Research On Power Quality Disturbance Analysis Methods Based On Deep Learning

With the rapid development of power electronics technology,the utilization ratio of frequency converter,rectifier and inverter is increasing gradually.These devices not only bring convenience to production and life,but also impact the power grid.In order to solve this problem,appropriate measures must be taken to improve the power quality.Firstly,it is necessary to effectively identify all kinds of power quality disturbance signal.In the face of more and more multiple disturbances in power system and the massive disturbance data brought by them,the traditional identification methods can no longer accurately identify all kinds of complex disturbance.Based on the above understanding,this thesis discusses and studies the power quality disturbance analysis methods from the perspective of deep learning.(1)The current research methods of power quality are summarized and the common power quality problems in the power grid are analyzed.And then it studies the types of deep learning methods.Finally,the research status of traditional power quality analysis methods and deep learning methods are expounded.(2)The definition of power quality and related standards at home and abroad are learned.Secondly,the cause of power quality disturbances and the classification of disturbances are analyzed,and the mathematical models of single power quality disturbance and multiple power quality disturbances are constructed-On this basis,the waveform of each disturbance signal is generated to provide effective data for subsequent recognition and classification.(3)Deep Belief Network(DBN)is used to analyze power quality disturbance.Firstly,a DBN model based on Restricted Boltzmann Machine(RBM)was established,and then the classifier was selected and the network training process was described.In the process of RBM training,an improved CD-k algorithm was proposed to address the defects of traditional training methods that could not give consideration to both accuracy and timeliness.Theoretical analysis proved that the method was feasible.Then,the optimal structure of DBN network is determined.Finally,simulation experiments prove that DBN network has a high identification efficiency in the field of power quality disturbance analysis.(4)Convolution Neural Network(CNN)is used to analyze power quality disturbance.In order to solve the problem that the effect of one-dimensional signal recognition by CNN is not ideal,a method of one-dimensional data to two-dimensional mapping is proposed in this paper.After that,the CNN structure and training process are discussed,and the CNN network structure is determined by combining the characteristics of multiple power quality disturbance types and large computation amount.Then the overall scheme of power quality analysis by CNN as disturbance identification network is proposed.After verifying that there is no over-fitting phenomenon in the network,the simulation experiment proves that CNN network has a high identification efficiency in the field of power quality disturbance analysis.(5)The characteristics and applications of the two deep learning networks in power quality analysis are compared and analyzed.Finally,the thesis summarizes the main research conclusions and the need for further research content.