Partial discharge positioning and identification method and system based on multi-modal fusion

Abstract

The embodiment of the invention provides a partial discharge positioning and identification method and system based on multi-modal fusion. The method comprises the following steps: acquiring an ultrahigh frequency signal of a UHF sensor array and an ultrasonic signal of an AE sensor array; calculating a signal starting point of an ultrahigh frequency signal in each UHF sensor, taking the UHF sensor with the smallest signal starting point as a reference sensor I, calculating a relative time difference with other UHF sensors, further solving liberation potential coarse positioning, and calculating a signal time window by combining the position of the AE sensor array; in the signal time window, ultrasonic signals are extracted, the relative time difference between other AE sensors and the second reference sensor is calculated, then a positioning model of the discharge potential is established, and fine positioning of the discharge potential is solved; and extracting a corresponding ultrahigh frequency feature vector, extracting an ultrasonic feature vector, and carrying out weighted fusion to identify the discharge type of the discharge potential.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, and in particular to a method and system for locating and identifying partial discharge based on multimodal fusion. Background Technology

[0002] During long-term operation, electrical equipment may experience partial discharge (PD) due to insulation aging, mechanical vibration, and environmental factors. Partial discharge is both an early sign of insulation degradation and a significant factor accelerating insulation damage. Failure to detect and locate discharge defects in a timely manner can lead to equipment failure or even major power accidents, threatening the safe and stable operation of the power grid. Therefore, partial discharge detection and location technology has always been one of the core research directions in the field of condition monitoring of high-voltage electrical equipment.

[0003] Traditional partial discharge detection methods mainly include pulsed current methods, ultrasonic methods (Acoustic Emission, AE), and ultra-high frequency (UHF) methods. While pulsed current methods offer high sensitivity, they are susceptible to electromagnetic interference and struggle with spatial localization. Ultrasonic methods achieve non-invasive measurement by detecting the acoustic signals generated by the discharge, but attenuation, refraction, and multipath effects during propagation within the equipment limit localization accuracy. UHF methods utilize electromagnetic signals excited by the discharge, offering advantages such as strong anti-interference capabilities and fast propagation speed; however, in complex equipment structures, electromagnetic wave reflection and scattering can affect localization accuracy. A single detection method often struggles to balance high-precision localization with strong anti-interference capabilities, resulting in significant uncertainty in on-site detection results.

[0004] In recent years, multimodal fusion technology has provided a new approach to partial discharge detection by combining sensor data with different physical characteristics to improve detection reliability and positioning accuracy. For example, simultaneously using ultrasonic and UHF signals can leverage the complementary advantages of AE (early-field acoustic wave) for high short-range positioning accuracy and UHF for strong long-range detection capabilities. However, effective fusion of multimodal data still faces many challenges. First, the different propagation mechanisms of ultrasonic and UHF signals result in significant differences in signal arrival time, amplitude attenuation, and other characteristics, making it difficult to directly apply traditional time-difference positioning algorithms. Second, the complex internal structure of power equipment means that the propagation paths of sound waves and electromagnetic waves may be affected by factors such as insulating media and metal casings, causing signal distortion and affecting the accuracy of the positioning model. Furthermore, most existing multi-sensor fusion methods focus only on the positioning aspect and fail to organically combine the spatial location information of the discharge point with discharge type identification, resulting in the detection system still relying on human experience for fault diagnosis and reducing the level of automation.

[0005] In discharge type identification, current methods mainly rely on partial discharge phase distribution (PRPD) maps or time-frequency analysis techniques, classifying discharges by statistically analyzing features such as discharge amplitude, phase, and frequency. However, feature extraction from a single sensor often fails to fully reflect the physical nature of the discharge. For example, corona discharge and surface discharge may appear similar in ultra-high frequency signals, but they differ significantly in the ultrasonic spectrum. Furthermore, traditional pattern recognition methods (such as support vector machines and artificial neural networks) are highly dependent on feature engineering and have limited generalization ability under small sample conditions, making them difficult to adapt to the complex and varied discharge types encountered in the field. Summary of the Invention

[0006] To address the problems existing in the prior art, embodiments of the present invention provide a method and system for partial discharge localization and identification based on multimodal fusion.

[0007] This invention provides a method for partial discharge localization and identification based on multimodal fusion, the method comprising:

[0008] The ultra-high frequency signals of the UHF sensor array deployed around the discharge point and the ultrasonic signals of the AE sensor array are collected, and the ultra-high frequency signals and ultrasonic signals are preprocessed.

[0009] Based on the AIC signal criterion, the signal starting point of the UHF signal in each UHF sensor is calculated. The UHF sensor with the smallest signal starting point is taken as the reference sensor one. The relative time difference between the other UHF sensors and the reference sensor one is calculated. Then, a mathematical model related to the discharge potential coordinates is established to obtain the coarse location of the discharge potential. Combined with the position of the AE sensor array, the signal time window is calculated.

[0010] Within the signal time window, the ultrasonic signal is extracted. Using the reference sensor two corresponding to the signal with the highest signal-to-noise ratio, the relative time difference between other AE sensors and the reference sensor two is calculated. Then, a positioning model of the discharge position is established. Using the coarse positioning of the discharge position as the initial iteration value, the fine positioning of the discharge potential is obtained.

[0011] The discharge pulses of the ultra-high frequency signal are collected, a PRPD spectrum is constructed, and the corresponding ultra-high frequency feature vector is extracted by combining it with a convolutional neural network. The time-frequency diagram of the ultrasonic signal of the second reference sensor is combined with the convolutional neural network training to extract the ultrasonic feature vector. The ultra-high frequency feature vector and the ultrasonic feature vector are weighted and fused to identify the discharge type of the discharge site.

[0012] In one embodiment, the calculation of the signal start point of the ultra-high frequency signal in each UHF sensor based on the AIC signal criterion includes:

[0013] Within the search window of each UHF sensor, the AIC function is calculated, including:

[0014] ,

[0015] in, For time series indexing, For window length, For variance, Let be a subsequence from 1 to j.

[0016] In one embodiment, the method further includes:

[0017] Find the UHF sensor with the smallest signal start point, and calculate the relative time difference between other UHF sensors and reference sensor one, including:

[0018]

[0019] in, For relative time difference, For reference to the relative time of sensor one, ,in, The number of channels in the UHF sensor array;

[0020] Establish the TDOA positioning equation system, and assume the coordinates of the discharge point are... , No. The coordinates of the UHF sensors are Then the distance between the two have:

[0021]

[0022] The following system of equations is then established:

[0023]

[0024] in , At the speed of light, For the first The distance difference between the first sensor and the second sensor;

[0025] The coarse location result of the discharge point is obtained by solving the problem.

[0026] .

[0027] In one embodiment, the method further includes:

[0028] Calculating the earliest arrival time of the UHF signal includes:

[0029]

[0030] in, This refers to the earliest arrival time of the UHF signal;

[0031] Estimate the distance from the coarse location of the discharge point to the farthest AE sensor, and calculate the signal time window, including:

[0032]

[0033] in, For protection intervals, To determine the distance from the coarse location of the discharge point to the farthest AE sensor, This refers to the speed at which sound waves propagate in the medium of the device.

[0034] In one embodiment, the method further includes:

[0035]

[0036] Where m is the second reference sensor, and k is another AE sensor. Let be the cross-correlation function between the two. The ultrasonic signal for each AE channel;

[0037] The calculation is accelerated using Fast Fourier Transform, including:

[0038]

[0039] And weighted:

[0040]

[0041] Search Record the time-shift index corresponding to the peak point with the largest absolute value. Then the time difference between the other AE sensors and the second reference sensor is:

[0042]

[0043] in The sampling rate.

[0044] In one embodiment, the method further includes:

[0045] Based on the conversion of the time difference set of reference sensor 2 into range difference, the equation includes:

[0046]

[0047] And define geometric distance , Construct a nonlinear least squares objective function:

[0048]

[0049] Based on the hyperbolic positioning principle, a mapping model from time difference measurement to spatial coordinates is established.

[0050] In one embodiment, the method further includes:

[0051] Iterative optimization based on the LM algorithm includes:

[0052]

[0053] in, The least squares objective function At the current point Jacobian matrix, The initial iteration value, It is the coordinate correction amount. For the residual vector, , , It is the damping factor;

[0054] The iterative optimization process begins with the initial iteration value and continues until... If the value is less than the preset threshold, or if the maximum number of iterations is reached, then... This is the final high-precision positioning result. .

[0055] This invention provides a partial discharge localization and identification system based on multimodal fusion, the system comprising:

[0056] The acquisition module is used to acquire the ultra-high frequency signals of the UHF sensor array deployed around the discharge position and the ultrasonic signals of the AE sensor array, and to preprocess the ultra-high frequency signals and ultrasonic signals.

[0057] The coarse positioning module is used to calculate the signal start point of the ultra-high frequency signal in each UHF sensor based on the AIC signal criterion. Taking the UHF sensor with the smallest signal start point as the reference sensor one, it calculates the relative time difference between the other UHF sensors and the reference sensor one, and then establishes a mathematical model related to the discharge potential coordinates to obtain the coarse positioning of the discharge potential. Combined with the position of the AE sensor array, it calculates the signal time window.

[0058] The fine positioning module is used to extract ultrasonic signals within the signal time window, calculate the relative time difference between other AE sensors and the reference sensor two with the signal with the highest signal-to-noise ratio, and then establish a positioning model of the discharge position. The coarse positioning of the discharge position is used as the initial iteration value to obtain the fine positioning of the discharge potential.

[0059] The identification module is used to collect the discharge pulses of the ultra-high frequency signal, construct the PRPD spectrum, extract the corresponding ultra-high frequency feature vector by combining it with a convolutional neural network, extract the ultrasonic feature vector by combining the time-frequency diagram of the ultrasonic signal of the second reference sensor with the convolutional neural network training, and perform weighted fusion of the ultra-high frequency feature vector and the ultrasonic feature vector to identify the discharge type of the discharge position.

[0060] This invention provides an electronic device, including a processor and a memory;

[0061] The processor is connected to the memory;

[0062] The memory is used to store executable program code;

[0063] The processor runs a program corresponding to the executable program code stored in the memory to perform the methods described in one or more embodiments.

[0064] This invention provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method for partial discharge localization and identification based on multimodal fusion.

[0065] In view of the above, in one or more embodiments of this specification, ultra-high frequency signals from a UHF sensor array deployed around the discharge site and ultrasonic signals from an AE sensor array are collected and preprocessed. Based on the AIC signal criterion, the signal starting point of the ultra-high frequency signal in each UHF sensor is calculated. The UHF sensor with the smallest signal starting point is used as reference sensor one. The relative time difference between other UHF sensors and reference sensor one is calculated, thereby establishing a mathematical model relating to the discharge site coordinates to obtain a coarse location of the discharge potential. Combined with the position of the AE sensor array, a signal time window is calculated. Within the signal time window, ultrasonic signals are extracted. Using reference sensor two corresponding to the signal with the highest signal-to-noise ratio, the relative time difference between other AE sensors and reference sensor two is calculated, thereby establishing a location model for the discharge site. The coarse location of the discharge site is used as the initial iteration value to obtain a fine location of the discharge potential. Discharge pulses of the ultra-high frequency signals are collected, and a PRPD spectrum is constructed. Combined with a convolutional neural network, the corresponding ultra-high frequency feature vectors are extracted. The time-frequency graph of the ultrasonic signal from reference sensor two is combined with convolutional neural network training to extract ultrasonic feature vectors. The ultra-high frequency feature vectors and ultrasonic feature vectors are weighted and fused to identify the discharge type of the discharge site. This allows for the determination of the approximate time window and region of discharge occurrence using UHF signals (coarse localization), and the precise time difference calculation using AE signals within this time window to achieve high-precision three-dimensional localization (fine localization). It combines long-range detection capability with short-range positioning accuracy, resolving the scale contradiction that cannot be addressed by a single method, and improving the system's resistance to false alarms and detection reliability. Furthermore, depth features characterizing the discharge type are extracted from both UHF and AE signals, fused at the feature layer, and input into a classifier to achieve automatic and accurate identification of the discharge type. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a flowchart of a partial discharge localization and identification method based on multimodal fusion provided in one embodiment of this specification.

[0068] Figure 2 This is a schematic diagram of a partial discharge localization and identification system based on multimodal fusion, provided in one embodiment of this specification.

[0069] Figure 3 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this specification. Detailed Implementation

[0070] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed merely to enable those skilled in the art to better understand and implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. The function and arrangement of the elements discussed may be changed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the various examples. For example, the described methods may be performed in a different order than described, and steps may be added, omitted, or combined. Furthermore, features described in some examples may be combined in other examples.

[0071] As used herein, the term "comprising" and its variations are open terms meaning "including but not limited to". The term "based on" means "at least partially based on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other definitions, whether explicit or implicit, may be included below. Unless explicitly indicated by the context, the definition of a term shall remain consistent throughout the specification.

[0072] like Figure 1 As shown, this embodiment of the invention provides a method for localizing and identifying partial discharges based on multimodal fusion, including:

[0073] Step S102: Collect the ultra-high frequency signal from the UHF sensor array deployed around the discharge position and the ultrasonic signal from the AE sensor array, and preprocess the ultra-high frequency signal and the ultrasonic signal.

[0074] Specifically, a UHF sensor array and an AE sensor array are arranged around the discharge sites where partial discharge may occur. The sensor arrays are arranged in the following manner to meet the computational requirements of subsequent steps:

[0075] UHF sensor array:

[0076] At least four (4-6 recommended) high-sensitivity UHF sensors (e.g., microstrip patch antennas or Archimedes' spiral antennas) can be used, with an effective frequency band typically between 300MHz and 1500MHz. The sensors are arranged non-contactly at specific measurement points on the power equipment under test (e.g., GIS basin insulators, transformer tank walls), forming a three-dimensional spatial array. The spatial coordinates of the sensors are... (in Precise measurements and data entry are performed in advance using a high-precision measuring instrument (such as a total station).

[0077] AE sensor array:

[0078] At least four (4-8 recommended) broadband acoustic emission sensors (such as PZT piezoelectric sensors) with resonant frequencies around 150kHz can be used. The sensors are tightly mounted on the surface of the device under test housing using an ultrasonic coupling agent (Vaseline or a special grease), forming another spatial array. Its coordinates... (in Similarly, accurate measurement and data entry are required.

[0079] In addition, during data acquisition, the signal lines of all UHF and AE sensors can be connected to the same timing unit to ensure that the sampling clocks of all channels are strictly synchronized, with a sampling error of less than 1 nanosecond. When the amplitude of any UHF sensor signal exceeds a preset threshold, a global trigger signal is generated, activating all channels to record data simultaneously.

[0080] Furthermore, the acquired signals require preprocessing to enhance the signal-to-noise ratio (SNR). For UHF signals, a bandpass filter of 300MHz to 1500MHz can be used to effectively filter out out-of-band interference such as power frequency harmonics and carrier communication. To further improve the SNR, discrete wavelet transform is used for soft-threshold denoising. A 5-level decomposition is performed using the 'sym4' wavelet as the basis function. The detail coefficients are processed using the 'rigrsure' (based on Stein's unbiased risk estimation) thresholding rule, and then the signal is reconstructed. This effectively preserves the signal's edge features (such as pulse start points). The preprocessed UHF signal is labeled as follows. .

[0081] For ultrasonic signals (AE signals), a bandpass filter of 20kHz~400kHz can be used to remove low-frequency mechanical vibrations and high-frequency electromagnetic interference. Subsequently, wavelet threshold denoising (e.g., using the 'db5' wavelet) is employed to primarily remove high-frequency scattering noise and periodic interference generated during propagation. The preprocessed AE signal is denoted as... .

[0082] Step S104: Calculate the signal start point of the UHF signal in each UHF sensor based on the AIC signal criterion. Take the UHF sensor with the smallest signal start point as the reference sensor one. Calculate the relative time difference between the other UHF sensors and the reference sensor one. Then, establish a mathematical model related to the discharge potential coordinates to obtain the coarse location of the discharge potential. Combined with the position of the AE sensor array, calculate the signal time window.

[0083] Specifically, the AIC (Akaike Information Criterion) segmentation method has high accuracy in identifying the starting point of weak signals for each channel in a UHF sensor array. The AIC function can be calculated within a short search window (typically 50 ns before and after the trigger point). The AIC function is defined as:

[0084] ,

[0085] in, It is a time series index. It is the window length. Represents variance. Signal origin. This is the time point corresponding to the minimum point of the AIC function (the global minimum point of the AIC function sequence summary). It is the first subsequence, which is the subsequence from the beginning of the sequence (index 1) to the candidate segment point j. If j is indeed the starting point of the signal, then this segment is pure noise or background signal. It is the second subsequence, from the candidate segmentation point j+1 to the end of the sequence N. If j is indeed the signal starting point, then this segment should be the signal plus noise, so as to accurately identify the absolute moment of arrival of the partial discharge electromagnetic pulse from the waveform of each UHF sensor.

[0086] Then, the UHF sensor with the smallest signal start point is selected as the reference sensor. Calculate the relative time difference between all other sensors and the reference sensor:

[0087] ,

[0088] in, ,

[0089] Then, a mathematical model relating to the discharge point coordinates is established, namely the TDOA positioning equation system. Let the discharge point coordinates be... , No. The coordinates of the sensors are Then the distance have:

[0090]

[0091] The following system of equations can be established:

[0092]

[0093] in , The speed of light (in GIS, it needs to be slightly lower than the speed of light in a vacuum, usually taken as...) ), For the first The distance difference between the first sensor and the second sensor.

[0094] The above-mentioned nonlinear equations can be solved using the Chan algorithm, which exhibits good statistical performance when the error follows a Gaussian distribution. The Chan algorithm transforms the nonlinear equations into a two-step least squares problem by introducing intermediate variables, ultimately obtaining the coarse location result of the discharge point. By utilizing the time difference between signals received by multiple spatially distributed UHF sensors and applying the principle of geometric hyperboloid intersection, the three-dimensional spatial location (coarse positioning) of the discharge point can be calculated.

[0095] Furthermore, given that the UHF signal arrives at all sensors almost instantaneously, while the AE signal propagates at the speed of sound and has a significant delay, the absolute time reference provided by UHF can be used to accurately predict the time range in which the AE signal should be located.

[0096] Calculate the earliest arrival time of a UHF signal :

[0097]

[0098] Estimate from the coarse location of the discharge Distance to the farthest AE sensor Combined with the speed of sound wave propagation in the device medium (For example, approximately 140-220 m / s in GIS SF6 and approximately 1400 m / s in transformer oil), then the time window for the AE signal is:

[0099]

[0100] in, A small guard interval (e.g., 1 μs) is used to compensate for minor errors in UHF-TOA extraction and velocity calibration. This allows the use of the precise absolute time reference provided by UHF and the spatial prior provided by coarse positioning to predict the time period during which the AE signal will appear, thus strictly limiting the analysis of the AE signal to this extremely short time window. This completely abandons the traditional threshold triggering mode for AE signals, compressing the search range from the millisecond level to the microsecond level, greatly avoiding interference from non-partial discharge sources.

[0101] Step S106: Within the signal time window, extract the ultrasonic signal, use the reference sensor two corresponding to the signal with the highest signal-to-noise ratio to calculate the relative time difference between other AE sensors and the reference sensor two, and then establish a positioning model of the discharge position. Using the coarse positioning of the discharge position as the initial iteration value, calculate the fine positioning of the discharge potential.

[0102] Specifically, after the signal time window in which the AE signal will appear is determined, the AE signal for each AE channel is... The process involves several steps. Because the acoustic signal frequency is low and the waveform starts oscillating slowly, a cross-correlation method is used to accurately calculate the time difference between sensor pairs, rather than absolute times. First, the signal-to-noise ratio (SNR) of each AE channel within the time window is calculated. The AE channel with the highest SNR is selected as the main channel (assumed to be channel m, serving as reference sensor two; typically, the sensor closest to the discharge point or with the best coupling has the highest SNR). Then, the cross-correlation function between each of the other channels k and the main channel m is calculated. :

[0103]

[0104] In practical digital processing, the Fast Fourier Transform (FFT) is used to accelerate calculations:

[0105]

[0106] To suppress noise, weighting is typically performed in the frequency domain (i.e., generalized cross-correlation), for example using PHAT (Phase Transform) weighting:

[0107]

[0108] PHAT weighting causes the cross-correlation function to exhibit a very sharp peak at the time delay, which is very effective for reverberation and fading in specific frequency bands.

[0109] turn up Record the time-shift index corresponding to the peak point with the largest absolute value. Then the sensor With the main sensor The time difference between them is:

[0110]

[0111] in It is the sampling rate. The relative time difference of arrival (TDOA) between sensor pairs is directly calculated, avoiding the inaccuracies of extracting the absolute time of arrival (TOA) from complex acoustic waveforms.

[0112] Furthermore, an AE-TDOA localization model for the discharge potential is established:

[0113] Based on the time difference set with sensor m as the reference Converting to distance difference, the equations include:

[0114]

[0115] in Define the geometric distance as the speed at which sound waves propagate within the current cutoff point of the device. , .

[0116] Construct a nonlinear least squares objective function:

[0117]

[0118] Based on the hyperbolic positioning principle, a mapping model from time difference measurement to spatial coordinates is established.

[0119] Furthermore, since AE-TDOA positioning is a highly non-convex problem, it contains numerous local minima. If the initial values ​​are improperly chosen (e.g., set as the coordinate origin or device center), the optimization algorithm is highly susceptible to getting trapped in erroneous local solutions, leading to positioning failure. Therefore, the UHF coarse positioning results... As a solution for nonlinear optimization algorithms initial value of iteration , It is already very close to the true solution. Starting from this point, we can ensure that the optimization process converges quickly and stably to the vicinity of the global optimum.

[0120] The Levenberg-Marquardt (LM) algorithm is used for iterative optimization. The LM algorithm combines the Gauss-Newton method and the steepest descent method, effectively handling nonlinear least squares problems. Its iterative formula is as follows:

[0121]

[0122] in, It is the objective function At the current point The Jacobian matrix. It is the coordinate correction amount.

[0123] It is the residual vector. ,

[0124] It is the damping factor, when When the value is very small, the LM algorithm is close to the Gauss-Newton method and converges quickly; when... When the value is large, it approximates gradient descent, exhibiting robust convergence.

[0125] Iterative process from Beginning, until If the value is less than a preset threshold (e.g., 1e-6 meters), or if the maximum number of iterations is reached, then the result is... This is the final high-precision positioning result. This allows us to utilize the high-quality initial values ​​provided by UHF. To avoid getting trapped in erroneous local minima during the optimization process, ensuring convergence to near the true solution, and utilizing the LM algorithm to balance convergence speed and stability, a high-precision solution is quickly obtained. This achieves centimeter-level positioning accuracy, far exceeding the coarse positioning accuracy of UHF. Furthermore, it boasts rapid convergence and high computational efficiency. It realizes the technical effect of UHF providing "direction guidance" while AE performs "precise aiming."

[0126] Step S108: Collect the discharge pulses of the UHF signal, construct the PRPD spectrum, combine it with a convolutional neural network, extract the corresponding UHF feature vector, combine the time-frequency diagram of the ultrasonic signal of the second reference sensor with the convolutional neural network training, extract the ultrasonic feature vector, perform weighted fusion of the UHF feature vector and the ultrasonic feature vector, and identify the discharge type of the discharge site.

[0127] Specifically, after the discharge event is located, the discharge type can be further identified. This first includes depth feature extraction of both UHF and AE signals, including:

[0128] UHF signal depth feature extraction.

[0129] The waveforms of the first pulse of the UHF signal and its subsequent pulses for a certain duration are captured. Hundreds of such discharge pulses are collected, and their phase-amplitude distributions are constructed to generate a PRPD (phase-resolved partial discharge) spectrum. This spectrum uses the power frequency phase (0°-360°) as the horizontal axis and the pulse amplitude as the vertical axis, with pulse density represented by color intensity.

[0130] Then, a lightweight convolutional neural network (CNN) is constructed for automatic feature extraction. The input to this CNN is the normalized PRPD spectral map (sized 224x224 pixels), and the output is a 256-dimensional vector as the depth feature representation of the UHF signal. This vector encapsulates key mode information such as the "cloud" shape, distribution phase, and amplitude concentration of different discharge types as manifested in the PRPD spectrum.

[0131] AE signal depth feature extraction.

[0132] Another CNN is constructed, whose input is a time-frequency plot of the complete waveform of the channel with the highest signal-to-noise ratio in the AE sensor array (e.g., generated via Short-Time Fourier Transform (STFT)). The time-frequency plot simultaneously reveals the time and frequency domain characteristics of the signal, making it ideal for characterizing the acoustic wave patterns of the discharge. Similarly, the output of the last fully connected layer of this network (a 128-dimensional vector) is used as the depth feature representation of the AE signal. This vector characterizes the physical features of the discharge acoustic signal, such as its frequency components, energy attenuation characteristics, and modes.

[0133] Then, the UHF and ultrasound feature vectors are weighted and fused to balance information completeness and the contribution of different modalities. First, the UHF and ultrasound feature vectors are weighted and fused together. and Perform L2 normalization separately, then project the results onto a unit sphere to eliminate feature scale differences.

[0134]

[0135]

[0136] Different types of discharges exhibit varying characterization capabilities for their UHF and AE signals. For example, corona discharges exhibit distinct UHF signal characteristics, while the AE signals of certain internal discharges may be more discriminative. Therefore, appropriate weights should be assigned to different modalities. Fusion weights should be assigned to the two modalities based on prior knowledge or attention weights accumulated during network training. and (For example, =0.6, =0.4). Finally, the pieces are joined together:

[0137]

[0138] Among them, the 384-dimensional fusion feature vector It not only includes bimodal information, but also highlights modal features that are more important for the current classification task through an attention mechanism.

[0139] Finally, the fused feature vectors are input into a fully connected neural network classifier for discharge type identification. The output layer uses the Softmax activation function to output a probability vector. Where C represents the total number of discharge types. The final identification result is the discharge type with the highest probability. Thus, by fusing electromagnetic and acoustic features, a more comprehensive description of the discharge event is formed, overcoming the limitation that single modes have similar features in certain types and are difficult to distinguish, and realizing the complementary enhancement of multimodal information.

[0140] This invention provides a method for partial discharge localization and identification based on multimodal fusion. The method collects ultra-high frequency (UHF) signals from a UHF sensor array deployed around the discharge site and ultrasonic signals from an AE sensor array. The UHF and ultrasonic signals are preprocessed. Based on the AIC signal criterion, the signal start point of the UHF signal in each UHF sensor is calculated. The UHF sensor with the smallest signal start point is used as reference sensor one. The relative time difference between the other UHF sensors and reference sensor one is calculated, and a mathematical model relating to the discharge site coordinates is established to obtain a coarse location of the discharge potential. Finally, the signal time is calculated in conjunction with the position of the AE sensor array. Within a signal time window, ultrasonic signals are extracted. Using the reference sensor with the highest signal-to-noise ratio (SNR) as the reference sensor, the relative time difference between other AE sensors and the reference sensor is calculated to establish a discharge location model. Coarse discharge location is used as the initial iteration value to calculate the release potential for fine location. Discharge pulses from UHF signals are collected, and a PRPD spectrum is constructed. Combined with a convolutional neural network, corresponding UHF feature vectors are extracted. The time-frequency map of the ultrasonic signal from the reference sensor is combined with the convolutional neural network training to extract ultrasonic feature vectors. The UHF and ultrasonic feature vectors are weighted and fused to identify the discharge type of the discharge location. This allows for the determination of the approximate time window and region of the discharge using UHF signals (coarse location), and precise time difference calculation using AE signals within this time window to achieve high-precision three-dimensional location (fine location). This approach combines long-range detection capability with short-range positioning accuracy, resolving the scale contradiction that a single method cannot simultaneously address, and improving the system's false alarm resistance and detection reliability. Furthermore, deep features characterizing the discharge type are extracted from both UHF and AE signals, fused at the feature layer, and input into a classifier to achieve automatic and accurate discharge type identification.

[0141] Please see Figure 2 , Figure 2 This is a schematic diagram of a partial discharge localization and identification system based on multimodal fusion provided in an embodiment of this application. Figure 2 As shown, the system includes:

[0142] The acquisition module S202 is used to acquire the ultra-high frequency signals of the UHF sensor array deployed around the discharge position and the ultrasonic signals of the AE sensor array, and to preprocess the ultra-high frequency signals and ultrasonic signals.

[0143] The coarse positioning module S204 is used to calculate the signal starting point of the UHF signal in each UHF sensor based on the AIC signal criterion, take the UHF sensor with the smallest signal starting point as the reference sensor one, calculate the relative time difference between the other UHF sensors and the reference sensor one, and then establish a mathematical model related to the discharge potential coordinates to obtain the coarse positioning of the discharge potential, and calculate the signal time window in combination with the position of the AE sensor array.

[0144] The fine positioning module S206 is used to extract the ultrasonic signal within the signal time window, calculate the relative time difference between other AE sensors and the reference sensor two with the reference sensor two corresponding to the signal with the highest signal-to-noise ratio, and then establish a positioning model of the discharge position. The fine positioning of the discharge position is obtained by taking the coarse positioning of the discharge position as the initial iteration value.

[0145] The identification module S208 is used to collect the discharge pulses of the ultra-high frequency signal, construct the PRPD spectrum, extract the corresponding ultra-high frequency feature vector by combining it with a convolutional neural network, extract the ultrasonic feature vector by combining the time-frequency diagram of the ultrasonic signal of the second reference sensor with the convolutional neural network training, and perform weighted fusion of the ultra-high frequency feature vector and the ultrasonic feature vector to identify the discharge type of the discharge position.

[0146] Those skilled in the art will clearly understand that the technical solutions of the embodiments of this application can be implemented by means of software and / or hardware. In this specification, "unit" and "module" refer to software and / or hardware that can independently complete or cooperate with other components to complete a specific function, wherein the hardware may be, for example, a field-programmable gate array (FPGA), an integrated circuit (IC), etc.

[0147] Each processing unit and / or module in the embodiments of this application can be implemented by an analog circuit that implements the functions described in the embodiments of this application, or by software that executes the functions described in the embodiments of this application.

[0148] See Figure 3 It shows a schematic diagram of the structure of an electronic device according to an embodiment of this application, which can be used to implement... Figure 1 The method in the illustrated embodiment. (As shown) Figure 3 As shown, the electronic device 300 may include: at least one processor 301, at least one network interface 304, user interface 303, memory 305, and at least one communication bus 302.

[0149] The communication bus 302 is used to enable communication between these components.

[0150] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0151] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0152] The processor 301 may include one or more processing cores. The processor 301 connects to various parts within the electronic device 300 using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 305, and by calling data stored in the memory 305. Optionally, the processor 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 301 may integrate one or a combination of several of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.

[0153] The memory 305 may include random access memory (RAM) or read-only memory. Optionally, the memory 305 may include a non-transitory computer-readable storage medium. The memory 305 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned processor 301. Figure 3 As shown, the memory 305, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and program instructions.

[0154] exist Figure 3In the illustrated electronic device 300, the user interface 303 is mainly used to provide an input interface for the user and acquire user input data; while the processor 301 can be used to call the image-based interactive application stored in the memory 305 and specifically perform the following operations: acquire the ultra-high frequency signals from the UHF sensor array deployed around the discharge potential and the ultrasonic signals from the AE sensor array, and preprocess the ultra-high frequency signals and ultrasonic signals; calculate the signal starting point of the ultra-high frequency signal in each UHF sensor based on the AIC signal criterion, take the UHF sensor with the smallest signal starting point as the reference sensor one, calculate the relative time difference between the other UHF sensors and the reference sensor one, and then establish a mathematical model related to the discharge potential coordinates. The process involves: coarsely locating the discharge potential and calculating the signal time window based on the position of the AE sensor array; extracting the ultrasonic signal within the signal time window; calculating the relative time difference between the other AE sensors and the reference sensor two, using the reference sensor two with the highest signal-to-noise ratio, to establish a discharge location model; and using the coarse discharge location as the initial iteration value to calculate the fine discharge potential location; collecting discharge pulses of ultra-high frequency signals to construct a PRPD spectrum; combining it with a convolutional neural network to extract the corresponding ultra-high frequency feature vectors; combining the time-frequency map of the ultrasonic signal of the reference sensor two with the training of the convolutional neural network to extract ultrasonic feature vectors; and weighted fusing the ultra-high frequency feature vectors and the ultrasonic feature vectors to identify the discharge type of the discharge location.

[0155] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0156] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0157] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0158] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0159] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0160] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0161] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0162] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0163] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A method for localizing and identifying partial discharges based on multimodal fusion, the method comprising: The ultra-high frequency signals of the UHF sensor array deployed around the discharge point and the ultrasonic signals of the AE sensor array are collected, and the ultra-high frequency signals and ultrasonic signals are preprocessed. Based on the AIC signal criterion, the signal starting point of the UHF signal in each UHF sensor is calculated. The UHF sensor with the smallest signal starting point is taken as the reference sensor one. The relative time difference between the other UHF sensors and the reference sensor one is calculated. Then, a mathematical model related to the discharge potential coordinates is established to obtain the coarse location of the discharge potential. Combined with the position of the AE sensor array, the signal time window is calculated. Within the signal time window, the ultrasonic signal is extracted. Using the reference sensor two corresponding to the signal with the highest signal-to-noise ratio, the relative time difference between other AE sensors and the reference sensor two is calculated. Then, a positioning model of the discharge position is established. Using the coarse positioning of the discharge position as the initial iteration value, the fine positioning of the discharge potential is obtained. The discharge pulses of the ultra-high frequency signal are collected, a PRPD spectrum is constructed, and the corresponding ultra-high frequency feature vector is extracted by combining it with a convolutional neural network. The time-frequency diagram of the ultrasonic signal of the second reference sensor is combined with the convolutional neural network training to extract the ultrasonic feature vector. The ultra-high frequency feature vector and the ultrasonic feature vector are weighted and fused to identify the discharge type of the discharge site.

2. The method according to claim 1, characterized in that, The calculation of the signal starting point of the ultra-high frequency signal in each UHF sensor based on the AIC signal criterion includes: Within the search window of each UHF sensor, the AIC function is calculated, including: ， in, For time series indexing, For window length, For variance, Let be a subsequence from 1 to j.

3. The method according to claim 2, characterized in that, The process involves using the UHF sensor with the smallest signal start point as reference sensor one, calculating the relative time difference between other UHF sensors and reference sensor one, and then establishing a mathematical model relating it to the discharge potential coordinates. This includes: Find the UHF sensor with the smallest signal start point, and calculate the relative time difference between other UHF sensors and reference sensor one, including: ， in, For relative time difference, For reference to the relative time of sensor one, ,in, The number of channels in the UHF sensor array; Establish the TDOA positioning equation system, and assume the coordinates of the discharge point are... , No. The coordinates of the UHF sensors are Then the distance between the two have: ， The following system of equations is then established: ， in , At the speed of light, For the first The distance difference between the first sensor and the second sensor; The coarse location result of the discharge point is obtained by solving the problem: 。 4. The method according to claim 1, characterized in that, The step of calculating the signal time window based on the position of the AE sensor array includes: Calculating the earliest arrival time of the UHF signal includes: ， in, This refers to the earliest arrival time of the UHF signal; Estimate the distance from the coarse location of the discharge point to the farthest AE sensor, and calculate the signal time window, including: ， in, For protection intervals, To determine the distance from the coarse location of the discharge point to the farthest AE sensor, This refers to the speed at which sound waves propagate in the medium of the device.

5. The method according to claim 1, characterized in that, The calculation of the relative time difference between other AE sensors and reference sensor two includes: ， Where m is the second reference sensor, and k is another AE sensor. Let be the cross-correlation function between the two. The ultrasonic signal for each AE channel; The calculation is accelerated using Fast Fourier Transform, including: ， And weighted: ， Search Record the time-shift index corresponding to the peak point with the largest absolute value. Then the time difference between the other AE sensors and the second reference sensor is: ， in The sampling rate.

6. The method according to claim 5, characterized in that, The establishment of the discharge location model includes: Based on the conversion of the time difference set of reference sensor 2 into range difference, the equation includes: ， And define geometric distance , Construct a nonlinear least squares objective function: ， Based on the hyperbolic positioning principle, a mapping model from time difference measurement to spatial coordinates is established.

7. The method according to claim 6, characterized in that, The step of using the coarse location of the discharge potential as the initial iteration value to calculate the fine location of the discharge potential includes: Iterative optimization based on the LM algorithm includes: ， in, The least squares objective function At the current point Jacobian matrix, The initial iteration value, It is the coordinate correction amount. For the residual vector, , , It is the damping factor; The iterative optimization process begins with the initial iteration value and continues until... If the value is less than the preset threshold, or if the maximum number of iterations is reached, then... This is the final high-precision positioning result. .

8. A partial discharge localization and identification system based on multimodal fusion, characterized in that, The system includes; The acquisition module is used to acquire the ultra-high frequency signals of the UHF sensor array deployed around the discharge position and the ultrasonic signals of the AE sensor array, and to preprocess the ultra-high frequency signals and ultrasonic signals. The coarse positioning module is used to calculate the signal start point of the ultra-high frequency signal in each UHF sensor based on the AIC signal criterion. Taking the UHF sensor with the smallest signal start point as the reference sensor one, it calculates the relative time difference between the other UHF sensors and the reference sensor one, and then establishes a mathematical model related to the discharge potential coordinates to obtain the coarse positioning of the discharge potential. Combined with the position of the AE sensor array, it calculates the signal time window. The fine positioning module is used to extract ultrasonic signals within the signal time window, calculate the relative time difference between other AE sensors and the reference sensor two with the signal with the highest signal-to-noise ratio, and then establish a positioning model of the discharge position. The coarse positioning of the discharge position is used as the initial iteration value to obtain the fine positioning of the discharge potential. The identification module is used to collect the discharge pulses of the ultra-high frequency signal, construct the PRPD spectrum, extract the corresponding ultra-high frequency feature vector by combining it with a convolutional neural network, extract the ultrasonic feature vector by combining the time-frequency diagram of the ultrasonic signal of the second reference sensor with the convolutional neural network training, and perform weighted fusion of the ultra-high frequency feature vector and the ultrasonic feature vector to identify the discharge type of the discharge position.

9. An electronic device, comprising a processor and a memory; The processor is connected to the memory; The memory is used to store executable program code; The processor runs a program corresponding to the executable program code stored in the memory to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of claims 1-7.

Images