Research On The Time-frequency Characteristic Analysis And Automatic Classification Method Of Power Quality Disturbances

In recent years,the development of the smart grid has became a national strategy of many developed countries,intellectualization has become the trend of international power grid development,and many intelligent power electronic devices were used in the smart grid,these devices has caused a lot of bad influence to the power quality undoubtedly.In order to improve the power quality effectively,the researchers not only need to monitor and record the disturbance signal,but also need to extract disturbance features from the large disturbance data to complete the automatic classification of disturbance signals,so the researchers can take corresponding measures to solve the power quality problems timely,it has important significance for the safe operation of the smart grid.Firstly,the paper introduced the definition and classification of power quality problems,and analyzed the reasons for paying close attention to the power quality disturbance widespread,and summarized the advantages and disadvantages of the time-frequency characteristic analysis method and automatic classification method of power quality disturbances.Secondly,this paper studied on the time-frequency characteristic analysis of power quality disturbances,and used the db4 wavelet made six floors of multi-resolution characteristic analysis of the different smart grid disturbance signals,and the difference of the energy of the wavelet transform signal in each layer and the standard signal energy was used as the feature vector,and did a preparation for the classification.Then,studied on the classification method of power quality disturbance signals,and proposed a classification method of power quality disturbance based on Fast Relevance Vector Machine(FRVM).First of all,divided the feature vectors into training samples set and test samples set,and used the training samples to train the FRVM,then used the test samples to test its classification performance.Finally,used MATLAB to simulation and analysis the proposed classification method based on wavelet transform and FRVM,the results show that,compared with the support vector machine(SVM),FRVM has less numbers of relevance vectors,shorter testing time,and has the ability to solve the large computational complexity and long training time problems of relevance vector machine(RVM).At the same time,its classification accuracy is better than the SVM and RVM method,and this method also can get high precision of classification accuracy in strong noise environment.