Method and system for extracting characteristic signal under low signal-to-noise ratio

Abstract

The invention discloses a method and a system for extracting a characteristic signal under a low signal-to-noise ratio. The method comprises the following steps: acquiring a pilot frequency signal of a detected mutual inductor; cascading a Gabor sparse complete dictionary with a Fourier sparse dictionary to construct an over-complete dictionary of the pilot frequency signals; performing sparse representation on the pilot frequency signal through the over-complete dictionary, and decomposing the pilot frequency signal by adopting a sparse representation dictionary learning algorithm K-SVD to generate a learning atomic dictionary; atoms matched with the pilot frequency signals are selected from the over-complete dictionary, and the matched atoms are linearly combined to obtain reconstructed pilot frequency calibration signals; and when the precision of the reconstructed pilot frequency calibration signal reaches a preset threshold value or reaches a preset number of iterations, extracting a feature signal in the pilot frequency signal. The signal-to-noise ratio of the collected signal is improved, and the characteristics of the frequency signal needing to be extracted under the frequency mixing condition can be accurately and efficiently obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a method and system for extracting characteristic signals under low signal-to-noise ratio. Background Art

[0002] The heterofrequency method of live calibration technology is a technology that first injects a cancellation signal with the same amplitude as the power frequency background signal and a phase difference of 180 degrees, and then injects a non-power frequency signal to calibrate the full range of the transformer. Due to the complex environment in which the transformer is located and the uncontrollable amplitude of the working current, the conventional calibration method cannot ensure that the calibration current is stable and controllable. Therefore, it is innovatively proposed to use a current different from the grid frequency, that is, a heterofrequency current, to carry out transformer calibration under live conditions.

[0003] However, the complex environment on site and the incomplete cancellation of background signals result in the final calibration signal of the tested transformer containing multiple components such as harmonics, noise, residual power frequency signals, and heterodyne signals, which cannot be consistent with the signal flowing into the standard transformer, making it impossible to calibrate the transformer in operation according to the verification regulations. Therefore, it is urgent to develop the extraction technology of heterodyne characteristic signals under low signal-to-noise ratio. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for extracting characteristic signals under low signal-to-noise ratio, comprising:

[0005] Collect the heterodyne signal of the transformer under test;

[0006] Constructing an overcomplete dictionary of the heterofrequency signal by cascading a Gabor sparse complete dictionary and a Fourier sparse dictionary;

[0007] The heterofrequency signal is sparsely represented by the overcomplete dictionary, and the heterofrequency signal is decomposed by using a sparse representation dictionary learning algorithm K-SVD to generate a learning atomic dictionary;

[0008] Atoms matching the heterofrequency signal are selected from the overcomplete dictionary, and a reconstructed heterofrequency calibration signal is obtained by linearly combining the matching atoms; when the accuracy of the reconstructed heterofrequency calibration signal reaches a preset threshold or reaches a preset number of iterations, a characteristic signal in the heterofrequency signal is extracted.

[0009] Furthermore, the expression of the different frequency signal of the measured transformer is:

[0010]

[0011] Where A is the amplitude, ω is the angular frequency of the signal, is the signal phase. t is the time

[0012] Furthermore, the heterodyne signal is sparsely represented by the overcomplete synthetic dictionary, wherein:

[0013] The Gabor sparse complete dictionary expression is:

[0014]

[0015] in, is the standard Gaussian window function; s a is the scale factor; u x is the displacement factor; ω f is the frequency modulation factor, is the signal phase, K r is the normalization factor, so that ||g r (t)|| 2 =1;

[0016] Discretize the Gabor sparse complete dictionary expression:

[0017]

[0018] Where λ is the atomic parameter set, where 0<p<log 2 N,0≤k≤2 1+p ,q,p,k∈Z, a=2,Δu=1 / 2,Δω=π,

[0019] For heterofrequency signals with a time window length of N, the corresponding overcomplete dictionary is:

[0020] ψ=[ψ G ψ F ]

[0021] Among them, ψ G is the Gabor sparse complete dictionary, g rn is the nth atom in the Gabor overcomplete dictionary, n>>; ψ F is the Fourier sparse complete dictionary;

[0022]

[0023] At this time, the heterodyne signal collected in the wth time window can be sparsely represented under the overcomplete synthetic dictionary as:

[0024]

[0025] Among them, Y i is the ith signal, X i is the ith atom.

[0026] Furthermore, a sparse representation dictionary learning algorithm K-SVD is used to decompose the error of the heterofrequency signal to generate a learning atomic dictionary, including:

[0027] Use the K-SVD algorithm to learn and obtain the learning atomic dictionary by solving the following formula:

[0028]

[0029] Among them, Q j is the sparse signal to be solved; P j is the coefficient matrix; D learn is a dictionary; T is a sufficiently small constant; M is the number of atoms.

[0030] Further, selecting atoms matching the heterofrequency signal from the overcomplete dictionary, and obtaining a reconstructed heterofrequency calibration signal by linearly combining the matching atoms, comprises:

[0031] Using the orthogonal projection method, select the atom x′ that matches the heterofrequency signal y′ from the overcomplete dictionary 0 , satisfying the following formula:

[0032] | <y′,x 0 ′>|=sup| <y′,x j ′>|

[0033] Among them, y' is the collected signal; x' j is the jth atom in the overcomplete dictionary; sup is the geometric supremum;

[0034] The matching atom matrix is ​​constructed by linearly combining the selected atoms;

[0035] Based on the matching atomic matrix, obtaining its column orthogonal projection operator, using the current signal residual and the orthogonal projection operator, constructing a sparse approximation method to update the residual signal;

[0036] The linear combination of the matching atom matrices plus the updated residual signal represents the reconstructed inter-frequency calibration signal.

[0037] Further, when the accuracy of the reconstructed inter-frequency calibration signal reaches a preset threshold or reaches a preset number of iterations, extracting a characteristic signal from the inter-frequency signal includes:

[0038] The atoms in the reconstructed heterofrequency standard signal are Schmidt orthogonalized, and the number of iterations is U;

[0039] Repeat the above steps until the accuracy of the residual signal reaches a preset threshold or reaches a preset number of iterations, terminate the above steps, and extract the characteristic signal from the heterofrequency signal.

[0040] The present invention also provides a system for extracting characteristic signals under low signal-to-noise ratio, comprising:

[0041] The heterodyne signal acquisition module is used to acquire the heterodyne signal of the transformer under test;

[0042] An overcomplete dictionary construction module, used for constructing an overcomplete dictionary of the heterofrequency signal by cascading a Gabor sparse complete dictionary and a Fourier sparse dictionary;

[0043] An atomic dictionary generation module, used for sparsely representing the heterofrequency signal through the overcomplete dictionary, and decomposing the heterofrequency signal using a sparse representation dictionary learning algorithm K-SVD to generate a learning atomic dictionary;

[0044] The characteristic signal extraction module is used to select atoms matching the heterofrequency signal from the overcomplete dictionary, and obtain a reconstructed heterofrequency calibration signal by linearly combining the matching atoms; when the accuracy of the reconstructed heterofrequency calibration signal reaches a preset threshold or reaches a preset number of iterations, extract the characteristic signal in the heterofrequency signal.

[0045] Furthermore, the expression of the different frequency signal of the measured transformer is:

[0046]

[0047] Where A is the amplitude, ω is the angular frequency of the signal, is the signal phase and t is the time.

[0048] Furthermore, the characteristic signal extraction module includes:

[0049] A matching submodule, for selecting an atom x′ that matches the hetero-frequency signal y′ from the overcomplete dictionary using an orthogonal projection method 0 , satisfying the following formula:

[0050] | <y′,x 0 ′>|=sup| <y′,x j ′>|

[0051] Among them, y' is the collected signal; x' j is the jth atom in the overcomplete dictionary; sup is the geometric supremum;

[0052] A matrix construction module is used to construct a matching atom matrix by linearly combining the selected atoms;

[0053] A residual updating module, used to obtain the column orthogonal projection operator of the matching atomic matrix based on the matching atomic matrix, and to construct a sparse approximation method to update the residual signal using the current signal residual and the orthogonal projection operator;

[0054] The reconstruction submodule is used to add the linear combination of the matching atomic matrix to the updated residual signal to represent the reconstructed heterofrequency calibration signal.

[0055] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in any one of the above claims are implemented.

[0056] The present invention adopts the collected reference signal and the measured signal to automatically generate an overcomplete atomic dictionary, obtains the atoms of the collected signal through the orthogonal matching pursuit (OMP) method, obtains the atomic linear combination that is finally matched after sparse decomposition, and obtains the hetero-frequency signal through the atomic linear combination; compared with the existing fast Fourier transform, wavelet transform and phase-locked technology, compared with the Fourier transform, the present invention does not need to subjectively judge the frequency range of the hetero-frequency signal, and has good adaptability; compared with the wavelet transform, it is not necessary to determine the wavelet type and the number of decomposition layers; compared with the phase-locked technology, the signal reconstruction accuracy is higher. Therefore, the present invention can improve the signal-to-noise ratio of the collected signal, and can accurately and efficiently obtain the characteristics of the frequency signal that needs to be extracted under the mixing condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a flow chart of a method for extracting characteristic signals under low signal-to-noise ratio provided by an embodiment of the present invention;

[0058] Figure 2 is a schematic diagram of a measured signal involved in an embodiment of the present invention;

[0059] Figure 3 is a schematic diagram of a characteristic signal after processing involved in an embodiment of the present invention;

[0060] Figure 4 is a schematic block diagram of a feature signal extraction system involved in an embodiment of the present invention;

[0061] Figure 5 It is a structural schematic diagram of a system for extracting characteristic signals under low signal-to-noise ratio provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] Many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present invention, so the present invention is not limited to the specific implementation disclosed below.

[0063] Example 1

[0064] The present invention aims to solve the problem that the existing filtering and weak signal extraction technology cannot effectively identify the inter-frequency calibration signal under the condition of low signal-to-noise ratio. To achieve the above object, the present invention provides a method for extracting characteristic signals under low signal-to-noise ratio, such as Figure 1 As shown, the following steps are included:

[0065] Step S101, collecting the heterofrequency signal of the transformer under test.

[0066] The signals collected by the measured transformer and the standard transformer are respectively input into the transformer tester. Before comparison, it is ensured that the collected measured transformer signals are subjected to heterodyne signal extraction.

[0067] The expression of the different frequency signal of the measured transformer is:

[0068]

[0069] Where A is the amplitude, ω is the angular frequency of the signal, is the signal phase and t is the time.

[0070] Step S102: constructing an overcomplete dictionary of the heterofrequency signal by cascading a Gabor sparse complete dictionary and a Fourier sparse dictionary.

[0071] The Gabor sparse complete dictionary expression is:

[0072]

[0073] in, is the standard Gaussian window function; s a is the scale factor; u x is the displacement factor; ω f is the frequency modulation factor, is the signal phase, K r is the normalization factor, so that ||g r (t)|| 2 =1;

[0074] Discretize the Gabor sparse complete dictionary expression:

[0075]

[0076] Where λ is the atomic parameter set, where 0<p<log 2 N,0≤k≤2 1+p ,q,p,k∈Z, a=2, Δu=1 / 2, Δω=π,

[0077] For heterofrequency signals with a time window length of N, the corresponding overcomplete dictionary is:

[0078] ψ=[ψ G ψ F ]

[0079] Among them, ψ G is the Gabor sparse complete dictionary, g rn is the nth atom in the Gabor overcomplete dictionary, n>>; ψ F is the Fourier sparse complete dictionary;

[0080]

[0081] At this time, the heterodyne signal collected in the wth time window can be sparsely represented under the overcomplete synthetic dictionary as:

[0082]

[0083] Among them, Y i is the ith signal, X i is the ith atom.

[0084] Step S103: The heterofrequency signal is sparsely represented by the overcomplete dictionary, and the sparse representation dictionary learning algorithm K-SVD is used to decompose the heterofrequency signal to generate a learning atomic dictionary.

[0085] Use the K-SVD algorithm to learn and obtain the learning atomic dictionary by solving the following formula:

[0086]

[0087] Among them, Q j is the sparse signal to be solved; P j is the coefficient matrix; D learn is a dictionary; T is a sufficiently small constant; M is the number of atoms.

[0088] Step S104, selecting atoms matching the heterofrequency signal from the overcomplete dictionary, and obtaining a reconstructed heterofrequency calibration signal by linearly combining the matching atoms; when the accuracy of the reconstructed heterofrequency calibration signal reaches a preset threshold or reaches a preset number of iterations, extracting a characteristic signal from the heterofrequency signal.

[0089] Using the orthogonal projection method, select the atom x that matches the heterofrequency signal y′ from the overcomplete dictionary 0 ′, satisfying the following formula:

[0090] | <y′,x 0 ′>|=sup| <y′,x j ′>|

[0091] Among them, y' is the collected signal; x' j is the jth atom in the overcomplete dictionary; sup is the geometric supremum;

[0092] The matching atom matrix is ​​constructed by linearly combining the selected atoms;

[0093] Based on the matching atomic matrix, obtaining its column orthogonal projection operator, using the current signal residual and the orthogonal projection operator, constructing a sparse approximation method to update the residual signal;

[0094] The linear combination of the matching atom matrix plus the updated residual signal represents the reconstructed hetero-frequency calibration signal. The atoms in the reconstructed hetero-frequency standard signal are Schmidt orthogonalized, and the number of iterations is U; 7 repeat the above steps until the accuracy of the residual signal reaches a preset threshold or reaches a preset number of iterations, terminate the above steps, and extract the characteristic signal in the hetero-frequency signal.

[0095] Example 2

[0096] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0097] This embodiment provides a method for extracting characteristic signals based on a sparse matrix, comprising the following steps:

[0098] The signals collected by the tested transformer and the standard transformer are respectively input into the transformer tester. Before comparison, ensure that the collected signals of the tested transformer are subjected to heterogeneous frequency signal extraction;

[0099] The expression of the different frequency signal is:

[0100]

[0101] An overcomplete synthetic dictionary is constructed by cascading the Gabor overcomplete sparse dictionary and the Fourier sparse dictionary to improve the sparse representation performance and reconstruction effect of the algorithm for the inter-frequency calibration signal. The expression of the Gabor overcomplete sparse dictionary is:

[0102]

[0103] in, is the standard Gaussian window function; s a is the scale factor; u x is the displacement factor; f is the frequency modulation factor, is the signal phase, K r is the normalization factor, so that ||g r(t)|| 2 =1.

[0104] Further preferably, the above formula is discretized:

[0105]

[0106] Where λ is the atomic parameter set, where 0<p<log 2 N,0≤k≤2 1+p ,q,p,k∈Z, a=2, Δu=1 / 2, Δω=π,

[0107] Further preferably, for the sampled data with a time window length of N, the corresponding over-complete dictionary is:

[0108] ψ=[ψ G ψ F ]

[0109] Among them, ψ G is the Gabor sparse complete dictionary, g rn is the nth atom in the Gabor overcomplete dictionary, n>>; ψ F is the Fourier sparse complete dictionary;

[0110]

[0111] At this time, the signal data collected in the wth time window can be sparsely represented under the overcomplete synthetic dictionary as:

[0112]

[0113] The overcomplete dictionary is learned using the K-SVD algorithm; learning the atomic dictionary requires solving the following equation:

[0114]

[0115] Among them, Q j is the sparse signal to be solved; P j is the coefficient matrix; D learn is a dictionary; T is a sufficiently small constant; the K-SVD algorithm solves the dictionary solution problem in the above formula; the dictionary is generally a training sample set, and the coefficient matrix is ​​unknown; is the number of atoms; among them, K-SVD is a dictionary learning algorithm for sparse representation. The reason why it is called K-SVD is that the algorithm uses SVD (singular value decomposition) for K iterations;

[0116] Set the iteration termination condition of the OMP algorithm;

[0117] Select the atom that best matches the reference signal and the denoised measured signal from the overcomplete atom dictionary D;

[0118] According to the overcomplete atom dictionary D, the orthogonal projection method is used to select the atom x′ that best matches the denoised measured signal y′. 0 , satisfying the following formula:

[0119] | <y′,x 0 ′>|=sup| <y′,x j ′>|

[0120] Among them, x′ j is the jth atom; sup is the geometric supremum;

[0121] The selected atoms are Schmidt orthogonalized, and the components on the orthogonalized atoms are subtracted from the signal to be decomposed to obtain the residual signal, and the above steps are repeated until the termination condition is met; and the signal is represented by the linear combination of the selected atoms; the measured signal y' after noise reduction is decomposed into two parts, one is the component on the best atom, and the other is the residual part, the expression is as follows:

[0122] y'= <y’,x 0 '>x 0 '+R 1

[0123] The Schmidt orthogonalization method is used to orthogonalize the matched atoms at each step of the decomposition, which can not only speed up the convergence speed, but also avoid the introduction of unnecessary components when the residual is projected on the atoms.

[0124] Let the initial value of residual R 0 =y', the number of iterations is U;

[0125] The linear combination of the best atoms selected from each decomposition is used to reconstruct the heterofrequency calibration signal and extract effective features.

[0126] Further preferably, the orthogonal projection unit includes an inner product maximum atom filter, a best matching atom matrix builder, an orthogonal projection operator calculator, a residual updater and a signal reconstructor;

[0127] The maximum inner product atom filter is used to select the atom with the largest absolute value of the inner product with the signal residual from the overcomplete atom dictionary; wherein the initial value of the signal residual is set to be the measured signal or the reference signal;

[0128] The best matching atom matrix builder is used to construct the best matching atom matrix using the atoms selected by the inner product maximum atom filter as columns;

[0129] The orthogonal projection operator calculator is used to obtain the orthogonal projection operator of the column space based on the best matching atomic matrix;

[0130] The residual updater is used to construct a sparse approximation method to update the signal residual using the current signal residual and the orthogonal projection operator;

[0131] The signal reconstructor is used to represent the heterofrequency signal by using the linear combination of the best matching atoms obtained in each iteration plus the signal residual of the last iteration. Figure 2 and Figure 3 They represent the heterodyne signal before and after processing respectively. Figure 4 A schematic block diagram of an application scenario of a feature signal extraction system provided in an embodiment of the present invention.

[0132] Example 3

[0133] Based on the same inventive concept, the present invention also provides a system 500 for extracting characteristic signals under low signal-to-noise ratio, such as Figure 5 As shown, including:

[0134] The heterodyne signal acquisition module is used to acquire the heterodyne signal of the transformer under test;

[0135] An overcomplete dictionary construction module, used for constructing an overcomplete dictionary of the heterofrequency signal by cascading a Gabor sparse complete dictionary and a Fourier sparse dictionary;

[0136] An atomic dictionary generation module, used for sparsely representing the heterofrequency signal through the overcomplete dictionary, and decomposing the heterofrequency signal using a sparse representation dictionary learning algorithm K-SVD to generate a learning atomic dictionary;

[0137] The characteristic signal extraction module is used to select atoms matching the heterofrequency signal from the overcomplete dictionary, and obtain a reconstructed heterofrequency calibration signal by linearly combining the matching atoms; when the accuracy of the reconstructed heterofrequency calibration signal reaches a preset threshold or reaches a preset number of iterations, extract the characteristic signal in the heterofrequency signal.

[0138] Furthermore, the expression of the different frequency signal of the measured transformer is:

[0139]

[0140] Where A is the amplitude, ω is the angular frequency of the signal, is the signal phase and t is the time.

[0141] Furthermore, the characteristic signal extraction module includes:

[0142] A matching submodule, for selecting an atom x′ that matches the hetero-frequency signal y′ from the overcomplete dictionary using an orthogonal projection method 0 , satisfying the following formula:

[0143] | <y′,x 0 ′>|=sup| <y′,x j ′>|

[0144] Among them, y' is the collected signal; x' j is the jth atom in the overcomplete dictionary; sup is the geometric supremum;

[0145] A matrix construction module is used to construct a matching atom matrix by linearly combining the selected atoms;

[0146] A residual updating module, used to obtain the column orthogonal projection operator of the matching atomic matrix based on the matching atomic matrix, and to construct a sparse approximation method to update the residual signal using the current signal residual and the orthogonal projection operator;

[0147] The reconstruction submodule is used to add the linear combination of the matching atomic matrix to the updated residual signal to represent the reconstructed heterofrequency calibration signal.

[0148] The present invention adopts the collected reference signal and the measured signal to automatically generate an overcomplete atomic dictionary, obtains the atoms of the collected signal through the orthogonal matching pursuit (OMP) method, obtains the atomic linear combination that is finally matched after sparse decomposition, and obtains the hetero-frequency signal through the atomic linear combination; compared with the existing fast Fourier transform, wavelet transform and phase-locked technology, compared with the Fourier transform, the present invention does not need to subjectively judge the frequency range of the hetero-frequency signal, and has good adaptability; compared with the wavelet transform, it is not necessary to determine the wavelet type and the number of decomposition layers; compared with the phase-locked technology, the signal reconstruction accuracy is higher. Therefore, the present invention can improve the signal-to-noise ratio of the collected signal, and can accurately and efficiently obtain the characteristics of the frequency signal that needs to be extracted under the mixing condition.

[0149] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0150] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0151] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0152] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalents that do not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for extracting characteristic signals under low signal-to-noise ratio, characterized in that: include: Collect the heterodyne signal of the transformer under test; Constructing an overcomplete dictionary of the heterofrequency signal by cascading a Gabor sparse complete dictionary and a Fourier sparse dictionary; The heterofrequency signal is sparsely represented by the overcomplete dictionary, and the heterofrequency signal is decomposed by using a sparse representation dictionary learning algorithm K-SVD to generate a learning atomic dictionary; Atoms matching the heterofrequency signal are selected from the overcomplete dictionary, and a reconstructed heterofrequency calibration signal is obtained by linearly combining the matching atoms; when the accuracy of the reconstructed heterofrequency calibration signal reaches a preset threshold or reaches a preset number of iterations, a characteristic signal in the heterofrequency signal is extracted.

2. The method according to claim 1, characterized in that The expression of the different frequency signal of the measured transformer is: Where A is the amplitude, ω is the angular frequency of the signal, is the signal phase and t is the time.

3. The method according to claim 1, characterized in that The heterofrequency signal is sparsely represented by the overcomplete synthetic dictionary, wherein: The Gabor sparse complete dictionary expression is: in, is the standard Gaussian window function; s a is the scale factor; u x is the displacement factor; ω f is the frequency modulation factor, is the signal phase, K r is the normalization factor, so that ||g r (t)||2=1; Discretize the Gabor sparse complete dictionary expression: Where λ is the atomic parameter set, where 0<p<log2N, 0≤k≤2 1+p ,q,p,k∈Z, a=2,Δu=1 / 2,Δω=π, For heterofrequency signals with a time window length of N, the corresponding overcomplete dictionary is: ψ=[ψ G ψ F ] Among them, ψ G is the Gabor sparse complete dictionary, g rn is the nth atom in the Gabor overcomplete dictionary, n>>; ψ F is the Fourier sparse complete dictionary; At this time, the heterodyne signal collected in the wth time window can be sparsely represented under the overcomplete synthetic dictionary as: Among them, Y i is the ith signal, X i is the ith atom.

4. The method according to claim 1, characterized in that: The sparse representation dictionary learning algorithm K-SVD is used to decompose the error of the heterofrequency signal to generate a learning atomic dictionary, including: Use the K-SVD algorithm to learn and obtain the learning atomic dictionary by solving the following formula: Among them, Q j is the sparse signal to be solved; P j is the coefficient matrix; D learn is a dictionary; T is a sufficiently small constant; M is the number of atoms.

5. The method according to claim 1, characterized in that Selecting atoms matching the hetero-frequency signal from the overcomplete dictionary, and obtaining a reconstructed hetero-frequency calibration signal by linearly combining the matching atoms, comprising: By using the orthogonal projection method, an atom x′0 matching the hetero-frequency signal y′ is selected from the overcomplete dictionary, satisfying the following formula: |<y′,x0′>|=sup|<y′,x j ′>| Among them, y is the collected signal; x′ j is the jth atom in the overcomplete dictionary; sup is the geometric supremum; The matching atom matrix is ​​constructed by linearly combining the selected atoms; Based on the matching atomic matrix, obtaining its column orthogonal projection operator, using the current signal residual and the orthogonal projection operator, constructing a sparse approximation method to update the residual signal; The linear combination of the matching atom matrices plus the updated residual signal represents the reconstructed inter-frequency calibration signal.

6. The method according to claim 1, characterized in that When the accuracy of the reconstructed inter-frequency calibration signal reaches a preset threshold or reaches a preset number of iterations, extracting a characteristic signal from the inter-frequency signal includes: The atoms in the reconstructed heterofrequency standard signal are Schmidt orthogonalized, and the number of iterations is U; Repeat the above steps until the accuracy of the residual signal reaches a preset threshold or reaches a preset number of iterations, terminate the above steps, and extract the characteristic signal from the heterofrequency signal.

7. A system for extracting characteristic signals under low signal-to-noise ratio, characterized in that: include: The heterodyne signal acquisition module is used to acquire the heterodyne signal of the transformer under test; An overcomplete dictionary construction module, used for constructing an overcomplete dictionary of the heterofrequency signal by cascading a Gabor sparse complete dictionary and a Fourier sparse dictionary; An atomic dictionary generation module, used for sparsely representing the heterofrequency signal through the overcomplete dictionary, and decomposing the heterofrequency signal using a sparse representation dictionary learning algorithm K-SVD to generate a learning atomic dictionary; The characteristic signal extraction module is used to select atoms matching the heterofrequency signal from the overcomplete dictionary, and obtain a reconstructed heterofrequency calibration signal by linearly combining the matching atoms; when the accuracy of the reconstructed heterofrequency calibration signal reaches a preset threshold or reaches a preset number of iterations, extract the characteristic signal in the heterofrequency signal.

8. The system according to claim 7, characterized in that The expression of the different frequency signal of the measured transformer is: Where A is the amplitude, ω is the angular frequency of the signal, is the signal phase and t is the time.

9. The system according to claim 7, characterized in that Feature signal extraction module, including: The matching submodule is used to select an atom x′0 matching the heterofrequency signal y′ from the overcomplete dictionary by using an orthogonal projection method, satisfying the following formula: |<y′,x0′>|=sup|<y′,x j ′>| Among them, y' is the collected signal; x' j is the jth atom in the overcomplete dictionary; sup is the geometric supremum; A matrix construction module is used to construct a matching atom matrix by linearly combining the selected atoms; A residual updating module, used to obtain the column orthogonal projection operator of the matching atomic matrix based on the matching atomic matrix, and to construct a sparse approximation method to update the residual signal using the current signal residual and the orthogonal projection operator; The reconstruction submodule is used to add the linear combination of the matching atomic matrix to the updated residual signal to represent the reconstructed heterofrequency calibration signal.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images