Study Of The Short-circuit Fault Design And The Diagnostic Method Of The Converter Transformer Winding

As one of the core equipment of UHVDC transmission system, the reliability and stability of the converter transformer directly affect the normal operation of the whole DC transmission system. Although the converter transformer has passed the assessment of factory test, it may damage during the process of transportation and installation.In addition, the converter transformer of higher voltage class and larger capacity needs field assembly, it is necessary to assess its insulating performance strictly in field. The impulse withstand voltage test is one of the main methods for insulation assessment. The major problem is that the sensitivity of the existing fault diagnosis methods is limited and difficult to diagnose small faults and judge fault types. The on-site repair is quite difficult. The experimental data of the converter transformer under fault conditions is insufficient. It is difficult to conduct in-depth study on the fault diagnosis method. Furthermore, the experimental study on the fault using the original model of converter transformer is of long period and high cost, and it is not easy to obtain valid data. Therefore, it has a certain theoretical and engineering application value to design scale model on the basis of the converter transformer structure, establish rational simulation model to simulate different faults, and study the fault diagnosis method under impulse voltage thus to provide reference basis for the field tests of converter transformer.The crux to judge the validity of the transformer fault simulation is whether the simulation model can reflect real properties of the converter transformer winding. Currently, the simulation models of the winding under impulse voltage mainly contain lumped parameter model, distributed parameter model and their hybrid model. These models were mainly used to study the voltage distribution of the winding under lightning impulse, and the propagation process of high-frequency impulse signal in winding, etc. However, because of the difficulty of measurement, the simulated results of large transformer are lack of experimental verification and the veracity of the model is difficult to verify. Aiming at the above-mentioned problems, in this paper, the scale model of the converter transformer is designed and manufactured according to the similarity theory, and the faults of different windings are simulated to conduct the study of impulse withstand voltage test. The improved distributed parameter model is established to conduct the simulation study under different fault conditions. A fault diagnosis method is proposed based on the learning vector quantization network. The research results in this thesis can provide reference for the fault diagnosis of the UHV converter transformer under impulse withstand voltage test. The main content and the obtained results are as follows:(1) The multi-conductor transmission line (MTL) model considered vice-side winding and the improved distributed parameter model are established. The distribution parameters of the transformer winding are solved and the simplified method of distribution parameter matrix is deduced. The intershield winding is choosen to calculate the winding voltage distribution under the lightning impulse voltage. The effect of vice-side winding on the voltage distribution of primary winding is analysed. The calculation results of two models are compared and analysed.(2) The similarity formulas of the converter transformer are derived according to the similarity theory. Taking the simplified duplex winding transformer as example, the similarity model is designed. The electric field and magnetic field distribution of the original model and similarity model under corresponding motivation are studied by simulation. The winding voltage waveforms of the original model and similarity model under corresponding impulse voltage are calculated by MTL model. The results indicate that the error between the original model and the similarity model is within 5%, which verifies the validity of the similarity formulas. The similarity model with typical structure of UHV converter transformer winding is designed and made with the scale of about 1/5.(3) The outgoing lines are used to simulate the winding fault. The impulse voltage tests are carried out on the similarity model of the converter transformer and the impulse voltage distribution of the winding is obtained. The turn-to-turn short circuit and gap discharge, and the earthing short circuit, etc. are simulated on the similarity model. The first terminal voltage and neutral point current waveforms under typical faults are obtained and the waveform characteristics under different faults are compared and analysed. The results indicate that the first terminal voltage and neutral point current waveforms of small faults under lightning impulse change little, which is difficult to diagnose the fault.(4) The test data under simulation fault conditons are analysed. The results show that the correlation of the transfer function can reflect fault characteristics. The training samples and testing samples are choosen from the test data. The genetic algorithm is used to optimize the initial weight vector of the learning vector quantization network. Then the network is trained with training samples. Finally the testing samples are used for testing. The diagnostic accuracy is 98.2% when the one turn short circuit faults are removed. The results verify the effectiveness of the fault diagnosis method.(5) The simulation model is established according to structure of the reduced-scale transformer. The voltage waveform of the winding under impulse voltage is calculated by the improved distributed parameter model, which is similar to the experimental waveform and thus verify the validity of the simulation model. More comprehensive faults than the experiment are simulated by this model. The feature datas of simulation fault are calculated. These datas are used for training and testing of the optimized learning vector quantization network. The fault diagnosis results are exactly right. It shows that the proposed diagnosis method can effectively determine different fault types.