Development of a wavelet transform and support vector machine based event recognition technique for power quality

Power quality will be a sought-after feature in power systems in the future as it has been in the recent years. Power quality problems cause such negativities as breakdowns and malfunctions in the system components and loads in the system. Therefore, it is essential that power quality problems in power systems are determined and so, probable negativities are prevented.In this thesis, an intelligent recognition technique is developed to determine the type of power system events which cause power quality problems. Three phase voltage signals are used to determine the types of power system events. In the first step of this recognition technique, normalization and segmentation processes are applied to the three phase voltage signals. In the second step, wavelet transform method is applied to the voltage signals and wavelet transform coefficients are obtained. Later, the two phase feature extraction process is applied to these coefficients and an effective feature vector which stands for the distinctive features of voltage signals and reduces the data size. In the last step of the intelligent recognition technique, power system event types are determined by using a support vector machine classifier which has a higher level of performance compared to the artificial neural network classifier in terms of both recognition time and performance level. The parameters of support vector machine classifiers are obtained by applying 10-fold cross validation test.The advanced analytical processing is applied to the fault events obtained from the output of the developed intelligent recognition technique. As a result of this, voltage dips in the system and power quality disturbance types are determined.Real power system data and simulation data obtained from the ATP/EMTP model are used to evaluate the performance of the developed technique. According to the results, the recognition technique is classified the three phase event types very accurately. According to the recognition results for noisy event data, the developed recognition technique has a robust structure. Besides, the intelligent recognition technique can function very accurately even if training is realized with little event data.