A Peak Analysis Method for Large Disturbance Signals in Primary Frequency Modulation

Abstract

This invention relates to the technical field of signal analysis and discloses a method for peak analysis of a primary frequency modulated (FM) large disturbance signal. The method includes: acquiring a FM large disturbance signal and performing filtering processing for multi-band noise separation; performing multi-scale decomposition on the filtered disturbance signal, extracting the dynamic entropy of the multi-scale signal sequence, and performing dynamic entropy denoising on the disturbance signal to obtain a denoised disturbance signal; calculating the potential peak position of the denoised disturbance signal, verifying the potential peak position under multiple conditions, and obtaining peak information for FM disturbance analysis. This invention is based on spectral energy distribution, selects specific wavelet bases for denoising different noise frequency bands, improves wavelet denoising performance, extracts the energy distribution complexity and structural complexity of the disturbance signal at different scales, more comprehensively captures the dynamic characteristics of the disturbance signal, improves denoising accuracy, and thus improves the accuracy of wind power grid connection fault identification and equipment life monitoring during peak analysis.

Description

Technical Field

[0001] This invention relates to the technical field of signal analysis, and in particular to a method for peak value analysis of a single-frequency modulated large disturbance signal. Background Technology

[0002] With a high proportion of new energy sources being connected to the grid, the primary frequency regulation of thermal power units needs to cope with more frequent and wider-ranging non-stationary and multi-harmonic superposition characteristics. For example, the randomness of wind power grid-connected loads leads to a wider range of fluctuations in the amplitude of grid frequency disturbances, and the disturbance signals often exhibit non-stationary and multi-harmonic superposition characteristics, resulting in significant disturbance information in the primary frequency regulation signal. Peak analysis is required for the large disturbance signals in frequency regulation to identify wind turbine faults and adjust the operating frequency based on the peak analysis results. However, the peak analysis process faces the following challenges: sensor data in the wind power environment is affected by external noise, environmental interference, and grid harmonics, leading to large peak extraction errors. Large disturbance signals in wind power simultaneously exhibit high-frequency short-time characteristics and low-frequency long-time variations, making it difficult to extract key features at a single scale. Traditional peak detection methods (such as fixed-time-window moving average and frequency domain filtering) rely on empirical threshold settings, which are prone to false peak detection under dynamic operating conditions. Summary of the Invention

[0003] In view of this, the present invention provides a peak analysis method for primary frequency modulation large disturbance signals. Based on the spectral energy distribution, wavelet bases of different frequency bands are adaptively selected, and specific wavelet bases are selected for noise reduction in different noise frequency bands. The potential peak positions are verified under multiple conditions by combining multi-scale signal sequences to obtain the peak information of primary frequency modulation large disturbance signals. Based on the peak information, frequency modulation disturbance analysis is performed to identify wind power grid connection faults and to monitor the lifespan of wind power grid-connected equipment, thereby reducing equipment losses.

[0004] To achieve the above objectives, the present invention provides a method for peak analysis of a single-frequency modulated large disturbance signal, comprising the following steps:

[0005] S1: Acquire a large frequency-modulated disturbance signal and perform multi-band noise separation filtering to obtain the filtered disturbance signal;

[0006] S2: Perform multi-scale decomposition on the filtered disturbance signal to obtain a multi-scale signal sequence of the disturbance signal, extract the dynamic entropy of the multi-scale signal sequence, perform dynamic entropy denoising on the disturbance signal, and obtain the denoised disturbance signal.

[0007] S3: Calculate the potential peak position of the denoised disturbance signal;

[0008] S4: Combine the denoised disturbance signal to perform multi-condition verification of the potential peak position, obtain the peak information of the primary frequency modulation large disturbance signal, and perform frequency modulation disturbance analysis based on the peak information.

[0009] As a further improvement of the present invention:

[0010] Optionally, the primary frequency modulation large disturbance signal is subjected to multi-band noise separation filtering processing, including:

[0011] The local trend of the primary frequency modulated large disturbance signal is calculated using the sliding window method. The primary frequency modulated large disturbance signal is then detrended. Finally, the frequency domain representation of the detrended frequency modulated large disturbance signal is obtained by using Fourier transform to perform spectral analysis.

[0012] The energy in the frequency domain is calculated in different frequency bands to form a spectral energy distribution. Using the spectral energy distribution as input features, the optimal wavelet basis sequence corresponding to the input features is generated using a wavelet basis action selection model. The detrended frequency-modulated large disturbance signal is subjected to wavelet transform processing for multi-band noise separation to obtain the filtered disturbance signal. The optimal wavelet basis sequence is the optimal wavelet basis of the detrended frequency-modulated large disturbance signal in different frequency bands.

[0013] Optionally, the wavelet basis action selection model is a reinforcement learning model, including a state space, an action space, and a policy matrix. The state space includes all spectral energy distributions, the action space includes multiple wavelet bases, and the policy matrix consists of policy values ​​between the spectral energy distribution and the wavelet base sequence, wherein the wavelet base sequence is a sequence composed of E wavelet bases. The wavelet basis action selection model takes the spectral energy distribution as the input feature and selects the wavelet base sequence with the largest policy value between the wavelet base and the input feature from the policy matrix as the optimal wavelet base sequence. The policy matrix is ​​the model parameters to be optimized.

[0014] A reward function is constructed to evaluate the multi-band noise separation performance of the wavelet basis sequences selected in the policy matrix. The variables of the reward function are the spectral energy distribution and the wavelet basis sequences. The expression of the reward function is as follows:

[0015] ;

[0016] ;

[0017] ;

[0018] in:

[0019] Represents the reward function, This represents the spectral energy distribution of the input reward function. This represents the wavelet basis sequence that is input to the reward function. Let e ​​represent the e-th wavelet basis in the wavelet basis sequence, where the e-th wavelet basis corresponds to the e-th frequency band, and the e-th wavelet basis is used to perform wavelet transform on the e-th frequency band;

[0020] This indicates the use of wavelet basis sequences S. The signal-to-noise ratio improvement of the corresponding signal after wavelet transform. This indicates the use of wavelet basis sequences S. The time consumed by performing wavelet transform on the corresponding signal. This indicates the smoothness of the wavelet basis sequence S;

[0021] The hyperparameter weights represent the signal-to-noise ratio improvement, latency, and smoothness, respectively.

[0022] Optionally, based on the reward function, a loss function for the policy matrix is ​​constructed:

[0023] ;

[0024] ;

[0025] in:

[0026] The loss function represents the policy matrix. This represents the i-th group of spectral energy distributions in the state space. This represents the j-th wavelet basis sequence constructed based on wavelet bases in the action space. This represents the number of spectral energy distributions in the state space. This indicates the number of wavelet basis sequences constructed;

[0027] Represents the policy matrix, ;

[0028] The policy matrix is ​​solved based on the loss function, and a wavelet basis action selection model is constructed using the solved policy matrix.

[0029] Optionally, the filtered perturbation signal is decomposed into multiple scales to obtain scale signals of the perturbation signal at different scales, forming a multi-scale signal sequence, including:

[0030] The center frequency and scale signal spectrum of the disturbance signal at A scales are initialized, and the Lagrange multipliers are initialized. The scale signal spectrum is iterated based on the center frequency and the Lagrange multipliers. The scale signal spectrum is weighted and calculated. The center frequency is iteratively updated. The Lagrange multipliers are updated using a dynamic step size update method until the iterative change of the scale signal spectrum is lower than a preset change threshold. The scale signal spectrum of the disturbance signal y at A scales is obtained, where A represents the number of scales. The inverse Fourier transform of the scale signal spectrum is performed to obtain the scale signals at A scales, forming a multi-scale signal sequence.

[0031] Optionally, singular value decomposition is performed on the multi-scale signal sequence to obtain the singular value sequence of the multi-scale signal sequence, and the energy entropy and singular value entropy of the multi-scale signal sequence are extracted respectively, including:

[0032] The formula for calculating the energy entropy is:

[0033] ;

[0034] in:

[0035] In the multi-scale signal sequence, the first... The spectrum of a scaled signal at each scale. Represents the spectrum of scaled signals energy, ;

[0036] This represents the energy entropy of the multi-scale signal sequence;

[0037] The formula for calculating the singular value entropy is:

[0038] ;

[0039] in:

[0040] Represents the first singular value in the sequence. A singular value, This represents singular value entropy.

[0041] Optionally, based on modal entropy and singular value entropy, the dynamic entropy of the multi-scale signal sequence is extracted, and the perturbation signal is subjected to dynamic entropy denoising to obtain a denoised perturbation signal, including:

[0042] The formula for calculating the dynamic entropy is:

[0043] ;

[0044] in:

[0045] Represents dynamic entropy;

[0046] The dynamic entropy is used to dynamically denoise the disturbed signal, wherein the dynamic denoising formula is:

[0047] ;

[0048] ;

[0049] in:

[0050] Indicates the inverse Fourier transform;

[0051] This indicates the preset control parameters. Represents the spectrum of scaled signals The corresponding transformation weights;

[0052] This represents the denoised disturbance signal after dynamic denoising.

[0053] Optionally, calculating the potential peak position of the denoised perturbation signal includes:

[0054] A peak detection sliding window is used to process the denoised perturbation signal, and the mean signal value within the sliding window is calculated. And the standard deviation of the signal value (std), generating an adaptive threshold for the sliding window. :

[0055] ;

[0056] in:

[0057] This indicates the preset adaptive control parameters;

[0058] The signal value within the sliding window is higher than the adaptive threshold. The signal location is used as the potential peak location.

[0059] Optionally, the step of combining the denoised perturbation signal to perform multi-condition verification of the potential peak position to obtain the peak information of the primary frequency modulated large perturbation signal includes:

[0060] Multi-condition verification methods include: in the denoised perturbation signal, before and after the potential peak position If all signal values ​​maintain a monotonic trend, then the first condition for verifying the potential peak position is met; the potential peak position is at least If the potential peak location is also a potential peak location in the signal of each scale, then the second condition verification of the potential peak location is passed;

[0061] The potential peak positions after the first condition verification and the second condition verification are taken as the peak values; the peak information of the peak values ​​is extracted, which is the signal value of the denoised disturbance signal at the peak value.

[0062] To address the above problems, the present invention provides an electronic device, the electronic device comprising:

[0063] Memory, storing at least one instruction;

[0064] Communication interfaces enable communication between electronic devices; and

[0065] The processor executes the instructions stored in the memory to implement the above-described method for analyzing the peak values ​​of a single-frequency large disturbance signal.

[0066] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the above-described method for analyzing peak values ​​of a single-frequency modulation large disturbance signal.

[0067] Compared with existing technologies, this invention proposes a peak value analysis method for single-frequency modulated large disturbance signals, which has the following advantages:

[0068] First, this scheme proposes a multi-band adaptive denoising method. It uses reinforcement learning to construct a wavelet basis action selection model. A reward function is constructed based on the smoothness of the wavelet basis sequence, the signal-to-noise ratio improvement, and the wavelet processing time. The policy matrix in the wavelet basis action selection model is optimized and solved. Wavelet bases in different frequency bands are adaptively selected according to the spectral energy distribution. Specific wavelet bases are selected for denoising in different noise frequency bands. A smooth wavelet basis sequence is set based on the reward function. The denoising amplitude is reduced near the peak signal to avoid important information being misfiltered and improve the filtering and denoising performance.

[0069] Meanwhile, this scheme proposes a dynamic feature capture and peak analysis method. By combining the energy entropy and singular value entropy of the spectrum, the dynamic entropy of the disturbance signal is calculated. The energy entropy can measure the energy distribution complexity of the disturbance signal at different scales, while the singular value entropy is used to measure the structural complexity of the disturbance signal and can effectively detect abnormal noise. By combining the two to construct an adaptive threshold, the dynamic features of the disturbance signal can be captured more comprehensively, improving the denoising accuracy. Based on the distribution characteristics of the denoised disturbance signal in the sliding window, the potential peak position is identified. The potential peak position is verified under multiple conditions by combining multi-scale signal sequences to obtain the peak information of the primary frequency modulation large disturbance signal. Based on the peak information, frequency modulation disturbance analysis is performed to identify wind power grid connection faults and monitor the lifespan of wind power grid-connected equipment, thereby reducing equipment losses. Attached Figure Description

[0070] Figure 1 This is a flowchart illustrating a method for peak analysis of a single-frequency modulated large disturbance signal according to an embodiment of the present invention.

[0071] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0072] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0073] This application provides a method for peak value analysis of a single-frequency modulated (SFMC) signal with large disturbances. The execution entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0074] Reference Figure 1 Embodiment 1 of the present invention is as follows:

[0075] A method for peak value analysis of a single-frequency modulated large disturbance signal includes the following steps:

[0076] S1: Acquire a large frequency-modulated disturbance signal and perform multi-band noise separation filtering to obtain the filtered disturbance signal.

[0077] The filtering process for multi-band noise separation of the primary frequency modulated large disturbance signal includes:

[0078] The local trend of the primary frequency modulated (FM) large disturbance signal is calculated using the sliding window method. The FM large disturbance signal is then detrended, and the frequency domain representation of the detrended FM large disturbance signal is obtained using Fourier transform. As an embodiment of the present invention, the representation of the primary FM large disturbance signal is as follows: :

[0079] ;

[0080] in:

[0081] Let N be the signal values ​​of the first-frequency large disturbance signal x. This represents the nth signal value in the first-order frequency-modulated large disturbance signal x. N represents the length of the first-frequency large disturbance signal x;

[0082] The detrending formula for the primary frequency modulation large disturbance signal x is as follows:

[0083] ;

[0084] ;

[0085] in:

[0086] Indicates signal value The detrending results The negative value information in the detrending processing formula is the local trend of the primary frequency modulation large disturbance signal, where the sliding window length is the length of the sliding window.

[0087] This indicates a frequency-modulated large disturbance signal after detrending processing;

[0088] The energy in the frequency domain is calculated in different frequency bands to form a spectral energy distribution. Using the spectral energy distribution as input features, an optimal wavelet basis sequence corresponding to the input features is generated using a wavelet basis action selection model. This sequence is then subjected to multi-band noise separation wavelet transform processing on the detrended frequency-modulated large disturbance signal to obtain a filtered disturbance signal. The optimal wavelet basis sequence represents the optimal wavelet basis for the detrended frequency-modulated large disturbance signal in different frequency bands. Specifically, the formula for calculating the spectral energy distribution is:

[0089] ;

[0090] ;

[0091] in:

[0092] This indicates a large frequency-modulated disturbance signal after detrending processing. The frequency domain representation, where f represents frequency information; Frequency domain representation Spectral energy distribution, Frequency domain representation Energy in the e-th frequency band E represents the total number of frequency bands, where the frequency range of the e-th frequency band is... , This represents the upper frequency limit of the e-th frequency band;

[0093] This represents the derivative of the frequency information f.

[0094] The wavelet basis action selection model is a reinforcement learning model, including a state space, an action space, and a policy matrix. The state space includes all spectral energy distributions, the action space includes multiple wavelet bases, and the policy matrix consists of policy values ​​between the spectral energy distribution and the wavelet base sequence, where the wavelet base sequence is a sequence composed of E wavelet bases. The wavelet basis action selection model takes the spectral energy distribution as the input feature and selects the wavelet base sequence with the largest policy value between it and the input feature from the policy matrix as the optimal wavelet base sequence. The policy matrix is ​​the model parameter to be optimized.

[0095] A reward function is constructed to evaluate the multi-band noise separation performance of the wavelet basis sequences selected in the policy matrix. The variables of the reward function are the spectral energy distribution and the wavelet basis sequences. The expression of the reward function is as follows:

[0096] ;

[0097] ;

[0098] ;

[0099] in:

[0100] Represents the reward function, This represents the spectral energy distribution of the input reward function. This represents the wavelet basis sequence that is input to the reward function. Let e ​​represent the e-th wavelet basis in the wavelet basis sequence, where the e-th wavelet basis corresponds to the e-th frequency band, and the e-th wavelet basis is used to perform wavelet transform on the e-th frequency band;

[0101] This indicates the use of wavelet basis sequences S. The signal-to-noise ratio improvement of the corresponding signal after wavelet transform. This indicates the use of wavelet basis sequences S. The time consumed by performing wavelet transform on the corresponding signal. This indicates the smoothness of the wavelet basis sequence S;

[0102] The hyperparameter weights, in order, represent the signal-to-noise ratio improvement, latency, and smoothness; specifically, .

[0103] Based on the reward function, construct the loss function for the policy matrix:

[0104] ;

[0105] ;

[0106] in:

[0107] The loss function represents the policy matrix. This represents the i-th group of spectral energy distributions in the state space. This represents the j-th wavelet basis sequence constructed based on wavelet bases in the action space. This represents the number of spectral energy distributions in the state space. This indicates the number of wavelet basis sequences constructed;

[0108] Represents the policy matrix, ;

[0109] The policy matrix is ​​solved based on the loss function, and a wavelet basis action selection model is constructed using the solved policy matrix. Specifically, the policy matrix is ​​solved using the Adam optimization algorithm.

[0110] S2: Perform multi-scale decomposition on the filtered disturbance signal to obtain a multi-scale signal sequence of the disturbance signal, extract the dynamic entropy of the multi-scale signal sequence, and perform dynamic entropy denoising on the disturbance signal to obtain the denoised disturbance signal.

[0111] The filtered perturbation signal is decomposed into multiple scales to obtain scaled signals of the perturbation signal at different scales, forming a multi-scale signal sequence, including:

[0112] The disturbance signal is represented as follows:

[0113] ;

[0114] in:

[0115] This represents the perturbation signal after filtering. N signal values ​​representing the disturbance signal y. This represents the nth signal value of the disturbance signal y;

[0116] The center frequencies and scale signal spectra of the perturbation signal at A scales are initialized, and the Lagrange multipliers are initialized. Based on the center frequencies and Lagrange multipliers, the scale signal spectrum is iteratively calculated, weighted, and the center frequencies are iteratively updated. The Lagrange multipliers are updated using a dynamic step-size update method until the iterative change in the scale signal spectrum is lower than a preset change threshold. This yields the scale signal spectrum of the perturbation signal y at A scales, where A represents the number of scales. An inverse Fourier transform is performed on the scale signal spectrum to obtain the scale signals at A scales, forming a multi-scale signal sequence. The [missing information - likely a typo, should be "the first" or "the second"]. The iterative formula for the spectrum of a scaled signal is:

[0117] ;

[0118] in:

[0119] Describes the result obtained in the t-th iteration. The spectrum of a scaled signal at each scale. Indicates the first The center frequency of the t-th iteration result of the scale signal at each scale. This indicates the preset maximum number of iterations. ;

[0120] Denotes the result of the t-th iteration of the Lagrange multipliers. express The spectrum;

[0121] This represents the mean center frequency of the result of the t-th iteration.

[0122] ;

[0123] The The iterative change is: ,in Describing the L1 norm, Denotes the square of the L1 norm;

[0124] The center frequency The iterative formula is:

[0125] ;

[0126] in:

[0127] Indicates center frequency information, The differential representing the center frequency information;

[0128] The Lagrange multipliers The iterative formula is:

[0129] ;

[0130] ;

[0131] in:

[0132] Represents the spectrum of scaled signals The corresponding scale signal;

[0133] Represents Lagrange multipliers Iteration step size, This indicates the preset initial step size.

[0134] Singular value decomposition is performed on the multi-scale signal sequence to obtain the singular value sequence of the multi-scale signal sequence. The energy entropy and singular value entropy of the multi-scale signal sequence are extracted, including:

[0135] The formula for calculating the energy entropy is:

[0136] ;

[0137] in:

[0138] In the multi-scale signal sequence, the first... The spectrum of a scaled signal at each scale. Represents the spectrum of scaled signals energy, ;

[0139] This represents the energy entropy of the multi-scale signal sequence;

[0140] The formula for calculating the singular value entropy is:

[0141] ;

[0142] in:

[0143] Represents the first singular value in the sequence. A singular value, This represents singular value entropy.

[0144] Specifically, the multi-scale signal sequence is converted into a multi-scale signal matrix, wherein the multi-scale signal matrix is ​​an A-row matrix, with each row corresponding to a set of scale signals. Singular value decomposition is performed on the multi-scale signal matrix to obtain A singular values.

[0145] Based on modal entropy and singular value entropy, the dynamic entropy of the multi-scale signal sequence is extracted, and the perturbation signal is subjected to dynamic entropy denoising to obtain the denoised perturbation signal, including:

[0146] The formula for calculating the dynamic entropy is:

[0147] ;

[0148] in:

[0149] Represents dynamic entropy;

[0150] The dynamic entropy is used to dynamically denoise the disturbed signal, wherein the dynamic denoising formula is:

[0151] ;

[0152] ;

[0153] in:

[0154] Indicates the inverse Fourier transform;

[0155] This indicates the preset control parameters. Represents the spectrum of scaled signals The corresponding transformation weights;

[0156] This represents the denoised disturbance signal after dynamic denoising. , Indicates the disturbance signal after denoising The nth signal value.

[0157] S3: Calculate the potential peak position of the denoised disturbance signal.

[0158] Calculating the potential peak location of the denoised perturbation signal includes:

[0159] A peak detection sliding window is used to process the denoised perturbation signal, and the mean signal value within the sliding window is calculated. And the standard deviation of the signal value (std), generating an adaptive threshold for the sliding window. :

[0160] ;

[0161] in:

[0162] This indicates the preset adaptive control parameters;

[0163] The signal value within the sliding window is higher than the adaptive threshold. The signal location is used as the potential peak location.

[0164] S4: Combine the denoised disturbance signal to perform multi-condition verification of the potential peak position, obtain the peak information of the primary frequency modulation large disturbance signal, and perform frequency modulation disturbance analysis based on the peak information.

[0165] The method of combining the denoised perturbation signal to perform multi-condition verification of the potential peak position yields the peak information of the primary frequency modulated large perturbation signal, including:

[0166] Multi-condition verification methods include: in the denoised perturbation signal, before and after the potential peak position If all signal values ​​maintain a monotonic trend, then the first condition for verifying the potential peak position is met; the potential peak position is at least If the potential peak location is also a potential peak location in the signal of each scale, then the second condition verification of the potential peak location is passed;

[0167] The potential peak positions after the first condition verification and the second condition verification are taken as the peak values; the peak information of the peak values ​​is extracted, which is the signal value of the denoised disturbance signal at the peak value.

[0168] As an embodiment of the present invention, the frequency modulation disturbance analysis includes: whether the peak information exceeds a preset peak threshold, the number of peaks exceeding the preset peak threshold, if there is peak information exceeding the preset peak threshold by 10%, it indicates that there is a wind power grid connection fault, if there is peak information exceeding the preset peak threshold by 30% and it occurs frequently, it may lead to gearbox fatigue failure, and key components need to be replaced in advance, if peak information exceeding the preset peak threshold occurs multiple times in a short period of time, it can be determined that a strong wind or storm is about to hit, and the wind turbine speed can be reduced in advance to reduce mechanical wear.

[0169] Example 2:

[0170] This scheme compares the peak value analysis methods for primary frequency modulation large disturbance signals, the fixed threshold peak value analysis method, the fixed threshold peak value analysis method based on wavelet denoising, and the fixed threshold peak value analysis method based on empirical mode decomposition. The fixed threshold peak value analysis methods based on wavelet denoising and empirical mode decomposition use wavelet denoising or empirical mode decomposition to reduce signal noise, and then use the fixed threshold peak value analysis method for peak detection. The experimental signal is a large power disturbance signal from a wind turbine. The comparative experimental results are shown in Table 1.

[0171] Table 1

[0172] Experimental methods accuracy Recall rate F1 score Peak value analysis method for large disturbance signals in primary frequency modulation 98.2% 97.6% 97.9% Fixed threshold peak analysis method 85.2% 75.2% 79.9% Fixed threshold peak analysis method based on wavelet denoising 91.3% 85.5% 88.3% Fixed threshold peak analysis method based on empirical mode decomposition 92.8% 88.6% 90.7%

[0173] As shown in Table 1, the peak analysis method for large disturbance signals with single-frequency modulation has better performance and better noise reduction performance for disturbance signals. The fixed threshold peak analysis method is sensitive to peak drift, resulting in low recall.

[0174] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0175] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0176] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0177] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for peak value analysis of a single-frequency modulated large disturbance signal, characterized in that, The method includes: S1: Acquire a large frequency-modulated disturbance signal and perform multi-band noise separation filtering to obtain the filtered disturbance signal; The primary frequency regulation disturbance signal is the frequency change signal generated by the wind turbine during the primary frequency regulation process when the power grid experiences a power disturbance. S2: Perform multi-scale decomposition on the filtered disturbance signal to obtain a multi-scale signal sequence of the disturbance signal, extract the dynamic entropy of the multi-scale signal sequence, perform dynamic entropy denoising on the disturbance signal, and obtain the denoised disturbance signal. S3: Calculate the potential peak position of the denoised disturbance signal; S4: Combine the denoised disturbance signal to verify the potential peak position under multiple conditions, obtain the peak information of the primary frequency modulation large disturbance signal, and perform frequency modulation disturbance analysis based on the peak information. The frequency modulation disturbance analysis includes wind power grid connection fault identification and equipment life monitoring. The filtering process for multi-band noise separation of the primary frequency modulated large disturbance signal includes: The local trend of the primary frequency modulated large disturbance signal is calculated using the sliding window method. The primary frequency modulated large disturbance signal is then detrended. Finally, the frequency domain representation of the detrended frequency modulated large disturbance signal is obtained by using Fourier transform to perform spectral analysis. The energy in the frequency domain is calculated in different frequency bands to form a spectral energy distribution. Using the spectral energy distribution as input features, the optimal wavelet basis sequence corresponding to the input features is generated using a wavelet basis action selection model. The detrended frequency-modulated large disturbance signal is subjected to wavelet transform processing for multi-band noise separation to obtain the filtered disturbance signal. The optimal wavelet basis sequence is the optimal wavelet basis of the detrended frequency-modulated large disturbance signal in different frequency bands. The wavelet basis action selection model is a reinforcement learning model, including a state space, an action space, and a policy matrix. The state space includes all spectral energy distributions, the action space includes multiple wavelet bases, and the policy matrix consists of policy values ​​between the spectral energy distribution and the wavelet base sequence, where the wavelet base sequence is a sequence composed of E wavelet bases. The wavelet basis action selection model takes the spectral energy distribution as the input feature and selects the wavelet base sequence with the largest policy value between it and the input feature from the policy matrix as the optimal wavelet base sequence. The policy matrix is ​​the model parameter to be optimized. A reward function is constructed to evaluate the multi-band noise separation performance of the wavelet basis sequences selected in the policy matrix. The variables of the reward function are the spectral energy distribution and the wavelet basis sequences. The expression of the reward function is as follows: ； ； ； in: Represents the reward function, This represents the spectral energy distribution of the input reward function. This represents the wavelet basis sequence that is input to the reward function. Let e ​​represent the e-th wavelet basis in the wavelet basis sequence, where the e-th wavelet basis corresponds to the e-th frequency band, and the e-th wavelet basis is used to perform wavelet transform on the e-th frequency band; This indicates the use of wavelet basis sequences S. The signal-to-noise ratio improvement of the corresponding signal after wavelet transform. This indicates the use of wavelet basis sequences S. The time consumed by performing wavelet transform on the corresponding signal. This indicates the smoothness of the wavelet basis sequence S; The hyperparameter weights represent the signal-to-noise ratio improvement, latency, and smoothness, respectively.

2. The method for peak value analysis of a single-frequency modulated large disturbance signal as described in claim 1, characterized in that, Based on the reward function, construct the loss function for the policy matrix: ； ； in: The loss function represents the policy matrix. This represents the i-th group of spectral energy distributions in the state space. This represents the j-th wavelet basis sequence constructed based on wavelet bases in the action space. This represents the number of spectral energy distributions in the state space. Indicates the number of wavelet basis sequences constructed; Represents the policy matrix, ; The policy matrix is ​​solved based on the loss function, and a wavelet basis action selection model is constructed using the solved policy matrix.

3. The method for peak value analysis of a single-frequency modulated large disturbance signal as described in claim 1, characterized in that, The filtered perturbation signal is decomposed into multiple scales to obtain scaled signals of the perturbation signal at different scales, forming a multi-scale signal sequence, including: The center frequency and scale signal spectrum of the disturbance signal at A scales are initialized, and the Lagrange multipliers are initialized. The scale signal spectrum is iterated based on the center frequency and the Lagrange multipliers. The scale signal spectrum is weighted and calculated. The center frequency is iteratively updated. The Lagrange multipliers are updated using a dynamic step size update method until the iterative change of the scale signal spectrum is lower than a preset change threshold. The scale signal spectrum of the disturbance signal y at A scales is obtained, where A represents the number of scales. The inverse Fourier transform of the scale signal spectrum is performed to obtain the scale signals at A scales, forming a multi-scale signal sequence.

4. The method for peak value analysis of a single-frequency modulated large disturbance signal as described in claim 3, characterized in that, Singular value decomposition is performed on the multi-scale signal sequence to obtain the singular value sequence of the multi-scale signal sequence. The energy entropy and singular value entropy of the multi-scale signal sequence are extracted, including: The formula for calculating the energy entropy is: ； in: In the multi-scale signal sequence, the first... The spectrum of a scaled signal at each scale. Represents the spectrum of scaled signals energy, ; This represents the energy entropy of the multi-scale signal sequence; The formula for calculating the singular value entropy is: ； in: Represents the first singular value in the sequence. A singular value, This represents singular value entropy.

5. The method for peak value analysis of a single-frequency modulated large disturbance signal as described in claim 4, characterized in that, Based on energy entropy and singular value entropy, the dynamic entropy of the multi-scale signal sequence is extracted, and the perturbation signal is subjected to dynamic entropy denoising to obtain the denoised perturbation signal, including: The formula for calculating the dynamic entropy is: ； in: Represents dynamic entropy; The dynamic entropy is used to dynamically denoise the disturbed signal, wherein the dynamic denoising formula is: ； ； in: Indicates the inverse Fourier transform; This indicates the preset control parameters. Represents the spectrum of scaled signals The corresponding transformation weights; This represents the denoised disturbance signal after dynamic denoising.

6. The peak value analysis method for a single-frequency modulated large disturbance signal as described in claim 5, characterized in that, Calculating the potential peak location of the denoised perturbation signal includes: A peak detection sliding window is used to process the denoised perturbation signal, and the mean signal value within the sliding window is calculated. And the standard deviation of the signal value (std), generating an adaptive threshold for the sliding window. : ； in: This represents the preset adaptive control parameters; The signal value within the sliding window is higher than the adaptive threshold. The signal location is used as the potential peak location.

7. The method for peak value analysis of a single-frequency modulated large disturbance signal as described in claim 1, characterized in that, The method of combining the denoised perturbation signal to perform multi-condition verification of the potential peak position yields the peak information of the primary frequency modulated large perturbation signal, including: Multi-condition verification methods include: in the denoised perturbation signal, before and after the potential peak position If all signal values ​​maintain a monotonic trend, then the first condition for verifying the potential peak position is met; the potential peak position is at least If the potential peak location is also a potential peak location in the signal of each scale, then the second condition verification of the potential peak location is passed; The potential peak positions after the first condition verification and the second condition verification are taken as the peak values; the peak information of the peak values ​​is extracted, which is the signal value of the denoised disturbance signal at the peak value.

Images