Study On The Power Quality Analysis Based On The Wavelet Packet Transform And Random Forest

The large amount of new energy and the diversity of power load have led to many power quality problems.In order to improve the power quality of the power system,an effective signal processing method must be adopted to detect the power quality(PQ)events.Therefore,this paper studies the parameter detection and disturbance classification of PQ events.In order to detect the key parameters of PQ events,this paper chooses the 4-layer Db4 wavelet packet algorithm(WPT),which is used to detect voltage swell and transient impulse signals.The simulation results show that the accuracy of the key parameters of transient disturbance signal detected by WPT has reached above 99.9%.For the steady harmonic detection,this paper proposes an improved WPT algorithm based on Hilbert transform(HT).By using HT for signal frequency shifting,the improved algorithm avoids serious aliasing leakage due to the wavelet filtering,therefore improves the overall precision of the harmonic detection.The simulation results verify that the measuring error of each harmonic component in the improved algorithm is below 0.01%,which indicates that the proposed method can improve the accuracy of wavelet packet algorithm effectively.This paper also studies the classification of PQ events,which contains two aspects: feature extraction and classification.Firstly,features are extracted using 4-layer Db4 WPT and appropriate mathematical statistical methods,then selected feature from them using the genetic algorithm(GA),which expresses the signal character with a low dimensional feature vector.Among them,this paper proposes a fitness function definition method of GA based on the Pearson correlation coefficient,as the feature selection evaluation index in the process of optimization.Finally,the random forests algorithm(RF)is used to classify PQ disturbances.An improved artificial fish algorithm is proposed to optimize the key parameters for the RF.Simulation experiments show that RF with the optimized parameters can maintain high classification performance,and its classification efficiency is greatly improved.In addition,the classification result of the real-life signal from IEEE database verifies that the proposed methods are suitable for PQ classification.