Wideband oscillation identification method for accessing new energy cluster to DC power grid

Abstract

The invention discloses a broadband oscillation identification method for a new energy cluster to access a DC power grid. Real-time current and voltage signals are collected from a synchronous vector measurement unit (PMU) of the power grid and then are subjected to data preprocessing; carrying out compressed sampling on the preprocessed signal through a CS technology; reconstructing the compressed low-dimensional data by adopting an SP algorithm; then performing mixed empirical mode decomposition (EEMD) on the recovery signal to obtain a group of intrinsic mode functions (IMFs) of the oscillation signal; redundant IMF components are removed by calculating each IMF energy coefficient; analyzing the IMF component containing the oscillation modal information by using an optimization Prony algorithm to obtain broadband oscillation information; and finally, predicting the dynamic change of the signal by using an optimized AFEKF algorithm, and effectively identifying the overall trend and change rule of the broadband oscillation by comparing the deviation between a predicted value and an actual observed value, thereby realizing accurate identification and early warning of the broadband oscillation in the power system. According to the invention, powerful support is provided for stable operation and optimal control of the new energy access DC power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of power system stability, and in particular to a method for identifying wide-band oscillations of a new energy cluster connected to a direct current grid. Background Art

[0002] The high proportion of access to new energy and power electronic equipment has led to changes in the pattern of the power system, and the degree of power electronics in the transmission system of large-scale diversified new energy clusters is gradually increasing. The weak inertia support and output volatility and uncertainty of new energy generation, the regional coupling of converters and AC / DC interconnection, and the wide access of power electronic loads have profoundly changed the dynamic behavior of the power grid. In the process of the power system developing towards power electronics, new oscillation problems in the power system have attracted widespread attention. After the power system is disturbed, multiple types of power electronic equipment and the power grid jointly participate in the periodic fluctuation of electrical quantities over time. The oscillation phenomenon with a wide oscillation frequency and frequency drift characteristics is considered to be a broadband oscillation. In recent years, due to the frequent occurrence of broadband oscillation accidents caused by power electronic equipment, broadband oscillation problems caused by new energy and power electronic equipment have occurred in countries around the world. With the rapid development of renewable energy such as wind power and solar energy, these new energy power generation usually uses DC output, which is converted into AC power by inverters and then connected to the power grid. Direct use of DC power grids can simplify this process and improve energy efficiency. DC power grids have many advantages over AC power grids, such as longer transmission distances, lower losses, and easy multi-terminal interconnection. However, the control of DC grid is more complicated, especially when new energy clusters are connected. Power electronic equipment represented by converters and inverters will be widely used. The multi-scale cascade and sequential switching characteristics of power electronic equipment will cause fundamental changes in the physical coupling mechanism of traditional power systems. Traditional electromagnetic equipment and power electronic equipment and their control systems have nonlinear strong coupling, which makes the dynamic behavior of power systems more complex and changeable.

[0003] It can be seen that the broadband oscillation problem of large-scale diversified new energy cluster transmission systems is a complex oscillation problem involving multiple types of power electronic equipment and multiple regions, involving frequency coupling and multi-mode aliasing, which seriously affects the safe and stable operation of the power grid. At present, there are few analyses of broadband oscillations of large-scale diversified new energy cluster transmission systems connected to the power grid, and the mechanism is not clear enough. It is urgent to carry out broadband analysis of large-scale diversified new energy cluster transmission systems connected to the power grid to reveal the broadband oscillation mechanism of large-scale diversified new energy cluster transmission systems under the power grid. The broadband oscillation problem is an important part of the stability problem of the power system, and the accurate detection and identification of the oscillation problem can obtain information such as the frequency, phase, amplitude, etc. of the oscillation, extract the characteristics of the oscillation, and provide data reference for accident tracing and oscillation suppression. Summary of the invention

[0004] The present invention proposes a broadband oscillation identification method for a new energy cluster connected to a DC power grid, which improves the identification accuracy and robustness of broadband oscillations and provides strong support for the stable operation and optimized control of the new energy connected to the DC power grid.

[0005] The technical solution adopted by the present invention is:

[0006] A method for identifying broadband oscillations when a new energy cluster is connected to a DC power grid is characterized in that real-time current and voltage signals are collected from a synchronous vector measurement device (PMU) in the power grid and data preprocessing is performed; the preprocessed signals are compressed and sampled by using a compressed sensing (CS) technology; after receiving low-dimensional oscillation data at a master station, the compressed low-dimensional data is reconstructed by using a subspace tracking (SP) algorithm; then the recovered signal is decomposed by a hybrid empirical mode decomposition (EEMD) to obtain a group of intrinsic mode functions (IMFs) of the oscillation signal; redundant IMF components are removed by calculating the energy coefficients of each IMF; the IMF components containing oscillation mode information are analyzed by using an optimized Prony algorithm to obtain broadband oscillation information, including frequency and amplitude information; finally, the dynamic changes of the signal are predicted by using an optimized adaptive filtering integrated Kalman filtering (AFEKF) algorithm, and the overall trend and change law of the broadband oscillation can be effectively identified by comparing the deviation between the predicted value and the actual observed value, thereby realizing accurate identification and early warning of broadband oscillations in the power system.

[0007] Further:

[0008] The measurement matrix A in the compressed sensing CS technology adopts a Gaussian matrix.

[0009] The hybrid empirical mode decomposition EEMD is used to decompose the reconstructed signal obtained by SP, and the intrinsic mode function IMF component containing broadband oscillation information is screened by using the energy coefficient: Gaussian white noise with uniform frequency distribution is added before the empirical mode decomposition EMD decomposition signal, so that the signals of different scales in the noisy signal are distributed on a reference scale that matches it, and the zero mean characteristic of white noise is used to cancel each other out after multiple averaging, and the average value of each group of IMF components is used as the final decomposition result; the structure of the hybrid empirical mode decomposition EEMD algorithm is as follows:

[0010] (1) Add a set of random Gaussian white noise sequences n(t) with a superposition mean of 0 and a constant standard deviation to the initial signal S(t) to obtain a new signal X(t), that is:

[0011] X(t)=S(t)+n(t) (15)

[0012] Mathematically, white noise is expressed as:

[0013] n(t)~N(0,σ 2) (16)

[0014] Where n(t) represents the noise signal at time t, Ν(0,σ 2 ) represents zero mean and variance σ 2 is a Gaussian distribution;

[0015] (2) Perform EMD decomposition on each X(t) to obtain the M-order intrinsic mode function IMF component, denoted as C i (t), where i = 1, 2, ..., M, and the remaining component or residual signal is denoted by r k (t), the expression is as follows:

[0016]

[0017] (3) Different Gaussian white noise n j (t) is added to the original signal X(t), and steps (1) and (2) are repeated N times, that is:

[0018]

[0019] In the formula, C ij (t) represents the i-th IMF component obtained by decomposing the j-th superposition of white noise;

[0020] (4) In order to eliminate the influence of noise, the IMFs of the same order are averaged to obtain the final IMF. The averaged IMF components of each order can be expressed as:

[0021]

[0022] In the formula, represents the jth IMF component after EEMD decomposition;

[0023] (5) The signal X(t) can be reconstructed through all the final IMFs:

[0024]

[0025] in is the remaining amount.

[0026] The adaptive filtering integrated Kalman filtering AFEKF algorithm is used to predict the oscillation information at the next moment according to the amplitude and frequency at the current moment, so as to determine the overall trend of the broadband oscillation.

[0027] The present invention combines compressed sensing (CS), subspace pursuit (SP), hybrid empirical mode decomposition (EEMD) and optimized Prony-AFEKF algorithm to give full play to their respective advantages. First, a Gaussian random sampling matrix is ​​selected to compress broadband oscillation data, overcoming the problems of Nyquist theorem, insufficient data sampling and insufficient data communication bandwidth transmission; secondly, after the main station receives the low-dimensional oscillation data, the subspace pursuit (SP) algorithm is used to reconstruct the compressed low-dimensional data so that the original signal is largely restored; finally, the restored signal is decomposed by EEMD to obtain a set of intrinsic mode functions (IMFs) of the oscillation signal. The energy coefficients of each IMF are calculated and the redundant IMF components are removed. The optimized Prony algorithm analyzes the IMF components containing the oscillation mode information to obtain broadband oscillation information. The present invention improves the identification accuracy and robustness of broadband oscillations by comprehensively using a variety of advanced signal processing and model fusion technologies, providing strong support for the stable operation and optimized control of new energy access to the DC power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of the present invention;

[0029] Figure 2 It is the topology diagram of the new energy cluster delivery system;

[0030] Figure 3 It is the flow chart of SP algorithm of the present invention;

[0031] Figure 4 The flowchart of broadband oscillation modal parameter identification of the Prony algorithm optimized by the present invention;

[0032] Figure 5 AFEKF algorithm flow chart of the present invention;

[0033] Figure 6 It is a model for connecting wind turbines to the grid;

[0034] Figure 7 Comparison between the original broadband oscillation signal and the reconstructed signal of the present invention;

[0035] Figure 8 is the EEMD decomposition result of the present invention;

[0036] Fig. 9 The Prony algorithm results are optimized for the present invention. DETAILED DESCRIPTION

[0037] The present invention is further described in detail below with reference to the accompanying drawings, specific derivations, and simulation experiments.

[0038] Reference Figure 1, a broadband oscillation identification method for a new energy cluster connected to a DC power grid, collects real-time current and voltage signals from a synchronous vector measurement device PMU in the power grid and performs data preprocessing; compresses and samples the preprocessed signals using the compressed sensing CS technology; after receiving the low-dimensional oscillation data at the master station, the compressed low-dimensional data is reconstructed using the subspace tracking SP algorithm; then the recovered signal is decomposed using the hybrid empirical mode decomposition EEMD to obtain a set of intrinsic mode functions IMFs of the oscillation signal; by calculating the energy coefficient of each IMF, the redundant IMF components are removed; the IMF components containing the oscillation mode information are analyzed using the optimized Prony algorithm to obtain broadband oscillation information, including frequency and amplitude information; finally, the optimized adaptive filtering integrated Kalman filtering AFEKF algorithm is used to predict the dynamic changes of the signal, and by comparing the deviation between the predicted value and the actual observed value, the overall trend and change law of the broadband oscillation can be effectively identified, thereby realizing accurate identification and early warning of broadband oscillation in the power system.

[0039] The present invention will be described in detail below.

[0040] 1. Information Collection

[0041] In the power grid monitoring system of the present invention, synchronous phasor measurement units (PMUs) are installed on the converter and the DC bus. These units are connected to the power grid via voltage transformers (PTs) and current transformers (CTs) to collect electrical signals from the power grid in real time, thereby obtaining core data including the amplitude, phase, and frequency of voltage and current. The collected signals are converted into digital signals through analog-to-digital conversion (ADC) processing, and the time synchronization accuracy of microseconds is achieved with the help of the built-in GPS synchronization system. This ensures that PMU data from different locations can maintain a high degree of consistency, providing a solid foundation for realizing high-precision and time-space consistency identification of broadband oscillation characteristics.

[0042] PMU has a high sampling rate of up to several thousand Hertz, which can capture the wide-band oscillation characteristics from subsynchronous to supersynchronous range, effectively supporting the real-time monitoring of various oscillation modes within the system. Its synchronous acquisition capability ensures that the oscillation signal in the power grid will not be lost due to the limitation of sampling bandwidth, thereby realizing the rapid and accurate identification of wide-band oscillation characteristics.

[0043] 2. Compressed Sensing (CS)

[0044] Compressed sensing theory is used to achieve data compression during the sampling process. Compressed sensing theory can sample at a rate much lower than the Nyquist sampling rate, under the premise that the signal is sparse or compressible, thereby greatly reducing the amount of data. The compression and sampling of broadband oscillation signals mainly involve the selection of sparse transformation bases and the calculation of the perception matrix.

[0045] For a sparse signal x∈R N and the basis space Ψ∈R N×N , because in R N Any signal x in the space can be obtained using the basis vector Ψ i Therefore, any signal x can be expressed in a sparse form:

[0046]

[0047] Where: θ is a sparse matrix with only M<<N non-zero elements. When the signal x is sparse, a measurement matrix A with M<<N is constructed to sample x, thereby obtaining the measurement dimension value y, that is:

[0048] y=Ax (2)

[0049] Calculate the perception matrix and substitute equation (1) into equation (2) to obtain the perception matrix Θ.

[0050] y=AΨθ=Θθ (3)

[0051] According to formula (3), the perception matrix Θ is determined by A and Ψ. The frequency domain sparsity of the wide-band oscillation signal exists at the oscillation point, so it has a certain sparsity in the frequency domain. The discrete data points are transformed into the frequency domain using discrete cosine transform (DCT). DCT is expressed as:

[0052]

[0053] In the process of reconstructing the compressed signal y, a Gaussian random matrix independent of the sparse matrix θ is selected as the measurement matrix A, and the measurement matrix A obeys the Gaussian normal distribution.

[0054] 3. Signal reconstruction based on subspace pursuit (SP) algorithm

[0055] The SP algorithm is used to reconstruct the compressed data obtained by the CS algorithm. The SP algorithm adopts a backward tracking strategy. In each iteration, the wrong atoms in the previous selection are removed and replaced with the newly identified correct atoms, so as to gradually approach the true support set. The specific steps of the SP algorithm are as follows:

[0056] (1) Input: compressed signal y, perception matrix Θ; Initialization: residual r0 = y, candidate vector C0 = φ, index set Π0 = φ, iteration number k = 1.

[0057] (2) Calculate the candidate vector: C k =Θ * r k-1 .

[0058] (3) Find candidate vector C kThe position labels of the N largest elements in the update index set Π i =sup(C k-1 | N )∪H, where H=sup(C k | N ).

[0059] (4) Update the estimated value:

[0060] (5) Select the best N sparse elements, And update the residual r k =y-θx k .

[0061] (6) If || r k-1 ||2<||r k ||2, or k<N, then The loop ends; otherwise k=k+1, and returns to step (3).

[0062] (7) Output: Estimated value Residual r.

[0063] The SP algorithm flow chart is as follows Figure 3 shown.

[0064] 4. Empirical Mode Decomposition (EMD)

[0065] The EMD method is to extract the intrinsic mode function (IMF) from the original signal. The characteristics of IMF are: ① the number of its extreme points and zero crossing points should be equal or at most 1 different; ② the mean of the two envelopes formed by connecting its local maximum and local minimum is zero at any point.

[0066] Suppose the original signal S(t), determine to find the positions of all local maxima and local minima on S(t), and after finding them, interpolate all the maximum points and all the minimum points through the interpolation function to form the two upper and lower envelopes of the given signal S(t) (the upper envelope is generally formed by the interpolation of the maximum value, and the lower envelope is generally formed by the interpolation of the minimum value). After obtaining the upper and lower envelopes, calculate the average value of these two envelopes l1 and l1′, recorded as M1(t), and then subtract the average value M1(t) obtained by the upper and lower envelopes from the given signal S(t) to obtain a new set of data recorded as F1 to calculate the average value,

[0067]

[0068] F1=S(t)-M1(t) (6)

[0069] After obtaining F1, observe F1. If it meets the two assumptions mentioned above to determine whether it is an IMF, there is no need to perform envelope averaging and other steps. Even if it can be used as an IMF separated from S(t), in most cases, the decomposed F1 does not meet these two assumptions. At this time, it is necessary to calculate the average M2(t) of the upper envelope l1 and the lower envelope l1′ of F1, and then subtract M2(t) from F1. The formula is as follows:

[0070]

[0071] F2=F1-M2(t) (8)

[0072] …

[0073]

[0074] F k =F k-1 -M k (t) (10)

[0075] Repeat the above process, and finally when the kth decomposition fails, a stable F will be obtained. k , and meets the two assumptions of the IMF, this F k This is the first intrinsic mode function IMF1 decomposed from S(t). Its period is the shortest among all the subsequent IMFs decomposed, and correspondingly, it is also the highest frequency among all IMFs. After the first IMF1 is separated, the original signal S(t) is subtracted from IMF1, and the formula is as follows:

[0076] r1=S(t)-IMF1 (11)

[0077] r2=r1-IMF2 (12)

[0078] …

[0079] r k =r k -IMF k (13)

[0080] In this way, we can get a new data sequence r1 without IMF1. We can repeat the process of obtaining IMF1 for this new data sequence r1, and then we can get the second intrinsic mode function. Then we can subtract IMF2 from r1 to get another new data sequence r2. This process is repeated until the last r k Unable to decompose further, k It is the residual data left after the original signal S(t) undergoes empirical mode decomposition, called the residual and recorded as R(t).

[0081] Therefore, the original signal can be reconstructed using all separated IMFs and the residual R(t):

[0082]

[0083] 5. Hybrid Empirical Mode Decomposition (EEMD)

[0084] EEMD is used to decompose the reconstructed signal obtained by SP and use the energy coefficient to filter the IMF components containing broadband oscillation information.

[0085] In order to avoid the modal aliasing effect during EMD decomposition, which may cause errors in subsequent signal analysis, Gaussian white noise with uniform frequency distribution is added before the EMD decomposition signal, so that the signals of different scales in the noisy signal are distributed on a reference scale that matches it, and the zero mean characteristic of white noise is used to cancel each other out after multiple averaging, thereby effectively suppressing the modal aliasing phenomenon and ensuring the continuity of each IMF component in the time domain after EMD decomposition, and taking the average value of each group of IMF components as the final decomposition result. The EEMD algorithm structure is as follows:

[0086] (1) Add a set of random Gaussian white noise sequences n(t) with a superposition mean of 0 and a constant standard deviation to the initial signal S(t) to obtain a new signal X(t), that is:

[0087] X(t)=S(t)+n(t) (15)

[0088] Mathematically, white noise can be expressed as:

[0089] n(t)~N(0,σ 2 ) (16)

[0090] Where n(t) represents the noise signal at time t, Ν(0,σ 2 ) represents zero mean and variance σ 2 is a Gaussian distribution.

[0091] (2) Perform EMD decomposition on each X(t) to obtain the M-order intrinsic mode function IMF component, denoted as C i (t), where i = 1, 2, ..., M, and the remaining component or residual signal is denoted by r k (t), the expression is as follows:

[0092]

[0093] (3) Different Gaussian white noise n j (t) is added to the original signal X(t), and steps (1) and (2) are repeated N times, that is:

[0094]

[0095] In the formula, C ij (t) represents the i-th IMF component obtained by decomposing the j-th superposition of white noise.

[0096] (4) In order to eliminate the influence of noise, the IMFs of the same order are averaged to obtain the final IMF. The averaged IMF components of each order can be expressed as:

[0097]

[0098] In the formula, represents the jth IMF component after EEMD decomposition.

[0099] (5) The signal X(t) can be reconstructed through all the final IMFs:

[0100]

[0101] in is the remaining amount.

[0102] 6. IMF Energy Distribution Screening

[0103] The energy of the IMF component reflects the importance of the component in the signal. Usually, stronger high-frequency oscillation signals are concentrated in a few IMFs, and the energy of these IMFs is relatively large. The energy calculation formula is as follows:

[0104]

[0105] Of which: IMF k (n) is the value of the kth IMF component at the nth moment; E k is the total energy of the kth IMF component.

[0106] Energy ratio, calculate the ratio of each IMF component relative to the total energy of the original signal to determine its contribution to the overall signal:

[0107]

[0108] Where: E total is the total energy of the original signal; P k is the energy proportion of the kth IMF component.

[0109] By calculating the correlation coefficient between the IMF component and the original signal, it is determined whether the IMF contains the main characteristics of the oscillation signal. The energy coefficient calculation formula is as follows:

[0110]

[0111] in:

[0112] s(n) is the value of the original signal at the nth moment, IMF k (n) is the value of the kth IMF component;

[0113] and are the average values ​​of the original signal and the kth IMF respectively;

[0114] ρ k The value range of is [-1,1]. The larger the absolute value, the stronger the correlation.

[0115] 7. Optimize Prony algorithm

[0116] The optimized Prony algorithm is used to identify broadband oscillation information using IMF components. The mathematical model used by the optimized Prony method is a set of P exponential functions with arbitrary amplitudes, frequencies, and attenuation factors. The input signal ((x0), (x1),…, (x N-1 )) comes from the dominant oscillation mode IMF, and in Prony analysis, its estimated value can be considered as a discrete time function form:

[0117]

[0118] Where: A i is the amplitude; θ i is the phase, f i is the frequency; α i is the attenuation factor; Δt is the sampling time interval.

[0119] The key to the Prony algorithm is to realize that the fit of equation (24) is a homogeneous solution to a constant linear difference equation. Through a series of mathematical transformations, the difference equation can be derived as:

[0120]

[0121] For the parameter α in the equation i Do the least squares estimation to minimize the sum of square errors, and then get a set of linear matrix equations:

[0122]

[0123] definition The normal equation of Prony algorithm can be obtained:

[0124]

[0125] Solving this equation, we can get the coefficients α1, α2, …, α p The estimated value of is obtained by the following formula to find the root (Prony pole) of the polynomial:

[0126] 1+a1z -1 +…+a p z -p =0 (28)

[0127] According to the recursive formulas (29) and (30), the parameters b1, b2, …, b p :

[0128]

[0129]

[0130] So the amplitude A i , phase θ i , frequency f i and the attenuation factor α i It can be obtained by the following formula:

[0131]

[0132] 8.AFEKF algorithm

[0133] The AFEKF algorithm is an advanced filtering algorithm that combines fuzzy logic and extended Kalman filtering. It adjusts the parameters of the extended Kalman filter, such as the process noise covariance matrix and the observation noise covariance matrix, in real time through the fuzzy logic system. This adaptive mechanism enables the algorithm to better cope with the uncertainty of the system model and the changes in noise characteristics, and improves the robustness and adaptability of the filter. In addition, the Kalman filter is also a recursive algorithm that can estimate the state of a dynamic system. Its main goal is to obtain the optimal estimate of the system state based on current and past observation data.

[0134] The AFEKF algorithm is used to predict the oscillation information at the next moment based on the amplitude and frequency at the current moment, so as to determine the overall trend of broadband oscillation. The estimation accuracy of the traditional extended Kalman filter algorithm is relatively high, but it is difficult to use in the estimation process with more complex environments, and it will produce divergence due to the error of the mathematical model of the nonlinear system, thus affecting the estimation accuracy. In response to the problems of the conventional extended Kalman filter algorithm in filter estimation, the adaptive extended Kalman filter (AFEKF) algorithm introduces a fading factor based on the EKF algorithm to weaken the proportion of old observations in the prediction process, increase the weight of new observations in the prediction correction process, and thus suppress the filter divergence.

[0135] The AFEKF state space expression is:

[0136]

[0137] Where: x k is the state vector; u k is the input frequency, amplitude, etc. vector; f(xk ,u k ) is the state transfer function; h(x k ,u k ) is the observation function; w k 、v k are the state noise value and measurement noise value in the system, both of which satisfy the normal distribution. The AFEKF algorithm linearizes the model locally and converts f(x k ,u k ) and h(x k ,u k )exist Performing Taylor expansion at ignoring higher-order terms yields:

[0138]

[0139] At the same time, the parameters in the nonlinear system can be transformed into:

[0140]

[0141] Substituting formula (31) into formula (30), we can obtain:

[0142]

[0143] Initialize the state variables, that is

[0144]

[0145] Where E(·) is the update process function.

[0146] State variable initialization mainly includes two steps: time update and state update:

[0147] 1) Time update

[0148] The prior state expression is:

[0149]

[0150] In the formula, is the prior estimate of the system at time k+1, and the prior estimate covariance expression is:

[0151]

[0152] Where: is the prior estimated covariance of the system at time k+1; P k + is the posterior estimated covariance at time k.

[0153] 2) Status Update

[0154] The Kalman gain calculation expression is:

[0155]

[0156] The posterior state estimation expression is:

[0157]

[0158] The posterior covariance error is updated as:

[0159]

[0160] AFEKF algorithm flow is as follows Figure 5 shown.

[0161] The Prony algorithm extracts key information such as the amplitude and frequency of the signal, and then uses the optimized AFEKF (Adaptive Filter Integrated Kalman Filter) algorithm to predict the dynamic changes of the signal. The AFEKF algorithm predicts the oscillation characteristics of the next moment, including the amplitude and frequency dynamic characteristics, by adaptively updating the state of the filter based on the parameters extracted at the current moment. Then, the prediction result is compared with the actual observed oscillation information. By comparing the deviation between the predicted value and the actual observed value, the overall trend of the broadband oscillation and its changing law can be effectively identified, thereby realizing the accurate identification and early warning of broadband oscillations in the power system. This process not only helps to analyze the long-term evolution trend of the oscillation signal, but also provides a scientific basis for the stability analysis and dispatching of the power system, and discovers potential risks and anomalies in advance.

[0162] 9. Experimental verification

[0163] A wind farm model consisting of 6 wind turbines was established. Each wind turbine is connected to the 25 kV distribution system through a corresponding transformer and outputs power to the 120 kV grid through a 30 km 25 kV feeder line. The simplified model is as follows Figure 6 shown.

[0164] The above model runs in discrete mode, with a sampling time of 1e-5s and a running time of 0.2s. The voltage waveform of wide-band oscillation 0.02-0.06s (a total of 12,000 points) is collected. Then the compressed sensing theory is applied to the running data, and the Gaussian random matrix is ​​used for compression sampling. Finally, the compressed sampled data is reconstructed according to the SP algorithm.

[0165] Figure 7 This is a comparison between the original broadband oscillation signal and the reconstructed signal. The solid line in the figure is the compressed and reconstructed signal, and the dotted line is the original signal. The reconstructed signal basically restores the basic shape of the original signal, and the reconstruction error is 0.1263.

[0166] The reconstructed signal is decomposed using EEMD, such as Figure 8 , and a set of IMFs are obtained. The energy coefficient is used to remove irrelevant components and retain the components containing broadband oscillation information.

[0167] Finally, the Prony algorithm is used to calculate the IMF components. In addition to a 50Hz base frequency, the measured signal also has 150Hz and 200Hz medium and high frequency oscillation information. The specific results are as follows: Fig. 9 , and use the AFEKF algorithm to predict the oscillation information at the next moment.

[0168] From the above results, it can be seen that through compressed sampling reconstruction, EEMD decomposition, Prony analysis and AFEKF algorithm, the oscillation information of the original signal with a large amount of noise signal removed can be obtained, and the trend of broadband oscillation can be determined, which shows the accuracy and effectiveness of this method.

Claims

1. A method for identifying broadband oscillations of a new energy cluster connected to a DC grid, characterized in that: Collect real-time current and voltage signals from the synchronous vector measurement device PMU in the power grid and perform data preprocessing; compressively sample the preprocessed signals using compressed sensing (CS) technology; After the master station receives the low-dimensional oscillation data, the subspace tracking SP algorithm is used to reconstruct the compressed low-dimensional data; then the recovered signal is decomposed by hybrid empirical mode decomposition EEMD to obtain a set of intrinsic mode functions IMFs of the oscillation signal; by calculating the energy coefficient of each IMF, the redundant IMF components are removed; the optimized Prony algorithm is used to analyze the IMF components containing the oscillation modal information to obtain broadband oscillation information, including frequency and amplitude information; finally, the optimized adaptive filtering integrated Kalman filtering AFEKF algorithm is used to predict the dynamic changes of the signal. By comparing the deviation between the predicted value and the actual observed value, the overall trend and changing law of the broadband oscillation can be effectively identified, thereby realizing accurate identification and early warning of broadband oscillation in the power system.

2. A method for identifying wide-band oscillation of a new energy cluster connected to a DC grid according to claim 1, characterized in that: The measurement matrix A in the compressed sensing CS technology adopts a Gaussian matrix.

3. The method for identifying wide-band oscillation of a new energy cluster connected to a DC grid according to claim 1, characterized in that: The hybrid empirical mode decomposition EEMD is used to decompose the reconstructed signal obtained by SP, and the intrinsic mode function IMF component containing broadband oscillation information is screened by using the energy coefficient: Gaussian white noise with uniform frequency distribution is added before the empirical mode decomposition EMD decomposition signal, so that the signals of different scales in the noisy signal are distributed on a reference scale that matches it, and the zero mean characteristic of white noise is used to cancel each other out after multiple averaging, and the average value of each group of IMF components is used as the final decomposition result; the structure of the hybrid empirical mode decomposition EEMD algorithm is as follows: (1) Add a set of random Gaussian white noise sequences n(t) with a superposition mean of 0 and a constant standard deviation to the initial signal S(t) to obtain a new signal X(t), that is: X(t)=S(t)+n(t) (15) Mathematically, white noise is expressed as: n(t)~N(0,σ 2 ) (16) Where n(t) represents the noise signal at time t, Ν(0,σ 2 ) represents zero mean and variance σ 2 is a Gaussian distribution; (2) Perform EMD decomposition on each X(t) to obtain the M-order intrinsic mode function IMF component, denoted as C i (t), where i = 1, 2, ..., M, and the remaining component or residual signal is denoted by r k (t), the expression is as follows: (3) Different Gaussian white noise n j (t) is added to the original signal X(t), and steps (1) and (2) are repeated N times, that is: In the formula, C ij (t) represents the i-th IMF component obtained by decomposing the j-th superposition of white noise; (4) In order to eliminate the influence of noise, the IMFs of the same order are averaged to obtain the final IMF. The averaged IMF components of each order can be expressed as: In the formula, represents the jth IMF component after EEMD decomposition; (5) The signal X(t) can be reconstructed through all the final IMFs: in is the remaining amount.

4. The method for identifying wide-band oscillation of a new energy cluster connected to a DC grid according to claim 1, characterized in that: The adaptive filtering integrated Kalman filtering AFEKF algorithm is used to predict the oscillation information at the next moment according to the amplitude and frequency at the current moment, so as to determine the overall trend of the broadband oscillation.

Images