| Power quality disturbances are complex and diverse.Detection and classification of various types of disturbances are the premise and basis for improving power quality.In this paper,the improvements and optimizations of S transform(ST)are studied to realize voltage disturbances parameter detection and feature extraction.Using multiple classifiers combination pattern recognition methods realizes the power quality disturbances synergetic classification.Firstly,aiming at the difficulties of optimizing parameter for generalized S-transform(GST),a parameter optimized method for GST is proposed and applied to the detection of the power quality disturbances.The value of parameter for the fundamental frequency is set independently to reveal the time disturbances,convenient to optimize the parameter for other frequencies on the concentration of frequency disturbances.So,the time-frequency disturbances characteristics can be obtained simultaneously in desirable resolution.The evaluating indicators for the determining parameter are put forward to provide the theoretical basis to determine parameter adaptively.The disturbances parameters such as starting and ending time,disturbances amplitude and harmonic components are further detected in this study.The experiment analysis and the application to real power engineering demonstrate that the method has good noise robustness and satisfactory accuracy.Secondly,a new method for the power quality disturbances recognition is proposed which combined the modified generalized S-transform(MGST)and the Extreme Learning Machine(ELM).To solve the problem of the GST that the window function is fixed and loses the gaussian characteristics in low frequency and high frequency,two adjusting parameters are introduced to the window function of the ST to control the time-frequency resolution more flexibly.The evaluation indexes are put forward to determine these parameters,providing the theoretical basis to define them adaptively.Four disturbances features are extracted based on the MGST as the input vector of ELM to realize the power quality disturbances classification.Experiments results of simulation data and engineering data showed that the method has higher classification accuracy with better anti-noise property.Thirdly,an adaptive optimal S transform(AOST)is proposed to extract the feature vectors of voltage sags in this pepar.With the effective window width matches the Fourier spectrum of the signals,the standard deviation of Gaussian window can be determined.Without additional parameters and iterative computations,the narrowest and widest window width of AOST are obtained to get the best frequency resolution and best time resolution,respectively.Based on the time-frequency matrix of AOST,five disturbances features are extracted to construct the feature vector.Fuzzy c-means,Gath-Geva and ELM classifiers are used to analyze the validity and redundancy of these features.Compared with ST,AOST provides higher time-frequency resolution to extract more precise feature vectors of eight types of voltage sags.Finally,a method of multi-feature extraction and multi-classifier synergetic analysis is proposed to classify the voltage sag disturbances.To extract more abundant feature information,a multi feature extraction method based on AOST,wavelet transform(WT)and Hilbert-Huang Transform(HHT)is proposed.The features separation is defined to evaluate their distinguishing ability.Based on correlation coefficients and mutual information,the whole correlation is put forward to sort these features,and use genetic algorithm and ELM for feature selection.The performance and diversity of classifiers are evaluated,and an adaptive weighted fusion multiple classifier is proposed to realize the synergetic classification of 20 typical power quality disturbances.The simulation results show that the classification accuracy,stability and anti-interference ability of multi-classifier are higher than that of single classifier. |
PDF Full Text Request