Research On Disturbances Detection And Identification For Power Quality

With the increasing number of new energy power generation grid and the application of non-linear devices such as soft starters,switching power supplies,UPS,the problem of power quality in China is becoming increasingly serious.On the other hand,the application of more and more high-precision electronic instruments in modern industrial production also puts higher requirements on power quality.However,the prerequisite for governance and improvement of power quality is to effectively classify it.Therefore,how to improve the accuracy and efficiency of power quality disturbance classification has become a research focus of scholars at home and abroad,and has also been closely watched by the power sector.This paper attempts to apply the sparse self-encoder algorithm and deep forest algorithm in deep learning to the classification of power quality to improve the accuracy of power quality disturbance identification.Firstly,the definition and standard of power quality are introduced,and the power quality disturbances existing in the current power grid and its causes are analyzed.The single disturbance,double disturbance and triple disturbance existing in the actual power grid are taken as the research object,and the experimental data used in this paper are generated by MATLAB,and the waveforms are briefly analyzed.Secondly,in order to replace the complex artificial feature design process in the existing power quality disturbance identification method,this paper introduces the sparse self-encoder in the deep learning algorithm and uses the stack sparse self-encoder to automatically extract the deep features in the power quality disturbance data.The low-latitude feature expression is obtained in the dimensional data,which simplifies the classification work and overcomes the shortcomings of the randomization of the weight coefficient initialization in the traditional neural network.In this paper,the sparse self-encoder model is trained with the data of Gaussian white noise added,and the anti-noise ability of its feature expression is improved.The results show that this method can accurately identify nine kinds of power quality disturbance signals,including two kinds of Multiple Disturbances,and has good anti-noise ability.Finally,this paper proposed a classification way for power quality disturbances based on deep neural network of sparse autoencoder.As a deep learning algorithm,sparse self-encoder requires a large number of experimental samples to train the model.Therefore,when the sample data is small,the classification effect of this method on power quality disturbance is not very ideal.Inspired by the deep learning surface layer learning and layer by layer learning ideas,the multi-granularity scanning and cascaded forest processes are added to the original random forest to form a deep forest algorithm with strong feature extraction ability similar to the deep learning algorithm.Unlike the sparse self-encoder,the deep forest algorithm benefit from the dimension expansion in the multi-granularity scanning process,so that the deep forest algorithm can mine the characteristics of the data as much as possible with little sample data.The experimental results show that the proposed method can accurately identify 20 power quality disturbances including double or even triple Multiple Disturbances with limited samples.