Transformer iron core insulation fault distinguishing method based on electromagnetic acoustic cross-modal fusion

Abstract

The invention discloses a transformer iron core insulation fault distinguishing method based on electromagnetic acoustic cross-modal fusion, which comprises the following steps: acquiring iron core grounding current, UHF electromagnetic signals and acoustic signals, fusing the acquired iron core grounding current, UHF electromagnetic signals and acoustic signals to obtain fusion features, processing the UHF electromagnetic signals and acoustic signals to obtain a transformer iron core insulation fault distinguishing result, and distinguishing the transformer iron core insulation fault distinguishing result. And respectively obtaining a final fault coordinate and an acoustic positioning coordinate, carrying out signal-to-noise ratio weighted fusion on the final fault coordinate and the acoustic positioning coordinate to obtain a fault three-dimensional coordinate after final fusion, processing the fusion feature and the fault three-dimensional coordinate after final fusion, and outputting a final diagnosis report. The cross-modal attention mechanism is introduced, the purified iron core current, the UHF electromagnetic signal and the acoustic signal are deeply fused, the iron core and insulation fault distinguishing accuracy is remarkably improved, and the fault diagnosis precision and stability are enhanced.

Description

Technical Field

[0001] This invention belongs to the field of power system transformer condition monitoring and fault diagnosis technology, specifically involving a method for distinguishing transformer core insulation faults by electromagnetic acoustic cross-modal fusion. Background Technology

[0002] Multi-point grounding of transformer cores and partial discharge of insulation are two typical faults. Misdiagnosis or failure to diagnose them can lead to equipment damage or even power grid accidents. Existing technologies have the following core limitations: Interference suppression lacks physical constraints: There are strong interferences such as power grid harmonics and equipment vibrations on site. Traditional filtering algorithms (such as wavelet thresholding and mean filtering) do not combine electromagnetic / acoustic signal propagation characteristics, resulting in distortion of core current harmonic characteristics and UHF discharge signal extraction. The signal-to-noise ratio is generally lower than 20dB, and fault characteristics are submerged by noise. Cross-modal feature fusion severs physical correlation: Core current (electrical signal), UHF (electromagnetic signal), and acoustic (vibration signal) are heterogeneous data. Existing methods simply splice together single signal features, ignoring the inherent physical correlation between "discharge current-electromagnetic radiation-mechanical vibration" (such as the positive correlation between discharge intensity and electromagnetic radiation energy and vibration amplitude), resulting in an accuracy of less than 85% in distinguishing between core and insulation faults. Low positioning accuracy and no trend prediction: It relies on a single UHF array for positioning, does not utilize the propagation stability of acoustic signals, and the positioning error exceeds 30cm; it can only "identify" faults in real time, but cannot predict the speed of fault development, resulting in delayed maintenance decisions.

[0003] Therefore, it is urgent to build a complete technical system that integrates "signal purification, cross-modal modeling, precise positioning, and trend prediction" to break through the accuracy and robustness bottlenecks of traditional methods. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for differentiating transformer core insulation faults by electromagnetic acoustic cross-modal fusion, aiming to solve the problems in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for differentiating transformer core insulation faults by electromagnetic acoustic cross-modal fusion, comprising: Step S1: Collect the core grounding current, UHF electromagnetic signal and acoustic signal, and fuse the collected core grounding current, UHF electromagnetic signal and acoustic signal to obtain the fused features; Step S2: Process the UHF electromagnetic signal and acoustic signal to obtain the final fault coordinates and acoustic location coordinates respectively. Perform signal-to-noise ratio weighted fusion on the final fault coordinates and acoustic location coordinates to obtain the final fused three-dimensional fault coordinates. Step S3: Process the fusion features and the final fused three-dimensional coordinates of the fault, and output the final diagnostic report.

[0006] Furthermore, the specific process of acquiring the core grounding current, UHF electromagnetic signal, and acoustic signal in step S1 is as follows: Core grounding current: is obtained by collecting the original current waveform of the transformer core grounding circuit using a broadband current transformer; UHF electromagnetic signal: It is obtained by collecting partial discharge pulses of transformer core from 6 array sensors, and a UHF electromagnetic signal array is obtained based on the UHF electromagnetic signal; Acoustic signal: It is obtained by collecting vibration signals emitted by the transformer core from 4 piezoelectric sensors.

[0007] Furthermore, the specific process of obtaining the UHF electromagnetic signal array is as follows: When the transformer is running, the UHF electromagnetic signal conditioning electronic unit monitors the signal amplitude of the 6-channel array sensor in real time. When the signal amplitude exceeds 3 times the background noise, the ADC module is triggered to sample. The signal amplitude is simultaneously sampled by the ADC module to obtain 6 sampling signals; the 6 sampling signals are then transmitted to the back-end processing unit via high-speed Ethernet. The background processing unit performs bandpass filtering, amplitude correction, and time delay compensation on the six sampled signals to obtain a UHF electromagnetic signal array.

[0008] Furthermore, the specific process for obtaining the fusion features is as follows: The core grounding current is decomposed using independent component analysis to obtain the decomposed current components. After decomposition The power spectral density satisfies At that time, harmonic interference is removed from the core grounding current to obtain the purified core current; The center frequency; Frequency offset; Current component Peak value; Current component The power spectral density; Constructing a spatial spectrum function based on the spatial orientation characteristics of a UHF electromagnetic signal array ; R is the conjugate transpose of the manifold vector of the UHF electromagnetic signal array; R is the signal covariance matrix. The manifold vector of the UHF electromagnetic signal array; Based on the constructed spatial spectrum function, the peak value of the search spatial spectrum is used to lock the target discharge direction of the UHF electromagnetic signal, suppress the non-target direction interference of the UHF electromagnetic signal, and obtain the suppressed UHF electromagnetic signal. Calculate the cross-correlation coefficient between the acoustic signal and the UHF electromagnetic signal. When the cross-correlation coefficient is greater than 0.6, it indicates that the acoustic signal is the mechanical vibration response of a real discharge. Remove the mechanical noise from the acoustic signal to obtain the acoustic signal after removal, which is the acoustic signal characteristic. The fractal dimension FD of the purified iron core current is extracted, which is the current characteristic, specifically: Calculating fractal dimension using box counting method ; For the number of boxes Take the logarithm; To determine the side length of the box Take the logarithm of the reciprocal; The slope of the linear fit; The wavelet packet energy spectrum of the suppressed UHF electromagnetic signal is extracted as follows: The suppressed UHF electromagnetic signal was subjected to 5-level wavelet packet decomposition, and the energy spectrum of the decomposed wavelet packets was calculated. ; The energy of the k-th wavelet packet subband after the suppressed UHF electromagnetic signal is decomposed into 5 layers of wavelet packets; The energy of the i-th wavelet packet subband after the suppressed UHF electromagnetic signal is decomposed into 5 layers of wavelet packets; This represents the proportion of the energy of the k-th wavelet packet subband in the total energy of all subbands. By fusing current characteristics and cross-modal attention mechanism The fusion features are obtained, specifically: Using current characteristics as a query, Acoustic signal features are used as key values, through the formula , ;in, For the first The output attention weight matrix of each attention head; For activation functions; For the first The query matrix corresponding to the current features in each attention head; For the first A person's attention / Transpose of the key matrix of acoustic signal features; For feature dimensions; Features of fusion; For the first A person's attention / Value matrix of acoustic signal characteristics; This is a transpose.

[0009] Furthermore, the specific process for obtaining the final fault coordinates is as follows: Establish a three-dimensional rectangular coordinate system; A three-dimensional rectangular coordinate system is established with the geometric center of the transformer tank as the origin (0,0,0). The three-dimensional rectangular coordinate system consists of an X-axis, a Y-axis, and a Z-axis. The X-axis is along the transformer axis, pointing from the high-voltage side to the low-voltage side. The Y-axis is perpendicular to the transformer axis in the horizontal plane, pointing towards the front of the tank. The Z-axis is vertical, with upward as positive, and perpendicular to the ground. The installation positions of the 6-channel array sensors in the UHF electromagnetic signal array are determined based on a three-dimensional Cartesian coordinate system, denoted as: ;in, ;Will As the first reference sensor; For the i-th sensor; Let be the coordinate value of the i-th sensor in the x-axis direction; For the i-th sensor in Coordinate values ​​along the axis; Let be the coordinate value of the i-th sensor in the z-axis direction; By synchronously sampling the partial discharge pulses of the transformer core using a 6-channel array sensor, the arrival time of the UHF electromagnetic signal at the fault point of the transformer core partial discharge pulse is extracted for each sensor: the arrival time of the UHF electromagnetic signal at the fault point is as follows: The time is The UHF electromagnetic signal at the fault point reaches the nth sensor. The time is ; ; ; The time it takes for the UHF electromagnetic signal at the fault point to reach the second sensor; The time it takes for the UHF electromagnetic signal at the fault point to reach the 6th sensor; Time difference calculation ; For sensors Arrival time With the first reference sensor Arrival time difference; The propagation speed of the UHF electromagnetic signal at the fault point in the transformer oil is calculated using the distance difference, where the distance difference is... ; For sensors With the first reference sensor The distance difference between the UHF electromagnetic signal and the fault location; The propagation speed of the UHF electromagnetic signal at the fault point in the transformer oil; Let the coordinates of the UHF electromagnetic signal at the fault point be... The fault point UHF electromagnetic signal is transmitted to the first reference sensor. The distance difference is ; , , These are the coordinates of the UHF electromagnetic signal at the fault point along the x, y, and z axes, respectively. , , The first reference sensor Coordinate values ​​along the z-axis, y-axis, and z-axis directions; The fault point UHF electromagnetic signal reaches the first One sensor The distance difference is ; , , The nth sensor Coordinate values ​​along the x, y, and z axes; The fault point's UHF electromagnetic signal and the nth sensor The geometric relationship of the time difference is as follows: ; Send the UHF electromagnetic signal from the fault point to the first reference sensor. coordinates Substituting the distance geometric relationship of the time difference, we obtain a nonlinear equation, which represents: ; Set the initial estimated fault point ; , , The initial estimated fault points are located at axis, axis, Coordinate values ​​along the axis; The nonlinear equations are simplified using a linear approximation method at the initially estimated fault points, resulting in the linearized equations: ; The coefficient matrix is ​​the initially estimated coordinates of the fault point; This is a correction vector for the initially estimated fault point coordinates; This is a constant vector representing the initially estimated coordinates of the fault point; in, ; This is the coordinate correction amount of the initially estimated fault point in the x-axis direction; This is the coordinate correction amount of the initially estimated fault point in the y-axis direction; This is the coordinate correction amount of the initially estimated fault point in the z-axis direction; The coefficient matrix of the initially estimated fault point coordinates is represented as follows: ; in, , , These are the numbers in coefficient matrix A, respectively. The elements in the first, second, and third columns of the row; From the initially estimated fault point to the nth sensor The distance; The initial estimated fault point is located at the first reference sensor. The distance; The constant vector representing the initially estimated fault point coordinates is: ; This is the i-th element in the constant term vector; Calculate the optimal approximate solution by solving the linearized equation using the pseudo-inverse matrix: ; The fault point coordinate correction vector is the one used for the optimal initial estimate. Perform the k-th iteration on the initially estimated fault point to obtain ; For the first The fault location coordinates are updated in the next iteration; Update fault location coordinates ; For the first The fault location coordinates are updated in the next iteration; when At this time That is, the final fault coordinates ;when If so, recalculate. ,until The calculation ends when the time is right. , , The final fault coordinates are respectively at axis, axis, Coordinate values ​​along the axis; For rice.

[0010] Furthermore, the specific process of obtaining acoustic positioning coordinates is as follows: Wideband acoustic emission sensors were selected and installed on the upper, middle, lower and side layers of the transformer tank wall; Upper layer: Near the center of the top of the fuel tank, coordinates ; It is the first broadband acoustic emission sensor; , , The first broadband acoustic emission sensor is located at... axis, axis, The coordinate values ​​along the axis; As a second reference sensor; Middle layer: Upper middle part of the fuel tank wall, coordinates ; It is the second broadband acoustic emission sensor; , , The second broadband acoustic emission sensor is located at... axis, axis, Coordinate values ​​along the axis; Lower layer: near the center of the bottom of the fuel tank, coordinates ; It is the third broadband acoustic emission sensor; , , The third broadband acoustic emission sensor is located at axis, axis, Coordinate values ​​along the axis; Side layer: Center of the side wall of the fuel tank, coordinates; It is the fourth wideband acoustic emission sensor; , , The fourth broadband acoustic emission sensor is located at axis, axis, Coordinate values ​​along the axis; When a fault inside the transformer generates vibration waves, these waves propagate towards the tank wall via oil. The broadband acoustic emission sensor, equipped with a high-speed ADC module, acquires the vibration signals in real time and uses a first-lead-edge detection algorithm to extract the arrival time of the acquired vibration signals at each broadband acoustic emission sensor. The arrival time of the vibration signal at the second reference sensor M1 is... The vibration signal reaches the d-th broadband acoustic emission sensor. The time is ; ; ; The time it takes for the vibration signal to reach the second broadband acoustic emission sensor; The time it takes for the vibration signal to reach the third broadband acoustic emission sensor; The time it takes for the vibration signal to reach the fourth broadband acoustic emission sensor; Let the fault point be The speed of sound propagating vibration waves in transformer oil =1500m / s; The fault point is at Coordinate values ​​along the axis; The fault point is at Coordinate values ​​along the axis; The fault point is at Coordinate values ​​along the axis; Calculate the vibration wave from the fault point arrive The propagation distance is: ; For the vibration wave from the fault point Propagation to the d-th broadband acoustic emission sensor The propagation distance; , , These are the d-th broadband acoustic emission sensors. exist axis, axis, Coordinate values ​​along the axis; Calculate the time difference between M2, M3, M4 and the reference sensor M1. ; The time difference between the vibration waves received by M2, M3, M4 and the reference sensor M1; Reference sensor The distance difference with M2, M3, and M4 is: ; The distance difference between reference sensors M1 and M2, M3, and M4; This is the distance the vibration wave travels from the fault point to the reference sensor M1; Based on the distance differences between reference sensors M1 and M2, M3, and M4, three independent nonlinear equations are constructed, representing: ; In the formula, The distance difference between reference sensors M1 and M2; The distance difference between reference sensors M1 and M3; The distance difference between reference sensors M1 and M3; Set initial coordinates Perform the k-th iteration on the initial coordinates to obtain... ; The coordinates of the point obtained after the k-th iteration; , , The initial coordinates are respectively at axis, axis, Coordinate values ​​along the axis; The three independent nonlinear equations are set in the initial coordinates. Simplifying the equations using a linear approximation method, we obtain the following system of linear equations: ; ; The correction vector for the initial coordinates; The coefficient matrix for the initial coordinates; A constant vector for the initial coordinates; , , The initial coordinates are respectively at axis, axis, Correction amount in the axial direction; Calculate the optimal approximate solution by solving the system of linear equations using the pseudo-inverse matrix. : ; The correction vector for the optimal initial coordinates; Update coordinates ; For the first The coordinates obtained after the next iteration; when At this time This refers to acoustic positioning coordinates. ,when Recalculate until The calculation ends when the time is right. , , The acoustic positioning coordinates are respectively at axis, axis, The coordinate values ​​along the axis.

[0011] Furthermore, the specific process for obtaining the final fused three-dimensional coordinates of the fault is as follows: Final fault coordinates Harmony and acoustic positioning coordinates The fusion process yields the final fused three-dimensional coordinates of the fault, represented as follows: ; In the formula, , , The three-dimensional coordinates of the fault after final fusion are respectively in axis, axis, Coordinate values ​​along the axis; The signal-to-noise ratio of the UHF signal; The signal-to-noise ratio of the acoustic signal.

[0012] Furthermore, the specific process of step S3 is as follows: Using a gated recurrent unit model to input fused features to predict the feature growth rate for the next hour. ; Combined with characteristic growth rate Cross-modal fusion feature threshold and And the location area, to construct a decision matrix; when Falling into the core fault area: <1.3 and >60%, positioned within ±5cm of the iron core post, and When the rate is <0.02 / min, the decision matrix outputs a planned maintenance diagnosis report; when Falling into the insulation fault area: >1.5 and >50%, positioned within ±10cm of the winding, and When the value is greater than 0.05 / min, the decision matrix outputs an emergency shutdown diagnostic report.

[0013] Compared with existing technologies, the present invention has the following advantages: (1) By fusing the power spectral density and kurtosis of the decomposed current components, this invention can accurately eliminate harmonic and noise interference, effectively improve the signal-to-noise ratio of the core grounding current, UHF electromagnetic signal and acoustic signal, and provide high-quality data for subsequent fault feature extraction and accurate diagnosis.

[0014] (2) The present invention introduces a cross-modal attention mechanism, which deeply integrates the purified core current, UHF electromagnetic signal and acoustic signal, significantly improves the accuracy of distinguishing between core and insulation faults, and enhances the accuracy and stability of fault diagnosis. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] like Figure 1 As shown, the present invention provides a technical solution: a method for distinguishing transformer core insulation faults by electromagnetic acoustic cross-modal fusion, comprising: Step S1: Collect the core grounding current, UHF electromagnetic signal and acoustic signal, and fuse the collected core grounding current, UHF electromagnetic signal and acoustic signal to obtain the fused features; Step S2: Process the UHF electromagnetic signal and acoustic signal to obtain the final fault coordinates and acoustic location coordinates respectively. Perform signal-to-noise ratio weighted fusion on the final fault coordinates and acoustic location coordinates to obtain the final fused three-dimensional fault coordinates. Step S3: Process the fusion features and the final fused three-dimensional coordinates of the fault, and output the final diagnostic report.

[0017] The specific process of acquiring the core grounding current, UHF electromagnetic signal, and acoustic signal in step S1 is as follows: Core grounding current: is obtained by acquiring the original current waveform (1kHz-1MHz, sampling rate 5MHz) of the transformer core grounding circuit using a broadband current transformer; UHF electromagnetic signal: It is obtained by collecting partial discharge pulses of transformer core (300MHz-1.5GHz, sampling rate 2GHz) from 6 array sensors, and a UHF electromagnetic signal array is obtained based on the UHF electromagnetic signal; Acoustic signal: It is obtained by collecting vibration signals emitted by the transformer core (1kHz-50kHz, sampling rate 100kHz) from 4 piezoelectric sensors.

[0018] The specific process for obtaining the UHF electromagnetic signal array is as follows: When the transformer is running, the UHF electromagnetic signal conditioning electronic unit monitors the signal amplitude of the 6-channel array sensor in real time. When the signal amplitude exceeds 3 times the background noise (the background noise is determined by collecting data for 1 hour during no-load operation, approximately -80dBm), the ADC module is triggered to sample. The signal amplitude is simultaneously sampled by an ADC module (sampling frequency 2GHz, resolution 12bit) to obtain 6 sampling signals; the 6 sampling signals are then transmitted to the back-end processing unit via high-speed Ethernet (1000Mbps). The background processing unit performs bandpass filtering (300MHz-1.5GHz), amplitude correction (applying gain correction coefficient), and time delay compensation (applying time delay compensation table) on the 6 sampled signals to obtain a UHF electromagnetic signal array.

[0019] The specific process for obtaining the fusion features is as follows: The core grounding current is decomposed using independent component analysis to obtain the decomposed current components. After decomposition The power spectral density satisfies At that time, harmonic interference is removed from the core grounding current to obtain the purified core current; The center frequency; Frequency offset; Current component Peak value; Current component The power spectral density.

[0020] Constructing a spatial spectrum function based on the spatial orientation characteristics of a UHF electromagnetic signal array ; R is the conjugate transpose of the manifold vector of the UHF electromagnetic signal array; R is the signal covariance matrix. The manifold vector of the UHF electromagnetic signal array; Based on the constructed spatial spectrum function, the peak value of the search spatial spectrum is used to lock the target discharge direction of the UHF electromagnetic signal, suppress the interference of the UHF electromagnetic signal in the non-target direction, and obtain the suppressed UHF electromagnetic signal.

[0021] Calculate the cross-correlation coefficient between the acoustic signal and the UHF electromagnetic signal. When the cross-correlation coefficient is greater than 0.6, it indicates that the acoustic signal is a mechanical vibration response of a real discharge. Remove the mechanical noise from the acoustic signal to obtain the acoustic signal after removal, which is the acoustic signal characteristic.

[0022] The fractal dimension FD of the purified core current is extracted, which is the current characteristic (quantizing waveform complexity: core fault (FD < 1.3), specifically: Calculating fractal dimension using box counting method ; For the number of boxes Take the logarithm; To determine the side length of the box Take the logarithm of the reciprocal; The slope is used for linear fitting; this is used to quantify the complexity of the current waveform, and thus distinguish between core faults and insulation faults (core fault FD < 1.3, insulation fault FD > 1.50). The wavelet packet energy spectrum of the suppressed UHF electromagnetic signal was extracted (distinguishing between frequency bands: core discharge energy is concentrated in 500-800MHz, and insulation discharge energy is concentrated in 800-1200MHz), specifically as follows: The suppressed UHF electromagnetic signal was subjected to 5-level wavelet packet decomposition, and the energy spectrum of the decomposed wavelet packets was calculated. ; The energy of the k-th wavelet packet subband after the suppressed UHF electromagnetic signal is decomposed into 5 layers of wavelet packets; The energy of the i-th wavelet packet subband after the suppressed UHF electromagnetic signal is decomposed into 5 layers of wavelet packets; The energy of the k-th wavelet packet subband is the proportion of the total energy of all subbands; this is determined by the wavelet packet energy spectrum. It can clearly distinguish the difference in frequency band between the iron core discharge energy concentrated in 500-800MHz and the insulation discharge concentrated in 800-1200MHz, providing a strong basis for fault diagnosis.

[0023] By fusing current characteristics and cross-modal attention mechanism The fusion features are obtained, specifically: Using current characteristics as a query, Acoustic signal features are used as key values, through the formula , ;in, For the first The output attention weight matrix of each attention head; For activation functions; For the first The query matrix corresponding to the current features in each attention head; For the first A person's attention / Transpose of the key matrix of acoustic signal features; For feature dimensions; Features of fusion; For the first A person's attention / Value matrix of acoustic signal characteristics; This is a transpose.

[0024] The specific process for obtaining the final fault coordinates is as follows: Establish a three-dimensional rectangular coordinate system; A three-dimensional rectangular coordinate system is established with the geometric center of the transformer tank as the origin (0,0,0). The three-dimensional rectangular coordinate system consists of the X-axis, Y-axis, and Z-axis. The X-axis is along the transformer axis, pointing from the high-voltage side to the low-voltage side. The Y-axis is perpendicular to the transformer axis in the horizontal plane, pointing to the front of the tank. The Z-axis is vertical, with upward as positive, and perpendicular to the ground.

[0025] The installation positions of the 6-channel array sensors in the UHF electromagnetic signal array are determined based on a three-dimensional Cartesian coordinate system, denoted as: ;in, ;Will As the first reference sensor; For the i-th sensor; Let be the coordinate value of the i-th sensor in the x-axis direction; For the i-th sensor in Coordinate values ​​along the axis; Let be the coordinate value of the i-th sensor in the z-axis direction; By synchronously sampling the partial discharge pulses of the transformer core using a 6-channel array sensor, the arrival time of the UHF electromagnetic signal at the fault point of the transformer core partial discharge pulse is extracted for each sensor: the arrival time of the UHF electromagnetic signal at the fault point is as follows: The time is The UHF electromagnetic signal at the fault point reaches the nth sensor. The time is ; ; ; The time it takes for the UHF electromagnetic signal at the fault point to reach the second sensor; The time it takes for the UHF electromagnetic signal at the fault point to reach the 6th sensor; Time difference calculation ; For sensors Arrival time With the first reference sensor Arrival time difference; The propagation speed of the UHF electromagnetic signal at the fault point in the transformer oil is calculated using the distance difference, where the distance difference is... ; For sensors With the first reference sensor The distance difference between the UHF electromagnetic signal and the fault location; The propagation speed of the UHF electromagnetic signal at the fault point in the transformer oil; Let the coordinates of the UHF electromagnetic signal at the fault point be... The fault point UHF electromagnetic signal is transmitted to the first reference sensor. The distance difference is ; , , These are the coordinates of the UHF electromagnetic signal at the fault point along the x, y, and z axes, respectively. , , The first reference sensor Coordinate values ​​along the z-axis, y-axis, and z-axis directions; The fault point UHF electromagnetic signal reaches the first One sensor The distance difference is ; , , The nth sensor Coordinate values ​​along the x, y, and z axes; The fault point's UHF electromagnetic signal and the nth sensor The geometric relationship of the time difference is as follows: ; Send the UHF electromagnetic signal from the fault point to the first reference sensor. coordinates Substituting the distance geometric relationship of the time difference, we obtain a nonlinear equation, which represents: ; Set the initial estimated fault point ; , , The initial estimated fault points are located at axis, axis, Coordinate values ​​along the axis; The nonlinear equations are simplified using a linear approximation method at the initially estimated fault points, resulting in the linearized equations: ; The coefficient matrix is ​​the initially estimated coordinates of the fault point; This is a correction vector for the initially estimated fault point coordinates; This is a constant vector representing the initially estimated coordinates of the fault point; in, ; This is the coordinate correction amount of the initially estimated fault point in the x-axis direction; This is the coordinate correction amount of the initially estimated fault point in the y-axis direction; This is the coordinate correction amount of the initially estimated fault point in the z-axis direction; The coefficient matrix of the initially estimated fault point coordinates is represented as follows: ; in, , , These are the numbers in coefficient matrix A, respectively. The elements in the first, second, and third columns of the row; From the initially estimated fault point to the nth sensor The distance; The initial estimated fault point is located at the first reference sensor. The distance; The constant vector representing the initially estimated fault point coordinates is: ; This is the i-th element in the constant term vector; Calculate the optimal approximate solution by solving the linearized equation using the pseudo-inverse matrix: ; The fault point coordinate correction vector is the one used for the optimal initial estimate. Perform the k-th iteration on the initially estimated fault point to obtain ; For the first The fault location coordinates are updated in the next iteration; Update fault location coordinates ; For the first The fault location coordinates are updated in the next iteration; when At this time That is, the final fault coordinates ;when If so, recalculate. ,until The calculation ends when the time is right. , , The final fault coordinates are respectively at axis, axis, Coordinate values ​​along the axis; For rice; Calculate the residual of the i-th equation ,express: ; when >0.05 Determine the final fault coordinates Due to interference, the final fault coordinates were removed. Recalculate the final fault coordinates until calculation <0.05 The calculation ends when the time is reached.

[0026] The specific process for obtaining acoustic positioning coordinates is as follows: Wideband acoustic emission sensors (operating frequency band 10kHz-500kHz, sensitivity ≥-65dB, resonant frequency 150kHz) are selected and installed on the upper, middle, lower and side layers of the transformer tank wall. Upper layer: Near the center of the top of the fuel tank, coordinates ; It is the first broadband acoustic emission sensor; , , The first broadband acoustic emission sensor is located at... axis, axis, The coordinate values ​​along the axis; As a second reference sensor; Middle layer: Upper middle part of the fuel tank wall, coordinates ; It is the second broadband acoustic emission sensor; , , The second broadband acoustic emission sensor is located at... axis, axis, Coordinate values ​​along the axis; Lower layer: near the center of the bottom of the fuel tank, coordinates ; It is the third broadband acoustic emission sensor; , , The third broadband acoustic emission sensor is located at axis, axis, Coordinate values ​​along the axis; Side layer: Center of the side wall of the fuel tank, coordinates; It is the fourth wideband acoustic emission sensor; , , The fourth broadband acoustic emission sensor is located at axis, axis, Coordinate values ​​along the axis; When an internal fault in the transformer (such as a loose core or insulation breakdown) generates vibration waves, these waves propagate towards the tank wall via oil. The broadband acoustic emission sensor, equipped with a high-speed ADC module (10MHz sampling rate), acquires the vibration signals in real time and uses a first-lead-edge detection algorithm to extract the arrival time of the acquired vibration signals at each broadband acoustic emission sensor. The arrival time of the vibration signal at the second reference sensor M1 is... The vibration signal reaches the d-th broadband acoustic emission sensor. The time is ; ; ; The time it takes for the vibration signal to reach the second broadband acoustic emission sensor; The time it takes for the vibration signal to reach the third broadband acoustic emission sensor; The time it takes for the vibration signal to reach the fourth broadband acoustic emission sensor; Let the fault point be The speed of sound propagating vibration waves in transformer oil =1500m / s; The fault point is at Coordinate values ​​along the axis; The fault point is at Coordinate values ​​along the axis; The fault point is at Coordinate values ​​along the axis; Calculate the vibration wave from the fault point arrive The propagation distance is: ; For the vibration wave from the fault point Propagation to the d-th broadband acoustic emission sensor The propagation distance; , , These are the d-th broadband acoustic emission sensors. exist axis, axis, Coordinate values ​​along the axis; Calculate the time difference between M2, M3, M4 and the reference sensor M1. ; The time difference between the vibration waves received by M2, M3, M4 and the reference sensor M1; Reference sensor The distance difference with M2, M3, and M4 is: ; The distance difference between reference sensors M1 and M2, M3, and M4; This is the distance the vibration wave travels from the fault point to the reference sensor M1; Based on the distance differences between reference sensors M1 and M2, M3, and M4, three independent nonlinear equations are constructed, representing: ; In the formula, The distance difference between reference sensors M1 and M2; The distance difference between reference sensors M1 and M3; The distance difference between reference sensors M1 and M3; Set initial coordinates Perform the k-th iteration on the initial coordinates to obtain... ; The coordinates of the point obtained after the k-th iteration; , , The initial coordinates are respectively at axis, axis, Coordinate values ​​along the axis; The three independent nonlinear equations are set in the initial coordinates. Simplifying the equations using a linear approximation method, we obtain the following system of linear equations: ; ; The correction vector for the initial coordinates; The coefficient matrix for the initial coordinates; A constant vector for the initial coordinates; , , The initial coordinates are respectively at axis, axis, Correction amount in the axial direction; Calculate the optimal approximate solution by solving the system of linear equations using the pseudo-inverse matrix. : ; The correction vector for the optimal initial coordinates; Update coordinates ; For the first The coordinates obtained after the next iteration; when At this time This refers to acoustic positioning coordinates. ,when Recalculate until The calculation ends when the time is right. , , The acoustic positioning coordinates are respectively at axis, axis, The coordinate values ​​along the axis.

[0027] The specific process for obtaining the final fused three-dimensional coordinates of the fault is as follows: Signal-to-noise ratio weighted fusion: This method addresses the low positioning accuracy of a single UHF electromagnetic signal, ultimately determining the fault coordinates. Harmony and acoustic positioning coordinates The fusion process yields the final fused three-dimensional coordinates of the fault, represented as follows: ; In the formula, , , The three-dimensional coordinates of the fault after final fusion are respectively in axis, axis, Coordinate values ​​along the axis; The signal-to-noise ratio of the UHF signal; The signal-to-noise ratio of the acoustic signal.

[0028] The specific process of step S3 is as follows: The GRU (Gated Recurrent Unit) model is used as input to fuse features from the past 30 minutes to predict the feature growth rate for the next hour. (Introducing total variation regularization to suppress prediction jitter), to avoid prediction results violating physical laws such as transformer heat conduction and insulation aging (e.g., excessively rapid growth exceeding the material's heat resistance limit), a loss function is constructed, representing: ; In the formula, The loss function; For the first The feature growth rate predicted by the gated cyclic unit model at each time step; For the first The true value of the feature growth rate at each time step; The regularization coefficient is used. This represents the physical upper limit of the characteristic growth rate; Combined with characteristic growth rate Cross-modal fusion feature threshold and And the location area, to construct a decision matrix; when Falling into the core fault area: <1.3 and >60%, positioned within ±5cm of the iron core post, and When the rate is <0.02 / min, the decision matrix outputs a planned maintenance diagnosis report; when Falling into the insulation fault area: >1.5 and >50%, positioned within ±10cm of the winding, and When the value is greater than 0.05 / min, the decision matrix outputs an emergency shutdown diagnostic report.

[0029] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for differentiating transformer core insulation faults using electromagnetic acoustic cross-modal fusion, characterized in that, include: Step S1: Collect the core grounding current, UHF electromagnetic signal and acoustic signal, and fuse the collected core grounding current, UHF electromagnetic signal and acoustic signal to obtain the fused features; Step S2: Process the UHF electromagnetic signal and acoustic signal to obtain the final fault coordinates and acoustic location coordinates respectively. Perform signal-to-noise ratio weighted fusion on the final fault coordinates and acoustic location coordinates to obtain the final fused three-dimensional fault coordinates. Step S3: Process the fusion features and the final fused three-dimensional coordinates of the fault, and output the final diagnostic report.

2. The method for distinguishing transformer core insulation faults by electromagnetic acoustic cross-modal fusion according to claim 1, characterized in that: The specific process of collecting the core grounding current, UHF electromagnetic signal, and acoustic signal in step S1 is as follows: Core grounding current: is obtained by collecting the original current waveform of the transformer core grounding circuit using a broadband current transformer; UHF electromagnetic signal: It is obtained by collecting partial discharge pulses of transformer core from 6 array sensors, and a UHF electromagnetic signal array is obtained based on the UHF electromagnetic signal; Acoustic signal: It is obtained by collecting vibration signals emitted by the transformer core from 4 piezoelectric sensors.

3. The method for distinguishing transformer core insulation faults by electromagnetic acoustic cross-modal fusion according to claim 2, characterized in that: The specific process of obtaining the UHF electromagnetic signal array is as follows: When the transformer is running, the UHF electromagnetic signal conditioning electronic unit monitors the signal amplitude of the 6-channel array sensor in real time. When the signal amplitude exceeds 3 times the background noise, the ADC module is triggered to sample. The signal amplitude is simultaneously sampled by the ADC module to obtain 6 sampling signals; the 6 sampling signals are then transmitted to the back-end processing unit via high-speed Ethernet. The background processing unit performs bandpass filtering, amplitude correction, and time delay compensation on the six sampled signals to obtain a UHF electromagnetic signal array.

4. The method for distinguishing transformer core insulation faults by electromagnetic acoustic cross-modal fusion according to claim 3, characterized in that: The specific process of obtaining the fusion features is as follows: The core grounding current is decomposed using independent component analysis to obtain the decomposed current components. After decomposition The power spectral density satisfies At that time, harmonic interference is removed from the core grounding current to obtain the purified core current; The center frequency; Frequency offset; Current component Peak value; Current component The power spectral density; Constructing a spatial spectrum function based on the spatial orientation characteristics of a UHF electromagnetic signal array ; R is the conjugate transpose of the manifold vector of the UHF electromagnetic signal array; R is the signal covariance matrix. The manifold vector of the UHF electromagnetic signal array; Based on the constructed spatial spectrum function, the peak value of the search spatial spectrum is used to lock the target discharge direction of the UHF electromagnetic signal, suppress the non-target direction interference of the UHF electromagnetic signal, and obtain the suppressed UHF electromagnetic signal. Calculate the cross-correlation coefficient between the acoustic signal and the UHF electromagnetic signal. When the cross-correlation coefficient is greater than 0.6, it indicates that the acoustic signal is the mechanical vibration response of a real discharge. Remove the mechanical noise from the acoustic signal to obtain the acoustic signal after removal, which is the acoustic signal characteristic. The fractal dimension FD of the purified iron core current is extracted, which is the current characteristic, specifically: Calculating fractal dimension using box counting method ; For the number of boxes Take the logarithm; To determine the side length of the box Take the logarithm of the reciprocal; The slope of the linear fit; The wavelet packet energy spectrum of the suppressed UHF electromagnetic signal is extracted as follows: The suppressed UHF electromagnetic signal was subjected to 5-level wavelet packet decomposition, and the energy spectrum of the decomposed wavelet packets was calculated. ; The energy of the k-th wavelet packet subband after the suppressed UHF electromagnetic signal is decomposed into 5 layers of wavelet packets; The energy of the i-th wavelet packet subband after the suppressed UHF electromagnetic signal is decomposed into 5 layers of wavelet packets; This represents the proportion of the energy of the k-th wavelet packet subband in the total energy of all subbands. By fusing current characteristics and cross-modal attention mechanism The fusion features are obtained, specifically: Using current characteristics as a query, Acoustic signal features are used as key values, through the formula , ;in, For the first The output attention weight matrix of each attention head; For activation functions; For the first The query matrix corresponding to the current features in each attention head; For the first A person's attention / Transpose of the key matrix of acoustic signal features; For feature dimensions; Features of fusion; For the first A person's attention / Value matrix of acoustic signal characteristics; This is a transpose.

5. The method for distinguishing transformer core insulation faults by electromagnetic acoustic cross-modal fusion according to claim 4, characterized in that: The specific process for obtaining the final fault coordinates is as follows: Establish a three-dimensional rectangular coordinate system; A three-dimensional rectangular coordinate system is established with the geometric center of the transformer tank as the origin (0,0,0). The three-dimensional rectangular coordinate system consists of an X-axis, a Y-axis, and a Z-axis. The X-axis is along the transformer axis, pointing from the high-voltage side to the low-voltage side. The Y-axis is perpendicular to the transformer axis in the horizontal plane, pointing towards the front of the tank. The Z-axis is vertical, with upward being positive and perpendicular to the ground. The installation positions of the 6-channel array sensors in the UHF electromagnetic signal array are determined based on a three-dimensional Cartesian coordinate system, denoted as: ;in, ;Will As the first reference sensor; For the i-th sensor; Let be the coordinate value of the i-th sensor in the x-axis direction; For the i-th sensor in Coordinate values ​​along the axis; Let be the coordinate value of the i-th sensor in the z-axis direction; By synchronously sampling the partial discharge pulses of the transformer core using a 6-channel array sensor, the arrival time of the UHF electromagnetic signal at the fault point of the transformer core partial discharge pulse is extracted for each sensor: the arrival time of the UHF electromagnetic signal at the fault point is as follows: The time is The UHF electromagnetic signal at the fault point reaches the nth sensor. The time is ; ; ; The time it takes for the UHF electromagnetic signal at the fault point to reach the second sensor; The time it takes for the UHF electromagnetic signal at the fault point to reach the 6th sensor; Time difference calculation ; For sensors Arrival time With the first reference sensor Arrival time difference; The propagation speed of the UHF electromagnetic signal at the fault point in the transformer oil is calculated using the distance difference, where the distance difference is... ; For sensors With the first reference sensor The distance difference between the UHF electromagnetic signal and the fault location; The propagation speed of the UHF electromagnetic signal at the fault point in the transformer oil; Let the coordinates of the UHF electromagnetic signal at the fault point be... The fault point UHF electromagnetic signal is transmitted to the first reference sensor. The distance difference is ; , , These are the coordinates of the UHF electromagnetic signal at the fault point along the x, y, and z axes, respectively. , , The first reference sensor Coordinate values ​​along the z-axis, y-axis, and z-axis directions; The fault point UHF electromagnetic signal reaches the first One sensor The distance difference is ; , , The nth sensor Coordinate values ​​along the x, y, and z axes; The fault point's UHF electromagnetic signal and the nth sensor The geometric relationship of the time difference is as follows: ; Send the UHF electromagnetic signal from the fault point to the first reference sensor. coordinates Substituting the distance geometric relationship of the time difference, we obtain a nonlinear equation, which represents: ； Set the initial estimated fault point ; , , The initial estimated fault points are located at axis, axis, Coordinate values ​​along the axis; The nonlinear equations are simplified using a linear approximation method at the initially estimated fault points, resulting in the linearized equations: ; The coefficient matrix is ​​the initially estimated coordinates of the fault point; This is a correction vector for the initially estimated fault point coordinates; This is a constant vector representing the initially estimated coordinates of the fault point; in, ; This is the coordinate correction amount of the initially estimated fault point in the x-axis direction; This is the coordinate correction amount of the initially estimated fault point in the y-axis direction; This is the coordinate correction amount of the initially estimated fault point in the z-axis direction; The coefficient matrix of the initially estimated fault point coordinates is represented as follows: ； in, , , These are the numbers in coefficient matrix A, respectively. The elements in the first, second, and third columns of the row; From the initially estimated fault point to the nth sensor The distance; The initial estimated fault point is located at the first reference sensor. The distance; The constant vector representing the initially estimated fault point coordinates is: ; This is the i-th element in the constant term vector; Calculate the optimal approximate solution by solving the linearized equation using the pseudo-inverse matrix: ; The fault point coordinate correction vector is the one used for the optimal initial estimate. Perform the k-th iteration on the initially estimated fault point to obtain ; For the first The fault location coordinates are updated in the next iteration; Update fault location coordinates ; For the first The fault location coordinates are updated in the next iteration; when At this time That is, the final fault coordinates ;when If so, recalculate. ,until The calculation ends when the time is right. , , The final fault coordinates are respectively at axis, axis, Coordinate values ​​along the axis; For rice.

6. The method for distinguishing transformer core insulation faults by electromagnetic acoustic cross-modal fusion according to claim 5, characterized in that: The specific process of obtaining acoustic positioning coordinates is as follows: Wideband acoustic emission sensors were selected and installed on the upper, middle, lower and side layers of the transformer tank wall; Upper layer: Near the center of the top of the fuel tank, coordinates ; It is the first broadband acoustic emission sensor; , , The first broadband acoustic emission sensor is located at... axis, axis, The coordinate values ​​along the axis; As a second reference sensor; Middle layer: Upper middle part of the fuel tank wall, coordinates ; It is the second broadband acoustic emission sensor; , , The second broadband acoustic emission sensor is located at... axis, axis, Coordinate values ​​along the axis; Lower layer: near the center of the bottom of the fuel tank, coordinates ; It is the third broadband acoustic emission sensor; , , The third broadband acoustic emission sensor is located at axis, axis, Coordinate values ​​along the axis; Side layer: Center of the side wall of the fuel tank, coordinates; It is the fourth wideband acoustic emission sensor; , , The fourth broadband acoustic emission sensor is located at axis, axis, Coordinate values ​​along the axis; When a fault inside the transformer generates vibration waves, these waves propagate towards the tank wall via oil. The broadband acoustic emission sensor, equipped with a high-speed ADC module, acquires the vibration signals in real time and uses a first-lead-edge detection algorithm to extract the arrival time of the acquired vibration signals at each broadband acoustic emission sensor. The arrival time of the vibration signal at the second reference sensor M1 is... The vibration signal reaches the d-th broadband acoustic emission sensor. The time is ; ; ; The time it takes for the vibration signal to reach the second broadband acoustic emission sensor; The time it takes for the vibration signal to reach the third broadband acoustic emission sensor; The time it takes for the vibration signal to reach the fourth broadband acoustic emission sensor; Let the fault point be The speed of sound propagating vibration waves in transformer oil =1500m / s; The fault point is at Coordinate values ​​along the axis; The fault point is at Coordinate values ​​along the axis; The fault point is at Coordinate values ​​along the axis; Calculate the vibration wave from the fault point arrive The propagation distance is: ; For the vibration wave from the fault point Propagation to the d-th broadband acoustic emission sensor The propagation distance; , , These are the d-th broadband acoustic emission sensors. exist axis, axis, Coordinate values ​​along the axis; Calculate the time difference between M2, M3, M4 and the reference sensor M1. ; The time difference between the vibration waves received by M2, M3, M4 and the reference sensor M1; Reference sensor The distance difference with M2, M3, and M4 is: ; The distance difference between reference sensors M1 and M2, M3, and M4; This is the distance the vibration wave travels from the fault point to the reference sensor M1; Based on the distance differences between reference sensors M1 and M2, M3, and M4, three independent nonlinear equations are constructed, representing: ; In the formula, The distance difference between reference sensors M1 and M2; The distance difference between reference sensors M1 and M3; The distance difference between reference sensors M1 and M3; Set initial coordinates Perform the k-th iteration on the initial coordinates to obtain... ; The coordinates of the point obtained after the k-th iteration; , , The initial coordinates are respectively at axis, axis, Coordinate values ​​along the axis; The three independent nonlinear equations are set in the initial coordinates. Simplifying the equations using a linear approximation method, we obtain the following system of linear equations: ; ; The correction vector for the initial coordinates; The coefficient matrix for the initial coordinates; A constant vector for the initial coordinates; , , The initial coordinates are respectively at axis, axis, Correction amount in the axial direction; Calculate the optimal approximate solution by solving the system of linear equations using the pseudo-inverse matrix. : ; The correction vector for the optimal initial coordinates; Update coordinates ; For the first The coordinates obtained after the next iteration; when At this time This refers to acoustic positioning coordinates. ,when Recalculate until The calculation ends when the time is right. , , The acoustic positioning coordinates are respectively at axis, axis, The coordinate values ​​along the axis.

7. The method for distinguishing transformer core insulation faults by electromagnetic acoustic cross-modal fusion according to claim 6, characterized in that: The specific process for obtaining the final fused three-dimensional coordinates of the fault is as follows: Final fault coordinates Harmony and acoustic positioning coordinates The fusion process yields the final fused three-dimensional coordinates of the fault, represented as follows: ； In the formula, , , The three-dimensional coordinates of the fault after final fusion are respectively in axis, axis, Coordinate values ​​along the axis; The signal-to-noise ratio of the UHF signal; The signal-to-noise ratio of the acoustic signal.

8. The method for distinguishing transformer core insulation faults by electromagnetic acoustic cross-modal fusion according to claim 7, characterized in that: The specific process of step S3 is as follows: Using a gated recurrent unit model to input fused features to predict the feature growth rate for the next hour. ; Combined with characteristic growth rate Cross-modal fusion feature threshold and And the location area, to construct a decision matrix; when Falling into the core fault area: <1.3 and >60%, positioned within ±5cm of the iron core post, and When the rate is <0.02 / min, the decision matrix outputs a planned maintenance diagnosis report; when Falling into the insulation fault area: >1.5 and >50%, positioned within ±10cm of the winding, and When the value is greater than 0.05 / min, the decision matrix outputs an emergency shutdown diagnostic report.

Images