Real-time detection method for subsynchronous oscillation of wind power plant

Abstract

The invention discloses a real-time detection method suitable for subsynchronous oscillation of a wind power plant. The method comprises the following steps: firstly, continuously acquiring grid-connected line current by adopting a sliding time window, then performing noise reduction processing on acquired data by using a singular spectrum analysis method, then adaptively determining a VMD modal number according to a grouping effective order solved in an SSA process, and performing VMD decomposition on a signal after noise reduction to obtain an effective modal; and then applying a symmetric differential energy operator to the related mode to solve instantaneous frequency and amplitude, and further solving to obtain system oscillation parameters. And finally, performing a simulation test based on the grid-connected data of the direct-driven wind power plant. The subsynchronous oscillation real-time detection method has high time-frequency measurement precision and good dynamic performance, can accurately and continuously reflect the system operation state change process, and has certain guiding significance for system stability evaluation, oscillation analysis and suppression research and disturbance source positioning research.

Description

Technical Field

[0001] The present invention relates to the field of power system stability analysis, and in particular to a method for real-time detection of sub-synchronous oscillations in wind farms. Background Art

[0002] In recent years, a high proportion of new energy and power electronic equipment has gradually become an important trend and key feature of the development of power systems. With the long-distance connection of large-capacity wind power, the stability problems caused by it have become more and more serious. Among them, the dynamic interaction between fast-responding power electronic equipment and the power grid will produce subsynchronous oscillation problems, which may affect the stable operation of the power grid and equipment. In addition, the oscillation signal presents multi-modal, nonlinear, and time-varying characteristics, making it increasingly difficult to identify.

[0003] At present, the subsynchronous oscillation parameter identification methods proposed by scholars can be summarized into two categories: signal model-based and data decomposition-based. Among them, the signal model-based method mainly establishes a detailed data model, and solves and identifies parameters through mathematical calculations or parameter estimation. Commonly used methods include eigenvalue analysis, Kalman filtering, and matrix bundle algorithm, but these methods have difficulties in model establishment and order determination, the identification accuracy cannot be guaranteed, and they cannot effectively reflect the oscillation dynamic characteristics of strong nonlinear power systems. With the large-scale use of wide-area measurement systems in dynamic monitoring of power systems, more and more algorithms are applied to oscillation signal parameter analysis, mainly including Fourier transform monitoring algorithms (fast Fourier transform, discrete Fourier transform, etc.), modal decomposition algorithms (empirical mode decomposition EMD, variational mode decomposition VMD, wavelet transform, etc.) and other monitoring methods (digital filter extraction of oscillation signals, random reduction technology solution, etc.). Among them, Fourier transform is prone to spectrum leakage, and it is impossible to obtain relevant information about the time when each frequency component appears. Empirical mode decomposition may cause modal aliasing, and wavelet transform is difficult to select basis functions. In order to improve the identification accuracy and calculation speed, many studies choose to fuse multiple algorithms, such as fusing EMD and Hilbert change HT to achieve instantaneous parameter solution of modal signals; fusing VMD and Prony analysis to improve Prony's noise resistance; fusing wavelet change and K-means clustering algorithm to achieve accurate frequency calculation, etc. Or introduce research methods in other disciplines, such as nonlinear modal decomposition and random reduction technology RDT commonly used in civil structure vibration analysis, to accurately decompose the signal, thereby providing a good data source for vibration parameter identification.

[0004] On the basis of the above research, the present invention proposes a real-time detection method for synchronous oscillation of wind farms. The continuously updated short-time window data is analyzed, and the singular spectrum analysis method is first used to realize signal noise reduction and adaptive parameter selection of modal decomposition. Then, the parameters of the obtained relevant effective modes are identified by combining the improved variational modal decomposition and the symmetric difference energy operator. The oscillation detection method proposed in the present invention has high time-frequency measurement accuracy and good dynamic performance, and can accurately and continuously reflect the change process of the system operation state. Summary of the invention

[0005] In order to realize the real-time identification of synchronous oscillation parameters of wind farms, this paper proposes an oscillation detection method that integrates noise reduction, modal decomposition and parameter identification.

[0006] The present invention is achieved by adopting the following technical solutions:

[0007] Firstly, a sliding time window is used to continuously acquire the grid-connected line current, and then the singular spectrum analysis method is used to perform denoising on the acquired data. Then, according to the effective order of the group obtained in the SSA process, the VMD mode number is adaptively determined, and the denoised signal is subjected to VMD decomposition to obtain the effective modes. Then, the symmetric difference energy operator is applied to the relevant modes to solve the instantaneous frequency and amplitude, and then the system oscillation parameters are solved.

[0008] As a preferred embodiment of the present invention, the scheme comprises the following steps:

[0009] Step A1: Collect relevant current, voltage or power signals in the power system as raw data, and analyze samples through window time series sliding;

[0010] Step A2: Using singular spectrum analysis method to perform noise reduction on the sampled data;

[0011] Step A3: Adaptively determine the number of decompositions according to the effective mode number obtained in the singular spectrum analysis grouping step, and perform VMD decomposition on the denoised signal;

[0012] Step A4: Use the three-point symmetric differential energy operator to calculate the instantaneous amplitude and frequency of the decomposed effective mode, obtain the relevant oscillation parameters, and then evaluate the system stability.

[0013] Furthermore, the step A1 comprises:

[0014] Step A11: Acquire the current data of the wind farm grid-connected line, set the sliding window length to 0.2s, the sliding step length to 0.02s, and the data sampling rate to the original data sampling rate, so as to achieve real-time acquisition of current data;

[0015] Step A12: To avoid errors in oscillation parameter identification caused by the port effect in VMD decomposition, a sub-window is selected for identification when using the energy operator to calculate the instantaneous parameters. The sub-window is a sliding window with the boundary indented by 0.01s each, and the sub-window changes continuously as the window slides.

[0016] Furthermore, the step A3 comprises:

[0017] Step A31: Determine the decomposition mode number M according to the SSA calculation result, and perform adaptive VMD decomposition on the denoised data;

[0018] Step A32: Set the correlation coefficient threshold to filter out valid modes, and its expression is:

[0019]

[0020] where u m (t) is the components decomposed by VMD, y i (t) is the original signal, R h The threshold δ was set to 1 / 10 of the maximum correlation coefficient obtained for each decomposition;

[0021] Step A33: For each IMF component obtained by decomposition, calculate the energy of its analysis sub-window in real time. When the energy of a component continues to increase in 3-4 consecutive time windows, it is necessary to focus on this mode. The energy formula is defined as follows:

[0022]

[0023] Among them, N is the number of samples in the corresponding analysis subwindow. In practical applications, in order to improve the running speed, the mode with continuously increasing energy can be directly analyzed.

[0024] Compared with the existing technology, this technology has the following beneficial effects:

[0025] The wind farm synchronous oscillation detection method provided by the present invention analyzes the continuously sliding wide time window data, and effectively improves the anti-noise ability and decomposition accuracy of VMD by combining SSA noise reduction and adaptive VMD decomposition. At the same time, the symmetric differential energy operator has a strong online application capability, and can quickly calculate and obtain the instantaneous parameters of the oscillation mode. The identification of the instantaneous parameters by the oscillation detection method can quickly and accurately reflect the system state transition and the SSO oscillation characteristic change process. The method does not require prior knowledge, is suitable for time-varying multi-modal signals, and can locate the system mutation point, intuitively and continuously reflect the system state change process, which is conducive to the further study of SSO characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solution and practical examples of the present invention, the used drawings are briefly introduced below.

[0027] Figure 1 This is a flow chart of the oscillation detection method proposed by the present invention;

[0028] Figure 2 It is a schematic diagram of window timing sliding;

[0029] Figure 3 This is the topological structure diagram of the direct-drive wind farm connected to the grid via series compensation;

[0030] Figure 4 The DC capacitor voltage waveform diagram and FFT analysis result diagram on the fan side;

[0031] Figure 5 It is a schematic diagram of line current waveform and window division;

[0032] Figure 6 It is the time spectrum of a window data after the wind speed changes;

[0033] Figure 7 It is the instantaneous amplitude-frequency diagram of the effective mode in a certain window after the wind speed changes. DETAILED DESCRIPTION

[0034] The present invention provides a method for real-time detection of subsynchronous oscillations in a wind farm. To make the purpose, technical solution and effect of the present invention clearer, the specific implementation scheme of the present invention is described in detail below with reference to the accompanying drawings and examples. The specific examples described in the present invention are only used to explain the present invention and are not used to limit the present invention.

[0035] 1. Introduction to the specific embodiments of the invention

[0036] Figure 1 This is a flow chart of the real-time detection method for wind farm sub-synchronous oscillation provided by the present invention. Figure 1 The present invention describes an oscillation detection method integrating noise reduction processing, modal decomposition and parameter identification, including:

[0037] Step A1: Collect relevant current, voltage or power signals in the power system as raw data, and analyze samples through window time series sliding;

[0038] Step A2: Using singular spectrum analysis method to perform noise reduction on the sampled data;

[0039] Step A3: According to the effective mode number obtained in the singular spectrum analysis grouping step, the number of decompositions is adaptively determined, and VMD decomposition is performed on the denoised signal;

[0040] Step A4: Use the three-point symmetric differential energy operator to calculate the instantaneous amplitude and frequency of the decomposed effective mode, solve the relevant oscillation parameters, and then evaluate the system stability.

[0041] Figure 2 The following is a schematic diagram of window timing sliding. In step A1, the sliding window is set to continuously obtain analysis data, including:

[0042] Step A11: Obtain the current data of the wind farm grid-connected line. It is generally believed that the sliding window contains at least two oscillation cycles corresponding to the lowest frequency. The present invention focuses on the components below 100 Hz. The window length is selected as 10 fundamental wave cycles. Therefore, the sliding window length is set to 0.2s, the sliding step is 0.02s, and the data sampling rate is the original data sampling rate, so as to realize the real-time acquisition of current data.

[0043] Step A12: Although mirror extension has been used to reduce the impact of endpoint effects in the following VMD decomposition process, there are still slight errors at the endpoints of the analyzed data. The resulting energy loss at the beginning and end will reduce the accuracy of the identification results. Therefore, it is necessary to continue to divide the sub-windows; when using the energy operator to calculate the instantaneous parameters, a sub-window is selected for identification. The sub-window is indented by 0.01s at the sliding window boundary, and the sub-window changes continuously as the window slides.

[0044] The step A2 performs SSA noise reduction analysis on the acquired data, including:

[0045] Step A21: Let the time series to be studied be [x1, x2, x3, …, x N ], where N is the sequence length. To achieve multi-dimensional mapping of one-dimensional sequences, the sliding window length L is set to lag the initial sequence, and the trajectory matrix is ​​obtained as follows:

[0046]

[0047] Wherein K=N-L+1, the matrix X is a L×N Hankel matrix, and the present invention selects L=N / 3 considering that the effective rank of the dominant signal matrix is ​​relatively low in practice.

[0048] Step A22: To effectively extract matrix features, perform singular value decomposition on the trajectory matrix X to obtain:

[0049] X L×K =U L×L S L×K V H K×K (4)

[0050] Where S = XX T is a diagonal matrix whose diagonal elements are non-negative and in descending order of singular values ​​σ i , 1≤i≤L,σ i Contains relevant characteristic information about the matrix X. At this time, X can be expressed as:

[0051]

[0052] Step A23: X can be divided into multiple non-intersecting subsets representing different signal components according to different signal characteristics. The signal can be extracted or processed by selecting a subset for reconstruction. This paper considers the characteristics of power grid signals and calculates singular value related parameters to assist in determining I. First, the singular value percentage is calculated:

[0053]

[0054] Where P is the number of selected p Reflects the change in the contribution of the selected order to the system, and then defines the order index R i :

[0055] R i =(G i+2 -G i+1 ) / (G i+1 -G i ) (7)

[0056] The increments of the true mode and the noise component are not of the same order of magnitude, so R i There will be large fluctuations near the real model order, that is, a minimum value. At this time, the next order at the lowest point of the curve is the real modal order Pr; the singular value corresponding to the noise signal is generally very small, and its increment is at the same order of magnitude, so R Pr and R Pr+1 Close to 1, the true model order can be determined based on the above two characteristics.

[0057] Step A24: Based on the grouping result of step A23, the singular vector of the useful signal can be reconstructed. Combining the time principal component and the time empirical orthogonal function, Y can be converted into a sequence yrc of length N by using diagonal averaging as follows:

[0058]

[0059] In summary, the reconstruction of the time series after denoising can be achieved.

[0060] The step A3 is to perform adaptive VMD decomposition and analysis mode selection on the noise-reduced signal, including:

[0061] Step A31: Since the parameters of the oscillation mode appear in pairs, the noise signal is considered to be separated, so the number of VMD decompositions M = [P / 2] + 1 is taken. VMD decomposition is to obtain the unconstrained augmented Lagrangian function shown in formula (9):

[0062]

[0063] In the formula, u m is the modal component to be solved, ω mis the instantaneous angular frequency of the corresponding component, λ is the Lagrange multiplier; δ(t) is the unit pulse function, * represents convolution, and X(t) is the decomposed original signal. The multiplier alternating direction algorithm is used to solve equation (9) through cyclic iteration. The calculation formulas for updating each parameter are as follows:

[0064]

[0065] in, represents the IMF component and Lagrange multiplier of the frequency domain state, and τ is the noise tolerance parameter. VMD decomposition can be achieved.

[0066] Step A32: The modes decomposed by VMD may contain false components. Set the correlation coefficient threshold to screen the valid modes. The expression is:

[0067]

[0068] Among them, u m (t) is the components decomposed by VMD, y i (t) is the original signal, R h The threshold δ is set to 1 / 10 of the maximum correlation coefficient obtained in each decomposition, which can achieve effective mode selection.

[0069] Step A33: For each IMF component obtained by decomposition, calculate the energy of its analysis sub-window in real time. When the energy of a component continues to increase in 3-4 consecutive time windows, it is necessary to focus on this mode. The energy formula is defined as follows:

[0070]

[0071] Among them, N is the number of samples in the corresponding analysis subwindow. In practical applications, in order to improve the running speed, the mode with continuously increasing energy can be directly analyzed.

[0072] The step A4 performs parameter identification on the effective modes obtained by VMD decomposition, including:

[0073] Step A41: According to the formula of the continuous signal energy operator, the energy operator in the discrete domain is derived as follows:

[0074] ψ[x(n)]=x 2 (n)-x(n-1)x(n+1) (13)

[0075] Then, we take ψ(x(n)) as input and apply the transfer function H(z)=z(1+z -1 +z -2 ) is solved, and the output is the energy operator after three-point symmetric difference as follows:

[0076]

[0077] Then, the instantaneous amplitude A of x(n) can be quickly obtained according to formula (15): m With instantaneous frequency F m .

[0078]

[0079] The damping factor of the mode can be obtained by fitting the instantaneous envelope with an exponential function, and then the damping ratio can be obtained according to the following formula:

[0080]

[0081] Combining the above solutions can realize the rapid identification of instantaneous parameters of multiple oscillation components.

[0082] Step A42: Evaluate the system stability based on the solved oscillation parameters. When the amplitude of the obtained oscillation mode exceeds a certain threshold (the threshold is generally determined by experiment), it indicates that the system may have a local oscillation mode in a weakly damped state, and the operation and dispatching personnel need to take relevant measures to avoid the global occurrence and spread of oscillation. Once negative damping is present, no matter how small the oscillation amplitude is, emergency control measures should be taken to avoid further instability.

[0083] 2. Technical feasibility verification of the present invention

[0084] Figure 3 The topological structure diagram of the direct-drive wind farm connected to the grid through series compensation used for simulation verification is shown in the figure. Figure 4 The FFT analysis result of the DC capacitor voltage waveform of the wind turbine system is shown in the figure. The integral coefficient of the outer loop of the d-axis voltage control of the grid-side converter is changed at 3s to stimulate oscillation, and the wind speed is changed at 4s to observe the effectiveness of the method of the present invention. The simulation results are shown in the figure. Figure 5 , Figure 6 and Figure 7 As shown, the effectiveness of the present invention is verified.

[0085] Figure 4 In order to change the grid-side controller parameters in 3 seconds to stimulate oscillation, FFT analysis is performed on the DC voltage signal on the wind turbine side. It can be seen that the system is limited by the amplitude limit link and presents a 16Hz constant amplitude oscillation after rapid oscillation. And Figure 4 a) The oscillation center of the current waveform after the oscillation occurs does not coincide with the steady-state operating value. This is because the electrical quantity contains a certain DC component. In the initial stage of the oscillation, as the input amplitude increases, it will first touch the upper limit of the limit, resulting in the asymmetric saturation of the limit link. The adverse effects of equal-amplitude oscillations accumulate linearly over time, so it is necessary to quickly identify and analyze the oscillation parameters to provide auxiliary information for oscillation suppression.

[0086] Figure 5The current waveform of the grid-connected line obtained under the same working conditions as above indicates the sliding direction and size of the window, and gives the oscillation decomposition at the relevant time. The sliding time window analysis of the data is set to start from 2.8s to observe the time-varying characteristics of the signal. SSO mode begins to appear in window 2 (2.82-3.02s), and the power frequency component decreases slightly at this time; the modal energy operator envelope calculated in window 3 (2.84-3.04s) fluctuates around 3s, which shows that this moment is the mutation point of the system state. Data analysis of window 4 (2.86-3.06s) shows that there are coupled sub / super synchronous components in the system, and the super synchronous components are large and change quickly. For super synchronous components, the system often has positive damping characteristics and generally does not cause oscillation, while sub synchronous components are prone to form negative damping, thereby causing SSO. Therefore, in engineering, generally only the sub synchronous frequency component is concerned to determine whether the system has SSO. Combined with the results of the eigenvalue analysis in Section 4.1, it can be seen that there is a super-synchronous component with positive damping of 84.34Hz that is complementary to 16.72Hz. Window 11 (3.00s-3.20s) calculates that the system adds a super-synchronous oscillation mode of about 80Hz, which begins to appear around 3.15s and slowly increases. According to the frequency coupling principle, there should be an SSO component of about 20Hz, but because its value is very small, it is not detected at this time. The amplitude of this component increases to a level that can be identified when the time window slides later. Window 12 (3.02-3.22s) is basically the amplification process of the sub / super-synchronous oscillation mode of interest, and the calculated oscillation parameters are the most accurate. The results are shown in Table 1.

[0087] Table 1 Window 12 recognition results

[0088]

[0089] According to the frequency coupling principle, the obtained 33.1857Hz is coupled with the 16Hz voltage oscillation frequency on the wind turbine side, which verifies the accuracy of the algorithm in this paper. It is also easy to know that the relevant parameters are constantly changing during the oscillation of the wind power system. In order to obtain more accurate identification results, appropriate delay is necessary and acceptable. The decomposition mode number K=5 is obtained from the data of window 15 (3.08-3.28s), and the threshold of the mutual correlation coefficient is set to 0.07, and the actual number of modes to be analyzed is determined to be 3. The oscillation change speed between 3.22-3.44s decreases rapidly, and then the line current is affected by the limiting link in the system and tends to oscillate with equal amplitude, and the oscillation parameters remain basically unchanged.

[0090] Figure 6In order to change the wind speed to 10m / s at 4s, the modal decomposition of the system is calculated in the 3.82-4.02s time window. It can be seen that there are 5 oscillation modes in the system at this time. The improved VMD decomposition has a bandpass filtering property similar to the wavelet transform, and can finely divide the signal according to the signal feature analysis results, that is, it has a strong parameter identification ability for complex multi-modal oscillation signals. Subsequently, the instantaneous envelope and frequency curves calculated from the 3.84-4.04s time window data both have mutation points around 4.02s, indicating that the system state changes at this time. As the time window slides, the number of modes obtained by VMD decomposition decreases, the small-amplitude SSO oscillation mode gradually disappears, and the instantaneous parameters of the sub / supersynchronous oscillation mode continue to change.

[0091] Figure 7 The figure is the instantaneous amplitude-frequency diagram of the relevant modes obtained by solving the 4.02-4.22s time window data after the wind speed changes. At this time, the dominant modes of the system are only the power frequency component and the supersynchronous component, and the amplitude and frequency fluctuate greatly during this period. It is easy to know that the system is in a transitional state at this stage. According to the IMF1 component being basically stable at 50Hz around 4.2s, combined with the fact that the instantaneous parameters calculated by the system in the next time window have a sudden change at 4.18-4.2s, it is inferred that the system gradually enters the next "steady state" at this time. Therefore, the identification of the instantaneous parameters by this method can intuitively, quickly and accurately reflect the changes in the system state and the changes in the SSO oscillation characteristics, and can adaptively screen the modes during the sliding solution process. In the early stage of oscillation, the non-dominant extremely small amplitude oscillation information is screened out in the denoising process, which reduces the computational complexity of VMD decomposition and SDEO solution while ensuring the accuracy of identification, which is conducive to the real-time online operation of the identification method.

[0092] The above simulation results show that the subsynchronous oscillation real-time detection method of the present invention can continuously and accurately obtain the system oscillation time-varying parameters. The above simulation results verify the effectiveness and feasibility of the identification method of the present invention.

[0093] In summary, the real-time detection method for wind farm synchronous oscillation described in the present invention does not require prior knowledge, is suitable for time-varying multimodal signal parameter identification, and can locate system mutation points, intuitively and continuously reflect the system state change process, which is conducive to further research on SSO characteristics.

[0094] Finally, it should be noted that the above examples of the present invention are merely examples for illustrating the present invention, and are not intended to limit the embodiments of the present invention. Although the applicant has described the present invention in detail with reference to the preferred embodiments, for those of ordinary skill in the art, other different forms of changes and modifications can be made based on the above description. It is impossible to list all the embodiments here. Any obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for real-time detection of synchronous oscillation of wind farms, characterized in that: The following steps are involved: S1: Collect relevant current, voltage or power signals in the power system as raw data and analyze samples through window time series sliding; S2: Use singular spectrum analysis method to reduce noise of sampled data; S3: According to the effective mode number obtained in the singular spectrum analysis grouping stage, the number of decompositions is adaptively determined, and VMD decomposition is performed on the denoised signal; S4: The instantaneous amplitude and frequency of the decomposed effective mode are calculated using the three-point symmetric differential energy operator to obtain the relevant oscillation parameters and then evaluate the system stability.

2. The method according to claim 1, characterized in that The step S1 comprises: S11: Acquire the current data of the wind farm grid-connected line, set the sliding window length to 0.2s, the sliding step to 0.02s, and the data sampling rate to the original data sampling rate, so as to achieve real-time acquisition of current data; S12: To avoid the error in oscillation parameter identification caused by the port effect in VMD decomposition, a sub-window is selected for identification when using the energy operator to calculate the instantaneous parameters. The sub-window is a sliding window with the boundary indented by 0.01s each, and the sub-window changes continuously as the window slides.

3. The method according to claim 1, characterized in that The step S3 comprises: S31: Determine the decomposition mode number M according to the SSA calculation result, and perform adaptive VMD decomposition on the denoised data; S32: Set the correlation coefficient threshold to filter out valid modes, and its expression is: where u m (t) is the components decomposed by VMD, y i (t) is the original signal, R h The threshold δ was set to 1 / 10 of the maximum correlation coefficient obtained for each decomposition; S33: For each IMF component obtained by decomposition, the energy of its analysis sub-window is calculated in real time. When the energy of a component continues to increase in 3-4 consecutive time windows, it is necessary to focus on this mode. The energy formula is defined as follows: Among them, N is the number of samples in the corresponding analysis subwindow. In practical applications, in order to improve the running speed, the mode with continuously increasing energy can be directly analyzed.

Images