Study On Power Quality Transient Disturbance Detection

The power industry and consumers’s concern over power quality is becomingmore and more apparent recently. Accurate detection and classification of powerquality problems are of great significance for the analysis and the comprehensivemanagement of power quality. All power problems fall into twocategories,namely,steady-state problems and transient disturbances. This paper mainlystudies power quality transient disturbance detection and classification.Firstly,an introduction of the status quo and development trend of power qualitydetection and classification both at home and abroad is given in this paper, followedby the definition and classification of power quality and power quality nationalstandards. New requirements and development prospects of power quality detectionin the context of Smart Grid are also provided.This paper then made a detailed study of applications of wavelet analysis,support vector machine and artificial neural network in the detection of power qualitydisturbances. The paper mainly focuses on theoretical basis of the various methodsand corresponding disturbance detection principles, with an emphasis on time domainlocation of transient disturbances in wavelet analysis and transient disturbanceclassification in support vector machine and artificial neural network applications.Most of the disturbance classifiers built on wavelet theory and neural networksuse energy distributions of wavelet coefficents as feature vectors and a single neuralnetwork to give final results.The performance of such classifiers can be furtherenhanced.With an good grasp of current disturbance classification methods avalable, thispaper advanced a new strategy in power disturbance classification,which is based onwavelet transform and BP neural network. This paper employs sym4as motherwavelet and performs11-level wavelet decomposition, and incorporates bothstatistical characteristics and wavelet transform domain energy distributions ofdisturbance signals into the feature vectors.By doing so, the resulting classifier canget more information and distinguish between different disturbance types moreeffectively. In respect to the fact that there are many uncertain factors,such as noiseand sample outliers, involved in the classifier construction process,outputs from threeindependent artificial neural networks are combined by using the least squares method to give the final discriminant results. The classifier performance is enhanced byutilizing multi-source infromation.Digital simulation shows that the proposedclassifier performs quite accurately, even with disturbance signals corrupted by noise.The resulting classifier can effectively identify such power quality disturbances asvoltage interruption, voltage sag, voltage swell, harmonics, oscillation transient, andflicker.