CN-121978630-A - Multi-snapshot partial discharge sound source positioning method based on non-convex joint sparse constraint

CN121978630ACN 121978630 ACN121978630 ACN 121978630ACN-121978630-A

Abstract

The invention relates to a multi-snapshot partial discharge sound source positioning method based on non-convex joint sparse constraint, and belongs to the technical field of partial discharge sound source positioning. According to the method, acoustic signals radiated to surrounding space by discharge of a microphone array acquisition device are utilized, a positioning problem is built into an optimization model of Laplace norm joint sparse constraint, a single snapshot model is cooperatively solved by utilizing differential convex programming and Nesterov accelerated soft threshold iteration, then a positioning algorithm capable of directly processing multi-sampling data is popularized by single snapshot according to joint sparsity, finally sound field distribution of a detection space is reconstructed, and the sound field distribution is visualized into an acoustic imaging graph, so that the position and the number of a discharge source are intuitively determined. According to the method, the Laplace non-convex sparse constraint is introduced, so that the spatial resolution and imaging quality of an algorithm are remarkably improved, and meanwhile, the positioning robustness of the multi-snapshot data in a noise and interference environment is effectively enhanced through joint solution of the multi-snapshot data.

Inventors

Zhu Jinchan
LIN BO
MA ZHENYU
LI XIAOSONG
WANG PING
LI CHUN
ZHANG XIAOBO
WEI XIAOXING
HOU MINGCHUN
YANG XIQING
Jing Maoheng
XIA GULIN

Assignees

重庆大学

Dates

Publication Date: 20260505
Application Date: 20260209

Claims (7)

1. A multi-snapshot partial discharge sound source positioning method based on non-convex joint sparse constraint is characterized by comprising the following steps: s1, establishing a partial-discharge acoustic signal model of a target detection area based on a planar microphone array; S2, modeling a partial-discharge sound source positioning problem as a least square model of Laplacian norm joint sparse constraint; S3, solving the model constructed in the step S2 by adopting a positioning algorithm, namely firstly, considering a single-measurement vector objective function, adopting a differential convex programming and a Nesterov accelerated soft threshold iterative algorithm to solve to obtain a single-snapshot version positioning algorithm; and S4, generating an acoustic imaging diagram based on the sound pressure distribution of the sound field based on the solving result of the positioning algorithm, and realizing the visual positioning of the partial-discharge sound source.
2. The multi-snapshot partial discharge sound source localization method according to claim 1, wherein in step S1, establishing a partial discharge sound signal model of the target detection area specifically includes the steps of: S11, establishing a microphone array with the center as an origin Is located in the plane of the array in a Cartesian coordinate system A plane surface of the glass fiber reinforced plastic plate, The axial direction is vertically directed to the target detection area, and the microphone array comprises Individual array elements whose set of coordinates is expressed as The target detection area is divided into At a distance of Is a uniform grid of (1), the grid coordinate set is Accordingly, the number of the components of the system, The coordinate set of the real sound sources is The real sound source can fall on the coordinates of any grid node, namely, the method satisfies And the larger the grid sound intensity is, the sound source exists at the coordinates The greater the probability of (2); s12, obtaining a continuous time from the microphone array Quick shot measuring signal Is a known quantity; sound intensity matrix of potential sound source at grid coordinates To be measured, wherein Representing complex number, the transfer matrix between array element and grid is Error matrix is Through Gaussian white noise simulation, then, the following linear partial-discharge acoustic signal model is constructed: Wherein the matrix Middle (f) Line 1 The elements of a column are defined as Wherein In units of imaginary numbers, Is the number of waves to be used, For the center frequency of the partial discharge signal, As the propagation velocity of sound in air, Representing an exponential function; represent the first Microphone and the first The euclidean distance between the individual grids, Denote the L 2 norm.
3. The method for positioning a multi-snapshot partial discharge sound source according to claim 2, wherein in step S2, the problem of positioning the partial discharge sound source is modeled as a least square model of Laplacian norms combined with sparse constraint, and specifically comprises the steps that the position of the sound source in sampling time is unchanged, and then the sound intensity matrix is obtained With uniform row sparsity, i.e The number of lines of the non-zero line vector is equal to the number of sound sources Based on this a priori information, the following objective function is constructed: Wherein, the Representing the Frobenius norm by the calculation method of , Representing matrix vectorization; is the (Laplace, L1) mixed norm, defined as , Is a matrix Middle (f) Line 1 The elements corresponding to the columns are arranged in a row, Is a control parameter; is a regularization parameter used to balance the degree of data fit and sparsity of the solution.
4. The multi-snapshot partial discharge sound source localization method according to claim 3, wherein in step S3, the sound intensity of the grid sound source is obtained by solving the objective function, and then the number and coordinates of the partial discharge sound sources are obtained by searching the local maxima of the grid sound intensity amplitude.
5. The multi-snapshot partial discharge sound source localization method according to claim 3, wherein in step S3, the specific steps of the localization algorithm for obtaining the single snapshot version are as follows: S301 pair sound intensity matrix Any row vector in (a) There is a single measurement vector objective function as follows: Wherein, the Regularization parameters of the single snapshot model; decomposing the original objective function into the difference of two convex functions by adopting the differential convex programming principle, and obtaining the approximate optimal solution of the overall objective function by processing the two sub-functions respectively, thereby rewriting the objective function as Wherein: S302, obviously, function Is a classical LASSO problem, in the first On each iteration, constant terms are ignored or introduced by the formulation, which translates equivalently into: Wherein, the For the iteration step size, Defined as a matrix Is the largest feature of (a); Representing a conjugate transpose operation; Pair function Under majorization-minimization framework, according to Is linearized into the first order concavity condition: Wherein, the Representing inner product, gradient matrix The elements in the method are calculated according to the following formula: Wherein, the Representing a complex modular length; Is a complex sign function defined as ; S303, to And Substituting the approximate expression of (a) into the overall objective function; similarly, the formulation was performed with the solution expressed as: Wherein, the Matrix defined for convenience of representation ; Representing the soft threshold operator(s), Middle (f) The values of the individual elements are defined as ; S304, accelerating the iterative solving process by adopting a Nesterov strategy and based on current and historical iterative information The following dynamic terms are constructed : Wherein, the Is a momentum parameter; replacing the object acted by the soft threshold operator with the result after the momentum acceleration to obtain the approximate optimal solution at the optimal first-order rate And repeating the soft threshold operation and the momentum acceleration step until the iteration stop condition is met, and obtaining the single snapshot version of the positioning algorithm.
6. The method for positioning a multi-snapshot partial discharge sound source according to claim 5, wherein in step S3, the specific steps of obtaining a positioning algorithm capable of directly processing multi-snapshot sampling data are as follows: S311 for multi-snapshot signals The objective function is generalized to the following multi-measurement vector form: Wherein, the Representing (L 2 ,L 1 ) a mixed norm by the method of calculation , Representation matrix Is the first of (2) A number of row vectors and, accordingly, In order to maintain consistency of the dimensions, According to Is calculated element by element; For the following The first of (3) Individual row vectors According to the Frobenius norm and the property of the (L 2 , L 1 ) mixed norm, the multi-measurement vector objective function is decoupled into a plurality of independent sub-functions Sub-objective functions corresponding to individual row vectors The method comprises the following steps: S312, th The solution at the time of iteration is expressed as , wherein, Is a scaling factor substituted into Is obtained by: Solving for For a pair of The first partial derivatives of (a) are: let it be zero, obtain At the position of Has the minimum value when I.e. Due to sound intensity Having a physical meaning other than negative, then the objective function is Has the minimum value when I.e. At the time of The place having the minimum value Expression of (c), th The solution at the time of iteration is: The expression is the popularization of a soft threshold operator from acting on elements to acting on row vectors, the soft threshold operator is named as a row vector threshold iterative operator, and the sign of the soft threshold operator is defined as Wherein the sign of the element is defined as When the line vector When only one element is contained in the vector, the row vector threshold iterative operator is equivalent to the soft threshold operator, and is uniformly written as: Similarly, the row vector threshold iteration operation is repeated And a momentum acceleration step of directly solving an original multi-snapshot objective function to obtain a multi-snapshot version of the positioning algorithm, wherein, Momentum acceleration terms corresponding to the multi-snapshot model are defined as ; S313, aiming at a multi-snapshot positioning algorithm, the regularization parameters are dynamically updated by adopting the following strategies: Wherein the symbols are Representation pair The first step after descending order Large values can be used as a benchmark for the threshold.
7. The system for realizing the multi-snapshot partial discharge sound source positioning method according to any one of claims 1-6 is characterized by comprising a microphone array, a partial discharge sound source positioning module and an acoustic imaging instrument, wherein the partial discharge sound source positioning module acquires a sound source target signal to be detected through the microphone array, then solves to obtain sound intensity distribution of a sound field by adopting the multi-snapshot partial discharge sound source positioning method based on non-convex joint sparse constraint, and finally sends the sound intensity distribution of the sound field to the acoustic imaging instrument to generate an acoustic imaging image so as to visualize the number and positions of the sound sources.

Description

Multi-snapshot partial discharge sound source positioning method based on non-convex joint sparse constraint Technical Field The invention belongs to the technical field of partial discharge sound source positioning, and relates to a multi-snapshot partial discharge sound source positioning method based on non-convex joint sparse constraint. Background The acoustic detection technology has the characteristics of non-invasiveness, high efficiency and electromagnetic interference resistance, and is an emerging power equipment state monitoring and fault detection means. The sound source positioning method based on the planar microphone array can realize sound field visualization, is gradually popularized in the field of electric power inspection in recent years, is used for detecting and positioning partial discharge of equipment, and has wide engineering application prospect. The beam forming algorithm (such as delay summation) is a typical sound source localization technology, and because of its simple principle and good robustness, the hand-held sound camera and other inspection equipment have been developed. However, this method is limited by the rayleigh criterion, has low spatial resolution, and is difficult to meet the high-precision requirement in the actual engineering scene. The deconvolution algorithm (such as DAMAS) effectively improves the resolution by deconvoluting the beam forming result, but relies on the complete cross spectrum matrix, has higher calculation complexity and is not beneficial to field application with higher real-time requirements. The method based on machine learning and deep learning improves the adaptability to complex environments through data driving, but the problems of scarcity of labeling data, limited model generalization capability and the like are faced in partial discharge detection, and the reliable application of the method in industrial scenes is restricted. In recent years, the rise of a compressed sensing theory promotes the rapid development of a compressed beam forming algorithm, and the method can still realize super-resolution positioning under the condition of few array elements based on sparsity priori of sound sources in space distribution, and is particularly suitable for typical sparse sound source scenes such as partial discharge (partial discharge usually occurs in a detection area at the same time in a few points). In terms of sparse characteristic modeling and constraint function selection, the common L 1 norm constraint is convex relaxation, and the sparse excitation capacity is limited, so that positioning deviation and sidelobe increase can be caused. Through research, the non-convex Laplace norm constraint function has strong sparse induction capability, and has good potential in inhibiting side lobes and improving multi-sound source resolution, but the related research is mainly focused on a theoretical level, and application verification and adaptation of partial discharge detection in power industry scenes are rare. In addition, the related research is concentrated on a single-measurement vector model, and the multi-sampling data cannot be effectively processed in a combined mode, so that the robustness is insufficient under the low signal-to-noise ratio or interference environment. Therefore, it is necessary to develop a sound source localization method with strong sparse excitation capability suitable for multi-snapshot data acquisition mode, so as to solve or alleviate the above problems and improve the accuracy and practicality of the local acoustic localization technology. Disclosure of Invention In view of the above, the invention aims to provide a multi-snapshot partial discharge sound source positioning method based on non-convex joint sparse constraint, which can be used for solving the problems of poor positioning precision and weak anti-interference capability of the traditional sound source positioning algorithm in partial discharge detection. According to the method, the resolution can be improved by suppressing side lobes, the environmental adaptability is enhanced by utilizing multi-snapshot sampling data, and accurate and robust local sound source detection and positioning are realized, so that the accuracy and the practicability of the local sound source positioning technology are improved. In order to achieve the above purpose, the present invention provides the following technical solutions: scheme 1. A multi-snapshot partial discharge sound source localization method based on non-convex joint sparse constraint specifically comprises the following steps: s1, establishing a partial-discharge acoustic signal model of a target detection area based on a planar microphone array; S2, modeling a partial-discharge sound source positioning problem as a least square model of Laplacian norm joint sparse constraint; S3, solving the model constructed in the step S2 by adopting a positioning algorithm, namely firstly, considering a sing