Identification of power quality disturbance types by using intelligent pattern recognition approachs

In this thesis, the powerful algorithms are developed to increase the efficiency and reliability of the power quality monitoring systems. Firstly, the comparatively analyses are done by using three different signal processing methods based on time-frequency representations to extract the different unique characteristics of power quality disturbance signals. It is shown that more effective analysis can be performed by discrete wavelet transform and s-transform methods. Then, two different reduced feature vectors are obtained without losing main distinguishing characteristics of power quality disturbance signals. The types of power quality disturbances are identified by applying separately to input of support vector machine pattern classifier of these feature vectors.The performance of the developed algorithms is evaluated by using real power system data, simulation data and synthetic data obtained from ATP/EMTP and mathematical models. The obtained results is shown that the most important advantages of these algorithms are the high accuracy classification of power quality disturbance signals including more than one of the disturbance type and different noise. Therefore, reduction of the size of feature space is to make more feasible for real time applications since complexity of the classifier is decreased and consequent computation time of the proposed systems is shorten.