Power Quality Disturbance Identification And Classification Based On Generalized S Transform And DDAGSVM Optimization

There are a large number of loads with non-linear,impact,and unbalanced electrical characteristics in the power system,which causes power quality problems such as the injection of a large number of pulses and flicker in the power grid.Detection efficiently and analysis of power quality problems is a hot topic of research today.The focus of this paper is to improve the generalized S transform,which will be applied to detect and locate power quality disturbance components as well as proposed an improved decision-directed acyclic graph SVM(DDAGSVM)recognition algorithm.The time-frequency distribution of the signal reflects the local energy distribution.The more concentrated the energy,the weaker the “modal aliasing” phenomenon of the signal and the better the time-frequency performance.For the self-adaptive optimization of window amplitude stretching factor and frequency scale stretching factor b,a new two-degree-of-freedom optimization algorithm based on time-frequency focusing criterion and new parameter algorithm is proposed in this paper,the parameters corresponding to the optimal time-frequency focusing measure are taken as the optimal parameters by combining Djurovic algorithm,Stankovic algorithm,and segmentation strategy,the optimal parameters are properly fitted in order to prevent faults.Improve the time-frequency focusing performance by optimizing the window width of each frequency frame,finally,the optimization of Generalized S Transform(OGST)is used to analyze signal perturbation components and extract feature quantities.The extreme point of the signal will be detected by Dyn dynamic measurement,which will be used to detected the disturbance components of power quality as well as the OGST transform in this paper.The power spectral density reflects the frequency domain energy distribution of the signal,the main frequency point can be detected by Dyn measure its power spectral density extreme point.Perturbation start and end moments will be obtained by Dyn measure,which will be used to detect the point of sharp change of the gradient because of the gradient difference between the distortion point and the normal point.The energy spectrum of frequency-squared mean of OGST time-frequency matrix reflects the frequency domain distribution of the signal,as well as the time-squared mean reflects the time domain distribution of the signal.For compound disturbances,the main frequency point and disturbance time are acquired by Dyn measure,which will be used to detected the energy spectrum of frequency-squared mean as well as time-squared mean.In view of the error increases because of the randomness of the selection and decision-making trend of the root node of the DDAGSVM decision classifier,In order to reduce error accumulation,this article introduces generalized KKT conditions and spatial discrimination,which reagrds the level of discrimination between classes as the basis for the selection of classifier nodes and takes the highest degree of differentiation as the root node.At the same time,the kernel function adjustment factor is optimized by the genetic algorithm(GA)so that building the recognition patterns.This paper compares the comprehensive performance of the three parameters of GSO-DDAGSVM,GA-DDAGSVM,and PSO-DDAGSVM in the identification of power disturbance signals by using three data sets: time-consuming,recognition accuracy,and convergence speed,finally GA-DDAGSVM algorithm is selected as power quality identification scheme,as well as the robustness of the algorithm in Gaussian white noise environment with SNR=20/30/40/50 dB and the recognition rate of compound disturbances are verified.