An improved variational mode decomposition method for wind turbine characteristic signal extraction

Abstract

The application discloses an improved variational mode decomposition method for feature signal extraction of a wind turbine. In order to overcome the problems of difficulty in effective signal extraction of the wind turbine and self-adaptability of the variational mode decomposition, optimal decomposition parameters are determined from three aspects of a correlation coefficient, a peak amplitude and a center frequency according to the characteristics of an intrinsic mode function obtained after the signal is processed by the variational mode decomposition. The improved variational mode decomposition method is a posteriori, so that the extracted intrinsic mode function contains less noise and interference information, and the accuracy and reliability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind turbine signal processing, and in particular to an improved variational mode decomposition method for extracting characteristic signals of a wind turbine. BACKGROUND

[0002] The wind power industry has entered a golden development period, and the internal structure of wind turbines has become increasingly complex, bringing more challenges to their daily operation and maintenance. During daily testing, vibration signals often contain signals from various components of the unit, and are mixed with a large amount of non-stationary signals, which makes it difficult to extract effective information using traditional signal processing methods. If a nonlinear signal processing method is used, effective information can be extracted from non-stationary signals using time-frequency transformation. Variational Mode Decomposition (VMD) can effectively reduce false information and enhance the extraction ability of effective signals when processing non-stationary signals. However, VMD requires the number of decompositions to be determined in advance. Too many decompositions will cause mode splitting and false modes, and too few decompositions will cause mode aliasing and mode omission, which has the problem of adaptability.

[0003] A non-stationary signal analysis method based on wideband Fourier decomposition is disclosed in a Chinese patent document, with publication number CN111767811A. The method includes wideband Fourier decomposition, which first converts the original time-domain signal to a frequency-domain signal through Fourier transform, then decomposes the frequency-domain signal into sparse narrow-band sub-signals through a Fourier spectrum bandwidth optimization algorithm, and converts the frequency-domain narrow-band sub-signals to time-domain bandwidth modal components through inverse Fourier transform. Finally, the characteristics of the signal are extracted through Hilbert transform. This scheme is based only on Fourier transform when performing signal decomposition analysis, while Fourier transform is a kind of overall transform that lacks time-domain positioning function. Therefore, when using this scheme to perform signal decomposition analysis, the obtained signal spectrum reflects the average value of a certain frequency component contained in the overall signal, which limits the time and frequency resolution of the results. SUMMARY

[0004] The application mainly solves the problems of difficulty in extracting effective signals of wind turbine and self-adaptability of variational mode decomposition in the prior art, and provides an improved variational mode decomposition method for extracting characteristic signals of wind turbine, collects vibration signals of the wind turbine and performs data preprocessing, processes the vibration signals by using the improved variational mode decomposition, so that the extracted intrinsic mode functions reach the best in terms of independence and noise reduction, and finally calculates characteristic frequencies from the extracted intrinsic mode functions; according to the characteristics of the intrinsic mode functions obtained after the signal is processed by the variational mode, optimal decomposition parameters are determined from three aspects of correlation coefficient, peak amplitude and center frequency; the scheme of the application is a post-improved variational mode decomposition method, so that the extracted intrinsic mode functions contain less noise and interference information, and the accuracy and reliability are improved.

[0005] The above technical problems of the application are mainly solved by the following technical scheme:

[0006] The application comprises the following steps: collecting vibration signals of wind turbine; performing detrend processing on the vibration signals to reduce interference; processing the vibration signals by using traditional variational mode decomposition (VMD) to obtain intrinsic mode functions (IMFs) under different decomposition numbers; adaptively determining the number K of variational mode decomposition from three aspects of correlation coefficient, peak amplitude and center frequency of the intrinsic mode functions, performing adaptive variational mode decomposition processing on the bus duct vibration signals to obtain independent intrinsic mode functions; analyzing the extracted intrinsic mode functions to determine the intrinsic mode functions containing useful information. The detrend processing on the signals after the signals are collected can reduce the collection error of the vibration signals; in addition, the vibration signals are processed by using the improved variational mode decomposition, so that the extracted intrinsic mode functions reach the best in terms of independence and noise reduction.

[0007] Preferably, the process of obtaining intrinsic mode functions under different parameters comprises the following steps: establishing a mathematical model of variational mode decomposition; converting the constrained variational problem into an unconstrained variational problem by using the augmented Lagrange method to obtain an augmented Lagrange expression; solving the unconstrained variational problem by using the alternating direction multiplier method to obtain an update formula; constantly updating the frequency center and wideband of each IMF component in the process of iteratively solving the variational model until the iteration stopping condition is met, and finally inversely Fourier transforming the solved mode to obtain K intrinsic modes IMF after VMD decomposition; letting K take 2 to 10 in turn and substituting into VMD to obtain intrinsic mode functions under different VMD decomposition numbers K.

[0008] Preferably, the mathematical model of variational mode decomposition is shown in the following expression:

[0009]

[0010] Wherein, f(t) represents the signal to be analyzed, K represents the number of inherent modal components, k represents the serial number of the mode, U k (t) represents the kth inherent modal component of the signal to be analyzed, ω k represents the center frequency of the kth inherent modal component of the signal to be analyzed, δ(t) represents Dirac, t represents time, represents the partial derivative of t, j represents an imaginary number.

[0011] As preferred, the augmented Lagrange expression is as follows:

[0012]

[0013] Wherein, α represents a quadratic penalty factor, λ(t) represents the Lagrange multiplier at time t. This step uses the augmented Lagrange method to convert the constrained variational problem into an unconstrained variational problem, which is convenient for subsequent data updating.

[0014] As preferred, the update formula is shown in the following expression:

[0015]

[0016]

[0017]

[0018] Wherein, ^ represents Fourier transform, * relationship represents convolution, k represents the serial number of the mode, n represents the number of iterations, τ represents the fidelity coefficient. This step continuously updates the frequency center and wideband of each IMF component in the process of solving the variational model iteratively until the iteration stopping condition is met, which makes the extracted inherent modal function contain less noise and interference information, increases data reliability and accuracy.

[0019] As preferred, the iteration stopping condition needs to meet the following expression:

[0020]

[0021] After meeting the expression, the update loop is exited.

[0022] As preferred, the number of variational modal decomposition is adaptively determined from three aspects of correlation coefficient, peak amplitude and center frequency of the inherent modal function, and the bus duct vibration signal is adaptively processed by variational modal decomposition to obtain each independent inherent modal function, which specifically includes the following contents: under the same decomposition parameter K, the independence and integrity of the obtained inherent modal function are evaluated from three aspects of correlation coefficient, peak amplitude and center frequency, and the decomposition effect is evaluated according to Kwc index, wherein Kwc is calculated according to the following expression:

[0023]

[0024] wherein c represents the correlation between the sum of all intrinsic mode functions in the VMD result and the original signal; represents the peak amplitude of the i th intrinsic mode function, ω i represents the center frequency difference between the 1 st IMF i and the IMF i+1 , c and ω i are all normalized, and N represents the number of modes. From the correlation coefficient, the peak amplitude and the center frequency, the Kwc index is calculated, so that the improved signal decomposition method has higher adaptive processing capability and accurate feature extraction capability.

[0025] As preferred, after the intrinsic mode function is evaluated, the Kwc index of the intrinsic mode function with the decomposition parameter K from 2 to 10 is traversed, and the K at the maximum Kwc is determined as the optimal parameter by drawing a graph.

[0026] The present application has the beneficial effects that: starting from the intrinsic mode function, the correlation, the peak value and the center frequency of the intrinsic mode function are analyzed, the Kwc index is calculated from the three directions, the adaptive decomposition number is selected, and the improved method has higher adaptive processing capability and accurate feature extraction capability. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is the flow chart of the method of the present application;

[0028] Figure 2 is the bus duct vibration signal in the embodiment;

[0029] Figure 3 is the bus duct vibration signal after pretreatment in the embodiment;

[0030] Figure 4 is the Kwc index change of the signal in the embodiment;

[0031] Figure 5 is the intrinsic mode function obtained after the vibration signal is improved and variational mode decomposition is performed in the embodiment;

[0032] Figure 6 is the original vibration signal spectrum and the spectrum of each intrinsic mode function. DETAILED DESCRIPTION

[0033] The technical solutions of the present application will be further specifically described below by means of embodiments, and in combination with the accompanying Figures 1-6 drawings.

[0034] Embodiment:

[0035] In this embodiment, an improved variational modal decomposition method for extracting characteristic signals of wind turbines is used, which uses the busbar vibration signals of wind turbines as experimental data for processing and analysis. Figure 1 As shown, the following steps are included.

[0036] Step 1: Collect the vibration signal of the wind turbine bus duct, the specific signal is as follows Figure 2 As shown;

[0037] Step 2: Detrend the collected signal to reduce the vibration signal acquisition error. This includes the following steps:

[0038] 2.1) Remove the trend term from the original vibration signal time domain waveform to reduce sensor error and increase the reliability of the vibration signal;

[0039] 2.2) Pre-filter the signal to eliminate some obvious interference signal frequency bands in advance. The pre-processed signal is as follows: Figure 3 As shown;

[0040] Step 3: Update the frequency and center bandwidth of each IMF component, perform inverse Fourier transform on the solved mode, and obtain K intrinsic mode IMFs after VMD decomposition. Specifically, the following steps are included:

[0041] 3.1) Initialization n=1,λ 1 , K = 2;

[0042] 3.2) Enter the update loop and continuously update according to the update formula λ n , the update formula is as follows:

[0043]

[0044]

[0045]

[0046] 3.3) When the data meets the stop condition, the update loop is exited and the update is completed. The stop condition meets the following expression:

[0047]

[0048] Step 4: Use traditional variational mode decomposition to process the busbar vibration acceleration signal, where the vibration signal is processed by traversing the decomposition number from 2 to 10 to obtain the natural mode function under different parameters;

[0049] Step 5: The number of VMD is determined by the correlation coefficient, peak amplitude and center frequency of the intrinsic mode function, and the bus duct vibration signal is processed by adaptive VMD to obtain each independent intrinsic mode function, the specific steps are as follows:

[0050] 5.1) Calculate the Kwc index of the intrinsic mode function under each different decomposition K, as shown in the following expression:

[0051]

[0052] Wherein, c represents the correlation between the sum of all intrinsic mode functions of VMD results and the original signal, ω i represents the peak amplitude of the i th intrinsic mode function, ω i represents the center frequency difference between the i th intrinsic mode function and the IMF i+1 , and c and ω i are normalized;

[0053] 5.2) Find the decomposition number K when Kwc is maximum, as shown in Figure 4 , wherein the Kwc index reaches the maximum value when K = 5;

[0054] 5.3) Adaptive VMD is performed using the optimal parameters, that is, when the decomposition number K = 5, the penalty coefficient α is set to the default parameter 3000, and the obtained intrinsic mode function is shown in Figure 5 ;

[0055] Step 6: Analyze the bus duct vibration signal as shown in Figure 6 (a) and the frequency spectrum of the extracted intrinsic mode function as shown in Figure 6 (b), each independent peak in the vibration signal can be individually represented from the intrinsic mode function, and the independent extraction of the complex signal is completed.

[0056] It should be understood that the embodiments are only used to illustrate the present application and not to limit the scope of the present application. In addition, it should be understood that after reading the content taught by the present application, those skilled in the art can make various modifications or modifications to the present application, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

Claims

1. An improved variational modal decomposition method for extracting characteristic signals of wind turbines, characterized in that: The following steps are involved: S1: Collect vibration signals of wind turbines; S2: Detrend the vibration signal; S3: Use traditional variational mode decomposition (VMD) to process the vibration signal and obtain the intrinsic mode functions (IMFs) under different decomposition numbers; S4: Adaptively determine the number of variational modal decompositions based on the correlation coefficient, peak amplitude, and center frequency of the intrinsic modal functions. Multiply the mean of the center frequency differences of the peak amplitudes of all intrinsic modal functions by the correlation coefficient to obtain the Kwc index of the intrinsic modal function under each different decomposition K. The K at which Kwc is maximized is determined as the optimal parameter. Adaptive variational modal decomposition is performed on the bus duct vibration signal using the optimal parameter to obtain each independent intrinsic modal function. S5: Analyze and extract the intrinsic mode functions, and determine the intrinsic mode functions containing useful information.

2. The improved variational modal decomposition method for extracting characteristic signals of wind turbines according to claim 1, characterized in that: The step S3 obtains the intrinsic mode functions under different decomposition numbers, which specifically includes the following steps: S3.1: Develop a mathematical model for variational mode decomposition; S3.2: Use the augmented Lagrangian method to transform the constrained variational problem into an unconstrained variational problem and obtain the augmented Lagrangian expression; S3.3: Use the alternating direction multiplier method to solve the unconstrained variational problem in S3.2 and obtain the updated formula; S3.4: During the iterative solution of the variational model, the frequency center and bandwidth of each intrinsic modal component are continuously updated until the iteration stopping condition is met. Finally, the solved mode is inverse Fourier transformed to obtain the K intrinsic modal components after variational modal decomposition. S3.5: Let K vary from 2 to 10, and substitute them into the mathematical model of variational mode decomposition in sequence to obtain the intrinsic mode functions obtained under different decomposition numbers K.

3. The improved variational modal decomposition method for extracting characteristic signals of wind turbines according to claim 2, characterized in that: The mathematical model of variational mode decomposition of S3.1 is shown as follows: Where f(t) represents the signal to be analyzed, K represents the number of intrinsic modal components, k represents the intrinsic modal function number, u k (t) represents the kth natural mode component of the signal to be analyzed, ω k Represents the center frequency of the kth natural mode component of the signal to be analyzed, δ(t) represents the Dirac function, t represents the time, represents the partial derivative with respect to t, and j represents an imaginary number.

4. The improved variational modal decomposition method for extracting characteristic signals of wind turbines according to claim 2, characterized in that: The augmented Lagrangian expression of S3.2 is as follows: Among them, α represents the quadratic penalty factor, and λ(t) represents the Lagrange multiplier at time t.

5. The improved variational modal decomposition method for extracting characteristic signals of wind turbines according to claim 2, characterized in that: The update formula in S3.3 is as follows: Among them, ^ represents Fourier transform, * represents convolution, n represents the number of iterations, and τ represents the fidelity coefficient.

6. The improved variational modal decomposition method for extracting characteristic signals of wind turbines according to claim 2, characterized in that: The iteration stopping condition described in S3.4 needs to satisfy the following expression: After the above expression is satisfied, the update loop is exited.

7. The improved variational modal decomposition method for extracting characteristic signals of wind turbines according to claim 1, characterized in that: Step S4 specifically includes the following contents: Under the same decomposition parameter K, the independence and integrity of the obtained intrinsic mode functions are evaluated from three aspects: correlation coefficient, peak amplitude, and center frequency. The decomposition effect is evaluated according to the Kwc indicator, where Kwc is calculated according to the following expression: Where c represents the correlation between the sum of all intrinsic mode functions in the variational mode decomposition result and the original signal; represents the peak amplitude of the i-th intrinsic mode function, ω i Indicates the IMF i with the IMF i+1 The center frequency difference between c and ω i All are normalized, and N represents the number of modes.

8. The improved variational modal decomposition method for extracting characteristic signals of a wind turbine generator set according to claim 7, characterized in that: After evaluating the intrinsic mode function, the Kwc index of the intrinsic mode function with decomposition parameter K ranging from 2 to 10 is traversed, and the K at the maximum Kwc is determined to be the optimal parameter.

Images