fault classification method using genetic algorithm correctly classified 1206815 fault cases and only 2785 faults were misclassified. The overall classification of 99.77% was obtained with GA based algorithm. In comparison to genetic algorithm, the particle swarm optimization (PSO) is a relatively recent computational intelligence technique based on swarm intelligence paradigm. Various works show that particle swarm optimization is equally well suited or even better than Genetic algorithm for solving a number of linear, non-linear and multi modal optimization problems. At the same time, a particle swarm algorithm is much simpler, easier to implement and has a fewer number of parameters that the user has to adjust. Modified PSO method is considered to be capable to reduce the effect of misclassication of fault type. The Modified PSO based fault classification algorithm misclassifies only 2515 fault cases and overall fault classification accuracy is further improved to 99.79%. Thus, the PSO based proposed technique can be used as an effective evolutionary alternative approach for classification of power system faults. The use of artificial neural network (ANN) in fault classification analysis has gained a lot of interest among researchers due to its ability to do parallel data processing, high accuracy and fast response. Therefore in later part of the thesis, author made an attempt to improve the fault classification accuracy by implementing various neural in network topologies. The probabilistic neural network (PNN) approach offers major advantages such as rapid training, easy to add or delete data from training set without lengthy retraining process. The training vectors are transformed into the weight vector. Hence, a wavelet transform based PNN method to classify the power system faults is proposed. The proposed algorithm extracts power system faults using DWT. The output signals of DWT are trained using PNN that classifies the power system fault transients. The DWT considerably simplifies the input signal of PNN. Extensive simulation studies have been conducted to verify the feasibility of the proposed classification scheme for various fault distances, inception angles, and fault resistances with symmetrical and unsymmetrical faults. The PNN based fault classification method misclassifies 6997 faults cases out of 1209600 fault cases. The overall performance with the use of PNN method is found to be 99.42%. Some ANN's have encountered the 'stability-plasticity dilemma' where they learn new things and some times destroy some of the old data clusters from old learning. The Adaptive Resonance Theory (ART) has a feature that prevents much of this from occurring. Hence, the neural network architecture called as ART2 is used for fault classification. The performance of ART2 based fault classifier for all types of faults is verified, it is observed that the ART2 provides very good classification of 99.908%) of average success for all type of operating conditions, misclassifying only 1110 fault cases out of 1209600 fault cases. One of the most popular neural networks is Back-Propagation Neural Network (BPNN) method for solving many non linear problems, but the original back-propagation networks suffer mainly from the drawbacks of slow convergence. Over the years, different improved variations of BPNN have been proposed to specifically address several important issues. Genetic algorithm has been also used in training ANNs recently, but in the training process, this algorithm needs various GA operators. Even though its training results may be better than the ones using the BP algorithm, but when the neural network structure is large, GA's convergent speed becomes very slow. The PSO algorithm is an alternative to GA in training the perceptrons of the ANN. The author has made an attempt to present an application of PSO based multi-layer perceptron neural network in classifying faults in transmission lines with the help of wavelet transform. In order to have good accuracy, it is evaluated with more number of fault cases at different system IV conditions, most of the training samples do not coincides with the testing data set. It is observed that the PSO-ANN provides very good classification with 99.91% of average success including all type of operating conditions and misclassifies only 1058 fault cases out of 1209600 fault cases. It is noted that testing cases of the PSO-based network are able to give a successful prediction rate of up to 99.91%, which is higher when compared to existing BPNN (99.88%) and SVM (96.01%) based methods. Since the inception of Classification and Regression Tree (CART) in the last decade, the methodology of its tree-structured adaptive non-parametric regression has been widely used in statistical data analysis. More recently, the Classification and Regression Tree (CART) methodology has caught the interest of a wide community of applied mathematicians and digital signal /image processing engineers. CART is nonparametric which implies that this method does not require specification of any functional form. It does not require variables to be selected in advance. CART algorithm will itself identify the most significant variables and eliminate non-significant ones. The author has also made an attempt to present the application of a new statistical decision-tree based fault classification technique using CART method. In order to classify the faults, the CART tree is used which must be trained with the help of training samples and CART algorithm, before implementing the proposed method. The learning database with the variety of faulted samples is used to improve the CART generalization capability. CART based algorithm tries to split all observations of the learning sample. It first makes more important splits and then tries to capture the overlapping structure. It isolates the similar observations from the rest at each split. With the aid of CART algorithm, the training sample generates a tree, which can be represented by 13 rules. The test samples are recognized using the rules extracted from the decision tree generated by CART algorithm. In total, this method misclassifies 352 fault cases out of 1209600 fault cases. The performance accuracy of this proposed method is 99.97%). The test result shows that the method based on proposed statistical CART algorithm is able to classify faults with very high precision under various fault conditions. In addition, this technique overcomes the problem of setting the detection thresholds inherent in the existing techniques by optimizing their settings. It can be concluded that the proposed fault classification technique is simple and can achieve very high accuracy. The author has also made an attempt to present a Comprehensive Adaptive Distance Relaying Scheme for Parallel Transmission Lines. Double circuit lines are commonly utilized in modern transmission network, but they are difficult to be protected. They have been chosen as the best solution in many cases due to their advantages such as increasing the reliability and power transmission capacity of the systems and are cost effective. Distance protection relaying algorithms are commonly applied for the protection of over head transmission lines. Their operation is based on measuring the input impedance of the line using the voltage and current signals at the relay location. However, when applied to parallel lines, the performance of conventional distance impedance relays to measure the apparent impedance at the relay point is affected by the mutual coupling between the lines, shunt capacitance, fault location and fault resistance. In this work, a new adaptive neural network scheme for the adaption of digital distance relay operating region setting for a two terminal parallel transmission line is presented. The detailed mathematical model is developed and the analytical behavior of the relay considering the variations in fault resistance, fault position, effect of mutual coupling between the parallel lines, effect of shunt capacitance and pre-fault power flow conditions is studied. The operating condition of parallel line could not be always constant; it will change from one to another condition during normal operation due to various reasons. One option /solution to the above problem is by changing the relay settings each time when the bus configuration or system operating condition is changed. To overcome this problem, the relay boundary setting should be made adaptive by considering the effects of shunt capacitance and mutual coupling both. This is achieved with the help of back propagation neutral network. The trained network based on BPNN and Levenberg- Marquardt optimization provides the quick, reliable and accurate response for all four boundary conditions. All the test results clearly show that the proposed adaptive technique is well suited for the protection of parallel lines without any mal-operation in relay trip signal. In summary, the thesis focuses on the detailed study on the problem of accurate fault classification of power transmission line. The solution proposes the algorithms with the help of emerging intelligent techniques like GA, PSO and artificial neural networks like PNN, ART, PSO-ANN and decision tress based CART methods. The simulation work has amply demonstrated that the decision tree based CART algorithm provides the better accuracy of fault classification amongst all classification methods. In addition to this VI work, a new adaptive neural network scheme for the adaptation of digital distance relay operating region setting (adaptive distance protection scheme) for a two terminal parallel transmission line is also presented along with the detailed mathematical modeling.
Other Identifiers: Ph.D
Appears in Collections:DOCTORAL THESES (Electrical Engg)

Files in This Item:
File Description SizeFormat 

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.