A switch cabinet ultrasonic partial discharge anti-interference monitoring method and system

Abstract

The application relates to a switch cabinet ultrasonic partial discharge anti-interference monitoring method and system, in particular to the technical field of power equipment state monitoring, which realizes synchronous collection through three ultrasonic sensors to form a rigid array with an included angle of 120 degrees, separates independent source signal components through centralization and whitening pretreatment and independent component analysis, inversely inverts the incident azimuth angle of each component by using the inverse matrix of the demixing matrix, and combines the effective area angle range in the cabinet and the background noise threshold to jointly judge, so that spatial discrimination of the cabinet partial discharge signal and the cabinet interference signal is realized, the high false alarm problem caused by the fact that a single probe cannot distinguish the sound source direction is fundamentally solved, and the accuracy and reliability of the switch cabinet ultrasonic partial discharge monitoring are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, and more specifically, to a method and system for ultrasonic partial discharge anti-interference monitoring of switchgear. Background Technology

[0002] In substations or distribution rooms with densely packed indoor switchgear ranging from 10 kV to 35 kV, partial discharge is one of the main causes of insulation degradation and even accidents in switchgear. Ultrasonic detection is widely used for partial discharge inspection and online monitoring of switchgear due to its advantages such as non-contact operation, high sensitivity, and live detection capability. However, this scenario presents a complex interference environment: the metal casing of the switchgear partially shields and reflects ultrasonic waves through multipath reflection; partial discharge sources such as insulators and contact boxes inside the cabinet coexist with those in adjacent cabinets outside; and the sensor mounting position is limited by the cabinet structure. External factors include corona discharge from other equipment in the substation, broadband mechanical vibrations from the operating mechanisms of fans and circuit breakers, human-caused knocking during inspections, and 50 Hz power frequency and its harmonics entering the acquisition link through power lines and signal lines. Under these conditions, single-probe ultrasonic sensors can only output signal amplitude and envelope, and cannot distinguish whether the sound source comes from inside or outside the cabinet, thus triggering a high false alarm dilemma of "alarm upon signal," which seriously affects the practicality and reliability of the monitoring system.

[0003] Existing single-probe ultrasonic partial discharge detection devices lack spatial resolution: their sensitive surface normal direction is fixed, and a single channel cannot acquire arrival time difference information, thus failing to invert the incident direction of the sound source; simultaneously, the traditional fixed threshold combined with bandpass filtering processing link lacks effective discrimination criteria for internal and external signals with similar frequencies and amplitudes; the system design does not construct a multi-channel spatial sampling array, the hardware only provides single-channel analog-to-digital conversion, and the algorithm layer lacks a source separation module based on direction-of-arrival estimation, representing a combined deficiency of insufficient sensing dimension and a single backend decision criterion; current ultrasonic partial discharge detection devices on the market... Most existing inspection instruments and online monitoring terminals use a single sensor combined with a fixed threshold and simple amplitude comparison. Some improved solutions introduce multiple sensors but only perform amplitude weighted fusion without utilizing array geometry for blind source separation. Therefore, there is an urgent need for a technical solution that, without increasing sensor directivity, uses a multi-probe spatial array and blind source separation algorithm to recover the independent sound source components from mixed observations and invert their incident azimuth angles, thereby achieving spatial discrimination between the effective area inside the cabinet and the interference area outside the cabinet. This would fundamentally solve the problem of internal and external signal confusion caused by the blindness of a single probe. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a method and system for ultrasonic partial discharge anti-interference monitoring of switchgear, thereby solving the problems mentioned in the background art.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: a method for ultrasonic partial discharge anti-interference monitoring of switchgear, specifically including the following steps:

[0006] Step S1: At the gap of the cabinet door of the switch cabinet to be tested, three ultrasonic sensors are fixedly installed in a rigid array with their sensitive axes coplanar and the included angle between each pair being 120 degrees. A unified clock source drives a multi-channel synchronous analog-to-digital converter to synchronously acquire the three signals output by the three ultrasonic sensors to obtain three observation signals.

[0007] Step S2: After centering and whitening preprocessing of the three observation signals, the independent component analysis algorithm is used to solve the unmixing matrix, so that the output components are statistically independent, thereby separating multiple independent source signal components.

[0008] Step S3: Use the inverse matrix of the demixing matrix as the mixing matrix estimate, take each column of the mixing matrix estimate as the mixing weight vector of the corresponding independent source signal component, and construct a set of candidate direction unit vectors covering 0 degrees to 359 degrees. By calculating the dot product between each mixing weight vector and each candidate direction unit vector, take the angle corresponding to the maximum value of the dot product as the incident azimuth angle of the independent source signal component.

[0009] Step S4: Compare the incident azimuth angle with the preset effective area angle range inside the cabinet. If the incident azimuth angle falls within the effective area angle range inside the cabinet and the time domain energy of the corresponding independent source signal component exceeds the preset background noise threshold, it is determined to be a partial discharge signal inside the cabinet and an alarm is output. Otherwise, it is determined to be an interference signal and suppressed.

[0010] In a preferred embodiment, the specific method for obtaining the three observation signals in step S1 is as follows:

[0011] During the offline calibration phase, three ultrasonic sensors are excited by a standard ultrasonic pulse of known direction, and three initial signals are recorded.

[0012] The generalized cross-correlation phase transform algorithm is used to estimate the relative time delay between the three ultrasonic sensors. The generalized cross-correlation phase transform algorithm obtains the peak position of the cross-correlation function as the time delay estimate by calculating the phase transform weighted integral of the cross power spectral density function of the two signals.

[0013] Then, taking one of the ultrasonic sensors as a reference, fractional delay compensation based on sinusoidal interpolation is performed on the three initial signals of the other two ultrasonic sensors to eliminate the fixed time delay difference caused by hardware clock skew and filter group delay differences between the ultrasonic sensors, so that the three signals are strictly aligned in the time domain, thereby obtaining three observation signals.

[0014] In a preferred embodiment, in step S1, in the rigid array composed of three ultrasonic sensors, the normal directions of the sensitive axes of the three ultrasonic sensors point to zero degrees, positive 120 degrees and negative 120 degrees respectively, and the three ultrasonic sensors are fixed on the same rigid base, which is attached to the central area of ​​the gap in the door of the switch cabinet to be tested.

[0015] A unified clock source drives a multi-channel synchronous analog-to-digital converter to synchronously digitize the analog signals output by the three ultrasonic sensors at a sampling rate of no less than one megahertz, ensuring that the sampling start times of the three signals are strictly aligned, thereby providing high-fidelity three observation signals.

[0016] In a preferred embodiment, the specific operations of the centering and whitening pretreatment in step S2 are as follows:

[0017] First, subtract the arithmetic mean of the three observation signals from each of the three observation signals to obtain the zero-mean observation signal;

[0018] Then, the autocorrelation matrix of the zero-mean observed signal is calculated, and the autocorrelation matrix is ​​decomposed into eigenvalues ​​to obtain the orthogonal eigenvector matrix and the eigenvalue diagonal matrix.

[0019] Next, the whitening matrix is ​​constructed by multiplying the negative half power of the eigenvalue diagonal matrix by the transpose of the orthogonal eigenvector matrix.

[0020] Finally, the whitening matrix is ​​multiplied by the zero-mean observation signal to obtain the whitened signal. The variance of each component of the whitened signal is one, thus eliminating the second-order correlation between the three observation signals.

[0021] In a preferred embodiment, in step S2, the independent component analysis algorithm employs a joint approximate diagonalization eigenma matrix algorithm based on fourth-order cumulants joint approximate diagonalization, specifically as follows:

[0022] First, calculate the fourth-order joint cumulants for each component of the whitened signal and construct a family of fourth-order cumulant matrices;

[0023] Then, through iterative optimization, an orthogonal rotation matrix is ​​found such that the sum of the off-diagonal elements of the matrix obtained by multiplying the transpose of the orthogonal rotation matrix by each fourth-order cumulant matrix and then by the orthogonal rotation matrix is ​​minimized, thus achieving joint approximate diagonalization;

[0024] Finally, the transpose of the orthogonal rotation matrix is ​​multiplied by the whitening matrix to obtain the demixing matrix, and the transpose of the orthogonal rotation matrix is ​​multiplied by the whitening signal to obtain the independent source signal components.

[0025] In a preferred embodiment, step S3, specifically calculating the dot product between each mixed weight vector and each candidate direction unit vector, is as follows:

[0026] First, each column of the estimated mixing matrix is ​​used as the mixing weight vector of the corresponding independent source signal component. The mixing weight vector contains three elements, which correspond to the three ultrasonic sensors respectively, and the three elements correspond one-to-one with the normal direction of the sensitive axis of the three ultrasonic sensors.

[0027] Then, with a step size of one degree, a set of candidate direction unit vectors from 0 degrees to 359 degrees is constructed. Each candidate direction unit vector consists of the cosine and sine values ​​of the angle.

[0028] Next, for each hybrid weight vector, based on the fact that the normal directions of the three ultrasonic sensors' sensitive axes point to 0 degrees, positive 120 degrees, and negative 120 degrees respectively, three unit vectors for the normal directions of the sensitive axes are defined: the first unit vector for the normal direction of the sensitive axes is the unit vector corresponding to the direction of 0 degrees, the second unit vector for the normal direction of the sensitive axes is the unit vector corresponding to the direction of positive 120 degrees, and the third unit vector for the normal direction of the sensitive axes is the unit vector corresponding to the direction of negative 120 degrees.

[0029] Multiply the first element of the mixed weight vector by the unit vector of the first sensitive axis normal direction, add the second element of the mixed weight vector by the unit vector of the second sensitive axis normal direction, and add the third element of the mixed weight vector by the unit vector of the third sensitive axis normal direction. The sum of the three products constitutes a two-dimensional projection composite vector.

[0030] Then, the dot product of the two-dimensional projection composite vector and each candidate direction unit vector is calculated. The dot product is calculated by multiplying the first element of the two-dimensional projection composite vector by the first element of the candidate direction unit vector, and adding the second element of the two-dimensional projection composite vector multiplied by the second element of the candidate direction unit vector.

[0031] Finally, record all the dot product values ​​corresponding to each mixed weight vector, and find the maximum value. The candidate direction angle corresponding to the maximum value is the incident azimuth angle of the independent source signal component.

[0032] In a preferred embodiment, in step S4, the preset background noise threshold is dynamically updated using an exponentially weighted moving average recursive estimation method, specifically as follows:

[0033] At the end of each preset analysis window, if no partial discharge signal is determined to be inside the cabinet within the analysis window, the time-domain energy of the independent source signal component within the analysis window is used as a background noise sample to update the background noise threshold. The new background noise threshold is equal to the preset forgetting factor multiplied by the current background noise threshold, plus one minus the preset forgetting factor multiplied by the time-domain energy of the current analysis window.

[0034] If the analysis window determines that the signal is a partial discharge signal inside the cabinet, no update will be made, and the current background noise threshold will remain unchanged.

[0035] In a preferred embodiment, in step S4, before comparing the incident azimuth angle with the preset effective area angle range inside the cabinet, the azimuth confidence factor is also calculated. Specifically, the operation is as follows:

[0036] Based on the ratio of the maximum to the second largest dot product in step S3, the azimuth confidence factor is calculated. The azimuth confidence factor is equal to the ratio of the maximum to the second largest dot product minus the preset peak confidence threshold, and then divided by the preset theoretical maximum ratio minus the preset peak confidence threshold. The result is limited to between zero and one.

[0037] When the ratio of the maximum value to the second largest value of the dot product is less than or equal to the preset peak confidence threshold, the azimuth confidence factor is zero, indicating that the azimuth is not reliable, and the independent source signal component is directly determined as an interference signal and suppressed.

[0038] When the ratio of the maximum to the second largest dot product is greater than or equal to the preset theoretical maximum ratio, the azimuth confidence factor is set to one, indicating that the azimuth is highly reliable.

[0039] In a preferred embodiment, step S4, which determines the signal as a partial discharge signal inside the cabinet and outputs an alarm, is implemented through a joint decision function. Specifically, the operation is as follows:

[0040] First, calculate the indicator function value of the angle range of the incident azimuth angle entering the preset effective area inside the cabinet. When the incident azimuth angle enters the preset effective area inside the cabinet angle range, take one value; otherwise, take zero value.

[0041] Then, the energy margin ratio is calculated, which is equal to the time-domain energy of the independent source signal component divided by the background noise threshold;

[0042] Next, the preset weighting coefficient is multiplied by the energy margin ratio, and then the preset weighting coefficient multiplied by the azimuth confidence factor is added, and then multiplied by the indicator function value of the incident azimuth angle entering the preset effective area angle range inside the cabinet to obtain the joint decision function value.

[0043] Finally, the joint decision function value is compared with the preset alarm threshold and the preset suspected threshold. If the joint decision function value is greater than or equal to the preset alarm threshold, it is determined to be a partial discharge signal in the cabinet and an alarm is output. If the joint decision function value is greater than or equal to the preset suspected threshold and less than the preset alarm threshold, it is marked as a suspected signal. If the joint decision function value is less than the preset suspected threshold, it is determined to be an interference signal and is suppressed.

[0044] This application also provides an ultrasonic partial discharge anti-interference monitoring system for switchgear, specifically including:

[0045] Acquisition module: Three ultrasonic sensors are mounted in a rigid array at a 120-degree angle in the gap of the cabinet door. The three signals are synchronously acquired using a unified clock source to obtain three observation signals.

[0046] Separation module: After centering and whitening the three observed signals, independent component analysis is used to solve the unmixing matrix and separate the statistically independent source signal components.

[0047] Azimuth module: The inverse of the unmixing matrix is ​​used as the mixing matrix estimate, and each column of the mixing matrix estimate is used as the mixing weight vector. Candidate direction unit vectors from 0 degrees to 359 degrees are constructed. The angle corresponding to the maximum value of the dot product is calculated as the incident azimuth angle of each independent source signal component.

[0048] Decision module: Compare the incident azimuth angle with the effective area inside the cabinet. If it falls within the effective area and the corresponding time domain energy exceeds the background noise threshold, it is determined to be a partial discharge signal and an alarm is triggered. Otherwise, it is determined to be interference suppression.

[0049] The beneficial effects of this invention are as follows: By using three ultrasonic sensors arranged in a rigid array at a 120-degree angle for synchronous acquisition, and after centering and whitening preprocessing, independent source signal components are separated by independent component analysis. Then, the incident azimuth angle of each component is obtained by inverting the inverse matrix of the unmixing matrix. Combined with the effective area angle range inside the cabinet and the background noise threshold, a joint decision is made, which realizes the spatial discrimination between partial discharge signals inside the cabinet and interference signals outside the cabinet. This fundamentally solves the problem of high false alarms caused by the inability of a single probe to distinguish the direction of the sound source, and significantly improves the accuracy and reliability of ultrasonic partial discharge monitoring of switchgear. Attached Figure Description

[0050] Figure 1 This is a flowchart of the method of the present invention;

[0051] Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0054] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0055] Example 1: This example provides the following... Figure 1 The present invention discloses an ultrasonic partial discharge anti-interference monitoring method for switchgear, which specifically includes the following steps:

[0056] Step S1: At the gap of the cabinet door of the switch cabinet to be tested, three ultrasonic sensors are fixedly installed in a rigid array with their sensitive axes coplanar and the included angle between each pair being 120 degrees. A unified clock source drives a multi-channel synchronous analog-to-digital converter to synchronously acquire the three signals output by the three ultrasonic sensors to obtain three observation signals.

[0057] Step S2: After centering and whitening preprocessing of the three observation signals, the independent component analysis algorithm is used to solve the unmixing matrix, so that the output components are statistically independent, thereby separating multiple independent source signal components.

[0058] Step S3: Use the inverse matrix of the demixing matrix as the mixing matrix estimate, take each column of the mixing matrix estimate as the mixing weight vector of the corresponding independent source signal component, and construct a set of candidate direction unit vectors covering 0 degrees to 359 degrees. By calculating the dot product between each mixing weight vector and each candidate direction unit vector, take the angle corresponding to the maximum value of the dot product as the incident azimuth angle of the independent source signal component.

[0059] Step S4: Compare the incident azimuth angle with the preset effective area angle range inside the cabinet. If the incident azimuth angle falls within the effective area angle range inside the cabinet and the time domain energy of the corresponding independent source signal component exceeds the preset background noise threshold, it is determined to be a partial discharge signal inside the cabinet and an alarm is output. Otherwise, it is determined to be an interference signal and suppressed.

[0060] In this embodiment, it is specifically necessary to explain how the three observation signals are obtained in step S1:

[0061] During the offline calibration phase, three ultrasonic sensors were excited using a standard ultrasonic pulse with a known direction, and three initial signals were recorded. The offline calibration phase was conducted when the switchgear was not in operation or during a power outage for maintenance. The standard ultrasonic pulse was generated by a calibrated ultrasonic transmitter with a center frequency of 40 kHz, a pulse width of 50 microseconds, and a repetition frequency of 1 Hz. The transmitter was placed 1 meter away from the center point of the rigid array, aligned with the array's normal direction. The three ultrasonic sensors simultaneously received the pulse, which was digitized by a multi-channel synchronous analog-to-digital converter at a sampling rate of 2 MHz. Three initial signals with a length of 1024 sampling points were recorded and denoted as the first initial signal, the second initial signal, and the third initial signal. To ensure calibration accuracy, the samples were collected 100 times and the average value was taken to suppress random noise.

[0062] The generalized cross-correlation phase transform algorithm is used to estimate the relative time delay between three ultrasonic sensors. This algorithm calculates the phase-transform weighted integral of the cross-power spectral density function of the two signals to obtain the peak position of the cross-correlation function as the time delay estimate. The specific calculation steps of the generalized cross-correlation phase transform algorithm are as follows: First, perform Fast Fourier Transform on the first and second initial signals respectively to obtain the first and second frequency domain signals. The number of transform points is set to 2048, and a Hanning window is used to suppress spectral leakage. Second, calculate the product of the conjugates of the first and second frequency domain signals to obtain the cross-power spectral density function. Third, divide the cross-power spectral density function by its amplitude spectrum to achieve phase-transform weighting, obtaining... The first step is to normalize the cross-power spectrum; the second step is to perform an inverse fast Fourier transform on the normalized cross-power spectrum to obtain the generalized cross-correlation function; the third step is to search for the peak position in the generalized cross-correlation function, and the time delay index corresponding to the peak is the relative time delay estimate between the first initial signal and the second initial signal, in units of sampling points; the fourth step is to divide the sampling point delay by the sampling rate of 2 MHz to convert it into a relative time delay estimate in seconds; the seventh step is to repeat the above steps to calculate the relative time delay estimate between the first initial signal and the third initial signal; to ensure the reliability of the time delay estimate, a peak confidence threshold is set. When the ratio of the peak value to the second peak value of the generalized cross-correlation function is greater than 3, the time delay estimate is considered valid; otherwise, the calibration is abandoned and the data is collected again.

[0063] Then, using one of the ultrasonic sensors as a reference, fractional delay compensation based on sinusoidal interpolation is performed on the three initial signals of the other two ultrasonic sensors to eliminate the fixed time delay difference caused by hardware clock skew and filter group delay differences between the ultrasonic sensors, so that the three signals are strictly aligned in the time domain, thereby obtaining three observation signals. The specific implementation process of fractional delay compensation based on sinusoidal interpolation is as follows: First, the first initial signal corresponding to the first ultrasonic sensor is used as the reference signal and is not processed; second, for the second initial signal corresponding to the second ultrasonic sensor, the fractional delay amount to be compensated is calculated based on its time delay estimate relative to the reference signal; this fractional delay amount is equal to the fractional part of the time delay estimate multiplied by the sampling period.

[0064] Then, the second initial signal is resampled using a sinusoidal interpolation kernel function. The interpolation kernel function uses a Singer function with a raised cosine window truncation and a window length of 8 sampling points. Specifically, for each output sampling point, 8 points near the corresponding integer index in the input signal are selected, multiplied by the coefficients of the corresponding positions in the interpolation kernel function, and accumulated to obtain the value of that output sampling point. This process is repeated to obtain the second compensated signal after time delay compensation. Similarly, the same fractional delay compensation is performed on the third initial signal to obtain the third compensated signal. Finally, the first initial signal, the second compensated signal, and the third compensated signal are used together as three observation signals. After compensation, the residual time delay difference between the three signals is less than 0.1 sampling periods, i.e., less than 50 nanoseconds, which meets the requirements of the linear instantaneous mixture model.

[0065] In step S1, the rigid array of three ultrasonic sensors has their sensitive axis normals pointing to 0 degrees, +120 degrees, and -120 degrees, respectively. The three ultrasonic sensors are fixed to the same rigid base, which is attached to the center area of ​​the gap in the switch cabinet door. The rigid base is made of aluminum alloy and is an equilateral triangular plate with sides of 80 mm and a thickness of 10 mm. The three ultrasonic sensors are installed at the three vertices of the triangle, with the sensor base parallel to the base plane and the sensitive axis normal perpendicular to the base plane outwards. Thermal grease is applied to the back of the base, which is then attached to the center area of ​​the switch cabinet door gap and fixed magnetically to ensure a tight fit between the base and the door, reducing acoustic impedance mismatch. This 120-degree uniform layout ensures consistent azimuth resolution across the entire array, and the column vectors of the mixing matrix exhibit maximum linear independence, which is beneficial for the convergence of the subsequent independent component analysis algorithm.

[0066] A unified clock source drives a multi-channel synchronous analog-to-digital converter (ADC) to synchronously digitize the analog signals output from three ultrasonic sensors at a sampling rate of at least 1 MHz, ensuring strict alignment of the sampling start times of the three signals and thus providing high-fidelity three-channel observation signals. The unified clock source uses a temperature-compensated crystal oscillator with a frequency stability of one part per million, and an output clock frequency of 2 MHz, which is directly used as the sampling clock for the multi-channel ADC. The multi-channel ADC uses a four-channel synchronous sampling chip, with three channels connected to the analog outputs of the three ultrasonic sensors respectively, and the fourth channel... One channel is reserved for environmental reference signals; the analog-to-digital converter adopts differential input mode with an effective number of bits not less than 12 and a full-scale input range of ±5V; on each sampling clock rising edge, the three channels start conversion simultaneously, and the conversion results are read out synchronously through the serial peripheral interface; this configuration ensures that the sampling time deviation of the three channels is less than 10 picoseconds, which is much smaller than the ultrasonic signal period, providing three observation signals with high fidelity and high synchronization accuracy for subsequent blind source separation; the sampling rate is selected as 2 MHz, which can cover the ultrasonic frequency band from 20 kHz to 200 kHz, and leaves a margin to prevent spectral aliasing.

[0067] In this embodiment, it is specifically necessary to explain the specific operations of centering and whitening preprocessing in step S2:

[0068] First, subtract the arithmetic mean of each of the three observation signals from the arithmetic mean of each of the three observation signals to obtain the zero-mean observation signal. The calculation process for the arithmetic mean of each of the three observation signals is as follows: sum all the sampling points of the first observation signal, and then divide by the total number of sampling points to obtain the arithmetic mean of the first observation signal; similarly, calculate the arithmetic mean of the second and third observation signals respectively; then subtract the arithmetic mean of the first observation signal from each sampling point of the first observation signal to obtain the first zero-mean signal; process the second and third observation signals in the same way to finally obtain the three zero-mean observation signals; the purpose of this operation is to eliminate the DC component in the signal and avoid the DC offset from interfering with the subsequent calculation of higher-order statistics; the total number of sampling points is 1024, which is consistent with the length recorded in step S1;

[0069] Then, the autocorrelation matrix of the zero-mean observed signal is calculated, and eigenvalue decomposition is performed on the autocorrelation matrix to obtain the orthogonal eigenvector matrix and the eigenvalue diagonal matrix. The calculation process of the autocorrelation matrix is ​​as follows: the three zero-mean observed signals are arranged into a 3x1024 matrix, with each row corresponding to one zero-mean observed signal; this matrix is ​​multiplied by its own transpose and then divided by the total number of sampling points, 1024, to obtain a 3x3 symmetric matrix, which is the autocorrelation matrix; the diagonal elements of the autocorrelation matrix represent the power of each zero-mean observed signal, and the off-diagonal elements represent the degree of cross-correlation between different signals.

[0070] The specific steps for eigenvalue decomposition of the autocorrelation matrix are as follows: First, solve for the characteristic polynomial of the autocorrelation matrix to obtain three eigenvalues, which are then sorted in descending order. Next, solve for the corresponding eigenvector for each eigenvalue and arrange the three eigenvectors column-wise to form an orthogonal eigenvector matrix. Finally, arrange the three eigenvalues ​​sequentially on the diagonal to obtain the eigenvalue diagonal matrix. The eigenvalue decomposition is implemented using the Jacobi iterative algorithm, with an iteration convergence threshold set to 10 to the power of -10. That is, iteration stops when the change in eigenvalue between two consecutive iterations is less than this threshold.

[0071] Next, the whitening matrix is ​​constructed by multiplying the negative half power of the eigenvalue diagonal matrix by the transpose of the orthogonal eigenvector matrix. The calculation process for the negative half power of the eigenvalue diagonal matrix is ​​as follows: for each eigenvalue in the eigenvalue diagonal matrix, take the square root of its reciprocal to obtain a new diagonal matrix, which is the negative half power of the eigenvalue diagonal matrix; for example, if the first eigenvalue is 4, then its negative half power is 0.5; then this matrix is ​​multiplied by the transpose of the orthogonal eigenvector matrix to obtain the whitening matrix; the whitening matrix is ​​a 3x3 matrix, and its function is to transform the zero-mean observation signal into a whitened signal in which the variance of each component is 1 and they are mutually uncorrelated.

[0072] Finally, the whitening matrix is ​​multiplied by the zero-mean observation signal to obtain the whitened signal. The variance of each component of the whitened signal is 1, eliminating the second-order correlation between the three observation signals. The specific operation of multiplying the whitening matrix by the zero-mean observation signal is as follows: the whitening matrix is ​​multiplied by the 3-row, 1024-column zero-mean observation signal matrix to obtain a 3-row, 1024-column matrix, with each row representing one whitened signal. The method to verify the whitening effect is to calculate the autocorrelation matrix of the whitened signal, which should be a 3-order identity matrix, i.e., all diagonal elements are 1 and all off-diagonal elements are 0. If there are off-diagonal elements with an absolute value greater than 0.01, it indicates that the whitening is incomplete, and the precision of eigenvalue decomposition needs to be readjusted or the number of sampling points increased. Whitening preprocessing is a key preliminary step in independent component analysis. It simplifies the estimation problem of the mixture matrix from a general matrix to an orthogonal matrix, greatly reducing the complexity of subsequent algorithms.

[0073] In step S2, the independent component analysis algorithm employs a joint approximate diagonalization eigenma matrix algorithm based on fourth-order cumulants joint approximate diagonalization. The specific operation is as follows:

[0074] First, calculate the fourth-order joint cumulant for each component of the whitened signal and construct a family of fourth-order cumulant matrices. The calculation process for the fourth-order joint cumulant is as follows: For the three components of the whitened signal, let the first component be the first whitened signal component, the second component be the second whitened signal component, and the third component be the third whitened signal component. The fourth-order joint cumulant is defined as the product of the fourth moments of the first, second, third, and fourth whitened signal components (when the four component indices are repeated) minus the combination of the second moments. The specific calculation formula is: the fourth-order joint cumulant is equal to the sum of the product of the fourth moment minus three times the second moment. Where the fourth moment is the time average of the product of the four components at the same time, and the second moment is the time average of the product of the two components at the same time. Since there are three components, the fourth-order joint cumulant has 3 to the power of 4, or 81 values, but due to symmetry, only a portion needs to be calculated.

[0075] The steps to construct a family of fourth-order cumulant matrices are as follows: Select a set of 3x3 basis matrices, totaling 9 basis matrices. Each basis matrix has only one element that is 1, and the rest are 0. For each basis matrix, calculate a 3x3 fourth-order cumulant matrix. The element in the i-th row and j-th column of this matrix is ​​equal to: calculating the fourth-order joint cumulant for the i-th and j-th components of the whitened signal and the two component indices corresponding to the non-zero elements in the basis matrix, multiplying it by the value of the non-zero element in the basis matrix, and then summing over all possible two component indices. Finally, 9 fourth-order cumulant matrices are obtained, forming a family of fourth-order cumulant matrices. This family of matrices contains all the fourth-order statistical information of the whitened signal and is used to measure the non-Gaussianity and statistical independence of the signal.

[0076] Then, through iterative optimization, an orthogonal rotation matrix is ​​found such that the sum of the off-diagonal elements of the matrix obtained by multiplying the transpose of the orthogonal rotation matrix by each fourth-order cumulant matrix and then by the orthogonal rotation matrix is ​​minimized, thus achieving joint approximate diagonalization. The specific process of iterative optimization is as follows: Initialize the orthogonal rotation matrix as an identity matrix; in each iteration, first calculate the product of the transpose of the current orthogonal rotation matrix and each fourth-order cumulant matrix, and then multiply it by the current orthogonal rotation matrix to obtain a set of transformed matrices; then calculate the sum of squares of the off-diagonal elements of each transformed matrix, and add the sum of squares of the off-diagonal elements of all transformed matrices to obtain the cost function value; next, update the orthogonal rotation matrix through gradient descent or Newton's method to reduce the cost function value; specifically, in each iteration, select a rotation angle, perform a planar rotation on two rows in the orthogonal rotation matrix, and update the elements corresponding to these two rows; traverse all possible row pairs and select the rotation angle that causes the cost function to decrease the fastest for updating;

[0077] The iteration termination condition is: the change in the cost function between two consecutive iterations is less than 10 to the power of -6, or the maximum number of iterations of 500 is reached; the final orthogonal rotation matrix satisfies that: after its transpose is multiplied by each fourth-order cumulant matrix and then multiplied by itself, the absolute value of the off-diagonal elements of the resulting matrix is ​​less than 0.01, which means that joint approximate diagonalization is achieved; this process is called the Joint Approximate Diagonalization Characteristic Matrix Algorithm, or JADE Algorithm for short.

[0078] Finally, the transpose of the orthogonal rotation matrix is ​​multiplied by the whitening matrix to obtain the unmixing matrix, and the transpose of the orthogonal rotation matrix is ​​multiplied by the whitening signal to obtain the independent source signal components. The calculation process of the unmixing matrix is ​​as follows: multiply the transpose of the orthogonal rotation matrix by the whitening matrix to obtain a 3x3 matrix, which is the unmixing matrix. Each row of the unmixing matrix corresponds to the extraction weight of an independent source signal component. Multiplying the unmixing matrix by the zero-mean observation signal yields the independent source signal components. The number of independent source signal components is at most three, corresponding to the partial discharge signal inside the cabinet, the interference signal outside the cabinet, and the environmental noise signal, respectively. To verify the separation effect, the cross-correlation function of the independent source signal components can be calculated. If the cross-correlation function between each component is close to zero near zero delay, it indicates successful separation. If there is obvious residual correlation, the construction method of the fourth-order cumulant matrix family needs to be adjusted or the number of iterations needs to be increased. The independent source signal components will be used as the input of step S3 to invert their respective incident azimuth angles.

[0079] In this embodiment, it is specifically necessary to explain the operation of calculating the dot product between each mixed weight vector and each candidate direction unit vector in step S3 as follows:

[0080] First, each column of the estimated mixing matrix is ​​used as the mixing weight vector for the corresponding independent source signal component. The mixing weight vector contains three elements, each corresponding to one of the three ultrasonic sensors, and each element corresponds one-to-one with the normal direction of the sensitive axis of the three ultrasonic sensors. The mixing matrix is ​​estimated by inverting the unmixing matrix obtained in step S2, i.e., the estimated mixing matrix is ​​equal to the inverse of the unmixing matrix. This inverse matrix is ​​a 3x3 square matrix, with each column corresponding to the mixing weight vector of an independent source signal component. For the first independent source signal component, its mixing weight vector is the first column of the inverse matrix, containing three elements, denoted as the first mixing coefficient, the second mixing coefficient, and the third mixing coefficient, respectively, corresponding to the first ultrasonic sensor pointing at 0 degrees, the second ultrasonic sensor pointing at +120 degrees, and the third ultrasonic sensor pointing at -120 degrees. Similarly, the mixing weight vector for the second independent source signal component is the second column of the inverse matrix, and the mixing weight vector for the third independent source signal component is the third column of the inverse matrix. These three mixing weight vectors carry the spatial orientation information of the three independent source signals, which are the basic data for subsequent azimuth inversion.

[0081] Then, with a step size of one degree, a set of candidate direction unit vectors from 0 degrees to 359 degrees is constructed. Each candidate direction unit vector consists of the cosine and sine values ​​of the angle. The construction process of the candidate direction unit vectors is as follows: starting from 0 degrees, an angle value is taken every 1 degree until 359 degrees, resulting in 360 candidate angles. For each candidate angle, the cosine and sine values ​​are calculated, and these two values ​​are combined into a two-dimensional vector, which is the candidate direction unit vector corresponding to that angle. For example, the candidate direction unit vector corresponding to 0 degrees is (1,0), the candidate direction unit vector corresponding to 90 degrees is (0,1), the candidate direction unit vector corresponding to 180 degrees is (-1,0), and the candidate direction unit vector corresponding to 270 degrees is (0, -1). These 360 ​​candidate direction unit vectors are evenly distributed in all directions within the horizontal plane, forming a complete azimuth search space.

[0082] Next, for each hybrid weight vector, based on the fact that the sensitive axis normal directions of the three ultrasonic sensors point to 0 degrees, +120 degrees, and -120 degrees respectively, three sensitive axis normal direction unit vectors are defined: the first sensitive axis normal direction unit vector is the unit vector corresponding to the 0-degree direction, the second sensitive axis normal direction unit vector is the unit vector corresponding to the +120-degree direction, and the third sensitive axis normal direction unit vector is the unit vector corresponding to the -120-degree direction; the specific values ​​of the three sensitive axis normal direction unit vectors are as follows: the first sensitive axis normal direction unit vector, i.e., the unit vector corresponding to the 0-degree direction, has a first component of 1 and a second component of 0; the second sensitive axis normal direction unit vector, i.e., the unit vector corresponding to the +120-degree direction, has a first component of -1 / 2 and a second component of -√3 / 2; the third sensitive axis normal direction unit vector, i.e., the unit vector corresponding to the -120-degree direction, has a first component of -1 / 2 and a second component of -√3 / 2; these three unit vectors are uniquely determined by the geometric layout of the three-probe 120-degree rigid array and remain unchanged in subsequent calculations;

[0083] The first element of the mixed weight vector is multiplied by the unit vector of the first sensitive axis normal direction, plus the product of the second element of the mixed weight vector and the unit vector of the second sensitive axis normal direction, plus the product of the third element of the mixed weight vector and the unit vector of the third sensitive axis normal direction. The sum of these three products constitutes a two-dimensional projected composite vector. The specific calculation process of the two-dimensional projected composite vector is as follows: taking the mixed weight vector of the first independent source signal component as an example, its first mixing coefficient is multiplied by the unit vector of the first sensitive axis normal direction to obtain a two-dimensional intermediate vector; its second mixing coefficient is multiplied by the unit vector of the second sensitive axis normal direction to obtain another two-dimensional intermediate vector; its third mixing coefficient is multiplied by the unit vector of the third sensitive axis normal direction to obtain a third two-dimensional intermediate vector; then these three two-dimensional intermediate vectors are added according to their corresponding components, that is, the first component is added to obtain the first component of the two-dimensional projected composite vector, and the second component is added to obtain the second component of the two-dimensional projected composite vector; this two-dimensional projected composite vector represents the weighted composite projection of the mixed weight vector in the three sensor directions, and its direction points to the estimate of the sound source incident direction;

[0084] Then, the dot product of the two-dimensional projection composite vector with each candidate direction unit vector is calculated. The dot product is calculated as follows: multiply the first element of the two-dimensional projection composite vector by the first element of the candidate direction unit vector, and add the second element of the two-dimensional projection composite vector multiplied by the second element of the candidate direction unit vector. The dot product calculation process is as follows: for the two-dimensional projection composite vector of the first independent source signal component, the dot product operation is performed with each of the 360 ​​candidate direction unit vectors in sequence. Taking the dot product with the 0-degree candidate direction unit vector as an example, multiply the first component of the two-dimensional projection composite vector by 1, and add the second component of the two-dimensional projection composite vector multiplied by 0 to obtain the dot product value. Repeat this process to obtain 360 dot product values, each dot product value corresponding to a candidate direction. The physical meaning of the dot product value is the projection length of the two-dimensional projection composite vector in the candidate direction. The larger the value, the more consistent the candidate direction is with the true incident direction of the sound source.

[0085] Finally, record all dot product values ​​corresponding to each mixed weight vector and find the maximum value. The candidate direction angle corresponding to the maximum value is the incident azimuth angle of the independent source signal component. The process of finding the maximum value is as follows: traverse the 360 ​​dot product values ​​corresponding to the first independent source signal component and record the maximum value and its corresponding candidate direction angle. To ensure the reliability of the azimuth angle estimation, a peak confidence threshold is set: when the ratio of the maximum value to the second largest value is greater than 1.5, the angle corresponding to the maximum value is considered reliable and is taken as the incident azimuth angle; if the ratio is less than or equal to 1.5, the azimuth angle estimation is considered unreliable, and the incident azimuth angle of the independent source signal component is marked as "direction uncertain" and processed as an interference signal in the subsequent step S4. Similarly, repeat the above process for the second and third independent source signal components to obtain their respective incident azimuth angles or "direction uncertain" labels. The peak confidence threshold of 1.5 is determined through experimental statistics and can effectively suppress false azimuth angles while ensuring the detection rate.

[0086] In this embodiment, it should be specifically noted that in step S4, the preset background noise threshold is dynamically updated using an exponentially weighted moving average recursive estimation method. The specific operation is as follows:

[0087] At the end of each preset analysis window, if no partial discharge signal is determined to be inside the cabinet within the analysis window, the time-domain energy of the independent source signal component within the analysis window is used as a background noise sample to update the background noise threshold. The new background noise threshold is equal to the preset forgetting factor multiplied by the current background noise threshold, plus one minus the preset forgetting factor multiplied by the time-domain energy of the current analysis window.

[0088] If the analysis window indicates a partial discharge signal within the cabinet, no update is performed, and the current background noise threshold remains unchanged. The initial background noise threshold is set to the average energy of the background noise without partial discharge collected during offline calibration. The length of the analysis window is 1024 sampling points, corresponding to the signal length recorded in step S1. At the end of each analysis window, the system reads the time-domain energy value of the independent source signal component within the current analysis window and records it as the current energy. It determines whether the current analysis window is identified as a partial discharge signal within the cabinet by the joint decision function in step S4. If it is not identified as a partial discharge signal, a threshold update is performed: the forgetting factor (ranging from 0.9 to 0.99, and 0.95 in this embodiment) is multiplied by the current background noise threshold, and then the forgetting factor (i.e., 0.05) is multiplied by the current energy, and the two are added together to obtain a new background noise threshold. If it is identified as a partial discharge signal, the update is skipped, and the background noise threshold remains unchanged.

[0089] The initial background noise threshold is obtained as follows: during the offline calibration phase when the switchgear is not in operation, a background noise signal with no partial discharge for at least 10 seconds is collected. The energy of each window is calculated with 1024 points as an analysis window, and the arithmetic mean of the energy of all windows is taken as the initial background noise threshold. This recursive update mechanism enables the threshold to automatically adjust with the slow changes in environmental noise (such as noise floor drift caused by diurnal temperature difference and load fluctuations) without the need for manual intervention and recalibration. The forgetting factor of 0.95 means that the weight of historical noise accounts for 95%, and the weight of the current noise sample accounts for 5%, which can smooth short-term fluctuations and track long-term trends.

[0090] In step S4, before comparing the incident azimuth with the preset effective area angle range inside the cabinet, the azimuth confidence factor is calculated. The specific operation is as follows:

[0091] Based on the ratio of the maximum to the second largest dot product in step S3, the azimuth confidence factor is calculated. The azimuth confidence factor is equal to the ratio of the maximum to the second largest dot product minus the preset peak confidence threshold, and then divided by the preset theoretical maximum ratio minus the preset peak confidence threshold. The result is limited to between zero and one.

[0092] When the ratio of the maximum value to the second largest value of the dot product is less than or equal to the preset peak confidence threshold, the azimuth confidence factor is zero, indicating that the azimuth is not reliable, and the independent source signal component is directly determined as an interference signal and suppressed.

[0093] When the ratio of the maximum to the second largest dot product is greater than or equal to the preset theoretical maximum ratio, the azimuth confidence factor is set to one, indicating that the azimuth is highly reliable. The ratio of the maximum to the second largest dot product is derived from the direction response consistency function in step S3. For each independent source signal component, step S3 calculates the dot product values ​​corresponding to 360 candidate directions and finds the maximum and second largest values. This ratio reflects the sharpness of the azimuth estimation: the larger the ratio, the more prominent the peak and the more reliable the azimuth estimation. In this embodiment, the peak confidence threshold is set to 1.5, and the theoretical maximum ratio is set to 10.

[0094] The specific calculation steps for the azimuth confidence factor are as follows: First, subtract the peak confidence threshold of 1.5 from the ratio of the maximum to the second largest dot product to obtain the difference; Second, subtract the peak confidence threshold of 1.5 from the theoretical maximum ratio of 10 to obtain a denominator of 8.5; Third, divide the difference from the first step by the denominator of the second step to obtain an intermediate value between negative and positive; Fourth, restrict the intermediate value to between 0 and 1, that is, if the intermediate value is less than 0, take 0; if the intermediate value is greater than 1, take 1; otherwise, take 1. The intermediate value itself; when the ratio is less than or equal to 1.5, the azimuth confidence factor is forced to 0. In this case, regardless of the energy level, the independent source signal component is directly identified as an interference signal and suppressed to avoid misjudgment caused by unreliable azimuth; when the ratio is greater than or equal to 10, the azimuth confidence factor is forced to 1, indicating that the azimuth estimation is extremely reliable; this confidence factor provides a quantitative indicator of the azimuth estimation quality for the subsequent joint decision function, enabling the system to dynamically adjust the decision weights according to the azimuth confidence.

[0095] In step S4, the process of determining a partial discharge signal inside the cabinet and outputting an alarm is implemented through a joint decision function. The specific operation is as follows:

[0096] First, calculate the indicator function value of the angle range of the incident azimuth angle entering the preset effective area inside the cabinet. When the incident azimuth angle enters the preset effective area inside the cabinet angle range, take one value; otherwise, take zero value.

[0097] Then, the energy margin ratio is calculated, which is equal to the time-domain energy of the independent source signal component divided by the background noise threshold;

[0098] Next, the preset weighting coefficient is multiplied by the energy margin ratio, and then the preset weighting coefficient multiplied by the azimuth confidence factor is added, and then multiplied by the indicator function value of the incident azimuth angle entering the preset effective area angle range inside the cabinet to obtain the joint decision function value.

[0099] Finally, the joint decision function value is compared with the preset alarm threshold and the preset suspected threshold. If the joint decision function value is greater than or equal to the preset alarm threshold, it is determined to be a partial discharge signal inside the cabinet and an alarm is output; if the joint decision function value is greater than or equal to the preset suspected threshold and less than the preset alarm threshold, it is marked as a suspected signal; if the joint decision function value is less than the preset suspected threshold, it is determined to be an interference signal and suppressed. The specific calculation process of the joint decision function is as follows: First, determine whether the incident azimuth angle falls within the effective area angle range inside the cabinet (±60 degrees centered on the cabinet door normal direction, i.e., from -60 degrees to +60 degrees); if it falls within, the indicator function value is 1; if it does not fall within, the indicator function value is 0. At this time, regardless of the energy and confidence level, the joint decision function value is 0, and it is directly determined to be an interference signal and suppressed.

[0100] The second step is to calculate the energy margin ratio, which is the time-domain energy of the independent source signal component divided by the current background noise threshold. If the energy margin ratio is less than 1, it means that the signal energy is below the noise threshold. At this time, even if the azimuth angle is correct, it is difficult to reliably determine that it is a partial discharge. The third step is to multiply the weighting coefficient (0.6 in this embodiment) by the energy margin ratio, and then multiply the weighting coefficient (0.4) by the azimuth confidence factor. The two are added together to obtain an intermediate value between 0 and a certain upper limit.

[0101] The fourth step is to multiply the intermediate value by the indicator function value to obtain the joint decision function value. The fifth step is to compare the joint decision function value with the alarm threshold of 1.5 and the suspected threshold of 0.8: if the joint decision function value is greater than or equal to 1.5, it is determined to be a partial discharge signal in the cabinet, and an alarm is immediately output. The alarm information includes the incident azimuth angle, energy value, confidence factor and joint decision function value of the component; if the joint decision function value is greater than or equal to 0.8 and less than 1.5, it is marked as a suspected signal and stored in the local database for subsequent analysis by maintenance personnel or for incremental learning.

[0102] If the joint decision function value is less than 0.8, it is judged as an interference signal, and no alarm is generated or stored in the database. This three-level decision mechanism avoids the "one-size-fits-all" problem of traditional binary decision, and significantly reduces the false alarm rate while ensuring a high detection rate. The weighting coefficient of 0.6 is biased towards energy evidence because, in actual measurements, the energy margin ratio contributes slightly more to distinguishing partial discharge from interference than the azimuth confidence level. This coefficient can be optimized and adjusted through field data. The alarm threshold of 1.5 and the suspected threshold of 0.8 are determined through receiver operating characteristic curve analysis. In a typical substation environment, this can reduce the false alarm rate by more than 90% and the missed alarm rate by less than 2%.

[0103] Example 2: This example provides the following... Figure 2 The ultrasonic partial discharge anti-interference monitoring system for switchgear shown here specifically includes:

[0104] Acquisition module: Three ultrasonic sensors are mounted in a rigid array at a 120-degree angle in the gap of the cabinet door. The three signals are synchronously acquired using a unified clock source to obtain three observation signals.

[0105] Separation module: After centering and whitening the three observed signals, independent component analysis is used to solve the unmixing matrix and separate the statistically independent source signal components.

[0106] Azimuth module: The inverse of the unmixing matrix is ​​used as the mixing matrix estimate, and each column of the mixing matrix estimate is used as the mixing weight vector. Candidate direction unit vectors from 0 degrees to 359 degrees are constructed. The angle corresponding to the maximum value of the dot product is calculated as the incident azimuth angle of each independent source signal component.

[0107] Decision module: Compare the incident azimuth angle with the effective area inside the cabinet. If it falls within the effective area and the corresponding time domain energy exceeds the background noise threshold, it is determined to be a partial discharge signal and an alarm is triggered. Otherwise, it is determined to be interference suppression.

[0108] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0109] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0110] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0111] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0113] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0114] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for ultrasonic partial discharge anti-interference monitoring of switchgear, characterized in that, Specifically, the following steps are included: Step S1: At the gap of the cabinet door of the switch cabinet to be tested, three ultrasonic sensors are fixedly installed in a rigid array with their sensitive axes coplanar and the included angle between each pair being 120 degrees. A unified clock source drives a multi-channel synchronous analog-to-digital converter to synchronously acquire the three signals output by the three ultrasonic sensors to obtain three observation signals. Step S2: After centering and whitening preprocessing of the three observation signals, the independent component analysis algorithm is used to solve the unmixing matrix, so that the output components are statistically independent, thereby separating multiple independent source signal components. Step S3: Use the inverse matrix of the demixing matrix as the mixing matrix estimate, take each column of the mixing matrix estimate as the mixing weight vector of the corresponding independent source signal component, and construct a set of candidate direction unit vectors covering 0 degrees to 359 degrees. By calculating the dot product between each mixing weight vector and each candidate direction unit vector, take the angle corresponding to the maximum value of the dot product as the incident azimuth angle of the independent source signal component. Step S4: Compare the incident azimuth angle with the preset effective area angle range inside the cabinet. If the incident azimuth angle falls within the effective area angle range inside the cabinet and the time domain energy of the corresponding independent source signal component exceeds the preset background noise threshold, it is determined to be a partial discharge signal inside the cabinet and an alarm is output. Otherwise, it is determined to be an interference signal and suppressed.

2. The ultrasonic partial discharge anti-interference monitoring method for switchgear according to claim 1, characterized in that: In step S1, the specific method for obtaining the three observation signals is as follows: During the offline calibration phase, three ultrasonic sensors are excited by a standard ultrasonic pulse of known direction, and three initial signals are recorded. The generalized cross-correlation phase transform algorithm is used to estimate the relative time delay between the three ultrasonic sensors. The generalized cross-correlation phase transform algorithm obtains the peak position of the cross-correlation function as the time delay estimate by calculating the phase transform weighted integral of the cross power spectral density function of the two signals. Then, taking one of the ultrasonic sensors as a reference, fractional delay compensation based on sinusoidal interpolation is performed on the three initial signals of the other two ultrasonic sensors to eliminate the fixed time delay difference caused by hardware clock skew and filter group delay differences between the ultrasonic sensors, so that the three signals are strictly aligned in the time domain, thereby obtaining three observation signals.

3. The ultrasonic partial discharge anti-interference monitoring method for switchgear according to claim 2, characterized in that: In step S1, in the rigid array composed of three ultrasonic sensors, the normal directions of the sensitive axes of the three ultrasonic sensors point to zero degrees, positive 120 degrees and negative 120 degrees respectively, and the three ultrasonic sensors are fixed on the same rigid base, which is attached to the central area of ​​the gap in the door of the switch cabinet to be tested. A unified clock source drives a multi-channel synchronous analog-to-digital converter to synchronously digitize the analog signals output by the three ultrasonic sensors at a sampling rate of no less than one megahertz, ensuring that the sampling start times of the three signals are strictly aligned, thereby providing high-fidelity three observation signals.

4. The ultrasonic partial discharge anti-interference monitoring method for switchgear according to claim 3, characterized in that: In step S2, the specific operations for centering and whitening preprocessing are as follows: First, subtract the arithmetic mean of the three observation signals from each of the three observation signals to obtain the zero-mean observation signal; Then, the autocorrelation matrix of the zero-mean observed signal is calculated, and the autocorrelation matrix is ​​decomposed into eigenvalues ​​to obtain the orthogonal eigenvector matrix and the eigenvalue diagonal matrix. Next, the whitening matrix is ​​constructed by multiplying the negative half power of the eigenvalue diagonal matrix by the transpose of the orthogonal eigenvector matrix. Finally, the whitening matrix is ​​multiplied by the zero-mean observation signal to obtain the whitened signal. The variance of each component of the whitened signal is one, thus eliminating the second-order correlation between the three observation signals.

5. The ultrasonic partial discharge anti-interference monitoring method for switchgear according to claim 4, characterized in that: In step S2, the independent component analysis algorithm employs a joint approximate diagonalization eigenma matrix algorithm based on fourth-order cumulants joint approximate diagonalization. The specific operation is as follows: First, calculate the fourth-order joint cumulants for each component of the whitened signal and construct a family of fourth-order cumulant matrices; Then, through iterative optimization, an orthogonal rotation matrix is ​​found such that the sum of the off-diagonal elements of the matrix obtained by multiplying the transpose of the orthogonal rotation matrix by each fourth-order cumulant matrix and then by the orthogonal rotation matrix is ​​minimized, thus achieving joint approximate diagonalization; Finally, the transpose of the orthogonal rotation matrix is ​​multiplied by the whitening matrix to obtain the demixing matrix, and the transpose of the orthogonal rotation matrix is ​​multiplied by the whitening signal to obtain the independent source signal components.

6. The ultrasonic partial discharge anti-interference monitoring method for switchgear according to claim 5, characterized in that: In step S3, the specific operation of calculating the dot product between each mixed weight vector and each candidate direction unit vector is as follows: First, each column of the estimated mixing matrix is ​​used as the mixing weight vector of the corresponding independent source signal component. The mixing weight vector contains three elements, which correspond to the three ultrasonic sensors respectively, and the three elements correspond one-to-one with the normal direction of the sensitive axis of the three ultrasonic sensors. Then, with a step size of one degree, a set of candidate direction unit vectors from 0 degrees to 359 degrees is constructed. Each candidate direction unit vector consists of the cosine and sine values ​​of the angle. Next, for each hybrid weight vector, based on the fact that the normal directions of the three ultrasonic sensors' sensitive axes point to 0 degrees, positive 120 degrees, and negative 120 degrees respectively, three unit vectors for the normal directions of the sensitive axes are defined: the first unit vector for the normal direction of the sensitive axes is the unit vector corresponding to the direction of 0 degrees, the second unit vector for the normal direction of the sensitive axes is the unit vector corresponding to the direction of positive 120 degrees, and the third unit vector for the normal direction of the sensitive axes is the unit vector corresponding to the direction of negative 120 degrees. Multiply the first element of the mixed weight vector by the unit vector of the first sensitive axis normal direction, add the second element of the mixed weight vector by the unit vector of the second sensitive axis normal direction, and add the third element of the mixed weight vector by the unit vector of the third sensitive axis normal direction. The sum of the three products constitutes a two-dimensional projection composite vector. Then, the dot product of the two-dimensional projection composite vector and each candidate direction unit vector is calculated. The dot product is calculated by multiplying the first element of the two-dimensional projection composite vector by the first element of the candidate direction unit vector, and adding the second element of the two-dimensional projection composite vector multiplied by the second element of the candidate direction unit vector. Finally, record all the dot product values ​​corresponding to each mixed weight vector, and find the maximum value. The candidate direction angle corresponding to the maximum value is the incident azimuth angle of the independent source signal component.

7. The ultrasonic partial discharge anti-interference monitoring method for switchgear according to claim 6, characterized in that: In step S4, the preset background noise threshold is dynamically updated using an exponentially weighted moving average recursive estimation method. Specifically, the operation is as follows: At the end of each preset analysis window, if no partial discharge signal is determined to be inside the cabinet within the analysis window, the time-domain energy of the independent source signal component within the analysis window is used as a background noise sample to update the background noise threshold. The new background noise threshold is equal to the preset forgetting factor multiplied by the current background noise threshold, plus one minus the preset forgetting factor multiplied by the time-domain energy of the current analysis window. If the analysis window determines that the signal is a partial discharge signal inside the cabinet, no update will be made, and the current background noise threshold will remain unchanged.

8. The ultrasonic partial discharge anti-interference monitoring method for switchgear according to claim 7, characterized in that: In step S4, before comparing the incident azimuth with the preset effective area angle range inside the cabinet, the azimuth confidence factor is calculated. The specific operation is as follows: Based on the ratio of the maximum to the second largest dot product in step S3, the azimuth confidence factor is calculated. The azimuth confidence factor is equal to the ratio of the maximum to the second largest dot product minus the preset peak confidence threshold, and then divided by the preset theoretical maximum ratio minus the preset peak confidence threshold. The result is limited to between zero and one. When the ratio of the maximum value to the second largest value of the dot product is less than or equal to the preset peak confidence threshold, the azimuth confidence factor is zero, indicating that the azimuth is not reliable, and the independent source signal component is directly determined as an interference signal and suppressed. When the ratio of the maximum to the second largest dot product is greater than or equal to the preset theoretical maximum ratio, the azimuth confidence factor is set to one, indicating that the azimuth is highly reliable.

9. The ultrasonic partial discharge anti-interference monitoring method for switchgear according to claim 8, characterized in that: In step S4, the process of determining a partial discharge signal inside the cabinet and outputting an alarm is implemented through a joint decision function. The specific operation is as follows: First, calculate the indicator function value of the angle range of the incident azimuth angle entering the preset effective area inside the cabinet. When the incident azimuth angle enters the preset effective area inside the cabinet angle range, take one value; otherwise, take zero value. Then, the energy margin ratio is calculated, which is equal to the time-domain energy of the independent source signal components divided by the background noise threshold; Next, the preset weighting coefficient is multiplied by the energy margin ratio, and then the preset weighting coefficient multiplied by the azimuth confidence factor is added, and then multiplied by the indicator function value of the incident azimuth angle entering the preset effective area angle range inside the cabinet to obtain the joint decision function value. Finally, the joint decision function value is compared with the preset alarm threshold and the preset suspected threshold. If the joint decision function value is greater than or equal to the preset alarm threshold, it is determined to be a partial discharge signal in the cabinet and an alarm is output. If the joint decision function value is greater than or equal to the preset suspected threshold and less than the preset alarm threshold, it is marked as a suspected signal. If the joint decision function value is less than the preset suspected threshold, it is determined to be an interference signal and is suppressed.

10. A switchgear ultrasonic partial discharge anti-interference monitoring system, applied to the switchgear ultrasonic partial discharge anti-interference monitoring method as described in any one of claims 1-9, characterized in that: Specifically, it includes: Acquisition module: Three ultrasonic sensors are mounted in a rigid array at a 120-degree angle in the gap of the cabinet door. The three signals are synchronously acquired using a unified clock source to obtain three observation signals. Separation module: After centering and whitening the three observed signals, independent component analysis is used to solve the unmixing matrix and separate the statistically independent source signal components. Azimuth module: The inverse of the unmixing matrix is ​​used as the mixing matrix estimate, and each column of the mixing matrix estimate is used as the mixing weight vector. Candidate direction unit vectors from 0 degrees to 359 degrees are constructed. The angle corresponding to the maximum value of the dot product is calculated as the incident azimuth angle of each independent source signal component. Decision module: Compare the incident azimuth angle with the effective area inside the cabinet. If it falls within the effective area and the corresponding time domain energy exceeds the background noise threshold, it is determined to be a partial discharge signal and an alarm is triggered. Otherwise, it is determined to be interference suppression.

Images