Power Quality Disturbances Analysis Based On S-transform And Pattern Recognition

Power quality disturbances analysis is one of the most important issue of powerquality control. It normally includes power quality signal compression, power qualitydisturbance recognition and detection. Different research result of power qualitydisturbance analysis are presented in this paper. A novel methods of power quality datacompression is proposed, the classifiers with different structure are constructed forsatisfying the diffferent requirements and a new method of disturbances detection andlocation is design under different disturbance characters.Each circle of Power quality (PQ) signal in the same event is similar. A novel PQdata compression method based on pattern similar measurement is designed in thisarticle. Firstly,6types of PQ signals are simulated to test wave distortion detectionability of different similar measurement. If a new PQ event happened, it will bedescribed by2circles which the distortion happened in, and the other circles in theevent are described as the recorded second circle. The novel approach could compressthe PQ signals in24hours without reduce the sampling rate. Compare to othercompression methods such as wavelet etc., the compression and reconstruction processof the new method is simple and more efficient. The high frequency features of PQsignals are also remained after compression. Simulation test shows that the new methodhas low distortion rate, short computation time and high compression rate. It’s verysatisfied for the requirement of PQ signal compression and analysis.Classification of power quality disturbance is one of the most important works inpower quality control. The process of classification is composed by two sequential stepsas signal features extraction and classifier design. The new method is proposed toclassify the power quality disturbances based on S-transform and modified MultilayerFeedforward Neural Network. Firstly, original signals is transformed by S-transform,4types of features are extracted from the result of S-transform. Then obtained features areutilized as input s into the modified Neural Network classifier based on Quasic-Newtonalgorithm and self-adaptive learning factor to realize the automatic classification ofpower disturbances. Simulation results show that with the merits of immune of noiseand improved learning ability, the proposed method can effectively classify fivedisturbance patterns including voltage sags, voltage swells, voltage interruptions,Oscillatory transients and harmonics.New approaches based on support vector machine （SVM）and ProbabilisticNeural Network (PNN) are designed for identification of power quality complexdisturbances and meeting the requirement of classification effectiveness and hardwarecost. Firstly, original power quality signals were processed by S-transform and featureswere extracted from the result of S-transform at different frequency areas. Then,2types of most distinguished feature were selected by statistic feature selection. The selectedfeatures were used as the input vector of SVM and PNN, the trained SVM and PNNbased classifier are using for power quality disturbances recognition. The proposedmethods reduce the computing costs of feature calculation, meanwhile saves the time oftraining and classification.8types of power quality disturbances including2types ofcomplex disturbances are accurate identified. The simulation results verify the validityof these methods.A novel high performance recognition method based on a generalized S-transformwith Hyperbolic window (Hyperbolic S-transform) and rule-based decision tree isproposed in this paper. The original power signals are analyzed by the HS-transform.4features are extracted to formulate decision rules. Finally, a rule-based decision tree isconstructed to classify power quality disturbances. The new approach has simpleclassification process and high classification effectiveness. It is very satisfied for thereal-time requirement in real work.After the modular matrix of power quality signal calculated by the HyperbolicS-transform, the disturbance is detected and located by the sum of amplitudes of eachsample point. The sum of amplitude is processed by threshold. Then, the points with thepeak value are extracted for disturbances detection. Finally, the different automaticdetection methods are designed for detecting different type of disturbances. The newdetection method realizes the accurate detection of7types of disturbances includingvoltage sag, swell, interruption, pulse, spike, notch and transient. The simulation onideal signals proves that the new method has the good adaptability and noise intensity.The extent of disturbances parameters has little impact on this proposed method.