Research On The Online Monitoring Method Of Power Transformer

Power transformers are one of the most important and expensive components in power systems,whose reliable and efficient operation is crucial for electricity supply.Therefore,it is very important to study the early fault detection and health assessment techniques of power transformer.With the application of advanced sensor technologies and the introduction of various new models and algorithms,condition-based maintenance is becoming the mainstream trend in the field of transformer health management.This dissertation mainly focuses on parameter identification algorithms and online condition monitoring techniques of power transformers based on electrical information.The innovative work of this dissertation are as follow:(1)The ill-posed problems that may occur during the online parameter identification of power transformers analyzed.Based on the measure of condition number and multicollinearity,the performance of various online parameter identification algorithms is analyzed.For two winding transformers,if the equivalent winding impedance of primary and the equivalent winding impedance of secondary are identified respectively based on the T-equivalent circuit,the condition number is always large,and the accuracy of the identification result is low.In contrast,when the short-circuit impedance is identified as a whole based on the simplified equivalent circuit,the accuracy of the identification result is higher.(2)An on-line parameter identification algorithm of three-winding transformer based on adaptive load matching technology is proposed.In order to improve the accuracy of parameter identification,the relationship between different load distributions of the transformer and the condition number of the coefficient matrix is explored.Using the condition number as a criterion,the number of unknown parameters is adjusted according to different load states of the transformer.Then,numerical simulation is used to verify the influence of condition number on accuracy of parameter identification,while the engineering data is used to demonstrate the performance of online monitoring.(3)The error transfer process in parameter identification algorithm is analyzed with particular attention on the coupling influence of measurement error and load fluctuation.When the measurement error exists,the identification results of the short-circuit impedance will change with the load fluctuation,and its distribution follows a Gaussian random process located in the domain of load current.Since the parameter identification algorithm contains the nonlinear operation of independent random variables,the identification results under a fixed load does not strictly follow a Gaussian distribution.However,because the relative standard deviation of random error is very small,the nonlinear operation of independent variables can be approximately linearized,so this distribution is approximate to Gaussian distribution.(4)A data-driven online monitoring algorithm for power transformers is proposed,which is robust to measurement error and load fluctuation.Based on the historical data,BP neural network is used to fit the mean value function and standard deviation function of the Gaussian process.Then,the fitted functions are used to standardize the online monitoring data to achieve error compensation.Based on the ring law of random matrix theory,the mean spectral radius(MSR)is calculated to judge the change of short circuit impedance,so as to reflect the health status of transformer.Finally,the validity of the proposed method is verified by a numerical simulation.