An ultra-high voltage equipment external insulation partial discharge positioning device and method

Abstract

The application relates to the field of extra-ultra high voltage equipment, in particular to an ultra-ultra high voltage equipment external insulation partial discharge positioning device and method. The device comprises: an ultrasonic sensor array which synchronously collects multi-channel ultrasonic signals; a signal processing unit which sequentially executes complete set empirical mode decomposition primary denoising and improved variable mode decomposition fine decomposition and mode discrimination on each channel signal to obtain a reconstructed signal; and a positioning unit which constructs a covariance matrix based on the reconstructed signal, separates signal and noise subspaces through eigenvalue decomposition, constructs a multiple signal classification spatial spectrum function, and determines the discharge source position through spectrum peak search. The application cascades the self-adaptive noise suppression of complete set empirical mode decomposition and the fine mode discrimination mechanism of improved variable mode decomposition, effectively solves the problems of low signal-to-noise ratio of partial discharge ultrasonic signals and difficulty in identifying effective modes in a strong noise environment, and significantly improves the quality of the reconstructed signal.

Description

Technical Field

[0001] This invention relates to the field of ultra-high voltage equipment, and in particular to a device and method for locating partial discharge of external insulation in ultra-high voltage equipment. Background Technology

[0002] As a core component of the power system, the reliability of the external insulation of ultra-high voltage (UHV) equipment directly affects the safe and stable operation of the power grid. Partial discharge is a major cause of aging, deterioration, and even breakdown of external insulation materials. Accurately locating partial discharge sources is a key technology for achieving equipment condition assessment, fault early warning, and predictive maintenance. Currently, partial discharge location methods based on ultrasonic signals have been widely used in power system operation and maintenance due to their non-invasiveness and real-time performance. However, the operating environment of UHV equipment is extremely complex. The ultrasonic signals generated by partial discharge have weak amplitudes and are easily interfered with by mechanical vibration noise, electromagnetic coupling noise, and environmental background noise, resulting in extremely low signal-to-noise ratios of the acquired multi-channel ultrasonic signals, posing a severe challenge to subsequent signal processing and location. In existing technologies, multi-signal classification algorithms, as a high-resolution spatial spectrum estimation method, can theoretically provide accurate direction-of-arrival (DOA) estimates, but their location performance is highly dependent on the quality of the input signal. When residual noise in the signal is strong or the effective signal features are not obvious, the covariance matrix estimation of multi-signal classification algorithms will produce large deviations, resulting in multiple spurious spectral peaks in the spatial spectrum, or even complete inability to identify the location of the actual discharge source, thus leading to misjudgment or location failure. Traditional filtering methods or single empirical mode decomposition and variational mode decomposition methods can suppress noise to a certain extent, but they are difficult to completely eliminate the interference of complex noise while effectively preserving the transient non-stationary characteristics of the partial discharge ultrasonic signal. Especially in high-noise environments, mode aliasing and spurious mode problems remain prominent, failing to provide high-quality reconstructed signals for high-resolution positioning algorithms. Therefore, how to achieve effective denoising, high-resolution feature extraction, and stable and reliable accurate positioning of the partial discharge ultrasonic signal of the external insulation of ultra-high voltage equipment in strong noise environments is an urgent problem to be solved in this field. Summary of the Invention

[0003] This invention provides a device and method for locating partial discharge of external insulation in ultra-high voltage equipment, aiming to solve the problem of how to effectively denoise, extract high-resolution features, and achieve stable and reliable accurate positioning of ultrasonic signals of partial discharge of external insulation in ultra-high voltage equipment under strong noise environment.

[0004] To achieve the above objectives, the following technical solution is adopted.

[0005] A partial discharge locating device for external insulation of ultra-high voltage equipment, comprising: An ultrasonic sensing array consists of M ultrasonic sensors arranged around the outer insulation of ultra-high voltage equipment. It is used to synchronously acquire multi-channel ultrasonic signals generated by partial discharge and obtain discrete-time signals of M channels. The signal processing unit, connected to the ultrasonic sensing array, is used to perform full set empirical mode decomposition (FMD) primary denoising on the discrete-time signal of each channel, and to perform improved variational mode decomposition (VMD) fine decomposition and mode discrimination on the signal after the FMD primary denoising, to obtain the reconstructed signal. The positioning unit, connected to the signal processing unit, is used to construct a covariance matrix based on the reconstructed signal, perform eigenvalue decomposition on the covariance matrix to obtain a signal subspace and a noise subspace, construct a multi-signal classification spatial spectrum function based on the noise subspace, and determine the spatial location corresponding to the maximum value of the multi-signal classification spatial spectrum function as the location of the local discharge source by performing spectral peak search on the multi-signal classification spatial spectrum function.

[0006] Optionally, the ultrasonic sensing array is a uniform circular array with a radius of r, the element spacing not exceeding half the wavelength of the highest frequency signal, and the number of ultrasonic sensors M greater than the number of partial discharge sources D.

[0007] Optionally, the signal processing unit includes: The primary denoising module adds amplitude-controlled Gaussian white noise to the discrete-time signal of each channel to construct multiple ensemble signals. Empirical mode decomposition (EMD) is performed on each ensemble signal to obtain the first-order intrinsic mode function (IMF). The IMFs obtained from the multiple ensemble signal decompositions are averaged to obtain the first-order average IMF. The first-order average IMF is removed from the discrete-time signal to obtain the first residual signal. The process of adding Gaussian white noise and performing EMD and averaging on the first residual signal is repeated to obtain the second to nth order average IMFs and the final residual term. Based on the spectral energy distribution of each order average IMF and its correlation with the discrete-time signal, average IMFs dominated by high-frequency random noise are removed. The remaining average IMFs are superimposed to obtain the complete ensemble EMD output signal. The fine decomposition module, connected to the primary denoising module, is used to take the output signal of the complete set empirical mode decomposition as the signal to be decomposed, establish a variational model of variational mode decomposition, and transform the variational model into an unconstrained optimization problem by introducing a quadratic penalty factor and Lagrange multipliers. Iterative solution is performed using the alternating direction multiplier method to obtain K modal components and their corresponding center frequencies. The spectral entropy of each modal component is calculated. When the spectral entropy of the newly added mode meets the preset spectral entropy threshold condition, the number of modes K is stopped from increasing. The current value of K is used as the final number of decomposed modes. For the K modal components obtained by decomposing with the final number of decomposed modes K, the energy proportion of each modal component is calculated. The K modal components are judged according to the spectral entropy, energy proportion, and instantaneous frequency stability of each modal component. The modal components with spectral entropy lower than the spectral entropy threshold, energy proportion higher than the energy proportion threshold, and instantaneous frequency stability are judged as effective modes, forming an effective mode set. All modal components in the effective mode set are superimposed to obtain the reconstructed signal.

[0008] Optionally, when the fine decomposition module establishes the variational model for variational mode decomposition, the objective of the variational model is to decompose the signal to be decomposed into K finite bandwidth modes and minimize the sum of the estimated bandwidths of each mode. The constraint condition of the variational model is that the sum of the modes is equal to the signal to be decomposed.

[0009] Optionally, when the fine decomposition module calculates the spectral entropy of each modal component, the spectral entropy of the k-th modal component is calculated based on the normalized power spectrum of that modal component, using the following formula:

[0010] in This is the normalized power spectrum of the k-th modal component.

[0011] Optionally, when the fine decomposition module calculates the energy proportion of each modal component, the formula for the energy proportion of the k-th modal component is as follows:

[0012] in This is the time-domain expression for the k-th modal component.

[0013] Optionally, the positioning unit includes: The covariance matrix construction module, connected to the signal processing unit, is used to assemble the reconstructed signals of each channel into an observation vector and calculate the estimated value of the covariance matrix of the observation vector. The subspace decomposition module, connected to the covariance matrix construction module, is used to perform eigenvalue decomposition on the estimated value of the covariance matrix to obtain M eigenvalues ​​and their corresponding eigenvectors. The M eigenvalues ​​are arranged in descending order, and the eigenvectors corresponding to the first D larger eigenvalues ​​form the signal subspace. The eigenvectors corresponding to the remaining MD smaller eigenvalues ​​form the noise subspace, where D is the number of local discharge sources. The peak search and localization module, connected to the subspace decomposition module, is used to construct a multi-signal classification spatial spectrum function using the noise subspace. The denominator of the multi-signal classification spatial spectrum function is the squared norm of the inner product of the array direction vector and the noise subspace. By performing peak search on the multi-signal classification spatial spectrum function, the spatial location corresponding to the maximum value of the multi-signal classification spatial spectrum function is determined as the location of the local discharge source.

[0014] A method for locating partial discharge on the external insulation of ultra-high voltage equipment includes the following steps: By synchronously acquiring multi-channel ultrasonic signals generated by partial discharge through an ultrasonic sensor array arranged around the external insulation of ultra-high voltage equipment, discrete-time signals of M channels are obtained. The discrete-time signal of each channel is subjected to primary denoising processing using complete ensemble empirical mode decomposition, and the signal after primary denoising processing using complete ensemble empirical mode decomposition is subjected to refined decomposition and mode discrimination processing using improved variational mode decomposition to obtain the reconstructed signal. A covariance matrix is ​​constructed based on the reconstructed signal. Eigenvalue decomposition is performed on the covariance matrix to obtain a signal subspace and a noise subspace. A multi-signal classification spatial spectrum function is constructed based on the noise subspace. By performing spectral peak search on the multi-signal classification spatial spectrum function, the spatial location corresponding to the maximum value of the multi-signal classification spatial spectrum function is determined as the location of the local discharge source.

[0015] Optionally, the discrete-time signal of each channel undergoes primary denoising processing using complete ensemble empirical mode decomposition (CEMD), and the signal after primary denoising processing using CEMD is then subjected to refined decomposition and mode discrimination processing using improved variational mode decomposition (VMD) to obtain the reconstructed signal. The specific steps include: Amplitude-controlled Gaussian white noise is added to the discrete-time signal of each channel to construct multiple ensemble signals. Empirical mode decomposition (EMD) is performed on each ensemble signal to obtain the first-order intrinsic mode function (IMF). The first-order IMFs obtained from the multiple ensemble signal decompositions are averaged to obtain the first-order average IMF. The first-order average IMF is removed from the discrete-time signal to obtain the first residual signal. The process of adding Gaussian white noise and performing EMD and averaging is repeated on the first residual signal to obtain the second to nth order average IMFs and the final residual term. Based on the spectral energy distribution of each order average IMF and its correlation with the discrete-time signal, the average IMF dominated by high-frequency random noise is removed. The remaining average IMFs are superimposed to obtain the complete ensemble EMD output signal. Using the complete set of empirical mode decomposition output signals as the signals to be decomposed, a variational model of variational mode decomposition is established. By introducing a quadratic penalty factor and Lagrange multipliers, the variational model is transformed into an unconstrained optimization problem. The alternating direction multiplier method is used for iterative solution to obtain K modal components and their corresponding center frequencies. The spectral entropy of each modal component is calculated. When the spectral entropy of the newly added mode meets the preset spectral entropy threshold condition, the addition of the mode number K stops. The current value of K is used as the final decomposed mode number. For the K modal components obtained by decomposing with the final decomposed mode number K, the energy ratio of each modal component is calculated. The K modal components are judged according to the spectral entropy, energy ratio and instantaneous frequency stability of each modal component. The modal components with spectral entropy lower than the spectral entropy threshold, energy ratio higher than the energy ratio threshold and instantaneous frequency stability are judged as effective modes and form an effective mode set. All modal components in the effective mode set are superimposed to obtain the reconstructed signal.

[0016] Optionally, in the step of establishing the variational model, the objective of the variational model is to decompose the signal to be decomposed into K finite bandwidth modes and minimize the sum of the estimated bandwidths of each mode. The constraint condition of the variational model is that the sum of the modes is equal to the signal to be decomposed. When calculating the spectral entropy of each mode component, the spectral entropy of the k-th mode component is calculated based on the normalized power spectrum of that mode component, using the following formula:

[0017] in Let be the normalized power spectrum of the k-th modal component; when calculating the energy proportion of each modal component, the formula for the energy proportion of the k-th modal component is:

[0018] in This is the time-domain expression for the k-th modal component.

[0019] Compared with the prior art, the present invention has the following beneficial effects: This invention proposes a device and method for locating partial discharge in the external insulation of ultra-high voltage (UHV) equipment. The device synchronously acquires multi-channel ultrasonic signals using an ultrasonic sensor array. A signal processing unit sequentially performs primary denoising processing using complete ensemble empirical mode decomposition (CEMD) and refined decomposition and mode discrimination processing on each channel signal to obtain a reconstructed signal. The location unit then constructs a covariance matrix based on the reconstructed signal, separates the signal subspace and noise subspace through eigenvalue decomposition, and constructs a multi-signal classification spatial spectrum function for peak search to determine the location of the partial discharge source. This technical solution cascades the adaptive noise suppression capability of CEMD with the refined mode discrimination mechanism of refined variational mode decomposition, effectively solving the problems of low signal-to-noise ratio and difficulty in identifying effective modes in partial discharge ultrasonic signals under strong noise environments, significantly improving the quality of the reconstructed signal. Furthermore, the high-quality reconstructed signal provides accurate covariance matrix estimation for the multi-signal classification location algorithm, thereby greatly improving the authenticity of the spatial spectrum peaks and suppressing the generation of false peaks. Ultimately, this achieves high-precision and high-stability location of partial discharge sources in the complex electromagnetic and acoustic noise environment of UHV systems. Furthermore, this invention specifically defines the ultrasonic sensing array as a uniform circular array and clarifies the constraint relationship between the element spacing and the wavelength, ensuring the array's spatial resolution and unambiguous direction-finding capability. The primary denoising module in the signal processing unit effectively eliminates mode aliasing and improves the robustness of the decomposition by adding controlled Gaussian white noise and performing multiple ensemble empirical mode decomposition and averaging operations. The fine decomposition module further introduces an adaptive mode number determination mechanism based on spectral entropy. When the spectral entropy of a newly added mode meets a preset threshold, the decomposition automatically stops, avoiding over- or under-decomposition problems caused by empirically setting the mode number. Simultaneously, it combines energy proportion and instantaneous frequency stability to comprehensively judge the mode components, ensuring that the reconstructed signal retains only effective modes that match the physical characteristics of partial discharge. The hierarchical arrangement of the covariance matrix construction module, subspace decomposition module, and spectral peak search module in the positioning unit provides a clear modular structure for the entire process from signal reconstruction to spatial spectrum estimation, facilitating engineering implementation. Furthermore, this invention also provides a positioning method corresponding to the device. This method organically integrates the above-mentioned steps of cascaded denoising, adaptive modal decision and high-resolution spatial spectrum estimation, forming a complete technical chain from signal acquisition to discharge power source location output. The logical connections between each step are close, and there are no isolated technical links. This not only ensures the reproducibility of the method, but also provides reliable technical support for online monitoring and fault early warning of partial discharge of external insulation of ultra-high voltage equipment. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the module structure of an embodiment of the external insulation partial discharge positioning device for ultra-high voltage equipment according to the present invention.

[0021] Figure 2 This is a schematic diagram of the steps of a method for locating partial discharge in the external insulation of ultra-high voltage equipment according to the present invention. Detailed Implementation

[0022] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0023] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0024] Example 1

[0025] like Figure 1 As shown, this embodiment provides a partial discharge locating device for the external insulation of ultra-high voltage equipment. This device is used for voltage levels of 1000KV. v AC or ±800K v The external insulation surface of DC and higher voltage transmission equipment, such as gas-insulated fully enclosed switchgear, transformers, bushings, etc., is prone to partial discharge during operation due to factors such as dirt, moisture, and electric field distortion. This partial discharge generates ultrasonic signals. The frequency range of these ultrasonic signals is typically between 20kHz and 200kHz, and the signal amplitude is weak, easily drowned out by mechanical vibration noise, electromagnetic coupling noise, and ambient background noise. The device in this embodiment is used for precise location of partial discharge sources in complex noisy environments.

[0026] The device comprises an ultrasonic sensor array. The ultrasonic sensor array consists of M ultrasonic sensors, where M is an integer greater than 1, ranging from 8 to 16. These ultrasonic sensors are arranged around the external insulation of the ultra-high voltage equipment. Each ultrasonic sensor can be an air-coupled piezoelectric ultrasonic sensor or a capacitive micromechanical ultrasonic sensor, with a center frequency selected from 40kHz to 150kHz, specifically determined based on the spectral characteristics of the partial discharge ultrasonic signal of the equipment under test. The ultrasonic sensor array adopts a uniform circular array configuration, i.e., the M ultrasonic sensors are arranged at equal intervals on a circle of radius r. The element spacing, i.e., the arc distance between adjacent sensors, does not exceed half the wavelength of the highest frequency signal to avoid spatial aliasing. The wavelength λ is determined by the relationship between the speed of sound c and the frequency f, λ = c / f, where the speed of sound c is taken as 340m / s. For example, when the highest frequency is 100kHz, the wavelength λ is 3.4mm, and the element spacing should not exceed 1.7mm. The total number M of ultrasonic sensors is greater than the number D of possible simultaneous partial discharge sources. D is typically 1 to 3, and M ranging from 8 to 16 ensures the array's ability to distinguish between multiple signal sources. The ultrasonic sensor array is used to synchronously acquire multi-channel ultrasonic signals generated by partial discharge, obtaining discrete-time signals from M channels. Each ultrasonic sensor is connected to a signal conditioning circuit, including a preamplifier, a bandpass filter, and an analog-to-digital converter. The analog-to-digital conversion of all channels is triggered by the same clock source to ensure accurate time delay relationships between the signals of each channel.

[0027] The device also includes a signal processing unit. The signal processing unit is electrically connected to the output of the ultrasonic sensing array. The signal processing unit sequentially performs full ensemble empirical mode decomposition (HEMD) primary denoising and improved variational mode decomposition (VMD) fine decomposition and mode discrimination processing on the discrete-time signal of each channel to obtain the reconstructed signal. The signal processing unit can be an embedded system composed of a field-programmable gate array (FPGA) and a digital signal processor (DSP), or it can be a computer or industrial controller running a dedicated algorithm.

[0028] The signal processing unit internally includes a primary denoising module and a fine decomposition module. The primary denoising module performs primary denoising processing using complete ensemble empirical mode decomposition. For a discrete-time signal of a certain channel, let the original signal be denoised as follows: Where the subscript m represents the m-th channel. The primary noise reduction module directs... Multiple sets of aggregated signals are constructed by repeatedly adding Gaussian white noise with controlled amplitude. Let the noise added in the i-th iteration be... Its amplitude is determined by the noise amplitude coefficient ε i Control, ε i The value range is 0.1 to 0.4 times the standard deviation of the original signal. Then the i-th set signal is represented as:

[0029] in Let represent the i-th noise-added signal of the m-th channel. Empirical mode decomposition (EMD) is performed on each set of signals to obtain the first-order intrinsic mode functions. Empirical Mode Decomposition (EMD) is an adaptive signal processing method that iteratively decomposes a signal into multiple intrinsic mode functions (EMFs). The specific process involves: finding local maxima and minima of the signal; fitting the upper and lower envelopes using cubic spline interpolation; calculating the envelope mean; subtracting the envelope mean from the original signal to obtain candidate components; and repeating this selection process until two conditions for EMFs are met: the number of extrema equals or differs from the number of zero-crossings by at most one, and the envelope mean is zero. Then, the first-order EMFs obtained from multiple ensemble signal decompositions are averaged to obtain the first-order averaged EMF. :

[0030] Where I represents the set degree, typically between 50 and 100. The 1 in IMF1 on the left side of the equals sign is a regular number used as part of the function name, while the 1 on the right side of the equals sign... In this context, the subscript 1 indicates the first order, and the superscript (i) indicates the i-th trial. The first-order average intrinsic mode function is removed from the original discrete-time signal to obtain the first residual signal:

[0031] in That is, the result of the previous calculation. Repeat the above process for the first residual signal: add controlled Gaussian white noise, perform empirical mode decomposition and averaging, to obtain the second-order average intrinsic mode function. Then subtract from the first residual signal The second residual signal is obtained. This process is repeated to obtain the average intrinsic mode functions from the third to the nth order, as well as the final residual term. The entire decomposition process can be represented as:

[0032] in Let K represent the k-th order average eigenmode function. c The total order of the eigenmode functions obtained from the decomposition is given by [the factorial order]. This is the final residual term. Decomposition stops when the residual signal becomes a monotonic function or its energy is less than 1% of the energy of the original signal.

[0033] The primary denoising module removes average intrinsic mode functions dominated by high-frequency random noise based on the spectral energy distribution of each order's average intrinsic mode function and its correlation with the original discrete-time signal. Specifically, for each... Perform a Fast Fourier Transform to obtain the power spectrum, and calculate its spectral centroid frequency. If the centroid frequency is higher than 100kHz and With the original signal If the Pearson correlation coefficient is less than 0.3, then it is determined that... Noise-dominant modes are discarded. The remaining average intrinsic mode functions are then superimposed to obtain the complete ensemble empirical mode decomposition output signal.

[0034] in This is the output signal of the m-th channel after primary denoising processing. To preserve the mode set, which is the set of intrinsic mode functions remaining after removing the IMF dominated by high-frequency random noise, this stage of processing initially suppresses strong background noise, especially high-frequency random noise, and alleviates the mode aliasing problem.

[0035] The fine decomposition module is connected to the primary denoising module and receives the output signal of the complete set of empirical mode decomposition. The signal to be decomposed is used as the input signal. The fine decomposition module performs improved variational mode decomposition. First, a variational model for variational mode decomposition is established. The goal of this model is to decompose the input signal into K... v A finite bandwidth modal component And minimize the sum of the estimated bandwidths of each modal component. The expression for the variational model is:

[0036] The constraints are:

[0037] in For the k-th modal component, ω k Let δ(t) be the center frequency of the k-th modal component, δ(t) be the Dirac function, and * denote convolution. To solve this constrained optimization problem, a quadratic penalty factor β and a Lagrange multiplier λ(t) are introduced, transforming the original problem into an unconstrained optimization problem, yielding the augmented Lagrange function. Then, the alternating direction multiplier method is used for iterative solution. In each iteration, each modal component is updated sequentially. The center frequency ω of each modal component k And the Lagrange multiplier λ. The iteration continues until the convergence condition is met, ultimately yielding K. v Modal components and its corresponding center frequency ω k .

[0038] The fine decomposition module introduces an adaptive method to determine the number of modes K. v The mechanism involves introducing modal spectral entropy H.k This is achieved through [method / method]. The formula for calculating spectral entropy is:

[0039] in This is the normalized power spectrum of the k-th modal component. The normalized power spectrum is calculated as follows: first calculate the modal components... The power spectrum is then divided by the sum of the power spectra at all frequency points, such that... Spectral entropy H k H reflects the degree of disorder in the spectrum. k The smaller the value, the more concentrated the frequency. H k A larger value indicates a more dispersed frequency range. Initialize the number of modes K. v =1, perform the above variational mode decomposition to obtain the current K. v K under value v There are several modal components. Then try to use K. v Increase by 1, decompose again, and calculate the spectral entropy H of the newly added modal component. Kv When the newly added modality satisfies: If the newly added mode is determined to be mainly composed of noise, then the addition of the mode number K is stopped. v Where η is a preset proportionality coefficient, typically taken as 1.2 to 1.5; This is the average of the spectral entropy of the first k-1 modes. With the current K... v The value is used as the final decomposition mode number.

[0040] After determining the final decomposition mode number K v Subsequently, the fine decomposition module uses this K v The value is re-performed by variational mode decomposition to obtain K. v Modal components k=1,…,K v Regarding this K v For each modal component, effective mode decision is still required. First, the energy percentage E of each modal component is calculated. k : The numerator represents the time-domain energy of the k-th modal component, and the denominator represents the energy of all K-th modal components. v The total energy of each modal component. Secondly, using the previously calculated spectral entropy H... k Next, the instantaneous frequency stability of each modal component is evaluated. The instantaneous frequency can be obtained by taking the phase derivative of the analytic signal of the modal component, which is constructed using the Hilbert transform. If the instantaneous frequency of a certain modal component fluctuates little over time, for example, the coefficient of variation of the instantaneous frequency is less than 0.3, then the instantaneous frequency of that mode is considered stable. In this embodiment, a spectral entropy threshold θ is set.H =0.5 and energy percentage threshold θ E =0.03. If a modal component simultaneously satisfies H k <θ H E k >θ E If the instantaneous frequency is stable, it is determined to be a valid mode. All valid modes constitute the valid mode set Ω. v Finally, the modal components of all modes in the effective mode set are superimposed to obtain the reconstructed signal:

[0041] The reconstructed signal eliminates noise-dominated modes and retains effective components that are highly correlated with the physical characteristics of the partial discharge ultrasonic signal, resulting in a high signal-to-noise ratio and time-frequency resolution.

[0042] The device also includes a positioning unit. The positioning unit is connected to the output of the signal processing unit and is used to construct a covariance matrix based on the reconstructed signal, perform eigenvalue decomposition on the covariance matrix to obtain a signal subspace and a noise subspace, construct a multi-signal classification spatial spectrum function based on the noise subspace, and determine the spatial location corresponding to the maximum value of the multi-signal classification spatial spectrum function as the location of the local discharge source by performing spectral peak search on the multi-signal classification spatial spectrum function.

[0043] The localization unit internally includes a covariance matrix construction module, a subspace decomposition module, and a spectral peak search and localization module. The covariance matrix construction module reconstructs the signals from each channel. Each observation vector is composed of elements with the same time index.

[0044] The superscript T indicates transpose. For a data block containing L sampling points, where L is typically 3 to 5 times M, the covariance matrix construction module calculates the estimated covariance matrix of the observed vectors:

[0045] in It represents the expected value of a mathematical expression, which is replaced by the sample average in actual calculations; the superscript H indicates the conjugate transpose.

[0046] The subspace decomposition module performs eigenvalue decomposition on the covariance matrix R, decomposing the covariance matrix into the sum of the signal subspace and the noise subspace: Where E s Λ is a matrix composed of signal eigenvectors in the signal subspace. s E is a diagonal matrix composed of the signal eigenvalues. nΛ is a matrix consisting of noise eigenvectors within the noise subspace. n Let R be a diagonal matrix composed of noise eigenvalues. Specifically, eigenvalue decomposition is performed on the covariance matrix R to obtain M eigenvalues ​​λ1≥λ2≥…≥λ M and the corresponding feature vectors v1, v2, ..., v M The first D larger eigenvalues ​​correspond to the signal energy of the partial discharge source, while the remaining M-DM-D smaller eigenvalues ​​correspond to the noise energy. Therefore, the signal subspace Es=[v1,v2,…,v…] D ], noise subspace E n =[v D+1 ,v D+2 ,…,v M The number of partial discharge sources, D, can be automatically estimated using the minimum description length criterion, or it can be set based on prior knowledge.

[0047] The peak search and localization module utilizes the noise subspace E n Construct a multi-signal classification spatial spectrum function. For a uniform circular array, the array direction vector a(θ) is composed of the elevation angle θ and the azimuth angle. It was jointly decided that, for simplicity, the angle parameter would be uniformly written as θ. The expression for the spatial spectral function is: The denominator is the squared norm of the inner product of the array direction vector and the noise subspace. Since the signal direction vector is orthogonal to the noise subspace, when the search direction θ is exactly aligned with the actual local discharge source, the denominator approaches 0, and the spatial spectral function reaches a maximum. The spectral peak search and localization module operates at elevation angles θ from 0° to 90° and azimuth angles... A two-dimensional scan is performed within a range of 0° to 360° to calculate the spatial spectral function value for each angle combination and identify all local maxima. The angle corresponding to each local maximum is the arrival direction of a local discharge source. Combining this with the spatial coordinates of the ultrasonic sensor array, the arrival direction can be further converted into three-dimensional spatial coordinates. Finally, the spectral peak search and localization module outputs the location information of the local discharge source, which can be displayed on a human-machine interface, stored in a database, or uploaded to a remote monitoring center.

[0048] In this embodiment, the signal processing unit and the positioning unit can be integrated into the same digital signal processor, field-programmable gate array, or microcontroller. The connections between the modules can be either wired electrical connections or wireless communication connections. The device may also include a triggering and synchronization unit, a power management unit, and a communication interface; these additional units do not affect the core positioning function.

[0049] Example 2

[0050] like Figure 2 As shown, this embodiment provides a method for locating partial discharge on the external insulation of ultra-high voltage (UHV) equipment. This method can be implemented using the device described in Embodiment 1, or it can be executed on other hardware platforms with the same functionality. The method includes the following steps: Simultaneously acquiring multi-channel ultrasonic signals generated by partial discharge using an ultrasonic sensor array arranged around the external insulation of the UHV equipment, obtaining M discrete-time signals; performing primary denoising processing using complete ensemble empirical mode decomposition (CEMD) on the discrete-time signals of each channel, and then performing refined decomposition and mode discrimination processing using improved variational mode decomposition (VMD) on the signals after primary denoising processing to obtain reconstructed signals; constructing a covariance matrix based on the reconstructed signals, performing eigenvalue decomposition on the covariance matrix to obtain signal subspace and noise subspace, and constructing a multi-signal classification spatial spectrum function based on the noise subspace; and determining the spatial location corresponding to the maximum value of the multi-signal classification spatial spectrum function as the location of the partial discharge source by performing spectral peak search on the multi-signal classification spatial spectrum function. Each step is described in detail below.

[0051] The first step is signal acquisition. M ultrasonic sensors are arranged around the outer insulation of the ultra-high voltage equipment, forming an ultrasonic sensor array. M is an integer greater than 1, ranging from 8 to 16. The ultrasonic sensor array is a uniform circular array with a radius of r, and the spacing between array elements does not exceed half the wavelength of the highest frequency signal. The wavelength λ is determined by the relationship between the speed of sound c and the frequency f, λ = c / f, where the speed of sound c is taken as 340 m / s. All sensors synchronously acquire the ultrasonic signals generated by partial discharge, with the sampling frequency set to 2.5 to 5 times the highest frequency signal. Each channel obtains a discrete-time signal, denoted as... , where t is the time variable, and m=1,...,M.

[0052] The second step involves performing primary denoising processing, specifically ensemble empirical mode decomposition, on the discrete-time signal of each channel. For a single channel signal... Perform the following sub-steps.

[0053] Sub-step 2.1: Initialize the residual signal The order index k=1.

[0054] Sub-step 2.2: Send the current residual signal N different groups of Gaussian white noise are added, the amplitude of each group of white noise is determined by the noise amplitude coefficient ε. i Control, ε i The value range is from 0.1 to 0.4. The i-th group of noisy signals is represented as:

[0055] in Let be the Gaussian white noise added for the i-th time.

[0056] Sub-step 2.3: Perform empirical mode decomposition on each group of noisy signals to extract the first-order intrinsic mode function, denoted as... The specific process of empirical mode decomposition is as follows: find the local maxima and minima of the signal, fit the upper and lower envelopes by cubic spline interpolation, calculate the envelope mean, subtract the envelope mean from the original signal to obtain candidate components, and repeat this screening process until the two conditions of the intrinsic mode function are met: the number of extreme points is equal to or differs by at most one from the number of zero crossings, and the envelope mean is zero.

[0057] Sub-step 2.4: Average the first-order intrinsic mode functions obtained from the decomposition of the N groups of noisy signals to obtain the k-th average intrinsic mode function. When k=1, the average intrinsic mode function is denoted as... The formula is:

[0058] Where I represents the set degree, typically between 50 and 100. The 1 in IMF1 on the left side of the equals sign is a regular number used as part of the function name, while the 1 on the right side of the equals sign... In this context, the subscript 1 indicates the first order, and the superscript (i) indicates the i-th trial. For k>1, the k-th order average intrinsic mode function is denoted as [equation missing]. .

[0059] Sub-step 2.5: Subtract the k-th order average eigenmode function from the current residual signal to obtain the new residual signal. When k=1:

[0060] When k>1:

[0061] Sub-step 2.6: Determine the residual signal Is it a monotonic function or is its energy less than the original signal? 1% of the energy. If so, stop the decomposition and let the total order be 1%. Otherwise, let k = k + 1 and return to sub-step 2.2.

[0062] Sub-step 2.7: All average intrinsic mode functions and the final residual term satisfy the following reconstruction relationship:

[0063] in Let k be the average eigenmode function. This is the final residual term.

[0064] Sub-step 2.8: For each average eigenmode function Calculate its power spectral density and compare it with the original signal. The correlation coefficient. If the energy of the power spectrum is mainly concentrated in the high-frequency band and the correlation coefficient is less than 0.3, then this... Modes marked as noise-dominant are discarded. The remaining average intrinsic mode functions are then superimposed to obtain the complete ensemble empirical mode decomposition output signal.

[0065] Where Ω c To preserve the mode set, which is the set of intrinsic mode functions remaining after removing the IMF dominated by high-frequency random noise.

[0066] The third step involves performing improved variational mode decomposition (VMD) on the output signal of the complete set empirical mode decomposition (EMD) and mode discrimination processing to obtain the reconstructed signal. This includes the following sub-steps.

[0067] Sub-step 3.1: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] As the signal to be decomposed, denoted as .

[0068] Sub-step 3.2: Establish the variational model for variational mode decomposition. The goal is to find... Modal components Each modal component has a finite bandwidth, and the sum of all modal components equals The objective function of the variational model is:

[0069] The constraints are:

[0070] in For the k-th modal component, ω k Let t be the center frequency of the k-th modal component, δ(t) be the Dirac function, and ∗ denote the convolution operation.

[0071] Sub-step 3.3: Introduce the quadratic penalty factor β and the Lagrange multiplier λ(t) to construct the augmented Lagrange function, transforming the original problem into an unconstrained optimization problem. Then, iteratively solve the problem in the frequency domain using the alternating direction multiplier method. Initialize the modal components. Center frequency and Lagrange multipliers The number of iterations is m=1. For each iteration, each modal component and its center frequency are updated sequentially. The update formula is:

[0072] Lagrange multipliers updated to ,in To update the step size, iteration continues until the convergence condition is met. ε is 10 -6 .

[0073] Sub-step 3.4: Adaptively determine the number of modes K v By introducing modal spectral entropy H k This is achieved through [method / method]. The formula for calculating spectral entropy is:

[0074] in The normalized power spectrum of the k-th modal component satisfies Initialize K v =1, perform the above variational mode decomposition to obtain the current K. v K under value v There are several modal components. Then try to use K. v Increase by 1, decompose again, and calculate the spectral entropy H of the newly added modal component. Kv When the newly added modality satisfies: If the newly added mode is determined to be mainly composed of noise, then the number of modes K is stopped. v Where η is a preset proportionality coefficient, typically taken as 1.2 to 1.5; This is the average value of the spectral entropy of the first k-1 modes. Using the current K... v The value is used as the final decomposition mode number.

[0075] Sub-step 3.5: Using the finally determined K v The value is a parameter, and a complete variational mode decomposition is performed to obtain K. v Modal components k=1,…,K v .

[0076] Sub-step 3.6: Perform effective mode decision for each modal component. First, calculate the energy percentage E of each modal component. k : The numerator represents the time-domain energy of the k-th modal component, and the denominator represents the energy of all K-th modal components. v The total energy of each modal component. Next, using the spectral entropy H calculated in sub-step 3.4... k Next, evaluate the instantaneous frequency stability of each modal component: for Perform a Hilbert transform to obtain the analytic signal, then take the time derivative of the phase of the analytic signal to obtain the instantaneous frequency. If the instantaneous frequency fluctuates little over time, for example, if the coefficient of variation of the instantaneous frequency is less than 0.3, then the instantaneous frequency of that mode is considered stable. In this embodiment, a spectral entropy threshold θ is set. H =0.5 and energy percentage threshold θ E =0.03. If a modal component simultaneously satisfies H k <θ H E k >θ E If the instantaneous frequency is stable, it is determined to be a valid mode. All valid modes constitute the valid mode set Ω. v .

[0077] Sub-step 3.7: Superimpose all modal components in the effective mode set to obtain the reconstructed signal:

[0078] The fourth step is to perform spatial positioning based on the reconstructed signal. This includes the following sub-steps.

[0079] Sub-step 4.1: Reconstruct the signals of each channel The observation vector is composed of elements based on the same time index:

[0080] For L sampling points, L is typically 3 to 5 times M. The estimated covariance matrix is ​​calculated as follows:

[0081] in It represents the expected value of a mathematical expression, which is replaced by the sample average in actual calculations; the superscript H indicates the conjugate transpose.

[0082] Sub-step 4.2: Perform eigenvalue decomposition on the covariance matrix R, decomposing it into the sum of the signal subspace and the noise subspace: Where E s Λ is a matrix composed of signal eigenvectors in the signal subspace. s E is a diagonal matrix composed of the signal eigenvalues. n Λ is a matrix consisting of noise eigenvectors within the noise subspace. n Let R be a diagonal matrix composed of noise eigenvalues. Specifically, eigenvalue decomposition is performed on the covariance matrix R to obtain M eigenvalues ​​λ1≥λ2≥…≥λ M and the corresponding feature vectors v1, v2, ..., v M The first D larger eigenvalues ​​correspond to the signal energy of the partial discharge source, while the remaining M−DM−D smaller eigenvalues ​​correspond to the noise energy. Therefore, the signal subspace Es=[v1,v2,…,v... D], noise subspace E n =[v D+1 ,v D+2 ,…,v M ].

[0083] Sub-step 4.3: Construct the multi-signal classification spatial spectrum function. For a uniform circular array, the array direction vector a(θ) consists of the elevation angle θ and the azimuth angle. It was jointly decided that, for simplicity, the angle parameter would be uniformly written as θ. The spatial spectral function is: The denominator is the squared norm of the inner product of the array direction vector and the noise subspace.

[0084] Sub-step 4.4: With the pitch angle θ ranging from 0° to 90° and the azimuth angle... Perform a two-dimensional scan within the range of 0° to 360°, with a step size of 1°. Calculate P for each angle combination. MUSIC(θ) Find all local maxima. The angle corresponding to each local maximum is the direction of arrival of a local discharge source.

[0085] Sub-step 4.5: Combining the spatial coordinates of the ultrasonic sensor array, convert the direction of arrival into three-dimensional spatial coordinates. Output the location information of the local discharge power source, which can be displayed in the form of azimuth, elevation, and distance, or directly marked on the three-dimensional model of the equipment's external insulation.

[0086] The method of this embodiment can be embedded in an embedded device as a software program to achieve fully automatic real-time positioning. It can also be used as an offline analysis tool to post-process pre-acquired signal data. The parameters used in the method include the noise amplitude coefficient ε. i Set number I, proportionality coefficient η, spectral entropy threshold θ H Energy percentage threshold θ E The embodiments described above can be adjusted according to the actual application scenario. As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or equivalent to this invention are included in this invention.

Claims

1. A partial discharge locating device for external insulation of ultra-high voltage equipment, characterized in that, include: An ultrasonic sensing array consists of M ultrasonic sensors arranged around the outer insulation of ultra-high voltage equipment. It is used to synchronously acquire multi-channel ultrasonic signals generated by partial discharge and obtain discrete-time signals of M channels. The signal processing unit, connected to the ultrasonic sensing array, is used to perform full set empirical mode decomposition (FMD) primary denoising on the discrete-time signal of each channel, and to perform improved variational mode decomposition (VMD) fine decomposition and mode discrimination on the signal after the FMD primary denoising, to obtain the reconstructed signal. The positioning unit, connected to the signal processing unit, is used to construct a covariance matrix based on the reconstructed signal, perform eigenvalue decomposition on the covariance matrix to obtain a signal subspace and a noise subspace, construct a multi-signal classification spatial spectrum function based on the noise subspace, and determine the spatial location corresponding to the maximum value of the multi-signal classification spatial spectrum function as the location of the local discharge source by performing spectral peak search on the multi-signal classification spatial spectrum function.

2. The partial discharge locating device for external insulation of ultra-high voltage equipment according to claim 1, characterized in that, The ultrasonic sensor array is a uniform circular array with a radius of r. The spacing between array elements does not exceed half the wavelength of the highest frequency signal. The number of ultrasonic sensors M is greater than the number of partial discharge sources D.

3. The partial discharge locating device for external insulation of ultra-high voltage equipment according to claim 1, characterized in that, The signal processing unit includes: The primary denoising module adds amplitude-controlled Gaussian white noise to the discrete-time signal of each channel to construct multiple ensemble signals. Empirical mode decomposition (EMD) is performed on each ensemble signal to obtain the first-order intrinsic mode function (IMF). The IMFs obtained from the multiple ensemble signal decompositions are averaged to obtain the first-order average IMF. The first-order average IMF is removed from the discrete-time signal to obtain the first residual signal. The process of adding Gaussian white noise and performing EMD and averaging on the first residual signal is repeated to obtain the second to nth order average IMFs and the final residual term. Based on the spectral energy distribution of each order average IMF and its correlation with the discrete-time signal, average IMFs dominated by high-frequency random noise are removed. The remaining average IMFs are superimposed to obtain the complete ensemble EMD output signal. The fine decomposition module, connected to the primary denoising module, is used to take the output signal of the complete set empirical mode decomposition as the signal to be decomposed, establish a variational model of variational mode decomposition, and transform the variational model into an unconstrained optimization problem by introducing a quadratic penalty factor and Lagrange multipliers. Iterative solution is performed using the alternating direction multiplier method to obtain K modal components and their corresponding center frequencies. The spectral entropy of each modal component is calculated. When the spectral entropy of the newly added mode meets the preset spectral entropy threshold condition, the number of modes K is stopped from increasing. The current value of K is used as the final number of decomposed modes. For the K modal components obtained by decomposing with the final number of decomposed modes K, the energy proportion of each modal component is calculated. The K modal components are judged according to the spectral entropy, energy proportion, and instantaneous frequency stability of each modal component. The modal components with spectral entropy lower than the spectral entropy threshold, energy proportion higher than the energy proportion threshold, and instantaneous frequency stability are judged as effective modes, forming an effective mode set. All modal components in the effective mode set are superimposed to obtain the reconstructed signal.

4. The partial discharge locating device for external insulation of ultra-high voltage equipment according to claim 3, characterized in that, When the fine decomposition module establishes the variational model for variational mode decomposition, the objective of the variational model is to decompose the signal to be decomposed into K finite bandwidth modes and minimize the sum of the estimated bandwidths of each mode. The constraint condition of the variational model is that the sum of the modes is equal to the signal to be decomposed.

5. The partial discharge locating device for external insulation of ultra-high voltage equipment according to claim 3, characterized in that, When the fine decomposition module calculates the spectral entropy of each modal component, the spectral entropy of the k-th modal component is calculated based on the normalized power spectrum of that modal component, using the following formula: in This is the normalized power spectrum of the k-th modal component.

6. The partial discharge locating device for external insulation of ultra-high voltage equipment according to claim 3, characterized in that, When the fine decomposition module calculates the energy percentage of each modal component, the formula for the energy percentage of the k-th modal component is as follows: in This is the time-domain expression for the k-th modal component.

7. The partial discharge locating device for external insulation of ultra-high voltage equipment according to claim 1, characterized in that, The positioning unit includes: The covariance matrix construction module, connected to the signal processing unit, is used to assemble the reconstructed signals of each channel into an observation vector and calculate the estimated value of the covariance matrix of the observation vector. The subspace decomposition module, connected to the covariance matrix construction module, is used to perform eigenvalue decomposition on the estimated value of the covariance matrix to obtain M eigenvalues ​​and their corresponding eigenvectors. The M eigenvalues ​​are arranged in descending order, and the eigenvectors corresponding to the first D larger eigenvalues ​​form the signal subspace. The eigenvectors corresponding to the remaining MD smaller eigenvalues ​​form the noise subspace, where D is the number of local discharge sources. The peak search and localization module, connected to the subspace decomposition module, is used to construct a multi-signal classification spatial spectrum function using the noise subspace. The denominator of the multi-signal classification spatial spectrum function is the squared norm of the inner product of the array direction vector and the noise subspace. By performing peak search on the multi-signal classification spatial spectrum function, the spatial location corresponding to the maximum value of the multi-signal classification spatial spectrum function is determined as the location of the local discharge source.

8. A method for locating partial discharge in the external insulation of ultra-high voltage equipment, characterized in that, Includes the following steps: By synchronously acquiring multi-channel ultrasonic signals generated by partial discharge through an ultrasonic sensor array arranged around the external insulation of ultra-high voltage equipment, discrete-time signals of M channels are obtained. The discrete-time signal of each channel is subjected to primary denoising processing using complete ensemble empirical mode decomposition, and the signal after primary denoising processing using complete ensemble empirical mode decomposition is subjected to refined decomposition and mode discrimination processing using improved variational mode decomposition to obtain the reconstructed signal. A covariance matrix is ​​constructed based on the reconstructed signal. Eigenvalue decomposition is performed on the covariance matrix to obtain a signal subspace and a noise subspace. A multi-signal classification spatial spectrum function is constructed based on the noise subspace. By performing spectral peak search on the multi-signal classification spatial spectrum function, the spatial location corresponding to the maximum value of the multi-signal classification spatial spectrum function is determined as the location of the local discharge source.

9. The method for locating partial discharge of external insulation in ultra-high voltage equipment according to claim 8, characterized in that, The specific steps for performing primary denoising processing of the discrete-time signal of each channel using complete ensemble empirical mode decomposition (CEMD) and then performing refined decomposition and mode discrimination processing of the signal after primary denoising processing using improved variational mode decomposition (VMD) to obtain the reconstructed signal include: Amplitude-controlled Gaussian white noise is added to the discrete-time signal of each channel to construct multiple ensemble signals. Empirical mode decomposition (EMD) is performed on each ensemble signal to obtain the first-order intrinsic mode function (IMF). The first-order IMFs obtained from the multiple ensemble signal decompositions are averaged to obtain the first-order average IMF. The first-order average IMF is removed from the discrete-time signal to obtain the first residual signal. The process of adding Gaussian white noise and performing EMD and averaging is repeated on the first residual signal to obtain the second to nth order average IMFs and the final residual term. Based on the spectral energy distribution of each order average IMF and its correlation with the discrete-time signal, the average IMF dominated by high-frequency random noise is removed. The remaining average IMFs are superimposed to obtain the complete ensemble EMD output signal. Using the complete set of empirical mode decomposition output signals as the signals to be decomposed, a variational model of variational mode decomposition is established. By introducing a quadratic penalty factor and Lagrange multipliers, the variational model is transformed into an unconstrained optimization problem. The alternating direction multiplier method is used for iterative solution to obtain K modal components and their corresponding center frequencies. The spectral entropy of each modal component is calculated. When the spectral entropy of the newly added mode meets the preset spectral entropy threshold condition, the addition of the mode number K stops. The current value of K is used as the final decomposed mode number. For the K modal components obtained by decomposing with the final decomposed mode number K, the energy ratio of each modal component is calculated. The K modal components are judged according to the spectral entropy, energy ratio and instantaneous frequency stability of each modal component. The modal components with spectral entropy lower than the spectral entropy threshold, energy ratio higher than the energy ratio threshold and instantaneous frequency stability are judged as effective modes and form an effective mode set. All modal components in the effective mode set are superimposed to obtain the reconstructed signal.

10. The method for locating partial discharge of external insulation in ultra-high voltage equipment according to claim 9, characterized in that, In the step of establishing the variational model, the objective of the variational model is to decompose the signal to be decomposed into K finite bandwidth modes and minimize the sum of the estimated bandwidths of each mode. The constraint of the variational model is that the sum of the modes is equal to the signal to be decomposed. When calculating the spectral entropy of each modal component, the spectral entropy of the k-th modal component is calculated based on the normalized power spectrum of that modal component, using the following formula: in The normalized power spectrum of the k-th modal component; When calculating the energy percentage of each modal component, the formula for the energy percentage of the k-th modal component is as follows: in This is the time-domain expression for the k-th modal component.

Images