Study On De-noising And Classification Of Power Quality Disturbance Signals Based On Wavelet Transform

The power quality disturbance caused by increasingly complex electricity grid environment jeopardize the work of precision instruments,damage the life of power equipment and cause irreparable economic losses to users.The key to control power quality disturbance is to accurately identify and analyze it,but the key information containing the disturbance characteristics is always mixed with signal noise during signal acquisition and transmission,which will reduce the accuracy of detection.A variety of power quality disturbance with complex composition make the disturbance characteristics wrongly removed as noise during de-noising,as well as the inaccurate feature extraction and classification.In order to solve the above problems,this paper studied the de-noising,feature extraction and classification of power quality disturbance.(1)This paper proposed an improved wavelet threshold de-noising algorithm.By calculating the peak-to-sum ratio(PSR)of wavelet coefficients of each layer,the noise content was accurately estimated,and the correction factor could adaptively adjust the universal threshold according to the noise distribution of different disturbance signals.Meanwhile,an improved threshold function was proposed,the variable parameter of which could adjust its soft and hard characteristics to determine a suitable threshold function.We used this algorithm to de-noise seven kinds of common power quality disturbance signals,and the simulation results showed that the improved de-noising algorithm could get a better signal-to-noise ratio(SNR)for various disturbance signals.Under different noise interference,the de-noising effects were stable,the signal waveforms were reconstructed well,and the disturbance characteristics were preserved,providing accurate and effective information for following power quality analysis.(2)In this paper,we used a new feature extraction method.Lagrange interpolation was used to generate a fractional delay filter to optimize the standard wavelet,and the signal feature matrix was extracted from the detail coefficients of each layer in wavelet decomposition.We constructed a new wavelet called as forward and reverse Lagrange fractional delay wavelet(LFDW),which could amplify the energy of detail part in the process of signal decomposition,highlighted the signal disturbance characteristics,and obtained the optimal delay wavelet suitable for a variety of disturbance by adjusting the delay coefficients.Compared with db4 wavelet,LFDW had a flatter frequency response and a better wavelet decomposition performance in our research.This method had fewer steps to extract features and was easy to implement,which was suitable for more power quality disturbance types.(3)Meanwhile,the LFDW+PNN_PSO classification method for power quality disturbance was proposed.We mapped the particles in the particle swarm to the smoothing factor of probabilistic neural network(PNN),defined the classification accuracy as the fitness function,and automatically optimized the smoothing factor value through the iterative operation of the particle swarm.This could make the network training faster and get a better classification performance.The simulation results showed that this method could effectively identify ten kinds of common power quality disturbance,such as transient,steady-state and composite one,with a high overall classification accuracy.