Broadband oscillation monitoring method and device

Abstract

The invention discloses a broadband oscillation monitoring method and device, and the method comprises the steps: collecting an initial signal of a to-be-detected power system line in real time, and carrying out the preprocessing of the initial signal; performing frequency interval division on the preprocessed signal, and performing adaptive filtering; separating a plurality of key components from the filtered signal; performing parameter estimation on the plurality of key components, extracting target modal information, and performing signal reconstruction; performing multi-modal identification on the reconstructed signal, and extracting a plurality of modal components; and judging whether a broadband oscillation phenomenon exists or not by using the plurality of modal components. According to the method, accurate separation and parameter estimation of broadband oscillation signals are realized, and the identification precision of oscillation frequency, amplitude and phase is improved.

Description

Technical Field

[0001] The present invention relates to the field of broadband oscillation of power systems, and in particular to a broadband oscillation monitoring method and device of TLS-ESPRIT based on improved VMD filtering. Background Art

[0002] In the contemporary power system environment with "double high" characteristics, new oscillation problems have emerged. This is due to the complex interaction between power electronic devices and grid components, which induces oscillation phenomena with broadband, time-varying and significant nonlinear characteristics. Unlike in the past, this new form of oscillation is mainly caused by the power electronic control mechanism, causing a 10-year -1 to 10 3 Electromagnetic oscillations in the Hertz frequency band, hence the name "broadband oscillations".

[0003] Broadband oscillation phenomenon As wind, solar, flexible DC and other power electronic equipment are increasingly connected to the grid, power systems often encounter a series of difficult problems when faced with managing and mitigating broadband oscillations, such as difficulty in monitoring and identification, complex causes, wide-ranging effects, and limitations of existing control strategies. Specifically, the wide spectrum and variable dynamic characteristics of broadband oscillations exceed the capture capabilities of conventional monitoring systems; their causes involve multiple aspects such as grid layout, new energy integration, and regulation strategies, requiring in-depth analysis and meticulous response. What's worse, such oscillations are prone to spread rapidly in the power grid, posing a severe test to the stability of the entire system. Summary of the invention

[0004] In order to solve the problems existing in the background technology, the present invention proposes a wide-band oscillation monitoring method and device, aiming to solve the problems faced by the prior art in wide-band oscillation monitoring, such as insufficient recognition accuracy, weak noise resistance and limited ability to process complex oscillation modes, and improve the monitoring and analysis capabilities of wide-band oscillations in power systems.

[0005] In one aspect, the present invention provides a broadband oscillation monitoring method, comprising:

[0006] Collect the initial signal of the power system line to be tested in real time and pre-process the initial signal;

[0007] Divide the preprocessed signal into frequency intervals and perform adaptive filtering;

[0008] Separate several key components from the filtered signal;

[0009] Parameters of the key components are estimated, target modal information is extracted, and signal reconstruction is performed;

[0010] Perform multimodal recognition on the reconstructed signal and extract several modal components;

[0011] The modal components are used to determine whether broadband oscillation occurs.

[0012] In one embodiment, the initial signal is preprocessed, comprising:

[0013] The initial signal is denoised using wavelet analysis technology, including:

[0014] By calculating each wavelet coefficient of the initial signal and applying soft threshold processing to the wavelet coefficients, wavelet coefficients below a first threshold are removed, and wavelet coefficients greater than or equal to the first threshold are retained, that is:

[0015] T λ (x) = sign(x)max(|x|-λ,0)

[0016] Where: λ is the preset first threshold, T λ (x) is the wavelet coefficient obtained after applying the first threshold, and x is the original wavelet coefficient.

[0017] In one embodiment, preprocessing the initial signal further comprises:

[0018] The initial signal is subjected to preliminary modal detection using fast Fourier transform to obtain initial modal information, which includes the frequency, amplitude, and initial phase of the resonance mode.

[0019] In one embodiment, the preprocessed signal is divided into frequency intervals and adaptively filtered, including:

[0020] Preliminarily dividing the preprocessed signal into a plurality of frequency intervals, the plurality of frequency intervals comprising a low frequency, a medium frequency, a high frequency, an extremely high frequency and an interference frequency band;

[0021] According to the multiple frequency intervals, an adaptive bandpass filter is designed to dynamically adjust the center frequency and bandwidth, including:

[0022] Set the initial upper and lower frequency limits of the bandpass filters in the low frequency, medium frequency, high frequency and ultra-high frequency ranges;

[0023] According to the initial upper frequency limit and lower frequency limit of each bandpass filter, the initial center frequency and the initial bandwidth of the bandpass filter are set, wherein the initial center frequency is the average of the initial upper frequency limit and the initial lower frequency limit, and the initial bandwidth is the difference between the average of the initial upper frequency limit and the initial lower frequency limit;

[0024] Obtain real-time spectrum information, and dynamically adjust the center frequency and bandwidth of each bandpass filter based on the spectrum information, where:

[0025] The adjusted center frequency is the sum of the initial center frequency and a frequency adjustment amount, and the frequency adjustment amount is determined according to a difference between the initial center frequency and a main frequency component extracted from the spectrum information;

[0026] The adjusted bandwidth is the sum of the initial bandwidth and the bandwidth adjustment amount, and the bandwidth adjustment amount is determined according to the difference between the initial bandwidth and the main bandwidth component extracted from the spectrum information.

[0027] In one embodiment, an objective function is constructed according to the adjusted center frequency and bandwidth, and the objective function is expressed as:

[0028]

[0029] Where: φ k (t) represents the kth mode function, a k (t) is the time-varying amplitude function of the kth mode, f H is the upper frequency limit, f L is the lower frequency limit; β and γ are regularization parameters used to control the smoothness and bandwidth of the mode to the extent that they are close to the preset target δ.

[0030] In one embodiment, all bandpass filters are cascaded, and the bandpass filter at the next stage fine-tunes the center frequency and bandwidth according to the output of the bandpass filter at the previous stage.

[0031] In one embodiment, a matching pursuit algorithm is used to separate the key components of the filtered signal layer by layer, including:

[0032] Set the initial residual to the filtered signal;

[0033] In each iteration, an atom is selected from a predefined atom library so that the atom has the highest matching degree with the current residual under a preset metric condition;

[0034] Calculate a coefficient between the atom and the current residual;

[0035] The next residual is obtained according to the difference between the current residual and the coefficient multiplied by the atom.

[0036] In one embodiment, the TLS-ESPRIT algorithm is used to estimate the parameters of the key component, including:

[0037] Taking the several key components as input, a Hankel matrix is ​​constructed;

[0038] The constructed Hankel matrix is ​​subjected to eigenvalue decomposition using the total least squares method;

[0039] From the obtained eigenvalues, the frequency components are extracted.

[0040] In one embodiment, performing multimodal identification on the reconstructed signal to extract modal components includes:

[0041] Calculating an autocorrelation function of a reconstructed signal and setting a threshold of the autocorrelation function to distinguish a modal component of the reconstructed signal from noise;

[0042] Different modal features are extracted and classified from the identified modal components, including frequency, amplitude and phase.

[0043] In one embodiment, determining whether there is a broadband oscillation phenomenon includes:

[0044] Determine whether there is a frequency band exceeding a preset frequency range in the frequencies of each modal component, and if there is a frequency band exceeding the preset frequency range, identify that broadband oscillation exists;

[0045] and / or

[0046] Calculate the power spectrum density of each modal component to determine the energy distribution; if there is an abnormal increase in the energy corresponding to the frequency band, it is identified as the presence of broadband oscillation;

[0047] Observe the change of the frequency of each modal component over time. If there is a rapid fluctuation or irregular change in frequency, it is identified as the presence of broadband oscillation.

[0048] In one embodiment, before determining whether there is a broadband oscillation phenomenon, the method further includes:

[0049] Apply the Hilbert transform to each modal component to determine the instantaneous frequency and instantaneous amplitude of the signal, where: the instantaneous frequency is determined by the ratio of the derivative of the instantaneous phase and 2π, that is:

[0050]

[0051] Wherein, f(t) represents the instantaneous frequency, the instantaneous phase Φ(t) is determined by the phase of the instantaneous complex signal z(t)=x(t)+jy(t) composed of the imaginary part y(t) obtained by the Hilbert transform of the real part x(t), and Φ(t)=arg(z(t)).

[0052] The instantaneous amplitude is determined by the modulus of the instantaneous complex signal amplitude, that is:

[0053]

[0054] Where a(t) represents the instantaneous amplitude.

[0055] In one embodiment, before applying Hilbert transform to each modal component to determine the instantaneous frequency and instantaneous amplitude of the signal, the method further comprises:

[0056] Screening modal components of complex frequency structures for adaptive filtering; and

[0057] In the adaptive filtering process, an iterative optimization algorithm is used to update the filter weights under the minimum mean square error criterion, where: the filter weight vector w n The update at the nth iteration is expressed as:

[0058] w[n+1]=w[n]+μe[n]x[n]

[0059] Where: e[n] is the current error, d[n] is the desired output, y[n] is the actual output of the filter, x[n] is the input vector, and μ is the learning rate.

[0060] Another aspect of the present invention further provides a broadband oscillation monitoring device, which adopts the above broadband oscillation monitoring method, and the broadband oscillation monitoring device comprises:

[0061] A preprocessing module is used to collect the initial signal of the power system line to be tested in real time and preprocess the initial signal;

[0062] An adaptive filtering module is used to divide the pre-processed signal into frequency intervals and perform adaptive filtering;

[0063] A key component extraction module is used to separate several key components from the filtered signal;

[0064] A reconstruction module is used to estimate the parameters of the key components, extract the target modal information, and reconstruct the signal;

[0065] A multimodal recognition module is used to perform multimodal recognition on the reconstructed signal and extract several modal components;

[0066] The broadband oscillation determination module is used to determine whether a broadband oscillation phenomenon exists by using the plurality of modal components.

[0067] In addition, another aspect of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above-mentioned wide-band oscillation monitoring method are implemented and the same technical effect can be achieved.

[0068] The present invention has outstanding substantive features and significant progress compared to the prior art, specifically:

[0069] The broadband oscillation monitoring method proposed in the present invention pre-processes the initial signal of the power system line to be tested in real time, divides the pre-processed signal into frequency intervals, and uses improved VMD filtering for adaptive filtering; at the same time, separates several key components from the filtered signal, and uses the TLS-ESPRIT algorithm to estimate the parameters of the key components and reconstruct the signal; and performs multi-modal recognition on the reconstructed signal to extract several modal components; and uses the several modal components to determine whether broadband oscillation phenomenon exists. This method realizes the accurate separation and parameter estimation of broadband oscillation signals, and improves the recognition accuracy of oscillation frequency, amplitude and phase. This method is applicable to various oscillation phenomena in modern power systems, especially broadband, time-varying and nonlinear oscillations caused by the complex interaction between power electronic devices and power grid components. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is a schematic diagram of the overall flow of a broadband oscillation monitoring method according to an embodiment of the present invention;

[0071] Figure 2 is a schematic block diagram of the overall structure of a broadband oscillation monitoring device according to an embodiment of the present invention;

[0072] Wherein, the accompanying drawings are marked as follows:

[0073] 300-Broadband Oscillation Monitoring Device;

[0074] 301-preprocessing module;

[0075] 302-adaptive filtering module;

[0076] 303-key component extraction module;

[0077] 304-Reconstruction module;

[0078] 305-Multimodal recognition module;

[0079] 306-Broadband oscillation determination module. DETAILED DESCRIPTION

[0080] In order to make the above features and effects of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings.

[0081] See also Figure 1 As shown, Figure 1 The figure shows the overall flow chart of the broadband oscillation monitoring method disclosed in one embodiment of the present invention.

[0082] A broadband oscillation monitoring method comprises at least the following steps:

[0083] Step S1: collect the initial signal of the power system line to be tested in real time, and pre-process the initial signal.

[0084] In one embodiment, a PMU (Phasor Measurement Unit) is used to monitor the initial signal of the power system line to be tested in real time, including voltage and current, and the original signal is sampled to ensure that the signal is digitized for further analysis.

[0085] Further, the signal quality is improved by preprocessing means. Specifically, in one embodiment, wavelet analysis technology, especially Daubechies wavelet basis, is used to perform denoising on the initial signal. Specifically, each wavelet coefficient of the initial signal is calculated, and soft threshold processing is applied to these wavelet coefficients, that is:

[0086] T λ (x) = sign(x)max(|x|-λ,0)

[0087] Where: λ is the preset first threshold, T λ (x) is the wavelet coefficient obtained after applying the first threshold, and x is the original wavelet coefficient.

[0088] In this formula, max(|x|-λ, 0) ensures that if |x|-λ is negative (i.e., x is less than the first threshold λ), it is set to 0, so that small values ​​are often regarded as noise. sign(x) is multiplied by max(|x|-λ, 0) to ensure that the processed value retains the same sign as the original value, but its absolute value is thresholded.

[0089] Soft threshold processing can "smooth out" the wavelet coefficients below the first threshold, while retaining the wavelet coefficients greater than or equal to the first threshold, although there will be some attenuation. Therefore, it is very useful for noise reduction and data compression, especially for large power systems containing a large number of microsecond power electronic equipment, ensuring that the quality of signal processing is not affected by noise.

[0090] In addition, in one embodiment, during the preprocessing process, the initial signal is further subjected to preliminary modal detection using fast Fourier transform to obtain initial modal information, which includes the frequency, amplitude, and initial phase of the resonant mode. This process provides basic data references for guiding the subsequent bandpass filter parameter settings and setting the threshold of signal processing.

[0091] Step S2: Divide the preprocessed signal into frequency intervals and perform adaptive filtering.

[0092] In one embodiment, a K-means clustering algorithm is used to perform cluster analysis on the frequency of the resonant mode in the initial modal information obtained by the FFT fast Fourier transform, and the pre-processed signal is initially divided into multiple frequency intervals according to the frequency range. Specifically, the multiple frequency intervals include low frequency f l 、Intermediate frequency f m , high frequency f h , ultra-high frequency f he And the interference frequency band f o There are five categories in total. Among them, the interference frequency band f o This is the result of poor clustering effect. These spectral feature points are those spectral components that are difficult to clearly classify into low frequency, medium frequency, high frequency or very high frequency. These feature points may frequently change the cluster to which they belong, or be far away from all cluster centers. Therefore, f o It is excluded as an interference term, and further noise reduction is performed on the basis of wavelet noise reduction processing.

[0093] Furthermore, based on the initial modal information provided by FFT modal detection, an adaptive bandpass filter is designed for adaptive filtering. The adaptive bandpass filter can dynamically adjust parameters such as center frequency and bandwidth to extract or separate the voltage / current signal in each frequency band. This not only improves processing efficiency, but also ensures the accuracy and real-time performance of signal decomposition.

[0094] In one embodiment, an improved VMD filter is designed. Based on the traditional VMD (Variational Mode Decomposition), the improved VMD filter dynamically adjusts the frequency lower limit f L and the upper frequency limit f H , and the lower frequency limit f L With the upper frequency limit f H The mean of is used as the center frequency, and the bandwidth and other parameters are automatically adjusted to achieve adaptive signal separation, especially within each frequency band, so that the filtering process is more in line with the real characteristics of the signal.

[0095] Specifically, first, according to the multiple frequency intervals divided as above, the initial upper and lower frequency limits of the bandpass filters in the low frequency, medium frequency, high frequency, and ultra-high frequency intervals are set, that is:

[0096] Low frequency bandpass filter: Initial frequency range Min(f l )-Max(f l )

[0097] Intermediate frequency bandpass filter: initial frequency range Min(f m )-Max(f m )

[0098] High frequency bandpass filter: initial frequency range Min(f h )-Max(f h )

[0099] Very high frequency bandpass filter: initial frequency range Min(f he )-Max(f he )

[0100] At the same time, the initial center frequency and initial bandwidth of each bandpass filter are further determined based on the initial upper frequency limit and lower frequency limit of each bandpass filter. The initial center frequency is the average of the initial upper frequency limit and the initial lower frequency limit, and the initial bandwidth is the difference between the initial upper frequency limit and the initial lower frequency limit.

[0101] In addition, in practical applications, the frequency characteristics of the signal may change with time or environment. Therefore, the adaptive bandpass filter further refines and dynamically adjusts the frequency range based on the above-mentioned preliminary divided frequency intervals, so as to more accurately separate and extract the frequency components in the signal. In this regard, in one embodiment, the spectrum information of the signal, especially the change of the frequency components, is monitored in real time. Specifically, the sliding window FFT can be used to obtain real-time spectrum information. The length of the sliding window is set to L, the sliding step is set to S, and the signal segment of each sliding window length of the original signal is FFT, and the spectrum within this time window is obtained as follows:

[0102]

[0103] You can get X n (k) The signal value in the window, and then the window slides according to the step size S to obtain a series of spectrum graphs X n (f).

[0104] Then, according to the spectrum information monitored in real time, the center frequency and bandwidth of each bandpass filter are dynamically adjusted, where:

[0105] For each bandpass filter, the adjusted center frequency is the sum of the initial center frequency and the frequency adjustment amount, and the frequency adjustment amount is determined according to the difference between the initial center frequency and the main frequency component extracted from the spectrum information. That is, the adjusted center frequency is:

[0106] f c (t+1)=f c (t)+Δf c , where: fc is the initial center frequency of the bandpass filter, Δf c is the frequency adjustment, Δf c =α(f main -f c), α is the adjustment coefficient, which is used to control the speed and amplitude of the adjustment.

[0107] For each bandpass filter, the adjusted bandwidth is the sum of the initial bandwidth and the bandwidth adjustment amount, and the bandwidth adjustment amount is determined according to the difference between the initial bandwidth and the main bandwidth component extracted from the spectrum information. That is, the adjusted bandwidth is: Bandwidth(t+1)=Bandwidth(t)+ΔBW, where: Bandwidth(t) is the initial bandwidth of the bandpass filter, ΔBW is the bandwidth adjustment amount, ΔBW=β(BW main -BW), β is the adjustment coefficient, which is used to control the speed and amplitude of the adjustment.

[0108] After adjusting the center frequency and bandwidth, the initial frequency range is provided for VMD optimization. At the same time, the results of the sliding window FFT are used to update the center frequency and bandwidth of the modal function in real time. Through this dynamic adjustment, it can be ensured that the modal function is always synchronized with the changes in the signal.

[0109] On this basis, subsequent optimized VMD processing is carried out.

[0110] The objective function of variational mode decomposition (VMD) is used to decompose a complex signal into a set of modal components with specific frequency characteristics. In one embodiment, the objective function is constructed according to the adjusted center frequency and bandwidth, and the objective function is expressed as:

[0111]

[0112] Where: φ k (t) represents the kth mode function, a k (t) is the time-varying amplitude function of the kth mode, f H is the upper frequency limit, f L is the lower frequency limit; β and γ are regularization parameters that control the smoothness and bandwidth constraints of the mode, and are used to control the degree to which the smoothness and bandwidth of the mode are close to the preset target δ.

[0113] Among them, the mode function φ k (t) and the amplitude function a k The initial state of (t) is determined based on the previous series of steps. β is used to control the smoothness of the modal function. An empirical initial value can be determined through theoretical analysis. For example, for signals with large noise, a larger β value is selected to suppress high-frequency noise. Similarly, an initial value of γ is set. γ is used to control the frequency domain bandwidth of the modal component to approach the preset target bandwidth δk. For signals with a wide spectrum, a larger γ value is selected to control the bandwidth. For the kth mode, its target bandwidth δ k Determined by the previous fast FFT and dynamic adjustment.

[0114] In addition, in one embodiment, the alternating direction multiplier method (ADMM) is used to iteratively optimize the objective function, and the amplitude function, the modal function, and the upper and lower limits of the frequency are iteratively optimized, and finally converge to obtain a local optimal solution to complete the decomposition process of VMD. Ultimately, when separating and extracting different modes of the signal, the goal of ensuring that the characteristics of each mode (such as frequency, bandwidth, sparsity, etc.) can accurately reflect the characteristics of the original signal is achieved.

[0115] In addition, in some embodiments, in order to process the different frequency components of the signal more carefully, a method of cascading multiple improved VMD filters is adopted, and the bandpass filters of the low frequency, medium frequency, high frequency, and extremely high frequency intervals set above are cascaded, and each bandpass filter is optimized for a different frequency band. By setting multiple VMD filters with different parameters, each bandpass filter is responsible for separating signal components within a specific frequency range, such as low frequency, medium frequency, high frequency, and extremely high frequency. The advantage of cascade processing is that it can improve the accuracy and real-time performance of signal separation through multi-level refinement.

[0116] The next filter adjusts parameters such as center frequency and bandwidth according to the output of the previous filter. The overall process can be viewed as an iterative optimization chain as follows:

[0117] x (i) (t)→VMD i (f L，i , f H,i params i )→y (i) (t)

[0118] x (i+1) (t) = y (i) (t)-Processed Component

[0119] Where: x (i) (t) and y (i) (t) respectively represent the input and output signals of the i-th level processing, VMDi represents the i-th filter processing and removing the signal components of a specific frequency band each time until the required processing accuracy or the predetermined number of cascades is achieved, and Processed Component represents the signal components from y (i) The specific frequency components separated from (t) actually involve the (i) (t) is appropriately processed to prepare for entering the next level of analysis or further filtering.

[0120] Specifically, first (1) the first-level VMD decomposition is performed on the divided frequency bands to obtain the first-level modal components; (2) the signal-to-noise ratio is calculated to evaluate the effect of the first-level decomposition; (3) according to the effect of the first-level VMD decomposition, the center frequency, bandwidth and other parameters are fine-tuned; (4) the steps of the first level are repeated, and the output signal of each level is further VMD processed, and the VMD parameters are continuously adjusted according to the decomposition effect until the set decomposition accuracy or the maximum number of cascades is reached.

[0121] In this embodiment, wavelet denoising technology and improved VMD filter are used to effectively remove various noises in the signal, including thermal noise, shot noise and flicker noise, thereby enhancing the quality of signal processing. The improved VMD filter can dynamically adjust the bandwidth and center frequency, adapt to the frequency characteristics of the signal, and is suitable for signal processing in different frequency bands. And by cascading multiple bandpass filters, the accuracy and real-time performance of signal separation can be improved through multi-level refinement.

[0122] Step S3: Separate several key components from the filtered signal.

[0123] In this embodiment, after filtering is completed in step S2, the filtered signal is further subjected to key component separation. In one embodiment, a matching pursuit algorithm is used to separate several key components of the signal layer by layer, especially to process transient and non-stationary parts, so as to improve the clarity of the signal structure.

[0124] The Matching Pursuit (MP) algorithm is an iterative signal decomposition method that is particularly suitable for processing non-stationary signals and signals with transient characteristics, such as broadband electromagnetic oscillations in power systems. The core idea of ​​the algorithm is to gradually extract the key components that best represent the signal structure from the signal through layer-by-layer decomposition, thereby improving the resolution and clarity of the signal. The MP algorithm is applied after the cascade VMD to further optimize the signal decomposition results and improve the reconstruction accuracy, especially when dealing with noisy and complex signals. The main steps are as follows:

[0125] First, at the beginning, the initial residual r0 of the signal is defined as the filtered signal x[n], that is, r0=x[n].

[0126] Then, in each iteration, an atom g[n] is selected from a predefined atom library (usually a redundant dictionary, such as a wavelet basis, a Fourier basis, or a Gabor basis, etc.) so that this atom is consistent with the current residual r n The best match under some metric (such as inner product). That is, find an atom such that

[0127]

[0128] Where: is the atomic library, and <.,.> represents the inner product.

[0129] Calculate the selected atom g[n] and the current residual r n The coefficient a between n , that is, the projection length of the signal in the direction of the atom, determines the contribution of the atom in the signal representation:

[0130]

[0131] According to the difference between the current residual and the coefficient multiplied by the atom, the residual signal is updated to obtain the next residual r n+1 , to reflect the remaining information beyond the extracted components, namely:

[0132] r n+1 =r n -a n g n

[0133] The above iteration process is repeated until a preset stopping criterion is met. In some embodiments, the preset stopping condition is set to the residual energy being lower than a set threshold, reaching a maximum number of iterations, or the residual change being small enough.

[0134] In this embodiment, a matching pursuit algorithm is used to separate several key components of the signal layer by layer, thereby improving the clarity and resolution of signal decomposition.

[0135] Step S4: Parameter estimation is performed on the key components, target modal information is extracted, and signal reconstruction is performed, where the target modal information includes frequency, amplitude, and phase information.

[0136] In this embodiment, after cascade VMD filtering and matching pursuit algorithm processing, several key components in the signal are extracted and the signal is decomposed into multiple modal functions. These modal functions contain the signal characteristics of different frequency components. Further, the TLS-ESPRIT algorithm is used to estimate the parameters of the key components, and the target modal information is accurately extracted by constructing the Hankel matrix and total least squares estimation to reconstruct the signal. Compared with the initial modal information, the target modal information is more accurate frequency, amplitude and phase information, completing the spectrum reconstruction of the signal.

[0137] First, a Hankel matrix H is constructed with the key components as input. If the sampling points of the key components are x n , (n=0.1.2.3....N-1), then the Hankel matrix can usually be expressed as

[0138]

[0139] Where: M is the number of rows of the Hankel matrix (which also determines the lag length), N is the total number of sampling points of the signal, and usually N>M.

[0140] Then, the constructed Hankel matrix is ​​eigenvalue decomposed using the total least squares (TLS) method. Unlike the traditional least squares method that only minimizes the sum of squared errors, the TLS method considers the errors of both the observation vector and the coefficient matrix, and generally provides a more robust estimate. After executing TLS-ESPRIT, the eigenvalue decomposition of the matrix can be obtained:

[0141] H=UΛU H

[0142] Where: U is a matrix containing orthogonal eigenvectors, Λ is a diagonal matrix whose diagonal elements are the eigenvalues ​​of the Hankel matrix, which contain the frequency information of the signal.

[0143] Finally, the frequency components of the signal can be further extracted from the obtained eigenvalues. For the case of discrete spectrum peaks, the frequency f k It can be calculated according to the kth eigenvalue through the following relationship:

[0144]

[0145] Where: T represents the sampling period, k is the index based on the eigenvalue sorting, reflecting the position of different frequency components in the signal.

[0146] In this embodiment, the target modal information of the signal is accurately estimated through the TLS-ESPRIT algorithm by using the total least squares method and the rotation invariance principle.

[0147] Step S5: Perform multimodal identification on the reconstructed signal to extract several modal components.

[0148] In some embodiments, after the signal is reconstructed through the above step S4, the autocorrelation function of the reconstructed signal is further calculated, and the threshold of the autocorrelation function is set to distinguish the modal component and noise of the reconstructed signal, and further verify the accuracy of the recognition result. For broadband oscillation monitoring, the calculation of the autocorrelation function (ACF) can be expressed as:

[0149]

[0150] Where: x[n] represents the sample value of the reconstructed signal, and τ is the lag time. In order to distinguish the modal component from the noise in the reconstructed signal, the threshold T of the autocorrelation function is set, and the screening condition can be expressed as: R xx (τ)>T, if the screening condition is met, it means that the modal component exists.

[0151] Then, different modal features are further extracted and classified for the identified modal components to provide a reliable data basis for multimodal identification. Once the signal modes are initially identified, the next step is to extract the features of these modes and classify them. The main features include frequency, amplitude and phase, which can be obtained directly or indirectly from the frequency domain or time-frequency analysis of the signal. The parameters in this case can be obtained from the TLS-ESPRIT algorithm, where:

[0152] The extracted frequency features are: Where: k is the frequency index, N is the number of sampling points, and T is the sampling period.

[0153] The extracted amplitude features are: A k =|X[k]|, where: X[k] is the kth coefficient of the discrete Fourier transform.

[0154] The extracted phase feature is: k =∠X[k].

[0155] Then, after extracting the frequency features, amplitude features, and phase features, a clustering algorithm is further used to perform modal classification:

[0156] Cluster=ClusteringMethod({A k ,φ k ,...})

[0157] Where: Set {A k ,φ k , ...} represents the set of eigenvectors of the mode, and ClusteringMethod represents the clustering algorithm.

[0158] Step S6: using the modal components to determine whether a broadband oscillation phenomenon exists.

[0159] In step S6 , Hilbert transform is first applied to the modal components to determine the instantaneous frequency and instantaneous amplitude of the signal.

[0160] In the in-depth analysis of broadband oscillations, the Hilbert transform is a powerful tool to understand the instantaneous characteristics of the signal. The calculation of the instantaneous frequency f(t) allows direct observation of the frequency variation of the signal at different points in time, which is essential for capturing nonlinear or non-stationary behavior in broadband oscillations. It helps to identify and track the evolution of oscillation modes over time.

[0161] Among them, the instantaneous frequency ft is determined by the ratio of the derivative of the instantaneous phase Φ(t) and 2π, that is:

[0162]

[0163] The instantaneous phase Φ(t) is determined by the phase of the complex signal z(t)=x(t)+jy(t) composed of the imaginary part y(t) obtained by Hilbert transform of the real part x(t), Φ(t)=argz(t)).

[0164] The instantaneous amplitude a(t) is the modulus of the instantaneous complex signal amplitude, that is:

[0165]

[0166] In addition, in some embodiments, before applying Hilbert transform to these modal components and determining the instantaneous frequency and instantaneous amplitude of the signal, adaptive filtering is also performed on the modal components of the complex frequency structure to further remove noise and ensure signal purity. When processing the modal components of the complex frequency structure, the filter parameters can be dynamically adjusted to better adapt to the signal characteristics. In one embodiment, an iterative optimization algorithm is used to update the filter weights under the minimum mean square error (LMS) criterion, where: The filter weight vector w n The update at the nth iteration is expressed as:

[0167] w[n+1]=w[n]+μe[n]x[n]

[0168] Where: e[n] is the current error, d[n] is the desired output, y[n] is the actual output of the filter, x[n] is the input vector, and μ is the learning rate.

[0169] In this embodiment, adaptive filtering is performed on the signal with a complex frequency structure to further remove noise, thereby ensuring the purity of the signal and the effectiveness of the processing, thereby enhancing the stability and reliability of the system.

[0170] In step S6, after applying Hilbert transform to each modal component to determine the instantaneous frequency and instantaneous amplitude of the signal, it is further determined whether a broadband oscillation phenomenon exists according to the instantaneous frequency and instantaneous amplitude of each modal component.

[0171] In some embodiments, specifically, whether there is a broadband oscillation phenomenon is determined from three perspectives: frequency distribution analysis, modal energy analysis, and time-varying characteristic analysis.

[0172] For frequency distribution analysis, by checking whether there is a frequency band exceeding a preset frequency range in the frequencies of each modal component identified, that is, a frequency band spanning a wider frequency range, especially focusing on those frequencies that are not within the traditional narrowband oscillation range (such as 0.1Hz to 2Hz commonly seen in power systems). Broadband oscillations usually involve multiple frequencies appearing simultaneously or successively, spanning a wider frequency range. If there is a frequency band exceeding the preset frequency range, it is identified as the presence of broadband oscillation.

[0173] For modal energy analysis, the power spectral density (PSD) of each modal component is calculated to evaluate its energy distribution. Broadband oscillation is often accompanied by abnormal increase in energy in certain frequency bands, which can be identified by comparing with historical data under normal operating conditions. If there is an abnormal increase in energy corresponding to one of the frequency bands, it is identified as the presence of broadband oscillation. The power spectral density is expressed as:

[0174] PSD(f k )=|X k | 2 , where: X k is the signal at frequency f k The spectral coefficients at .

[0175] For time-varying characteristic analysis, the instantaneous frequency information obtained by Hilbert transform is used to observe the frequency variation of each modal component over time. Broadband oscillation may appear as rapid fluctuation or irregular change of frequency, which is related to nonlinear interaction or external disturbance within the system. Therefore, if there is rapid fluctuation or irregular change of frequency, it is identified as the presence of broadband oscillation.

[0176] In summary, the broadband oscillation monitoring method disclosed in the present invention realizes accurate separation and parameter estimation of broadband oscillation signals through the combination of improved VMD filtering and TLS-ESPRIT algorithm, and improves the recognition accuracy of oscillation frequency, amplitude and phase. The method is applicable to various oscillation phenomena in modern power systems, especially broadband, time-varying and nonlinear oscillations caused by complex interactions between power electronic devices and grid components.

[0177] In addition, corresponding to the above-mentioned broadband oscillation monitoring method, another embodiment of the present invention also discloses a broadband oscillation monitoring device, referring to Figure 2 , Figure 2 FIG. 3 shows the architecture of the broadband oscillation monitoring device 300, which can be implemented by Figure 1 The various processes of the broadband oscillation monitoring method are shown.

[0178] A broadband oscillation monitoring device 300, the device at least comprising:

[0179] The preprocessing module 301 is used to collect the initial signal of the power system line to be tested in real time and preprocess the initial signal.

[0180] The adaptive filtering module 302 is used to divide the pre-processed signal into frequency intervals and perform adaptive filtering.

[0181] The key component extraction module 303 is used to separate a number of key components from the filtered signal.

[0182] The reconstruction module 304 is used to estimate parameters of the key components, extract target modal information, and perform signal reconstruction.

[0183] The multimodal recognition module 305 is used to perform multimodal recognition on the reconstructed signal and extract a plurality of modal components.

[0184] The broadband oscillation determination module 306 is used to determine whether a broadband oscillation phenomenon exists by using the plurality of modal components.

[0185] In addition, corresponding to the above-mentioned wide-band oscillation monitoring method, another embodiment of the present invention also discloses an electronic device, including a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the above-mentioned wide-band oscillation monitoring method are implemented and the same technical effect can be achieved.

[0186] The processor is the processor in the electronic device in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0187] Another embodiment of the present invention further provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the above-mentioned steps of wide-band oscillation monitoring are implemented and the same technical effect can be achieved.

[0188] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be applied, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0189] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0190] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A broadband oscillation monitoring method, characterized in that: Include: Collect the initial signal of the power system line to be tested in real time and pre-process the initial signal; Divide the preprocessed signal into frequency intervals and perform adaptive filtering; Separate several key components from the filtered signal; Parameters of the key components are estimated, target modal information is extracted, and signal reconstruction is performed; Perform multimodal recognition on the reconstructed signal and extract several modal components; The modal components are used to determine whether broadband oscillation occurs.

2. The method according to claim 1, characterized in that The initial signal is preprocessed, including: The initial signal is denoised using wavelet analysis technology, including: By calculating each wavelet coefficient of the initial signal and applying soft threshold processing to the wavelet coefficients, wavelet coefficients below a first threshold are removed, and wavelet coefficients greater than or equal to the first threshold are retained, that is: T λ (x)=sign(x)max(|x|-λ,0) Where: λ is the preset first threshold, T λ (x) is the wavelet coefficient obtained after applying the first threshold, and x is the original wavelet coefficient.

3. The method according to claim 1 or 2, characterized in that: Preprocessing the initial signal also includes: The initial signal is subjected to preliminary modal detection using fast Fourier transform to obtain initial modal information, which includes the frequency, amplitude, and initial phase of the resonance mode.

4. The method according to claim 1, characterized in that: The preprocessed signal is divided into frequency intervals and adaptively filtered, including: Preliminarily dividing the preprocessed signal into a plurality of frequency intervals, the plurality of frequency intervals comprising a low frequency, a medium frequency, a high frequency, an extremely high frequency and an interference frequency band; According to the multiple frequency intervals, an adaptive bandpass filter is designed to dynamically adjust the center frequency and bandwidth, including: Set the initial upper and lower frequency limits of the bandpass filters in the low frequency, medium frequency, high frequency and ultra-high frequency ranges; According to the initial upper frequency limit and lower frequency limit of each bandpass filter, the initial center frequency and the initial bandwidth of the bandpass filter are set, wherein the initial center frequency is the average of the initial upper frequency limit and the initial lower frequency limit, and the initial bandwidth is the difference between the average of the initial upper frequency limit and the initial lower frequency limit; Obtain real-time spectrum information, and dynamically adjust the center frequency and bandwidth of each bandpass filter based on the spectrum information, where: The adjusted center frequency is the sum of the initial center frequency and a frequency adjustment amount, and the frequency adjustment amount is determined according to a difference between the initial center frequency and a main frequency component extracted from the spectrum information; The adjusted bandwidth is the sum of the initial bandwidth and the bandwidth adjustment amount, and the bandwidth adjustment amount is determined according to the difference between the initial bandwidth and the main bandwidth component extracted from the spectrum information.

5. The method according to claim 4, characterized in that Also includes: An objective function is constructed based on the adjusted center frequency and bandwidth. The objective function is expressed as: Where: φ k (t) represents the kth mode function, a k (t) is the time-varying amplitude function of the kth mode, f H is the upper frequency limit, f L is the lower frequency limit; β and γ are regularization parameters used to control the smoothness and bandwidth of the mode to the extent that they are close to the preset target δ.

6. The method according to claim 4, characterized in that Also includes: All bandpass filters are cascaded, and the bandpass filter at the latter stage fine-tunes the center frequency and bandwidth according to the output of the bandpass filter at the previous stage.

7. The method according to claim 1, characterized in that The matching pursuit algorithm is used to separate the key components of the filtered signal layer by layer, including: Set the initial residual to the filtered signal; In each iteration, an atom is selected from a predefined atom library so that the atom has the highest matching degree with the current residual under a preset metric condition; Calculate a coefficient between the atom and the current residual; The next residual is obtained according to the difference between the current residual and the coefficient multiplied by the atom.

8. The method according to claim 1, characterized in that The TLS-ESPRIT algorithm is used to estimate the parameters of the key component, including: Taking the several key components as input, a Hankel matrix is ​​constructed; The constructed Hankel matrix is ​​subjected to eigenvalue decomposition using the total least squares method; From the obtained eigenvalues, the frequency components are extracted.

9. The method according to claim 1, characterized in that: Perform multimodal identification on the reconstructed signal and extract the modal components, including: Calculating an autocorrelation function of a reconstructed signal and setting a threshold of the autocorrelation function to distinguish a modal component of the reconstructed signal from noise; Different modal features are extracted and classified from the identified modal components, including frequency, amplitude and phase.

10. The method according to claim 1, characterized in that Determine whether there is a broadband oscillation phenomenon, including: Determine whether there is a frequency band exceeding a preset frequency range in the frequencies of each modal component, and if there is a frequency band exceeding the preset frequency range, identify that broadband oscillation exists; and / or Calculate the power spectrum density of each modal component to determine the energy distribution; if there is an abnormal increase in the energy corresponding to the frequency band, it is identified as the presence of broadband oscillation; Observe the change of the frequency of each modal component over time. If there is a rapid fluctuation or irregular change in frequency, it is identified as the presence of broadband oscillation.

Images