Research On Partial Discharge Pattern Recognition Method Of Large Hydrogenerator

Large generator is one of the most important equipment in the power system.The fault of large generators will result in huge economic losses and a negative impact on society,the safe and stable operation of large generators is important to the safety and stability of the power system.Statistics indicate that stator winding insulation fault is the most common fault of generators.After long-term research on the insulation condition monitoring techniques,partial discharge(PD)becomes the most widely accepted and used insulation condition monitoring indicator.During the operation of generators,the insulation of stator windings is gradually aging under the comprehensive action of multiple aging stresses,such as thermal,electrical,ambient and mechanical stresses.The operation condition of large hydro-generators which play an important role in the peaking and frequency regulation of the power grid is especially harsh,so the large hydro-generators face a more severe insulation aging problem.In addition,the stator winding structure of large hydro-generators is complex,which is specific in large size of the stator,a large number of winding bars,the strong noise,the attenuation of multiple discharge signals in transmission and the cross-coupling phenomenon make it more difficult to monitor,identify and locate the stator insulation PD.To sum up,the PD on-line monitoring of large hydro-generator is still a great worldwide challenge.PD signal is rich in insulation condition related information,including the type and number of insulation defects and the development trend of insulation health state.However,the PD signal processing and analysis still face many challenges at present,which are manifested in the following two aspects:(1)the measured PD signal contains a large amount of noise and interference,and the frequency component of the PD signal attenuates when it propagates from the discharge point to the detection point,leading to a low SNR of the measurement signal of PD signal,(2)there are usually multiple discharge sources and various types of discharges in large generators at the same time,the occurrence location,development speed and damage to the insulation of different types of discharges are different,so it is difficult to evaluate the insulation health of generators directly according to mixed signals.To meet the above challenge,this paper studies the denoising and pattern recognition method of PD signal of large hydro-generators.The research is centered on the signal denoising,disturbance suppression and multi-source PDs separation,to improve the PD signal analysis level of generators,provide strong support for generator insulation condition evaluation and maintenance decisions and ensure the safe and stable operation of generator and power system.The main research contents and innovative achievements of this paper are as follows:(1)To accurately extract the pulse waveforms from the generator measured PD signal with strong noise and plenty of pulses,a wavelet denoising method based on the rule to maximize the SNR of coefficients is proposed.With the idea of distinguishing noise coefficient and signal coefficient by wavelet threshold at each level,the calculation method of SNR of decomposition coefficient is deduced.The mother wavelet is optimized according to the principle of maximizing the SNR of decomposition coefficient at each level,which effectively removes the noise in each frequency band and improves the denoising performance of wavelet denoising.Besides,a cut-off rule of wavelet decomposition based on energy criterion is constructed,which greatly simplifies the determination method of decomposition level and improves the computational efficiency of the whole process of PD signal denoising.(2)A method of suppressing disturbance pulses based on the concept of one-class in PD signal is proposed.First,a method for PD feature selection by combining Gram-Schmidt orthogonalization(GSO)transform and maximal information coefficient(MIC)is studied,pulse features of high correlation and low redundancy are selected to fully represent the pulse type information,which lays a foundation for effective suppression of disturbance pulses.Then,a method of disturbance pulse suppression based on the feature grid space was developed,which realized the rapid and effective suppression of multi-type disturbance pulse according to the distribution of pulse in grid space.At last,a disturbance suppression method based on support vector data description(SVDD)is proposed,which improves the generalization ability of disturbance suppression methods.The results show that the proposed method can effectively suppress the multi-type disturbance pulses in the PD signals,greatly simplifies the PD pulse data set and improves the accuracy of subsequent multi-source PD separation results.(3)A multi-source PD signal separation method based on an improved DPC algorithm is proposed.In consideration of the characteristic of irregular shape and uneven density of PD data set,this paper introduces the density peaks clustering(DPC)algorithm to separate the multi-source PDs,the distribution of pulse data set is accurately described based on local density and distance.According to the demand to accurately determine the number of pulse sources,the original DPC algorithm of clustering center is improved,the distribution of each pulse in the pulse set was comprehensively described by the self-defined decision variable,and whether two pulses belong to the same cluster are determined based on the density reachable relationship.This method effectively improved the classification accuracy of the uneven pulse set.Application results show that the improved DPC algorithm can effectively improve the accuracy of multi-source PD signal separation results,and is appropriate for the PD pulse set with uneven density and distribution.(4)Based on the research results of PD pattern recognition method,this paper carries out the application and practice of generator PD monitoring.First,a method based on the equivalent phase center is studied to analyze the discharge phase of a single PD source,the problem of discharge phase location caused by three-phase signal overlapping and phase to phase cross-coupling phenomenon is effectively solved.Then,the PD measurement system is designed from three aspects: real-time processing of high-speed data stream,monitoring data management strategy and system software architecture.Finally,the effectiveness and engineering application value of the research in this paper is verified by several application cases of large hydro-generators in hydropower stations,reflecting the important role in the insulation health state evaluation,decision-making of generator maintenance and ensuring the safe and stable operation of generators.