Composite Power Quality Disturbance Identification Based On Improved S-Transform And Random Forest

In view of the poor time-frequency adaptive ability of the traditional S transform,the shape,main lobe width and side lobe height of the window function are too single,and the detection resolution of signals in different frequency bands is insufficient to meet the demand.This paper proposes a new improved S transform.(Modified S-Transform,MST),by improving the energy concentration of the S-transform,introducing four auxiliary parameters to control the shape of the Gaussian window function,optimizing the adaptive ability of the scale factor,and making the improved S-transformation better.Time-frequency resolution performance.By improving the S-transform to detect the power quality disturbance signal,and acquiring the mode time-frequency matrix,the time-frequency curve can be extracted,and the time-domain and frequency-domain characteristic information of the disturbance signal can be comprehensively reflected.In this paper,a system for evaluating power quality disturbance signals based on eigenvalues such as amplitude and phase kurtosis and skewness is established.By improving S-transformation,31 feature quantities of each disturbance signal are comprehensively extracted.In the node division,the Gini importance degree of each feature in the candidate feature subset is divided,and the feature with the greatest Gini importance is selected as the segmentation feature of the node.The importance ranking of all features is obtained.The tradeoff is used as the pattern recognition classification.The number of feature quantities input by the device is selected to classify and identify the feature combination that makes the disturbance recognition result higher.Based on Bayesian optimization,iteratively selects the hyperparameters in the Random Forest(RF)algorithm.This paper optimizes the four most important parameters in the process of random forest construction: the largest number of decision trees,the maximum number of features,and the maximum Depth and internal nodes subdivide the minimum number of samples required,through 8 kinds of single electrical energy such as standard signal and voltage sag,voltage swell,voltage interruption,voltage flicker,voltage harmonics,transient oscillation,voltage spike and voltage gap A total of 9 kinds of signals are classified and identified by the quality disturbance signal,Comparison of performance between random forest based on Bayesian optimization and traditional random forest pattern recognition classifier,which verifies the superiority of the improved algorithm.The highest recognition accuracy can reach 99.63%.In addition,based on the concept of multi-label classification,the stochastic forest algorithm with improved S-transformation and Bayesian optimization is applied to classify and identify the power quality disturbance signals of dual,multiple and a provincial actual power system.Among them,dual power quality disturbance recognition The upper limit accuracy rate is about 96%,the upper limit accuracy rate of the multiple power quality disturbance signals is about 93.5%,and the actual power system identification accuracy rate is 94.2%.Finally,this paper compares the recognition effect with the XBGoost and Lib SVM algorithms.The simulation results show that the improved algorithm designed in this paper has high classification accuracy,strong anti-interference ability,and low training signal and low signal-to-noise ratio(Signal-to-Under the condition of noise Radio,SNR),the classification results have obvious advantages.