Research On Analysis Of Power Quality Disturbance Data Based On CS-CNN Theory

With the increasing of non-linear loads and electrically sensitive devices,higher power quality is required for many precision equipment.To improve power quality,it is necessary to analyze various signals in power systems that affect power quality.In general,power quality analysis includes power quality signal compression,classification,detection,and positioning.Based on power quality disturbance compression and classification & identification methods,a new method for power quality disturbance analysis based on compressed sensing and convolutional neural network is proposed in this paper.The power quality analysis methods based on the Nyquist sampling theorem often generate a large amount of data,which is not good for data storage transmission.Compressed Sensing(CS)theory compensates the limitations of the Nyquist sampling theorem.By compressing and sampling from the original signals,a small amount of measurement signals are obtained,and then the original signal is recovered and reconstructed,which greatly reduces the requirements for storage space.There is a prominent problem in the reconstruction of power quality data based on the principle of compressed sensing.When sparse representation is performed using orthogonal bases formed by ordinary functions,the best sparse representation cannot be obtained adaptively.To overcome this important issue,for the first time,we applied the K-singular value decomposition dictionary learning method into the field of power quality data reconstruction,and developed a new algorithm for power quality data reconstruction.In addition,several common types of power quality signals were simulated using this new algorithm.The results show that new algorithm can be used to reconstruct the signal when the compression ratio is 25%.The good reconstruction performance can meet the analysis requirements in practical application.The analysis of power quality disturbance data is complex.The traditional data classification methods usually require complex noise removal processing,and also require the feature extraction and pattern recognition.The traditional data classification methods including feature extraction and pattern recognition,which is very complicated.The effect of the previous stage directly affects the results of the later processing.The Convolutional Neural Network has strong self-learning ability and characterization ability for feature extraction.It can directly use the signal as the input object to identify the data.Therefore,using CNN to automatically learn the texture gradient feature of the signal can avoid the error caused by manual extraction of features.For the first time,this paper applies the convolutional neural network into the field of power quality data classification and identification.Through the construction of the CNN model,the power quality disturbance data classification is systematically studied.The experimental results show that CNN can be used for power quality disturbance data classification and recognition.The accuracy of pedestrian detection based on the new method training model can reach 97%.