Research On Fault Diagnosis Method Of Fiber Optic Current Transformer Based On Deep Learning

Optical fiber current transformer(FOCT)provides primary side current information for relay protection,energy metering and power quality monitoring devices of power system in UHV and DC power transmission projects.Due to aging,temperature,vibration and other factors,optical fiber current transformers will fail,affecting the safe and reliable operation of the power system.It is urgent to carry out technical research on fault diagnosis and fault warning of optical fiber current transformer.A fault diagnosis method of optical fiber current transformer based on empirical mode decomposition and deep residual shrinkage network is proposed,and the fault identification accuracy is improved.By analyzing the basic working principle of fiber optic current transformer,the fault mode and output signal mathematical model are determined.By decomposition of fault signals and analysis of time-frequency domain and sample entropy characteristics,the fault data set of output signals is obtained and constructed after denoising.By using the fault data set,the fault diagnosis model is trained and the fault diagnosis function is realized of FOCT.The main research contents are as follows:(1)The fault mode and fault model of FOCT are analyzed.The working principle of FOCT is studied.Combined with the on-site faults,the fault modes and characteristics of internal devices are analyzed.The corresponding mathematical model of fault output signal is established,which provides theoretical bases for noise reduction and fault diagnosis of fault signal of FOCT.(2)Fault signal noise reduction algorithm of FOCT is studied.By time domain and frequency domain analysis,the output signal of FOCT is analyzed,and the fault signal characteristics are studied.The complete ensemble empirical mode dcomposition adaptive noise(CEEMDAN)is used to decompose the output signal,and the fault component is obtained.On the basis of the frequency domain characteristics of each component and the sample entropy(SE)values,the characteristics of each component are determined.The power frequency and low frequency components are determined as the characteristic vectors representing the fault modes.The obtained high-frequency noise components are eliminated.The power frequency and low frequency fault components are reconstructed and superimposed,and the fault signal data set is constructed after denoising.(3)The fault diagnosis model of FOCT is established by dedp networks.The basic principle of convolutional neural network(CNN)is studied.The two kinds of deep residual shrinkage networks(DRSNs)are compared and analyzed.Combined with the advantages of residual shrinkage units,an improved deep residual shrinkage network(IDRSN)is designed by stacking the two residual shrinkage units.The network structure and parameters are determined.A fault diagnosis model of FOCT based on IDRSN is constructed.(4)Simulation and verification.The fault data sets before and after denoising are respectively used to test the improved deep residual shrinkage network model.Accuracy rate,recall rate and F1 index were used to comprehensively evaluate the results.The comprehensive data index after noise reduction reaches 98%,about 2% higher than before noise reduction.The feasibility of the proposed noise reduction algorithm and fault diagnosis model is proved.The popular learning T-SNE algorithm is used to visualize the output of each layer of the model,and the effectiveness of the proposed model is verified.The t-SNE algorithm of manifold learning is used to visualize the output of each layer of the model,the effect of fault classification is observed Compared with other commonly used models,the comprehensive index is improved by 2%-6%,and the validity of the proposed model in the field of fault diagnosis of optical fiber current transformer is verified.