Design Of Transformer On-line Monitoring And Fault Diagnosis System

With the development and popularization of technologies such as the Internet of Things,cloud computing,and intelligent algorithms,data collection technology has become mature enough to realize equipment health management and fault diagnosis prediction based on artificial intelligence technology and big data analysis,and then become capable of self-learning And the intelligent system of self-growth ability.As one of the important equipment of the power system,the power transformer’s operation reliability and safety directly affect the safety of the entire power system and the economic benefits of the system.The research on the working status of the transformer in this paper mainly analyzes from two aspects: online monitoring of transformer operating parameters and fault diagnosis based on DGA data,and designs and implements a transformer monitoring and fault diagnosis system.The first is the research on online monitoring of transformers.Through studying the application of online monitoring transformers with processors such as single-chip microcomputer and DSP as the core,it is found that there are some shortcomings in monitoring,such as insufficient parameter collection,complicated circuit design,insufficient flexibility,and unstable system operation.FPGA,as one of the current three mainstream processors,has super performance and flexibility in hardware and software,and is more reliable than single-chip,DSP and other applications in circuits.In this paper,an online monitoring system for distribution transformers is constructed based on FPGA’s flexible implementation of circuit operation mode functions.Through the determination of the system’s software and hardware circuit schemes,the welding of the circuit board and the construction of related circuit models are completed,and the online monitoring of transformers is realized.Simulation experiment.Mainly realize the collection of temperature,voltage,current,power factor,frequency and other parameters of power transformer windings to realize online monitoring of distribution transformers,and use FPGA to analyze the data to determine whether the distribution transformer is in normal operation.Alarm function.The actual circuit measurement results show that the system has the characteristics of high accuracy,good stability,economy and practicality,and high degree of visualization,which can meet the needs of many practical projects.The second is the research on transformer fault diagnosis.This paper analyzes the data of the gas in the transformer oil to identify the fault type of the transformer.In order to improve the accuracy of transformer fault diagnosis,a special concentration normalization method combined with cross-validation RBF neural network algorithm is proposed to diagnose five common fault types in transformers.First,establish the missing value and outlier detection processing model in the big data platform,then use the feature concentration normalization method to normalize the gas component samples,and divide the processed sample data into training set and test set randomly,and apply them separately To the transformer fault diagnosis model.In the transformer fault diagnosis model,in view of the limited sample data of transformer faults,the poor generalization ability of RBF neural network and the prone to overfitting,a K-fold cross-validation method was established to improve the generalization ability of the network and the accuracy of RBF network classification.Classification recall rate.Finally,a classification algorithm evaluation model is established,and the classification effect of the overall algorithm model is evaluated using indicators such as ROC,PR curve and K-S curve.The results of experimental analysis show that the average accuracy of transformer fault diagnosis and classification under this classification algorithm model can reach 90.84%.Compared with the traditional RBF neural network,random forest(RF)and gradient boosting decision tree classification(GBDT)algorithm classification,The characteristic concentration normalization method combined with cross-validation to improve the RBF neural network can improve the accuracy of transformer fault diagnosis,avoid falling into the local optimum,and effectively improve the fitting degree and stability of the network model.