Classification Of Power Quality Disturbance Based On Deep Learning

At present,with the rapid development of various new technologies in the power field,the power source and power load in the power grid are increasing,which has a certain impact on power quality.Therefore,various power electronic devices and power users in the power grid have put forward higher requirements on power quality.In order to control and improve power quality,it is first necessary to accurately identify power quality disturbances.This article first explains the research background and research significance of power quality disturbances,analyzes the power quality disturbance identification methods commonly used at home and abroad,and introduces the definition and standards of power quality under different conditions and the types of disturbances under different classification conditions.For the reasons,specific characteristics and consequences,the mathematical model of the disturbance signal is established and the disturbance signal sample data required by the research is generated in MATLAB.Secondly,it compares the sampling and compression effects of the compressed sensing algorithm and the traditional signal sampling algorithm,and introduces the specific implementation steps of the compressed sensing algorithm.By selecting the appropriate sparse transform base,measurement matrix and reconstruction algorithm,the disturbances studied in this paper are analyzed.The signal is compressed and sensed to obtain sparse vectors of different disturbance signals and reconstructed signals.Analyze the network structure of one-dimensional convolutional neural network,introduce the basic structure of neurons,the specific operation steps of convolution and pooling,and the training process of neural network.Aiming at the sparse vector of the power quality disturbance signal,a classification model based on one-dimensional convolutional neural network is established in Tensorflow/keras,the sparse vector of the disturbance signal is input,and the model is trained to realize the recognition of the disturbance.Finally,the realization process of phase space reconstruction and the selection method of delay time and embedding dimension are introduced.According to the characteristics of power quality disturbance signals,the values of delay time and embedding dimension are determined.The basic principles and evaluation parameters of the recursive graph algorithm based on phase space reconstruction are analyzed,and the effects of time and amplitude parameters in the disturbance mathematical model on the recursive trajectory graph are studied.The two influences each other to determine the characteristic area.Used to distinguish different types of disturbance signals.And compared the difference between traditional recurrent neural network and long and short-term memory neural network,analyzed the weight parameter learning algorithm of the network,and established the classification model based on long and short-term memory neural network in Tensorflow/keras for the recursive graph of power quality disturbance,Input the recursive graph corresponding to the disturbance signal into the model,and realize the recognition of different types of disturbance signals by training the classification model.Figure [24] Table [8] Reference [83]...