Research On Key Techniques Of Partial Discharge On-line Monitoring Of Generator Stator Winding

The generator stator is the key component of the generator set and the component most prone to insulation failure.Insulation faults in the stator windings,if not detected and dealt in time,will lead to more serious grounding or short-circuit faults and even serious consequences that the generator cannot be repaired.The life of generator stator mainly depends on the insulation level of stator.However,partial discharge(PD)will lead to stator insulation damage.Therefore,it is still of great engineering significance to study the key technologies of on-line PD monitoring of generator stator.In this paper,the high frequency current sensor is selected as the sensor to obtain the PD signal,and the problem of eliminating interference and pattern recognition in the on-line PD monitoring is studied.The specific work of this paper is as follows:Aiming at the problem of suppressing narrow-band interference of PD signals in the generator stator,this paper presents a new denoising method based on Hankel matrix and singular value decomposition(SVD).Firstly,Hankel matrix is chosen as the trajectory matrix of SVD.Then the rule of SVD of narrow-band interference is studied.On the basis of this rule,the singular values corresponding to narrow-band interference are found by using K-means algorithm,and the narrow-band interference is reconstructed.Finally,by subtracting the narrow-band interference from original signal,the PD signal containing only white noise is obtained.The proposed method is applied to noisy PD signals,and its results are compared with those of improved FFT threshold method and wavelet denoising method.The results show that: this method has better suppression effect on narrow-band interference,it can effectively retain the details of the original signal,and the waveform similarity is higher.Aiming at the problem of noise suppression of PD signals in the generator stator,this paper presents a new de-noising method based on two-dimensional SVD.In this method,two-dimensional time-frequency matrix is obtained by generalized S-transform as the trajectory matrix of SVD.It overcomes the limitation of the traditional one-dimensional time-domain signal to construct the trajectory matrix,and can reflect the time-frequency information of the PD signal;Then,the singular values corresponding to the white noise is selected and set to 0 by the singular entropy incremental curvature spectrum threshold method to remove the white noise.The proposed method is applied to noisy PD signals,and its results are compared with those of adaptive SVD(ASVD)de-noising method and traditional wavelet de-noising method.The results show that: this method has better noise reduction effect,higher signal-to-noise ratio(SNR)and less waveform distortion.Aiming at the problem of pattern recognition of generator PD signals,this paper presents a new method based on GS-SVD and QPSO-SVM.The time-frequency matrix of PD signal is obtained by generalized S-transform,and the data of the matrix is compressed by SVD.A feature extraction method based on GS-SVD is proposed.Then the QPSO-SVM algorithm is used to classify the characteristics of PD signals;Compared with PSO-SVM algorithm and GA-SVM algorithm,the results show that: the proposed method has the advantages of strong learning ability,high diagnosis accuracy and good robustness.