Transformer voiceprint denoising method and system based on edge computing

Abstract

The present invention discloses a transformer voiceprint denoising method and system based on edge computing. For the current power system or the servers of the platform, large-scale neural networks are processed to achieve the purpose of denoising, which has a large operation load, low efficiency, and the data analysis accuracy of extracting voiceprint features therein is generally average. The present invention determines a complete voiceprint sampling signal by processing the environmental noise characteristics in the transformer voiceprint sampling signal at the edge end, converts it into a frequency-domain signal and obtains the spectrum of the corresponding segmented signal, performs feature processing, constructs a transformer voiceprint denoising processing model and evaluates the spectrum data of the segmented signal, and the regional-level detection background automatically adjusts the gain according to the transformer denoising spectrum data of each segment. The present invention can achieve faster transformer denoising processing work to obtain a more accurate noise estimation result, which is beneficial to subsequent fault detection and diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer data processing, in particular to a transformer acoustic fingerprint denoising method and system based on edge computing. Background Art

[0002] Transformer acoustic fingerprint technology is a technology that uses sound signals to monitor and diagnose the operating status of power transformers. It analyzes the acoustic and vibration signals generated during the operation of the transformer, extracts the acoustic fingerprint features therein, and thus evaluates the health status of the transformer.

[0003] The vibration noise of the windings and iron core in the transformer is the main component of the transformer body noise. The sound generated by the transformer is related to the applied voltage; the main frequency components of the iron core vibration and winding vibration sounds are 100 hz; in addition, there are also influences from environmental noises such as cooling systems like fans and on-load tap-changers; for traditional transformer acoustic fingerprint denoising methods, most are to use the servers of the power system or platform for large-scale processing of BP neural network algorithms or CNN convolutional neural network algorithms to achieve the purpose of denoising. Their computing load is large, the efficiency is low, and the data analysis accuracy for extracting the acoustic fingerprint features therein is average. At the same time, traditional fault inspection and diagnosis rely on offline analysis in the later stage. Even for online monitoring, a large amount of data is transmitted to the server for processing and analysis. However, at present, there is a lack of edge-side processing of transformer acoustic fingerprint data. For this reason, we propose a transformer acoustic fingerprint denoising method and system based on edge computing. Summary of the Invention

[0004] In view of the above problems existing in the existing transformer acoustic fingerprint data processing, the present invention is proposed.

[0005] Therefore, one of the objectives of the present invention is to provide a transformer acoustic fingerprint denoising method and system based on edge computing. Through the edge-side processing of transformer acoustic fingerprint data, a transformer acoustic fingerprint denoising processing model based on lightweight edge computing at the edge is constructed, which can achieve faster transformer denoising processing work to obtain a more accurate noise estimation result, which is beneficial to subsequent fault detection and diagnosis.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] On the one hand, the present invention provides a transformer acoustic fingerprint denoising method based on edge computing, including:

[0008] S101. After the edge side obtains the transformer acoustic fingerprint sampling signal through the sensor, the environmental noise characteristics in the transformer acoustic fingerprint sampling signal are processed by the adopted filtering algorithm, and then a complete acoustic fingerprint sampling signal is determined. The complete acoustic fingerprint sampling signal is segmented into several segmented transformer acoustic fingerprint signals, and the width of each segmented transformer acoustic fingerprint signal is determined.

[0009] S102, after windowing each segment of the transformer voiceprint signal, Fourier transform is performed on the transformer voiceprint signal in the window, so that each segment of the transformer voiceprint signal is converted from a time domain signal to a frequency domain signal and a spectrum of a corresponding segmented signal is obtained;

[0010] S103. The edge end performs machine learning annotation on the spectrum of the segmented signal, determines the preset core vibration and winding vibration sound characteristics and the characteristics of the environmental noise influence and the corresponding parameter information according to the annotation processing results, builds a transformer voiceprint denoising processing model and performs spectrum data evaluation of the segmented signal, generates a frequency domain signal evaluation coefficient corresponding to the spectrum data of the segmented signal, calculates the energy value of the segmented frequency corresponding to the spectrum data of the segmented signal according to the frequency domain signal evaluation coefficient, that is, represents the characteristic parameters of the sound signals of different transformer noises, performs transformer voiceprint denoising based on the energy value of the segmented frequency and generates transformer denoised spectrum data of each segment, and transmits the transformer denoised spectrum data of each segment to the regional detection background;

[0011] S104, after the regional detection background automatically adjusts the gain according to the transformer denoising spectrum data of each segment, it is input into the neural network and cyclically identified and verified, and then the voiceprint is identified and matched according to the transformer voiceprint feature vector to complete the determination of the transformer fault, and at the same time, feedback is provided to verify whether the transformer voiceprint denoising meets the standard and output the corresponding control information to the edge end;

[0012] S105. The edge end receives corresponding control information and performs corresponding control work based on the transformer.

[0013] As a preferred solution of the present invention, in S102, the transformer voiceprint signal in the window is subjected to Fourier transform. Given a transformer voiceprint signal x(t) in a continuous time domain, its Fourier transform is expressed as follows:

[0014]

[0015] Where x(f) is a complex-valued function with frequency f, x(t) is the transformer voiceprint signal at the tth time point in the continuous time domain, i is an imaginary unit, f is the frequency in the continuous time domain, and dt is the differential sign.

[0016] As a preferred solution of the present invention, in S103, the transformer voiceprint denoising processing model is constructed and the spectrum data of the segmented signal is evaluated to generate the frequency domain signal evaluation coefficient corresponding to the spectrum data of the segmented signal, which is specifically as follows:

[0017]

[0018] Among them, Gi is the evaluation coefficient of the i-th frequency domain signal; respectively represent the weight coefficients of the three specification gradients of the edge transformer device; respectively represent the weight coefficients of the α, β, and χ indices corresponding to the three characteristics of core vibration, winding vibration, and environmental noise vibration; U(xi) represents the parameter value of the spectral data of the segmented signal.

[0019] As a preferred embodiment of the present invention, wherein: in S103, calculate the energy value of the segmented frequency corresponding to the spectral data of the segmented signal according to the frequency domain signal evaluation coefficient, specifically as follows:

[0020] E i = G i * ∫X i (t) 2 dt;

[0021] Wherein, E i is the energy value of the i-th frequency domain signal and serves as the characteristic parameter of the transformer noise sound signal, and X i (t) is the original data value of the transformer voiceprint signal on the i-th frequency domain signal.

[0022] As a preferred embodiment of the present invention, wherein: it further includes the analysis of the weights of core vibration and winding vibration, specifically as follows:

[0023] Assume that the magnitude of the load current flowing through the transformer winding is The magnitude of the electromagnetic force on the transformer winding is

[0024] Wherein, I is the value of the load current of the transformer winding, and I m is the amplitude of the load current, is the initial phase value, is the cosine quantity of the load current, p is the electromagnetic force coefficient, and it is taken as a constant under steady-state operating conditions, and F is the electromagnetic force on the winding;

[0025] The winding electromagnetic force is composed of a constant and a sine quantity, the frequency is twice the power supply frequency, and the magnitude is proportional to the square of the load current;

[0026] When the current contains harmonics, the frequency components of the electromagnetic force are the harmonic multiple frequencies, sum frequencies, and difference frequencies, and obtain the magnitude distribution information of the axial electromagnetic force and the radial electromagnetic force;

[0027] According to the information of the harmonic multiple frequency, sum frequency and difference frequency, and the magnitude distribution of the axial electromagnetic force and the radial electromagnetic force, a two-dimensional feature matrix of the corresponding harmonic category is constructed, which is the two-dimensional feature matrix of the multiple frequency, the two-dimensional feature matrix of the sum frequency, and the two-dimensional feature matrix of the difference frequency. That is, the two-dimensional feature matrix of the multiple frequency is expressed as F1(μ, γ), the two-dimensional feature matrix of the sum frequency is expressed as F2(μ, γ), and the two-dimensional feature matrix of the difference frequency is expressed as F3(μ, γ); where (μ, γ) are the electromagnetic force characteristic data in the axial and radial directions respectively;

[0028] Based on the two-dimensional feature matrix of the harmonic category, an axial and radial weight analysis model of the core vibration and the winding vibration is established, and the weight coefficients of the core vibration and the winding vibration are generated according to the analysis of the axial and radial weight analysis model of the core vibration and the winding vibration.

[0029] As a preferred embodiment of the present invention, wherein: the filtering algorithm in S101 adopts the least mean square error algorithm.

[0030] As a preferred embodiment of the present invention, wherein: the voiceprint recognition and matching in S104 use the Gaussian mixture model or the support vector machine for voiceprint recognition.

[0031] As a preferred embodiment of the present invention, wherein: the axial and radial weight analysis model of the core vibration and the winding vibration is specifically as follows:

[0032]

[0033] Wherein, n represents the i-th data of the transformer under the current harmonic category, that is, the i-th data of the two-dimensional feature matrix of the multiple frequency, sum frequency and difference frequency of the harmonic, j represents the index of the weight analysis, that is, the axial index of the core vibration and the winding vibration, k represents the radial index, and y ij represents the axial electromagnetic force characteristic data of the core vibration and the winding vibration under the current harmonic, and y ik represents the radial electromagnetic force characteristic data of the core vibration and the winding vibration under the current harmonic, p ij 、p ik are the axial and radial weight analysis results of the core vibration and the winding vibration respectively.

[0034] On the one hand, the present invention provides a system for a transformer voiceprint denoising method based on edge computing, including:

[0035] An edge-end acquisition module, which is used to process the environmental noise characteristics in the transformer voiceprint sampling signal by using a filtering algorithm after acquiring the transformer voiceprint sampling signal through a sensor, and determine a complete voiceprint sampling signal;

[0036] The edge - end processing module is used to perform windowing processing on each segment of the transformer sound - print signal, and then perform Fourier transform on the transformer sound - print signal within the window, so that each segment of the transformer sound - print signal is converted from a time - domain signal to a frequency - domain signal and the spectrum of the corresponding segmented signal is obtained;

[0037] The edge - end analysis module is used to perform machine - learning annotation on the spectrum of the segmented signal, determine the characteristics of the preset core vibration, winding vibration sound, and the influence of environmental noise and the corresponding parameter information according to the annotation processing result, construct a transformer sound - print denoising processing model and perform spectrum data evaluation on the spectrum of the segmented signal, generate a frequency - domain signal evaluation coefficient corresponding to the spectrum data of the segmented signal, and calculate the energy value of the segmented frequency corresponding to the spectrum data of the segmented signal according to the frequency - domain signal evaluation coefficient;

[0038] The regional - level detection background is used to automatically adjust the gain according to the transformer denoised spectrum data of each segment and then input it to the transformer cloud service platform;

[0039] The transformer cloud service platform is used to receive the upload information from the regional - level detection background, input it into the neural network and perform cyclic recognition and verification, then identify and match the sound - print according to the transformer sound - print feature vector to complete the determination of the transformer fault. At the same time, it feeds back and verifies whether the transformer sound - print denoising meets the standard and outputs the corresponding regulation information to the edge - end, and sends the corresponding regulation information to the transformer to perform the corresponding regulation work.

[0040] The beneficial effects of the present invention: Through the processing of the transformer sound - print data at the edge - end, the present invention constructs a transformer sound - print denoising processing model based on lightweight edge computing at the edge - end, which can more quickly perform the denoising processing work of the transformer to obtain a more accurate noise estimation result, which is beneficial to subsequent fault detection and diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to these drawings. Among them:

[0042] Figure 1 It is the flow block diagram of the method of the present invention;

[0043] Figure 2 It is another judgment flow block diagram of the method of the present invention;

[0044] Figure 3 It is the modular structure schematic diagram of the system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions of the embodiments of the present invention in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present invention.

[0046] Embodiment 1

[0047] Referring to Figure 1 and Figure 2 , which is an embodiment of the present invention. This embodiment provides a transformer acoustic fingerprint denoising method based on edge computing, including:

[0048] S101. After the edge device obtains the transformer acoustic fingerprint sampling signal through a sensor, the minimum mean square error algorithm, which can be used as the filtering algorithm, is adopted to process the environmental noise characteristics in the transformer acoustic fingerprint sampling signal, and then a complete acoustic fingerprint sampling signal is determined. The complete acoustic fingerprint sampling signal is segmented into several segmented transformer acoustic fingerprint signals, and the width of each segmented transformer acoustic fingerprint signal is determined.

[0049] S102. After each segmented transformer acoustic fingerprint signal is windowed, the Fourier transform is performed on the transformer acoustic fingerprint signal within the window, so that each segmented transformer acoustic fingerprint signal is converted from a time-domain signal to a frequency-domain signal, and the spectrum of the corresponding segmented signal is obtained.

[0050] S103. The edge device performs machine learning annotation on the spectrum of the segmented signal, determines the characteristics of the preset core vibration and winding vibration sounds and the characteristics and corresponding parameter information affected by environmental noise according to the annotation processing results, constructs a transformer acoustic fingerprint denoising processing model and conducts spectrum data evaluation on the spectrum of the segmented signal, generates a frequency-domain signal evaluation coefficient corresponding to the spectrum data of the segmented signal, calculates the energy value of the segmented frequency corresponding to the spectrum data of the segmented signal according to the frequency-domain signal evaluation coefficient, which represents the characteristic parameters of different noise sound signals of the transformer. Based on the energy value of the segmented frequency, transformer acoustic fingerprint denoising work is carried out, and the transformer denoising spectrum data of each segment is generated. The transformer denoising spectrum data of each segment is transmitted to the regional-level detection background.

[0051] S104. After the regional-level detection background automatically adjusts the gain according to the transformer denoising spectrum data of each segment, it is input into the neural network and repeatedly recognized and verified. According to the transformer acoustic fingerprint feature vector, acoustic fingerprint recognition and matching are performed to determine the transformer fault. At the same time, it is fed back to verify whether the transformer acoustic fingerprint denoising meets the standard and corresponding regulation information is output to the edge device, where Gaussian mixture model or support vector machine is used for acoustic fingerprint recognition and matching.

[0052] S105. The edge end receives corresponding control information and performs corresponding control work based on the transformer.

[0053] Specifically described in one embodiment, in S102, the transformer voiceprint signal in the window is Fourier transformed. Given a transformer voiceprint signal x(t) in a continuous time domain, its Fourier transform is expressed as follows:

[0054]

[0055] Where x(f) is a complex-valued function with frequency f, x(t) is the transformer voiceprint signal at the tth time point in the continuous time domain, i is an imaginary unit, f is the frequency in the continuous time domain, and dt is the differential sign.

[0056] In one embodiment, it should be emphasized that in S103, the transformer voiceprint denoising processing model is constructed and the spectrum data of the segmented signal is evaluated to generate the frequency domain signal evaluation coefficient corresponding to the spectrum data of the segmented signal, which is specifically as follows:

[0057]

[0058] Among them, G i is the evaluation coefficient of the i-th frequency domain signal; They represent the weight coefficients of the three specification gradients of the edge-end transformer equipment respectively; They represent the weight coefficients of the α, β and χ indicators corresponding to the three characteristics of core vibration, winding vibration and environmental noise vibration respectively; U(xi) represents the parameter value of the spectrum data of the segmented signal.

[0059] In addition, in S103, the energy value of the segmented frequency corresponding to the spectrum data of the segmented signal is calculated according to the frequency domain signal evaluation coefficient, as follows:

[0060] E i =G i *∫X i (t) 2 dt;

[0061] Among them, E i is the energy value of the ith frequency domain signal and is used as the characteristic parameter of the transformer noise sound signal. i (t) is the original data value of the transformer voiceprint signal in the i-th frequency domain signal.

[0062] Example 2

[0063] Further elaborating, in one embodiment, in addition to causing axial vibration, the alternating electromagnetic force also causes radial vibration of the winding. Compared with the research on the axial electromagnetic force of the winding, there has been less research on the radial vibration electromagnetic force in the past. At the same time, there is a lack of correlation analysis of the axial and radial vibrations caused by electromagnetic force at the edge end at the present stage. Due to reasons such as the increase in transformer capacity, more and more researchers have proposed to comprehensively consider axial and radial vibrations in order to obtain more accurate noise estimation results. Therefore, based on one embodiment, a second embodiment is given. In the second embodiment, it also includes an analysis of the weights of core vibration and winding vibration, as follows:

[0064] Assume that the magnitude of the load current flowing through the transformer winding is The magnitude of the electromagnetic force on the transformer winding is

[0065] where I is the value of the load current of the transformer winding, I m is the amplitude of the load current, is the initial phase value, is the cosine quantity of the load current, p is the electromagnetic force coefficient, and it is taken as a constant under steady-state operating conditions, F is the electromagnetic force on the winding;

[0066] The electromagnetic force of the winding consists of a constant and a sine quantity, the frequency is twice the power supply frequency, and the magnitude is directly proportional to the square of the load current;

[0067] When there are harmonics in the current, the frequency components of the electromagnetic force are the harmonic multiples, sum frequencies, and difference frequencies, and the magnitude distribution information of the axial and radial electromagnetic forces is obtained;

[0068] According to the harmonic multiples, sum frequencies, and difference frequencies and the magnitude distribution information of the axial and radial electromagnetic forces, corresponding two-dimensional characteristic matrices of harmonic categories are constructed, namely the two-dimensional characteristic matrix of multiples, the two-dimensional characteristic matrix of sum frequencies, and the two-dimensional characteristic matrix of difference frequencies. That is, the two-dimensional characteristic matrix of multiples is expressed as F1(μ, γ), the two-dimensional characteristic matrix of sum frequencies is expressed as F2(μ, γ), and the two-dimensional characteristic matrix of difference frequencies is expressed as F3(μ, γ); where (μ, γ) are the electromagnetic force characteristic data in the axial and radial directions respectively;

[0069] Based on the two-dimensional characteristic matrices of harmonic categories, an axial and radial weight analysis model of core vibration and winding vibration is established, and weight coefficients of core vibration and winding vibration are generated according to the analysis of the axial and radial weight analysis model of core vibration and winding vibration.

[0070] According to the content of the above Embodiment 2, this embodiment preferably gives the specific axial and radial weight analysis models of core vibration and winding vibration as follows:

[0071]

[0072] Among them, n represents the i-th data of the transformer under the current harmonic category, that is, the i-th data of the two-dimensional characteristic matrix of the harmonic multiple frequency, sum frequency and difference frequency. j represents the index of weight analysis, that is, the axial index of core vibration and winding vibration. k represents the radial index, and y ij represents the axial electromagnetic force characteristic data of core vibration and winding vibration under the current harmonic, and y ik represents the radial electromagnetic force characteristic data of core vibration and winding vibration under the current harmonic, and p ij and p ik are the weight analysis results of the axial and radial directions of core vibration and winding vibration respectively.

[0073] As Figure 2 shown, based on the above two embodiments, it can be seen that after processing and feature analysis of the collected voiceprint data, a transformer voiceprint denoising processing model for edge-side lightweight edge computing can be constructed. By performing spectral data evaluation on the segmented signals in the voiceprint denoising processing model, the frequency-domain signal evaluation coefficients corresponding to the spectral data of the segmented signals are generated. According to the frequency-domain signal evaluation coefficients, the energy values of the segmented frequencies corresponding to the spectral data of the segmented signals are calculated. According to the energy values of the segmented frequencies, transformer voiceprint denoising work is carried out and the transformer denoised spectral data of each segment is generated. Then, the transformer denoised spectral data of each segment is transmitted to the regional-level detection background. In this way, the amount of data processed by the platform can be effectively reduced, and through the accurate analysis and judgment of the energy values of the segmented frequencies, voiceprint denoising work on core vibration, winding vibration, and environmental noise can be carried out on transformers of different specifications at three levels. And while uploading to the platform for fault identification after automatic gain adjustment by the regional-level detection background, it is verified whether the transformer voiceprint denoising meets the standard and corresponding regulation information is output to the edge side, and corresponding regulation information is made. The overall process design is reasonable, and compared with traditional voiceprint denoising, it is faster in response, greatly reduces the central computing volume, has high accuracy, and is conducive to subsequent fault detection and diagnosis.

[0074] At the same time, as Figure 3 shown, this embodiment also gives a corresponding system based on the above-mentioned transformer voiceprint denoising method based on edge computing, including an edge computing terminal 10, a regional-level detection background 20, and a transformer cloud service platform 30. Among them, the edge computing terminal 10 of this system is provided with an edge-side acquisition module, an edge-side processing module, and an edge-side analysis module, and their functions are as follows:

[0075] The edge-side acquisition module is used to process the environmental noise characteristics in the transformer voiceprint sampling signal by using a filtering algorithm after obtaining the transformer voiceprint sampling signal through a sensor, and then determine a complete voiceprint sampling signal;

[0076] An edge processing module, for performing a windowing process on each segment of the transformer voiceprint signal, and then performing a Fourier transform on the transformer voiceprint signal in the window, so that each segment of the transformer voiceprint signal is converted from a time domain signal to a frequency domain signal and a spectrum of a corresponding segmented signal is obtained;

[0077] The edge analysis module is used to perform machine learning annotation on the spectrum of the segmented signal, determine the preset core vibration and winding vibration sound characteristics and the characteristics of the environmental noise impact and the corresponding parameter information according to the annotation processing results, build a transformer voiceprint denoising processing model and evaluate the spectrum data of the segmented signal, generate the frequency domain signal evaluation coefficient corresponding to the spectrum data of the segmented signal, and calculate the energy value of the segmented frequency corresponding to the spectrum data of the segmented signal according to the frequency domain signal evaluation coefficient;

[0078] The regional detection backend 20 is used to automatically adjust the gain according to the transformer denoising spectrum data of each section and input it to the transformer cloud service platform;

[0079] The transformer cloud service platform 30 is used to receive the uploaded information from the regional detection background, input it into the neural network and after cyclic identification and verification, identify and match the voiceprint according to the transformer voiceprint feature vector to complete the determination of the transformer fault, and at the same time feedback to verify whether the transformer voiceprint denoising meets the standard and output the corresponding control information to the edge end, and send the corresponding control information transmission value transformer to perform corresponding control work.

[0080] To sum up, the present invention constructs a transformer voiceprint denoising processing model based on lightweight edge computing at the edge through processing of transformer voiceprint data at the edge, which can realize faster transformer denoising processing to obtain more accurate noise estimation results, which is beneficial to subsequent fault detection and diagnosis.

[0081] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0082] In the description of this specification, the descriptions with reference to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0083] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of these features. In the description of the present application, "a plurality of" means two or more unless otherwise specifically defined.

[0084] Any process or method description represented in a flowchart or otherwise described herein may be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a specific logical function or process. And the scope of the preferred embodiments of the present application includes additional implementations, where the functions may be performed in a substantially simultaneous manner or in a reverse order according to the involved functions, rather than in the order shown or discussed.

[0085] The logic and / or steps represented in a flowchart or otherwise described herein, for example, may be considered as a sequenced list of executable instructions for implementing a logical function, and may be specifically implemented in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device).

[0086] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the method in the above embodiments can be completed by a program instructing relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0087] In addition, each functional unit in various embodiments of the present application may be integrated into one processing module, may exist physically alone for each unit, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. The storage medium may be a read-only memory, a magnetic disk, an optical disc, or the like.

[0088] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various changes or substitutions thereof, and these should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A transformer acoustic fingerprint denoising method based on edge computing, characterized in that, include: S101, after the edge end obtains the transformer voiceprint sampling signal through the sensor, the filtering algorithm is used to process the environmental noise characteristics in the transformer voiceprint sampling signal, and a complete voiceprint sampling signal is determined by dividing it into several segments of the transformer voiceprint signal and determining the width of each segment of the transformer voiceprint signal; S102, after windowing each segment of the transformer voiceprint signal, Fourier transform is performed on the transformer voiceprint signal in the window, so that each segment of the transformer voiceprint signal is converted from a time domain signal to a frequency domain signal and a spectrum of a corresponding segmented signal is obtained; S103, the edge end performs machine learning annotation on the spectrum of the segmented signal, determines the preset core vibration and winding vibration sound characteristics and the characteristics of environmental noise influence and corresponding parameter information according to the annotation processing results, builds a transformer voiceprint denoising processing model and performs spectrum data evaluation of the segmented signal, and generates a frequency domain signal evaluation coefficient corresponding to the spectrum data of the segmented signal, as follows: Among them, G i is the evaluation coefficient of the i-th frequency domain signal; respectively represent the weight coefficients of the three specification gradients of the edge transformer device; respectively represent the weight coefficients of the α, β, and χ indices corresponding to the three characteristics affected by core vibration, winding vibration, and environmental noise vibration; U(xi) represents the parameter value of the spectrum data of the segmented signal; The energy value of the segmented frequency corresponding to the spectrum data of the segmented signal is calculated according to the frequency domain signal evaluation coefficient, which represents the characteristic parameters of different noise sound signals of the transformer. The transformer voiceprint denoising work is performed based on the energy value of the segmented frequency and the transformer denoising spectrum data of each segment is generated. The transformer denoising spectrum data of each segment is transmitted to the regional detection background; S104, after the regional detection background automatically adjusts the gain according to the transformer denoising spectrum data of each segment, it is input into the neural network and cyclically identified and verified, and then the voiceprint is identified and matched according to the transformer voiceprint feature vector to complete the determination of the transformer fault, and at the same time, feedback is provided to verify whether the transformer voiceprint denoising meets the standard and output the corresponding control information to the edge end; S105. The edge end receives corresponding control information and performs corresponding control work based on the transformer.

2. The edge-computing-based transformer voiceprint denoising method according to claim 1, wherein In S102, the transformer voiceprint signal in the window is subjected to Fourier transform. Given a transformer voiceprint signal x(t) in a continuous time domain, its Fourier transform is expressed as follows: Where x(f) is a complex-valued function with frequency f, x(t) is the transformer voiceprint signal at the tth time point in the continuous time domain, i is an imaginary unit, f is the frequency in the continuous time domain, and dt is the differential sign.

3. The method for transformer voiceprint denoising based on edge computing according to claim 1, characterized in that In S103, the energy value of the segmented frequency corresponding to the spectrum data of the segmented signal is calculated according to the frequency domain signal evaluation coefficient, as follows: E i = G i * ∫ |X i (t)| 2 dt; Among them, E i is the energy value of the i-th frequency domain signal and serves as a characteristic parameter of the transformer noise sound signal, and X i (t) is the original data value of the transformer voiceprint signal on the i-th frequency domain signal.

4. The edge-computing-based transformer voiceprint denoising method according to claim 1, wherein It also includes the analysis of the core vibration and winding vibration weights, as follows: Assume that the magnitude of the load current flowing through the transformer winding is The magnitude of the electromagnetic force on the transformer winding is Among them, I is the load current value of the transformer winding, I m is the amplitude of the load current, is the initial phase value, is the cosine quantity of the load current, p is the electromagnetic force coefficient, and it is taken as a constant under steady-state operating conditions, and F is the electromagnetic force on the winding; The electromagnetic force of the winding is composed of a constant and a sinusoidal quantity, the frequency is twice the power supply frequency, and the magnitude is proportional to the square of the load current; When the current contains harmonics, the frequency components of the electromagnetic force are divided into harmonic multiples, sum frequencies and difference frequencies, and the distribution information of the axial electromagnetic force and the radial electromagnetic force is obtained; According to the information on the harmonic multiple frequencies, sum frequencies, difference frequencies, and the magnitude distributions of the axial electromagnetic force and the radial electromagnetic force, a two-dimensional feature matrix for the corresponding harmonic categories is constructed, namely the two-dimensional feature matrix for the multiple frequencies, the two-dimensional feature matrix for the sum frequencies, and the two-dimensional feature matrix for the difference frequencies. That is, the two-dimensional feature matrix for the multiple frequencies is denoted as F1(μ, γ), the two-dimensional feature matrix for the sum frequencies is denoted as F2(μ, γ), and the two-dimensional feature matrix for the difference frequencies is denoted as F3(μ, γ); where (μ, γ) are the electromagnetic force characteristic data in the axial and radial directions respectively. Based on the two-dimensional feature matrix of the harmonic categories, a weight analysis model for the axial and radial directions of the core vibration and the winding vibration is established, and the weight coefficients of the core vibration and the winding vibration are generated according to the analysis of the weight analysis model for the axial and radial directions of the core vibration and the winding vibration.

5. The method for denoising transformer voiceprint based on edge computing according to claim 1, characterized in that The filtering algorithm described in S101 uses the least mean square error algorithm.

6. The method for transformer voiceprint denoising based on edge computing according to claim 1, wherein The voiceprint recognition and matching described in S104 use the Gaussian mixture model or the support vector machine for voiceprint recognition.

7. The method for transformer voiceprint denoising based on edge computing according to claim 4, characterized in that The weight analysis model for the axial and radial directions of the core vibration and the winding vibration is specifically as follows: Among them, n represents the i-th data of the transformer under the current harmonic category, that is, the i-th data of the two-dimensional feature matrix of the harmonic multiple frequency, sum frequency and difference frequency. j represents the index of weight analysis, that is, the axial index of core vibration and winding vibration. k represents the radial index, and y ij represents the axial electromagnetic force characteristic data of core vibration and winding vibration under the current harmonic, and y ik represents the radial electromagnetic force characteristic data of core vibration and winding vibration under the current harmonic, and p ij and p ik are the weight analysis results of the axial and radial directions of core vibration and winding vibration respectively.

8. A system applied to the edge computing-based transformer voiceprint denoising method as described in claim 1, characterized in that, It includes: An edge-end acquisition module, which is used to process the environmental noise characteristics in the transformer voiceprint sampling signal by using the filtering algorithm after obtaining the transformer voiceprint sampling signal through a sensor, and then determine a complete voiceprint sampling signal. An edge-end processing module, which is used to perform windowing processing on each segment of the transformer voiceprint signal, and then perform Fourier transform on the transformer voiceprint signal within the window, so that each segment of the transformer voiceprint signal is converted from a time-domain signal to a frequency-domain signal and the spectrum of the corresponding segmented signal is obtained. An edge-end analysis module, which is used to perform machine learning annotation on the spectrum of the segmented signal, determine the characteristics of the preset core vibration and winding vibration sounds and the influence of environmental noise and the corresponding parameter information according to the annotation processing result, construct a transformer voiceprint denoising processing model and perform spectrum data evaluation on the spectrum of the segmented signal, generate a frequency-domain signal evaluation coefficient corresponding to the spectrum data of the segmented signal, and calculate the energy value of the segmented frequency corresponding to the spectrum data of the segmented signal according to the frequency-domain signal evaluation coefficient. A regional-level detection background, which is used to automatically adjust the gain according to the de-noised spectrum data of each segment of the transformer and then input it to the transformer cloud service platform. A transformer cloud service platform, which is used to receive the uploaded information from the regional-level detection background, input it into the neural network and perform cyclic recognition and verification, then identify and match the voiceprint according to the transformer voiceprint feature vector, complete the determination of the transformer fault, and at the same time feedback and verify whether the transformer voiceprint denoising meets the standard and output the corresponding regulation information to the edge-end, and send the corresponding regulation information transmission value to the transformer to perform the corresponding regulation work.

Images