Transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge calculation

Abstract

The invention discloses a transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge calculation, and the method comprises the steps: obtaining transformer voiceprint sampling data, carrying out the smooth gain equalization processing of the transformer voiceprint sampling data, and determining the transformer voiceprint sampling data as first voiceprint data; the edge computing end identifies and collects transformer abnormal state voiceprint features according to the first voiceprint data; constructing a transformer abnormal state diagnosis model, and generating voiceprint abnormal state evaluation result data of the second voiceprint data through analysis; according to the transformer abnormal state diagnosis method, edge calculation and cloud calculation are combined, a complex voiceprint diagnosis analysis task is processed by utilizing powerful calculation and storage capabilities of the cloud, and gain equalization processing of transformer voiceprint sample smoothing is carried out by utilizing an edge calculation end; the communication burden and time delay of the edge node are reduced, and the processing effect of the transformer voiceprint sample data is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer data processing, and in particular to a transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing. Background Art

[0002] A large number of acoustic and vibration signals are generated during the operation of power equipment, and these signals contain rich status information. Once a defect or failure occurs in the equipment, its soundprint characteristics will also change. By using acoustic fingerprint recognition technology, operation and maintenance personnel can promptly detect abnormal conditions of the equipment and accurately identify the causes of defects or failures. At present, the main acoustic diagnosis methods for transformers are: noise diagnosis method and ultrasonic diagnosis method. The noise diagnosis method uses the noise emitted by the equipment to diagnose faults. When the state of the parts or components of the equipment changes due to wear, aging, etc., the characteristics of its acoustic signal will also change accordingly. By monitoring these characteristics, the status of the equipment can be evaluated and faults can be discovered in a timely manner.

[0003] However, the current transformer voiceprint processing lacks lightweight edge computing to identify and fusion diagnose abnormal samples of abnormal voiceprints emitted by the transformer equipment, resulting in redundant and inefficient cloud processing data. Therefore, in order to address the problems of a large number of transformers, wide distribution, and average efficiency in identifying abnormal samples, we propose a transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing. Summary of the invention

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

[0005] Therefore, one of the objects of the present invention is to provide a transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing, which uses lightweight edge cloud collaboration to perform smooth gain equalization processing on the transformer voiceprint recognition stage, which is beneficial to the subsequent voiceprint denoising processing. At the same time, through the state diagnosis of abnormal voiceprints and the main cause analysis, the transformer fault type can be identified more accurately, which is beneficial to the subsequent fusion diagnosis and other work.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The present invention provides a transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing, including:

[0008] The transformer voiceprint sampling data generated by the transformer is obtained through the edge computing end, and the transformer voiceprint sampling data is subjected to feature extraction and cluster analysis based on the transformer voiceprint feature to generate transformer voiceprint sample data; at the same time, the edge computing end performs smooth gain equalization processing on the transformer voiceprint sample data according to each segment of preset transformer denoising spectrum data to form voiceprint sample data after the transformer denoising spectrum is formed, which is determined as the first voiceprint data;

[0009] The edge computing end receives the first voiceprint data and identifies and aggregates the transformer abnormal state voiceprint feature data according to the first voiceprint data to form transformer abnormal voiceprint feature data as the second voiceprint data, and generates a corresponding transformer abnormal voiceprint data set according to the second voiceprint data;

[0010] The cloud computing end responds to the second voiceprint data and performs priority scheduling and aggregation of task processing, and at the same time builds a transformer abnormal state diagnosis model according to the second voiceprint data, and generates voiceprint abnormal state evaluation result data of the second voiceprint data through analysis;

[0011] Based on the voiceprint abnormal state assessment result data, a similarity analysis is performed with each fault voiceprint spectrum feature in the transformer fault feature database, and the transformer abnormal state diagnosis result data is obtained according to the similarity analysis result, and the transformer abnormal state diagnosis result data is transmitted to the edge computing end for feedback.

[0012] As a preferred solution of the present invention, the edge computing end performs smooth gain equalization processing on the transformer voiceprint sample data according to each segment of preset transformer denoising spectrum data, as follows:

[0013] Step S11, the edge computing end obtains the transformer voiceprint sample data of each segment and determines the spectrum information of the segmented signal in the transformer voiceprint sample data processing corresponding to the previous time period based on the complete transformer voiceprint sampling data signal collected in the previous time period in history;

[0014] Step S12, based on the preset smoothing factor of the historical voiceprint sampling signal, the preset smoothing factor is used as the weight value of future gain equalization, and the preset smoothing factor is updated through weighted average processing, and then determined as the current smoothing factor information;

[0015] Step S13, according to the spectrum information of the segmented signal corresponding to the previous time period and the current smoothing factor information, and based on the adjustment threshold range, the transformer denoised spectrum data after gain equalization processing is generated by calculation and analysis, and then the transformer denoised spectrum data after gain equalization processing is input into the BP neural network for wavelet threshold denoising processing.

[0016] As a preferred solution of the present invention, the specific steps of the wavelet threshold denoising are as follows:

[0017] Step S131, selecting a wavelet basis and determining a number of decomposition levels according to the abnormal transformer soundprint signal characteristics;

[0018] Step S132, performing wavelet decomposition on the noisy signal in the transformer denoised spectrum data to obtain a plurality of wavelet coefficients;

[0019] Step S133, determining a preset threshold function and selecting a threshold, and quantizing the wavelet coefficients with the noisy signal;

[0020] Step S134, after quantization, reconstruct each of the wavelet coefficients to obtain the final transformer abnormal voiceprint denoising result of the transformer voiceprint sample data.

[0021] As a preferred solution of the present invention, the threshold function preset by the determination is specifically divided into a hard threshold and a soft threshold, and the hard threshold and the soft threshold are respectively shown in the following formulas:

[0022]

[0023] in, is the hard threshold calculation value, is the soft threshold calculation value, W j,k is the wavelet coefficient of the original signal, λ is the threshold, and sgn is the sign function.

[0024] As a preferred solution of the present invention, wherein: the selection threshold adopts one of unbiased likelihood estimation, fixed value estimation, heuristic threshold estimation and extreme value threshold estimation;

[0025] Among them, the threshold selection of the unbiased likelihood estimation adopts the Stein unbiased likelihood method to realize the estimation of the threshold; the threshold selection of the heuristic threshold estimation is a compromise calculation of the unbiased likelihood estimation and the fixed threshold estimation. Specifically, when the signal-to-noise ratio is lower than the preset value, the fixed threshold method is adopted, and when the signal-to-noise ratio is higher than the preset value, the result with better effect between the unbiased likelihood estimation and the fixed threshold estimation is taken as the final value; the threshold selection of the extreme value threshold estimation is to select the threshold by adopting the maximum or minimum value.

[0026] As a preferred solution of the present invention, the selection threshold is estimated by a fixed value, wherein the calculation of the fixed value estimation is shown in the following formula:

[0027]

[0028] Among them, σ is the standard deviation of the noise, N is the length of the noisy signal, and ln is the natural logarithmic function expression.

[0029] As a preferred solution of the present invention, the priority scheduling and aggregation of task processing, wherein task processing aggregation is to aggregate tasks with similar computing requirements or data dependencies into batches for parallel execution.

[0030] As a preferred solution of the present invention, the transformer abnormal state diagnosis model is calculated using a Gaussian kernel function, as follows:

[0031]

[0032] Where exp is the Gaussian kernel function expression, x and y are both input vectors, ||xy|| is the Euclidean distance between input vector x and input vector y, δ is a positive parameter used to control the width of the Gaussian kernel, and K(x,y) is the analysis result of the Gaussian kernel function.

[0033] As a preferred solution of the present invention, the fault types include current loop heating, partial discharge, insulation moisture, core fault, winding fault, transformer oil deterioration, and insulation aging.

[0034] As a preferred solution of the present invention, wherein: the transformer abnormal state diagnosis result data is obtained according to the similarity analysis result, and the main cause analysis of the fault type is also performed by determining the internal constraint index;

[0035] The determination of the intrinsic constraint index is specifically carried out by constructing an abnormal state set of fault types, and classifying the transformer fault characteristic parameters into samples of the abnormal state set; the abnormal state types in the abnormal state sets of different fault types are numbered, and then the characteristic parameters of the abnormal state types with close similarity are classified into them, and at the same time, different abnormal state type characteristic parameters are allowed to appear repeatedly in abnormal state sets of different fault types, and the corresponding intrinsic constraint index is calculated and determined through the sensitive distance of the characteristic parameters of each abnormal state type.

[0036] Beneficial effects of the present invention: The present invention uses lightweight edge cloud collaboration to perform smooth gain equalization processing on the transformer voiceprint recognition stage, which is beneficial to the subsequent voiceprint denoising processing. At the same time, through the status diagnosis of abnormal voiceprints and the main cause analysis, it can more accurately complete the identification of transformer fault types, which is beneficial to subsequent fusion diagnosis and other work. In addition, the present invention combines edge computing with cloud computing, using the powerful computing and storage capabilities of the cloud to process complex voiceprint diagnosis and analysis tasks, and using the edge computing end to perform smooth gain equalization processing on transformer voiceprint samples, reducing the communication burden and delay of edge nodes, and improving the processing effect of transformer voiceprint sample data. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0038] in:

[0039] Figure 1 It is a flow chart of a transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing in an embodiment of the present invention;

[0040] Figure 2 It is a flow chart of performing smooth gain equalization processing on transformer voiceprint sample data in an embodiment of the present invention;

[0041] Figure 3 Flow chart of wavelet threshold denoising in an embodiment of the present invention; DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0043] In order to improve the accuracy and robustness in dynamic noisy environments and complex working conditions, this embodiment proposes a transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing to deal with problems caused by environmental noise, mixed faults and unknown faults. The experimental results show that the proposed state fusion monitoring and fault diagnosis system can quickly distinguish the faulty working state and diagnose six typical faults and mixed faults of power transformers in different noisy environments with an accuracy rate of more than 97.83%.

[0044] Reference Figure 1-Figure 3 , is an embodiment of the present invention, which provides a transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing, including:

[0045] refer to Figure 1As shown, the transformer voiceprint sampling data generated by the transformer is obtained through the edge computing end, and the transformer voiceprint sampling data is subjected to feature extraction and cluster analysis based on the transformer voiceprint feature to generate the transformer voiceprint sample data; at the same time, the edge computing end performs smooth gain equalization processing on the transformer voiceprint sample data according to the preset transformer denoising spectrum data of each segment, and forms the voiceprint sample data after the transformer denoising spectrum is formed, which is determined as the first voiceprint data;

[0046] The edge computing end receives the first voiceprint data and identifies and collects the transformer abnormal state voiceprint feature data according to the first voiceprint data to form transformer abnormal voiceprint feature data, which is determined as the second voiceprint data, and generates a corresponding transformer abnormal voiceprint data set according to the second voiceprint data;

[0047] The cloud computing end responds to the second voiceprint data and performs priority scheduling and aggregation of task processing, and at the same time builds a transformer abnormal state diagnosis model according to the second voiceprint data, and generates voiceprint abnormal state evaluation result data of the second voiceprint data through analysis;

[0048] Based on the voiceprint abnormal state assessment result data, a similarity analysis is performed with the various fault voiceprint spectrum features in the transformer fault feature database, and the transformer abnormal state diagnosis result data is obtained based on the similarity analysis results. The transformer abnormal state diagnosis result data is then transmitted to the edge computing end for feedback.

[0049] refer to Figure 2 As shown, in this embodiment, preferably, the edge computing end performs smooth gain equalization processing on the transformer voiceprint sample data according to each segment of preset transformer denoising spectrum data, as follows:

[0050] Step S11, the edge computing end obtains the transformer voiceprint sample data of each segment and determines the spectrum information of the segmented signal in the transformer voiceprint sample data processing corresponding to the previous time period based on the complete transformer voiceprint sampling data signal collected in the previous time period in history;

[0051] Step S12, based on the preset smoothing factor of the historical voiceprint sampling signal, the preset smoothing factor is used as the weight value of future gain equalization, and the preset smoothing factor is updated through weighted average processing, and then determined as the current smoothing factor information;

[0052] Step S13, according to the spectrum information of the segmented signal corresponding to the previous time period and the current smoothing factor information, and based on the adjustment threshold range, the transformer denoised spectrum data after gain equalization processing is generated by calculation and analysis, and then the transformer denoised spectrum data after gain equalization processing is input into the BP neural network for wavelet threshold denoising processing.

[0053] refer to Figure 3As shown, in the above embodiment, the specific steps of wavelet threshold denoising are as follows:

[0054] Step S131, selecting a wavelet basis and determining a number of decomposition levels according to the abnormal transformer soundprint signal characteristics;

[0055] Step S132, performing wavelet decomposition on the noisy signal in the transformer denoised spectrum data to obtain a plurality of wavelet coefficients;

[0056] Step S133, determining a preset threshold function and selecting a threshold, and quantizing the wavelet coefficients with the noisy signal;

[0057] Step S134, after quantization, reconstruct each wavelet coefficient to obtain the final transformer abnormal voiceprint denoising result of the transformer voiceprint sample data.

[0058] It should be emphasized in this embodiment that the preset threshold function is determined. Specifically, the threshold function is divided into a hard threshold and a soft threshold. The hard threshold and the soft threshold are respectively shown in the following formulas:

[0059]

[0060] in, is the hard threshold calculation value, is the soft threshold calculation value, W j,k is the wavelet coefficient of the original signal, λ is the threshold, and sgn is the sign function.

[0061] In this embodiment, preferably, the threshold is selected using one of unbiased likelihood estimation, fixed value estimation, heuristic threshold estimation and extreme value threshold estimation;

[0062] Among them, the threshold selection of unbiased likelihood estimation adopts Stein unbiased likelihood method to realize the estimation of threshold; the threshold selection of heuristic threshold estimation is a compromise calculation of unbiased likelihood estimation and fixed threshold estimation. Specifically, when the signal-to-noise ratio is lower than the preset value, the fixed threshold method is adopted. When the signal-to-noise ratio is higher than the preset value, the result with better effect between unbiased likelihood estimation and fixed threshold estimation is taken as the final value; the threshold selection of extreme value threshold estimation is to select the threshold by using maximum or minimum value.

[0063] Specifically, in this embodiment, the threshold is selected using a fixed value estimate, wherein the calculation of the fixed value estimate is shown in the following formula:

[0064]

[0065] Among them, σ is the standard deviation of the noise, N is the length of the noisy signal, and ln is the natural logarithmic function expression.

[0066] Preferably, the present embodiment performs priority scheduling and aggregation of task processing, wherein task processing aggregation is aggregating tasks with similar computing requirements or data dependencies into batches for parallel execution.

[0067] In addition, in this embodiment, preferably, the transformer abnormal state diagnosis model is calculated using a Gaussian kernel function, as follows:

[0068]

[0069] Where exp is the Gaussian kernel function expression, x and y are both input vectors, ||xy|| is the Euclidean distance between input vector x and input vector y, δ is a positive parameter used to control the width of the Gaussian kernel, and K(x,y) is the analysis result of the Gaussian kernel function.

[0070] In this embodiment, preferably, the fault types include current loop heating, partial discharge, insulation moisture, core fault, winding fault, transformer oil deterioration, and insulation aging.

[0071] This embodiment further illustrates that the transformer abnormal state diagnosis result data is obtained according to the similarity analysis result, and also includes performing a main cause analysis of the fault type by determining the internal constraint index;

[0072] Specifically, the intrinsic constraint index is determined by constructing an abnormal state set of fault types and classifying the samples of the abnormal state set according to the characteristic parameters of transformer faults; the abnormal state types in the abnormal state sets of different fault types are numbered, and then the characteristic parameters of the abnormal state types with close similarity are classified into them, and at the same time, different abnormal state type characteristic parameters are allowed to appear repeatedly in abnormal state sets of different fault types, and the corresponding intrinsic constraint index is calculated and determined through the sensitive distance of the characteristic parameters of each abnormal state type.

[0073] In summary, this embodiment can perform smooth gain equalization processing on the transformer voiceprint recognition stage by using lightweight edge cloud collaboration, which is beneficial to the subsequent voiceprint denoising processing. At the same time, through the status diagnosis of abnormal voiceprints and the main cause analysis, it can more accurately complete the identification of transformer fault types, which is beneficial to subsequent fusion diagnosis and other work. In addition, the present invention combines edge computing with cloud computing, uses the powerful computing and storage capabilities of the cloud to process complex voiceprint diagnosis and analysis tasks, and uses the edge computing end to perform smooth gain equalization processing on transformer voiceprint samples, reducing the communication burden and delay of edge nodes, and improving the processing effect of transformer voiceprint sample data.

[0074] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction 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 any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0075] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0076] Any process or method description in the flow chart or otherwise described herein can be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0077] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0078] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0079] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can 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 can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0080] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing, characterized in that: include The transformer voiceprint sampling data generated by the transformer is obtained through the edge computing end, and the transformer voiceprint sampling data is subjected to feature extraction and cluster analysis based on the transformer voiceprint feature to generate transformer voiceprint sample data; at the same time, the edge computing end performs smooth gain equalization processing on the transformer voiceprint sample data according to each segment of preset transformer denoising spectrum data to form voiceprint sample data after the transformer denoising spectrum is formed, which is determined as the first voiceprint data; The edge computing end receives the first voiceprint data and identifies and aggregates the transformer abnormal state voiceprint feature data according to the first voiceprint data to form transformer abnormal voiceprint feature data as the second voiceprint data, and generates a corresponding transformer abnormal voiceprint data set according to the second voiceprint data; The cloud computing end responds to the second voiceprint data and performs priority scheduling and aggregation of task processing, and at the same time builds a transformer abnormal state diagnosis model according to the second voiceprint data, and generates voiceprint abnormal state evaluation result data of the second voiceprint data through analysis; Based on the voiceprint abnormal state assessment result data, a similarity analysis is performed with each fault voiceprint spectrum feature in the transformer fault feature database, and the transformer abnormal state diagnosis result data is obtained according to the similarity analysis result, and the transformer abnormal state diagnosis result data is transmitted to the edge computing end for feedback.

2. The transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing according to claim 1 is characterized in that: The edge computing end performs smooth gain equalization processing on the transformer voiceprint sample data according to each segment of preset transformer denoising spectrum data, as follows: Step S11, the edge computing end obtains the transformer voiceprint sample data of each segment and determines the spectrum information of the segmented signal in the transformer voiceprint sample data processing corresponding to the previous time period based on the complete transformer voiceprint sampling data signal collected in the previous time period in history; Step S12, based on the preset smoothing factor of the historical voiceprint sampling signal, the preset smoothing factor is used as the weight value of future gain equalization, and the preset smoothing factor is updated through weighted average processing, and then determined as the current smoothing factor information; Step S13, according to the spectrum information of the segmented signal corresponding to the previous time period and the current smoothing factor information, and based on the adjustment threshold range, the transformer denoised spectrum data after gain equalization processing is generated by calculation and analysis, and then the transformer denoised spectrum data after gain equalization processing is input into the BP neural network for wavelet threshold denoising processing.

3. The transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing according to claim 2 is characterized in that: The specific steps of wavelet threshold denoising are as follows: Step S131, selecting a wavelet basis and determining a number of decomposition levels according to the abnormal transformer soundprint signal characteristics; Step S132, performing wavelet decomposition on the noisy signal in the transformer denoised spectrum data to obtain a plurality of wavelet coefficients; Step S133, determining a preset threshold function and selecting a threshold, and quantizing the wavelet coefficients with the noisy signal; Step S134, after quantization, reconstruct each of the wavelet coefficients to obtain the final transformer abnormal voiceprint denoising result of the transformer voiceprint sample data.

4. The transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing according to claim 3 is characterized in that: The threshold function preset in the determination is specifically divided into a hard threshold and a soft threshold, and the hard threshold and the soft threshold are respectively shown in the following formulas: in, is the hard threshold calculation value, is the soft threshold calculation value, W j,k is the wavelet coefficient of the original signal, λ is the threshold, and sgn is the sign function.

5. The transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing according to claim 3 is characterized in that: The selection threshold adopts one of unbiased likelihood estimation, fixed value estimation, heuristic threshold estimation and extreme value threshold estimation; Among them, the threshold selection of the unbiased likelihood estimation adopts the Stein unbiased likelihood method to realize the estimation of the threshold; the threshold selection of the heuristic threshold estimation is a compromise calculation of the unbiased likelihood estimation and the fixed threshold estimation. Specifically, when the signal-to-noise ratio is lower than the preset value, the fixed threshold method is adopted, and when the signal-to-noise ratio is higher than the preset value, the result with better effect between the unbiased likelihood estimation and the fixed threshold estimation is taken as the final value; the threshold selection of the extreme value threshold estimation is to select the threshold by adopting the maximum or minimum value.

6. The transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing according to claim 5 is characterized in that: The selection threshold is estimated by a fixed value, wherein the calculation of the fixed value estimate is shown in the following formula: Among them, σ is the standard deviation of the noise, N is the length of the noisy signal, and ln is the natural logarithmic function expression.

7. The transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing according to claim 1 is characterized in that: The priority scheduling and aggregation of task processing, wherein task processing aggregation is to aggregate tasks with similar computing requirements or data dependencies into batches for parallel execution.

8. The transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing according to claim 1 is characterized in that: The transformer abnormal state diagnosis model is calculated using a Gaussian kernel function, as follows: Where exp is the Gaussian kernel function expression, x and y are both input vectors, ||xy|| is the Euclidean distance between input vector x and input vector y, δ is a positive parameter used to control the width of the Gaussian kernel, and K(x,y) is the analysis result of the Gaussian kernel function.

9. The transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing according to claim 1 is characterized in that: The fault types include current loop heating, partial discharge, insulation moisture, core fault, winding fault, transformer oil deterioration, and insulation aging.

10. The transformer voiceprint recognition and abnormal state diagnosis method based on lightweight edge computing according to claim 9 is characterized in that: The transformer abnormal state diagnosis result data is obtained based on the similarity analysis results, and the main cause analysis of the fault type is also performed by determining the internal constraint index; The determination of the intrinsic constraint index is specifically carried out by constructing an abnormal state set of fault types, and classifying the transformer fault characteristic parameters into samples of the abnormal state set; the abnormal state types in the abnormal state sets of different fault types are numbered, and then the characteristic parameters of the abnormal state types with close similarity are classified into them, and at the same time, different abnormal state type characteristic parameters are allowed to appear repeatedly in abnormal state sets of different fault types, and the corresponding intrinsic constraint index is calculated and determined through the sensitive distance of the characteristic parameters of each abnormal state type.

Images