Classification and Detection of Power Quality Disturbances using Discrete Wavelet Transform and Support Vector Machine
The aim of this paper is to perform the detection and classification of the Power Quality Disturbance using Discrete Wavelet Transform and Support Vector Machine. A novel approach is used for the feature selection where multiple linear regression models are used for the selection of statistically significant features and obtain an optimal feature vector. The work was conducted on synthetic data that was generated by introducing faults in the three-phase voltage supply working at a frequency of 50Hz and sampled at 10 kHz. The fault event data points generated are linearly distributed so multiple linear regression is appropriate to apply on the dataset. This proposed methodology involves the detection of various three-phase disturbances incorporated in the supply voltage signals. The feature extraction is performed using a discrete wavelet transform with a 9-level multiresolution analysis. The 10 sub-bands after performing DWT are then used to evaluate various features. Thus creating a feature vector of 90 features using the detailed and approximation coefficient for each phase. The features for all the fault events are recorded. These events are normalized for accurate classification of disturbances. The feature selection methodology is proposed for improving classification results and achieving high levels of accuracy by using a support vector machine.
Keywords - Power Quality, Discrete Wavelet Transform, Power Quality Disturbance, Support Vector Machine, Multiple Linear Regression