The Research Of Power Quality Analysis Technology Based On Frequency Segmentation And Local Feature Selection

In recent years,the intermittent and random output characteristics of distributed generators such as photovoltaic and wind power are directly connected to the distribution network,which affects the voltage quality of the power grid.Meanwhile,in order to realize the distributed power consumption and flexible scheduling,a lot of nonlinear power electronic equipment is widely used and system operation(power supply and compensation capacitor switching)are frequently carried out in the distribution network.As a result,these lead to an increase in the number of power quality transient disturbance events.The accurate identification of the power quality disturbance signal is the premise and foundation of the effective comprehensive evaluation of the power quality and the accurate positioning of the disturbance source.It is of great significance to improve the power quality of the actual power system.In order to improve the performance of feature extraction,the optimal multi-resolution fast S transform is used to carry out signal processing,and partial features are extracted by frequency domain segmentation.Moreover,the lack of feature selection and optimal decision tree automatic construction method in complex power quality disturbances identification,a novel power quality feature selection and optimal decision tree construction method based on classification and regression tree was proposed.Firstly,12 kinds of power quality disturbance signals including 6 kinds of complex disturbances are simulated by a mathematical model called Matlab8.5.Subsequently,the disturbance signals are processed by optimal multi-resolution fast S transform,and 67 kinds of commonly used power quality features are extracted on the basis of time-frequency characteristic analysis and frequency domain segmentation to constitute the original feature set.Then,the optimal feature subset is determined by Gini importance and sorted according to an embedded feature selection method based on the Gini index.Therefore,the dimension of the feature vector is reduced and the structure of the classifier is simplified.Finally,subtree evaluation methods based on one standard error rule of the cross-validation evaluation are applied to determine the complexity parameter value,which is used for cost complexity pruning.After that,the optimal decision tree can be constructed automatically.The experimental results show that the new method can effectively reduce the computational complexity of S transform through optimal multi-resolution fast S transform,and extract effective features based on time-frequency analysis and frequency domain segmentation.According to the training set,optimal decision tree is constructed automatically,and the optimal feature subset selection is realized in the training process.In addition,the classification efficiency is improved under these influences.The optimal decision tree can accurately identify 12 kinds of power quality signals with 6 kinds of complex disturbances in different noise environments.The classification accuracy is higher than probabilistic neural network,extreme learning machine and support vector machine.The new method has good robustness and anti-noise performance.