The Faults Signal Analysis And Application In The HVDC System Based On Wavelet Packet And General Regression Neural Network

With the rapid development of multi-infeed HVDC transmission system in China,the transmission capacity is enlarged,but the transmission system becomes increasingly complicated.The interaction between AC and DC systems is more complex,any AC or DC faults can affect the stable operation of the whole system.The present situation has put forward higher requirements for the fault diagnosis system of HVDC transmission system,and it is necessary to study the fault signal analysis and fault diagnosis method of DC transmission system.Because the transient faults signal of DC transmission system has a large amount of faults information,it is very important to analyze and extract the transient characteristic of fault signal and apply it to fault diagnosis.In this thesis,the WPD is used to analyze the fault signal,combine with the GRNN to form a fault diagnosis method,which is applied to the fault location of DC line grounding fault and the identification of commutation failures,and obtains good results.The main research contents focus on the following aspects:1)The difference between the transient signals of the two kinds of faults is analyzed theoretically,and distinguish method between the two faults are proposed according to the difference of electrical quantity change caused by different faults.The simulation model of HVDC transmission system was established in PSCAD for analyzing the transient change of electrical quantities and verifying the validity of the distinguish method.2)In order to locate line faults accurately,a new method of DC line faults location based on WPD and GRNN is proposed.Due to the large difference of transient energy of travelling wave spectrum in different fault distances,the fault transient energy is extracted by WPD,and the GRNN is used for fitting the nonlinear relation of energy and distance realizes the fault location.The fault location model based on BPNN is established with the same samples,and the results of the two models are compared.3)In order to identify commutation failures and find reasons which cause commutation failures accurately,a new method of commutation failures identification based on WPD and GRNN is proposed.Using the WPD quantify different faults feature,two output structures of GRNN model are used to identify the different faults which cause commutation failures,comparing the accuracy of different output structures.Meanwhile,the results of GRNN and BPNN identification models under different training samples are compared,and the adaptability of the two models to different training samples is verified.