Virtual Array Extension And Signal Optimum Processing Of Acoustic And Electromagnetic Sensor Arrays For Power Equipment Condition Monitoring

Power equipment is the basic unit of the power system.The safe and stable operation of power equipment is the key to the safety and reliability of the power system.Power equipment condition on-line monitoring not only monitor all kinds of abnormal electrical and non-electrical quantities of the equipment,but also conduct a comprehensive and systematic analysis of the power equipment operating status in order to find potential faults.It has been extensively studied in recent years.The power equipment online condition monitoring method based on the acoustic and electrical sensor array has the advantages of wide coverage range,high monitoring efficiency and strong anti-interference performance.Pilot applications have been conducted in several substations all around the world.However,during the actual application process,there are some practical problems that need to be solved urgently,including: 1)the insufficient performance of the sensor array for multi-source signal recognition;2)the excessive requirement of sampling frequency for ultra-high frequency signal sampling;3)the interference to the sensor array caused by the noise in substation sites and 4)the errors of array signal time delay calculation and spatial localizations.Aiming at the application difficulties above,this paper analyzed and studied the principle of sensor array-based signal processing and proposed a virtual array extension and signal optimum processing method for power equipment monitoring,trying to give solutions to the problems above.Aiming at the problem of insufficient performance of sensor array multi-source signal recognition and separation,we found that the monitoring performance of the array not only depends on the algorithm performance,but also depends on the array hardware parameters(ie,the array aperture,the number of sensors,etc.).Improving the sensor array performance by hardware approaches always leads to the increase in cost.Thus,we proposed an array virtual extension method based on high order cumulants which can improve the sensor array performance without the need for hardware change or extra investment.The method replace the low-order cumulant in the array with the high-order cumulant established by Taylor expansion.Since the latter contains the high-order terms that are omitted in the low-order cumulant,it has more detailed directional information about the signal source which is utilized to improve the directional performance of the sensor array.In addition,since the high-order cumulant only contains the information of non-Gaussian components,the expanded array shows better anti-Gaussian noise performance.We proposed two type of extension for different substation application scenarios,including the aperture extension and the sensor number extension,which can respectively increase the multi-source signal directional resolution of the sensor array and reduce the sidelobe level of the array spectrum.Aiming at the problem of excessive sampling rate requrirement of high-frequency electromagnetic signals,this paper studies the principles of sensor array signal under-sampling and reconstruction methods.Aiming at the key problem of signal sampling mode,an under-sampling and re-construction method based on compressed sensing which can reconstruct the original signal with sampling rate lower than the Nyquist sampling law is proposed.The difference between this method and the traditional analog signal sampling method is that its sampling mode is changed from the original equal interval sampling to non-equal interval sampling of Gaussian random variables,converting the frequency aliasing of signal under-sampling into the energy leak in the entire frequency band.Afterwards,the signal reconstruction algorithm is utilized to obtain the original signal.Subsequently,based on signal under-sampling,a steering vector enhancement method is also proposed.Through signal sparse representation,observation matrix selection and signal reconstruction,the original 2-dimensional steering vector is upgraded to 3-dimensional and 4-dimensional steering vectors,which can improve the estimation accuracy of the sensor array.The experimental results further verify the effectiveness of the proposed method.The above research provides theoretical support for the subsequent localization of the power equipment signal source.Aiming at the problem of interference of the complex environmental noise in substation sites,this paper proposed a signal source direction of arrival(DOA)estimation method based on power maximum likelihood estimation which improve the anti-interference ability of the sensor array.Different from the traditional signal processing method that depends on signal waveform,this article uses the statistical method of maximum likelihood estimation to analyze the distribution characteristics of the signal source,so as to transform the traditional waveform processing into a new perspective of statistical analysis.The proposed method significantly improves the accuracy of signal source DOA estimation and shows better anti-noise ability especially under the condition of low signal-to-noise ratio(SNR in the range of [-5d B,5d B]).Field tests conducted in a 110 k V substation(with SNR around 5d B)show that the accuracy of signal source DOA estimation can be improved by 70% compared with traditional methods.Aiming at the problem of the localization error of the sensor array,this paper proposes a machine learning-based error correction method which consists of two multiple Radial Basic Function(RBF)networks RBFt and RBFp.Among them,RBFt is a time correction network for the correction of time delay calculation error,and RBFp is a spatial correction network,for the correction of the signal source’s spatial position error.Both networks adopt a multi-neural network structure to improve its error correction ability.Field tests have proved that the correction network RBFt and RBFp have strong capabilities to learn and simulate the error distribution.After error correction and compensation,the signal source distance positioning error of the sensor array is reduced from 3m to 0.5m,and the direction angle estimation error is reduced from 12° to less than 5°.The generality experiment also shows that the method has good adaptability to different time delay calculation algorithms and different types of sensor arrays.