Research On Power Quality Disturbance Classification Based On Deep Learning

With the increasing requirements of power quality for distribution system and production users,smart grid is constantly improving,which its performance are improved performance.In order to improve power quality and effectively control various disturbance problems,it is necessary to classify the disturbance types in power system quickly and effectively.Because the existing power quality disturbance classification and recognition algorithm analyzes massive data,the recognition accuracy is not high.Deep learning has the advantages of efficient processing and recognition.The application of deep learning to power quality disturbance classification has attracted the attention of many researchers.This thesis first introduces the mechanism and characteristic parameters of several disturbances in power system,and briefly describes the traditional classification algorithm of power quality disturbances.These algorithms have their own advantages and disadvantages.In this part,the method of combined S-transform andmulti-class support vector machine(SVM)is also studied to classify power quality disturbances.However,in the process of traditional recognition algorithms,due to the large amount of power quality data and many categories,the recognition rate is low,and the amount of calculation is large and the speed is slow.In order to solve this problem,the convolutional neural network(CNN)in deep learning is used for classification of power quality disturbances,which has the strong ability of representation learning and effectively avoids the recognition error caused by artificial experience.So,the signals of power quality disturbances are input into the CNN model,and classified by autonomous learning.Finally,the CNN and Sparse Autoencoder(SAE)are also combined to classify power quality disturbances.In the combined method,SAE is firstly used for power quality data by unsupervised learning.And the high-dimensional and sparse features are extracted while the data is compressed to reduce the dimension.Then while the sparse features are input into the CNN model for further processing,the classification of power quality disturbances is completed by deep learning algorithm.The experimental results show that the above two deep learning algorithms with autonomous classification and unsupervised learning have good accuracy and robustness for the classification of power quality disturbances,and the classification accuracy of the model of SAE combined with CNN is higher than that of CNN.