Stufy On Sparse Power Quality Disturbance Recognition On The Basis Of Deep Learning

With the burgeon of smart grids and alternative energy grid-connected technologies,a large number of power electronic devices have been connected to power systems in recent years,which results the increment of nonlinear,shock,and volatility loads,distorts the current and voltages waveform,deteriorates the power quality problem.In addition,the modern industrial and residential electrical equipment controlled by computers,microprocessors and power electronics have been extensively applied.these precision control devices are susceptible to power quality disturbance signals.The power quality problem has become increasingly severe.Therefore,it is necessary to quickly and accurately detect disturbances and identify disturbances type,which is of great significance for the treatment of power quality pollution.Based on the previous studies,this paper carried out the study on power quality disturbance recognition based on compressed sensing and deep learning.In order to represent original signal by less data.the sparsity of power quality disturbance was firstly analyzed,then a power quality sparse model combined stationary wavelet transform and compressive sensing was constructed.This model can effectively reduce dimensionality and has a certain denoisy performance.Based on this sparse model,this paper carried out study on power quality disturbances recognition and designed three methods.1)A power quality recognition based on CS-SWT algorithm is designed.In this method,the sparse vector quadratic feature factors was obtained by the sparse model.And seven indicators are selected as feature vectors and inputted into the constructed BP neural network for identification.The simulation results show that for several typical single and double disturbances,this method can accurately identify some non-stationary signals such as pulses and harmonics,with a good recognition accuracy and noise immunity.2)A sparse power quality recognition based on advanced deep belief network(DBN)is designed.Considered the advantages of DBN in handling high-dimensional data,DBN was firstly constructed,expecting to obtain disturbance information by utilizing the feature self-learning ability of RBM.Secondly,the disturbance sparse matrix obtained form the sparse model was directly inputted into the deep belief network to classify the disturbance.In addition,in order to improve the accuracy of disturbance model,cross entropy algorithm was applied to seek the best parameter of DBN.Compared with the first method,The simulation results demonstrate that this method can achieve the equivalent recognition effect in the recognition of single disturbance.For multiple disturbances,the results show that this method can effectively avoid the poor feature extraction problem.Furthermore,the identification time is shortened in a way.3)A sparse power quality recognition based on staked sparse de-noise auto-encoder(SSDAE)is designed.Inspired by the recognition idea of method 2,this method appliedstacked auto-encoder as the deep learning model.Adding sparse constraints and noise reduction in Auto-Encoder,the stacked sparse denoise auto-encoder(SSDAE)was firstly built.Then the sparse disturbance vector was directly inputted into SSDAE to achieve intelligent disturbance recognition.With the comparison of the method 1 and method 2,the simulation results show that this proposed method has the most shorted time among this three methods.Compared with method 2,the recognition accuracy and noise immunity are increased correspondingly.At the same time,this method was tested by applying three kinds of measured disturbance data,and its effectiveness was verified.On this basis,a GUI recognition interface for power quality disturbance recognition was built,which can intuitively display train and test results of the disturbance recognition model.According to the above study results,it provides some ideas for disturbance recognition combined compressed sensing and deep learning.