Research On Power Quality Disturbance Detection And Recognition Methods Based On Deep Learning

With the rapid development of the power grid,more and more power electronic devices and non-linear loads are connected to the power grid,and the level of power quality is greatly affected,and the level of power quality determines the industrial level and people’s living standards.In order to solve the power quality problem,the power quality disturbance signal should be effectively identified and classified,and corresponding countermeasures should be taken to avoid the power quality problem causing greater harm to the power grid.In the face of increasingly complex power quality disturbance signal types in power systems,traditional methods have been unable to effectively identify various types of disturbance signals.Based on the above problems,this thesis uses the deep learning method to deeply study and discuss the power quality disturbance problem:(1)Introduce the research background and significance of power quality problem,analyze the research status of disturbance recognition and detection methods at home and abroad,and introduce the application of deep learning method in the field of pattern recognition.(2)Firstly,the power quality is summarized,and the standards of power quality at home and abroad are introduced.Then,the types of common power quality disturbance signals are given,and a comprehensive mathematical model of the corresponding single disturbance and corresponding composite disturbance signals is constructed.Finally,using MATLAB to simulate and construct the integrated mathematical model,it provides sufficient samples for subsequent classification research.(3)The S-transform and Convolutional Neural Network(CNN)are used to classify the power quality disturbance signals.Firstly,the continuous S transform and the discrete S transform are introduced.The time-frequency mode matrix of the disturbance signal is extracted by S transform and the corresponding three-dimensional mesh map is generated.Finally,the time-frequency mode matrix of the disturbance signal is cropped according to the threedimensional mesh map.This is the input to the CNN.Then the principle.structure and training process of CNN are introduced,and the optimal parameters of CNN are determined according to the corresponding simulation experiments.Then the overall flow of the energy quality disturbance signal classification by the algorithm combining S transform and CNN is given.Finally,the advantages of this method in the field of power quality classification are verified by simulation experiments.(4)The method of combining Deep Belief Network(DBN)and Extreme Learning Machine(ELM)to classify and study power quality disturbance signals.Firstly,the DBN model with Restricted Boltzmann Machine(RBM)as the core is introduced.During the training of DBN,an improved(Quantum Particle Swarm Optimization,QPSO)algorithm is proposed to improve the DBN Training speed.Then the concept of ELM and the training process are introduced,and the parameters of ELM model are determined according to the corresponding simulation experiments.Finally,the optimal parameters of the DBN model were determined,and the effectiveness of the method was verified.(5)Compare the application scope and corresponding characteristics of two deep learning methods in the field of power quality classification.Finally,the thesis is summarized and the contents that need further research are introduced.