Research On Disturbance Detection And Recognition Method For Power Quality

With the implementation of "carbon neutrality" policy,more and more new energy power generation is connected to the grid,and various nonlinear power electronic components are widely used.Therefore,the power quality disturbance of power system becomes more and more serious,which makes the power supply quality face severe challenge.In this case,rapid detection and recognition of power quality disturbance signals is of great significance to the safe and stable operation of power system and power quality management.Based on the advantages of deep learning in efficient processing of massive data and automatic feature extraction,this paper studies deep learning.The main work of this paper is as follows:(1)This paper introduces the research background and significance of power quality disturbance detection and recognition,and investigates the research status at home and abroad.At the same time,it summarizes the related problems of power quality,including the relevant definitions,standards and classification basis.Seven kinds of single disturbance and six kinds of composite disturbance are established by simulation,and the disturbance data set is established to provide basic samples for subsequent work.(2)Aiming at the problem of network degradation or even non-convergence caused by network deepening,a power quality disturbance recognition method based on improved residual network is proposed in this paper.It is suggested that the convolution layer is used to extract the features of the basic disturbance samples,and the power quality disturbance signals are classified by activation function layer,pooling layer and full connection layer.The residual network model is further studied by the residual block deepening network hierarchy and the comparison of simulation experiments.This paper explores the optimal configuration of residual network,including the number of residual blocks,learning rate,activation function,batch size,etc.The results show that compared with the traditional convolutional neural network(CNN),the improved residual network has better classification recognition rate and stability.(3)In order to solve the problem that single-scale convolution feature characterization is not strong,this paper proposes a power quality disturbance recognition method based on multi-scale fusion selective convolutional neural network(MSS-CNN).Firstly,convolution operations are carried out with convolution kernels of different scales to extract global and local features under different receptive fields and enrich sample features.Then,global average pooling is adopted to fuse the maximum weight eigenvalues obtained at different scales and output new feature maps.Finally,the disturbance signals are classified through the full connection layer.The experimental results show that MSS-CNN has high classification accuracy.Under the noise of 30 d B,the average recognition rate can still reach99%,which fully proves the excellent anti-noise performance of MSS-CNN.At the same time,the measured data are introduced to verify the validity of the model.(4)To solve the problem of poor classification effect caused by small sample size,a power quality disturbance data enhancement method based on deep convolutional generative adversarial network(DC-GAN)was proposed.DC-GAN introduces CNN into generative adversarial networks to improve the robustness of generative adversarial networks.And the generative model is responsible for learning the probability distribution of real disturbed samples through game training of generative model and adversarial model.The experiment uses the trained generation model to generate a new disturbed data set,and uses CNN as a discriminator and classifier to calculate the accuracy of the generated disturbed data set and the classification accuracy of the test set data.The results show that DC-GAN can effectively learn the probability distribution of disturbed samples and improve the classification accuracy of disturbed data.Problems such as network degradation,non-convergence and poor characterization of single-scale convolution are worth studying.From this point of view,this paper proposes a model of improved residual network and multi-scale fusion selection convolutional network,which has a high recognition rate.To solve the problem of too few disturbance data,a power quality disturbance data enhancement scheme for DC-GAN was proposed.This scheme can effectively expand the number of samples and solve the problem of insufficient real data of power system.