Improved node inertia online evaluation method suitable for power system
Through the MOESP and PEM composite algorithm combined with the data of the synchronous phasor measurement unit, the node inertia evaluation of power system is optimized, which solves the problems of poor robustness and low accuracy in the existing technology, and achieves higher evaluation accuracy and robustness, reducing the The risk of large-scale collapse of the power system.
Patent Information
- Application Number
- CN202410826419.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-25
- Publication Date
- 2024-11-08
AI Technical Summary
The existing technology cannot effectively estimate the inertia of the power system nodes online, resulting in poor robustness, cumbersome calculations and low accuracy, and is unable to provide reliable guidance for power grid scheduling in a timely manner, increasing the risk of large-scale collapse of the power system.
Using a composite algorithm based on multivariable output error state space (MOESP) and prediction error minimization (PEM) and combined with the node data feedback from the synchronous phasor measurement unit in the quasi-steady state of the power system, the global optimal system state space model is obtained through iterative optimization. , extract the inertia constant and damping coefficient to improve robustness and accuracy.
It improves the robustness of node inertia evaluation and the accuracy of evaluation results, overcomes the coupling problems of inertia constant and damping coefficient, and achieves more accurate node inertia monitoring and evaluation of power system nodes.
Smart Images

Figure CN118920443A_ABST
Abstract
Description
Technical Field
[0001] The invention discloses an improved node inertia online evaluation method applicable to a power system, relates to AC power distribution system technology, power grid planning and dispatching technology, and belongs to the technical field of power generation, power transformation or power distribution. Background Art
[0002] With the rapid development of new energy related technologies, more and more non-synchronous power sources such as photovoltaics and wind power are connected to the power grid, resulting in a large number of power electronic converters in the system, such as photovoltaic grid-connected converters, energy storage converters, etc. However, these power electronic devices decouple the active power and system frequency in the system. When the power system is subjected to external active power shocks, the non-synchronous power source cannot respond to the active power changes in the line like the traditional synchronous power source, and actively absorb the active power changes, resulting in the non-synchronous power source being disconnected from the grid due to excessive changes in the system frequency, and even causing large-scale paralysis of the power system. Therefore, it is necessary to monitor the node inertia in the power system in real time to ensure that effective measures are taken in time to suppress the node inertia drop and reduce the risk of large-scale collapse of the power system.
[0003] Most existing studies are based on large active-frequency events such as synchronous generator fault removal, which are ex post estimates and cannot estimate node inertia online and provide effective and reliable guidance to grid dispatchers in advance. Although some researchers have proposed online inertia evaluation methods based on parameter identification using quasi-steady-state data of power systems, most of these studies have problems such as poor robustness, cumbersome calculations, and low accuracy due to multiple factors. Therefore, how to improve the robustness of online node inertia evaluation methods and the accuracy of evaluation results is an urgent problem to be solved. Summary of the invention
[0004] In view of the deficiencies mentioned in the above technical background, the purpose of the present invention is to propose an improved node inertia online evaluation method suitable for power systems, construct a composite algorithm based on multivariable output error state space (MOESP) and prediction error minimization (PEM), and introduce it into the node inertia parameter identification of the power system, combine the node data fed back by the synchronous phasor measurement unit in the quasi-steady state of the power system for system identification, and use the iterative characteristics of the PEM algorithm to optimize the initial system state space model obtained by the MOESP algorithm to ensure the robustness of the power system node inertia evaluation process and the accuracy of the evaluation results.
[0005] The present invention adopts the following technical solutions to achieve the above-mentioned invention objectives:
[0006] An improved node inertia online evaluation method applicable to a power system comprises the following steps:
[0007] Step 1: collect active power changes and frequency changes at the nodes to be evaluated in the power system, and pre-process the collected data;
[0008] Step 2: Using random subspace algorithm to obtain the local optimal system state space model;
[0009] Step 3: Using the local optimal system state space model parameters obtained in step 2 as initial values, the prediction error minimization algorithm is used to iteratively optimize the model parameters to obtain the global optimal system state space model;
[0010] Step 4: Convert the global optimal system state space model obtained in step 3 into a discrete transfer function, perform continuous processing on the discrete transfer function, perform order reduction processing on the time domain transfer function obtained by continuous processing, and extract the inertia constant and system damping coefficient at the node to be evaluated from the time domain function after order reduction processing.
[0011] As a further optimization scheme of the improved node inertia online evaluation method suitable for power systems, step one performs preprocessing of the collected data including standardization and detrending.
[0012] As a further optimization scheme for the improved node inertia online evaluation method suitable for power systems, the method for normalizing the collected data in step 1 is: Calculate the per-unit value of the active power change and the per-unit value of the frequency change at the node to be evaluated, where: is the per-unit value of the frequency change at the nth node to be evaluated, busn; f busn is the measured value of the frequency at the nth node to be evaluated, busn; f 0 is the system rated frequency; is the per unit value of the active power change at the nth node to be evaluated busn; busn is the measured value of active power at the nth node to be evaluated, busn; P 0,busn is the active power of the nth node to be evaluated busn in steady state operation; S gn,busn is the apparent power of the synchronous generator connected to the nth node busn to be evaluated.
[0013] As a further optimization scheme of the improved node inertia online evaluation method suitable for power systems, the method for detrending the collected data in step 1 is:
[0014] The time series of the active power change at the node to be evaluated after the per-unit processing or the time series of the frequency change at the node to be evaluated after the per-unit processing is x(t), t = 1, 2, ..., N, and the average value of x(t) is calculated. The cumulative outlier value x'(t) of N is the length of the time series;
[0015] Divide x'(t) into m non-overlapping intervals, each of which is N / m long, and use the least squares method to fit the local trend x of each subsequence. v (t), x v (t) = a 0 +a 1 t+a 2 t 2 +…+a α t α v=1,2,…,m,α 0 ,α 1 ,α 2 ,……,α α are polynomial coefficients, α is the order of polynomial fitting;
[0016] Remove the local trend of each subsequence from the time series x(t) to obtain the final time series x*(t).
[0017] The input sequence u(k) is constructed according to the per-unit value of the active power change at the node to be evaluated after detrending, and the output sequence y(k) is constructed according to the per-unit value of the frequency change at the node to be evaluated after detrending, where k=1,2,…,N.
[0018] As a further optimization scheme of the improved node inertia online evaluation method suitable for power systems, the specific method of obtaining the local optimal system state space model using the random subspace algorithm in step 2 is:
[0019] According to the output sequence and the input sequence, an output Hankel matrix Y and an input Hankel matrix U are constructed respectively. With the current sampling time g as the limit, the output Hankel matrix Y is divided into the past output Hankel matrix Y p and the future output Hankel matrix Y f , divide the input Hankel matrix U into the past input Hankel matrix U p and the future input Hankel matrix U f ;
[0020] The output Hankel matrix and the input Hankel matrix are compressed by QR decomposition to obtain U fThe row space of Y f The orthogonal projection matrix O of the row space g ;
[0021] For the orthogonal projection matrix O g Perform singular value decomposition and use the generalized matrix Г g Represents the orthogonal projection matrix Calculate the generalized energy observable matrix Г g , Using the generalized observable matrix Г g Calculate the system matrix A and the output matrix C, Calculate the input matrix B and the feedforward matrix D, Among them, S is O g The diagonal matrix composed of the eigenvalues of T O g The singular vectors composed of the eigenvectors of 1 , S 2 is a submatrix of S, U 1 , U 2 is a submatrix of U, V T The subarray of is the state sequence W g The non-stationary Kalman filter estimate of are the Kalman filter estimates of the system state at the g-th sampling moment, the g+1-th sampling moment, and the g+j-1-th sampling moment, respectively. is the pseudo-inverse operation of the Moore-Penrose matrix, Г g (1,:) is Г g The first line of g (1:g-1,:) is Г g The first g-1 rows of Г g (g+1:2g,:) is Г g The last g-1 row, [L 1 L 2 L 3 ···L g ]=U 2 , is the identity matrix, n y is the order of the output sequence y(k).
[0022] As a further optimization scheme of the improved node inertia online evaluation method suitable for power systems, the specific method of obtaining the global optimal system state space model in step 3 is:
[0023] The parameter vector θ is used to parameterize the local optimal system state space model parameters obtained in step 2. The system state space model is expressed as: in, is the estimated value of the system state at the kth sampling moment, x(k+1) is the system state at the k+1th sampling moment, e(k) is the system noise at the kth sampling moment, A(θ) is the parameterized system matrix, B(θ) is the parameterized input matrix, C(θ) is the parameterized output matrix, D(θ) is the parameterized feedforward matrix, and K(θ) is the parameterized noise matrix;
[0024] With the goal of minimizing the mean square error of the system spatial state model, the parameter vector θ is iteratively solved. According to Δθ p θ when it approaches 0 infinitely p Get the best estimate of the parameter vector θ Among them, θ p+1 is the p+1th iteration value of the parameter vector θ, Δθ p is the increment of the parameter vector θ at the pth iteration, J(θ p ) is θ p The Jacobian matrix at V N (θ p ) is the pth iteration value θ based on the parameter vector θ p The obtained mean square error of the system spatial state model is ε(k,θ p ) is the pth iteration value θ based on the parameter vector θ p Predict output sequence of error.
[0025] As a further optimization scheme of the improved node inertia online evaluation method suitable for power systems, step 4 converts the global optimal system state space model obtained in step 3 into a discrete transfer function G(z). Among them, z represents the discrete domain, Y(z) is the output sequence in the discrete domain, and U(z) is the input sequence in the discrete domain.
[0026] As a further optimization scheme for the improved node inertia online evaluation method suitable for power systems, step 4 uses a zero-order holder to continuously process the discrete transfer function, and continuously converts the discrete transfer function G(z) into a time domain transfer function G(s). Where s represents the Laplace operator, Y(s) is the time domain function of the output sequence, U(s) is the time domain function of the input sequence, and k 0,k 1 ,……,k h-1 ,k h is the coefficient of the time domain function of the output sequence, l 0 ,l 1 ,……,l h-1 ,l h are the coefficients of the time domain function of the input sequence.
[0027] As a further optimization scheme for the improved node inertia online evaluation method suitable for power systems, step 4 reduces the order of the time domain transfer function obtained by continuous processing. The time domain function after the reduced order processing is H is the inertia constant at the node to be evaluated, and D is the damping coefficient at the system.
[0028] A computer-readable storage medium stores a computer program, which executes the steps of the above-mentioned improved node inertia online evaluation method when the computer program is running.
[0029] Compared with the prior art, the technical solution provided by the present invention has the following significant beneficial effects:
[0030] (1) The present invention introduces the detrending fluctuation analysis algorithm into the data preprocessing of inertia assessment, which effectively eliminates the accumulated deviation information generated in the measurement process, avoids its influence on the inertia assessment process, and improves the accuracy of node inertia assessment.
[0031] (2) The present invention uses MOESP and PEM to construct a composite algorithm and introduces it into the node inertia parameter identification. The iterative characteristics of the PEM algorithm are used to make up for the defect that the MOESP algorithm may produce random errors. The iterative process of the PEM algorithm is shortened by the MOESP algorithm. The two complement each other, making the node inertia evaluation method have better robustness and higher evaluation accuracy.
[0032] (3) The present invention overcomes the coupling problem of inertia constant and damping coefficient in the previous inertia evaluation process, and can simultaneously evaluate the node inertia and damping coefficient, further improving the accuracy of the node inertia evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0034] Figure 1 It is a simple power system structure consisting of 3 synchronous generators and 9 buses.
[0035] Figure 2 It is a flow chart of the improved online evaluation method of node inertia.
[0036] Figure 3 This is a detrended data chart of the node frequency and active power deviation per unit value collected from Bus2.
[0037] Figure 4 This is a diagram showing the validation results of the state-space model obtained by the first evaluation using the Bus2 bus.
[0038] Figure 5 It is a schematic diagram of the continuous estimation deviation of the inertia of Bus1, 2, and 3 nodes.
[0039] Figure 6 It is a schematic diagram of the continuous estimation deviation of the damping coefficients of Bus1, 2, and 3 nodes.
[0040] Explanation of symbols in the figure: ΔP* is the per unit value of the active power deviation at the node, and the benchmark is the apparent power S of each synchronous unit gn,bus n ; Δf* is the per-unit value of the frequency deviation at the node, and the reference is the industrial frequency of 50Hz. DETAILED DESCRIPTION
[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0042] like Figure 2 As shown, the improved node inertia online evaluation method applicable to the power system proposed in the present invention includes the following four steps.
[0043] Step 1: Collect the active power change and frequency change at the node to be evaluated in the power system and pre-process the collected data
[0044] After selecting the nodes bus1, bus2, ..., busn to be evaluated in the power system, the active power and frequency change of each node to be evaluated are obtained through a synchronous phasor measurement unit, and then the active power change and frequency change at each node to be evaluated are normalized and detrended.
[0045] 1.1 The frequency and active power changes at the nodes to be evaluated in the power system collected in standard form
[0046]
[0047] in, is the per-unit value of the frequency change at the nth node to be evaluated, busn; f busn is the measured value of the frequency at the nth node to be evaluated, busn; f 0 is the system rated frequency; is the per unit value of the active power change at the nth node to be evaluated busn; busn is the measured value of active power at the nth node to be evaluated, busn; P 0,busn is the active power of the nth node to be evaluated busn in steady state operation; S gn,busn is the apparent power of the synchronous generator connected to the nth node busn to be evaluated.
[0048] 1.2 Detrending the normalized data:
[0049] ① Assume that there is a time series x(t) (t=1,2,…,N), and calculate its cumulative outlier x'(t) relative to the mean:
[0050]
[0051] in, is the average value of the original time series x(t):
[0052]
[0053] ② Divide the time series x'(t) into m non-overlapping intervals with an interval length of N / m, and use the least squares method to fit the local trend x of each subsequence v (t):
[0054] x v (t) = a 0 +a 1 t+a 2 t 2 +…+a α t α v=1,2,…,m
[0055] Where α is the order of the polynomial fit.
[0056] ③ Eliminate the local trend of the time sequence x(t) (t=1,2,…,N) and generate the final sequence x*(t):
[0057]
[0058] The change in active power after normalization ΔP bus n and frequency change Δf bus n After the above processing, the data is processed to obtain the following input and output sequence
[0059]
[0060] Among them, the input u(k) is the processed active power change data; the output y(k) is the processed frequency change data.
[0061] Step 2: Use the random subspace algorithm MOESP to obtain the local optimal system state space model M 0 Using the random subspace algorithm MOESP to obtain the system state space model M 0 , the algorithm flow is as follows:
[0062] First, using the input-output sequence Construct the corresponding Hankel matrices respectively. Take the output sequence y(k) as an example to construct the following output Hankel matrix Y:
[0063]
[0064] Among them, the output Hankel matrix Y is divided into the past output Hankel matrix Y with the current sampling time g as the limit. p and the future output Hankel matrix Y f Two parts; j is the number of sampling data (j=1,2,…,N).
[0065] Similarly, the input Hankel matrix U is constructed from the input sequence u(k), thereby obtaining the past input Hankel matrix U p and the future input Hankel matrix U f .
[0066] Secondly, use QR decomposition to compress the data of the Hankel matrix:
[0067]
[0068] Among them, R 11 , R 21 , R 22 , R 31 , R 32 , R 33 is a submatrix of the upper triangular matrix R, is a submatrix of the normal orthogonal matrix Q, R 32 That is U f The row space of Y f The matrix of the orthogonal projection of the row space is denoted as O g .
[0069] Again, the projection matrix O is decomposed using singular value decompositiong Decomposition, while the projection matrix O g The generalized observable matrix Г g To express:
[0070]
[0071] Among them, S is O g The diagonal matrix composed of the eigenvalues of T O g The singular vectors composed of the eigenvectors of 1 , S 2 is a submatrix of S, U 1 , U 2 is a submatrix of U, V T The subarray, U 2 S 2 V 2 T Characterizes the noise pollution in the system and its magnitude is relative to U 1 S 1 V 1 T Small, negligible, does not affect the matrix accuracy, and can also achieve the purpose of noise reduction to a certain extent; is the state sequence W g The non-stationary Kalman filter estimate of They are the Kalman filter estimates of the system state at the g-th sampling moment, the g+1-th sampling moment, and the g+j-1-th sampling moment respectively.
[0072] According to the above two formulas, the generalized energy matrix Г can be calculated g :
[0073]
[0074] Finally, the system matrix A and output matrix C are calculated using the generalized energy observable matrix:
[0075]
[0076] in, represents the pseudo-inverse operation of the Moore-Penrose matrix; Г g (1,:) is Г g The first line of ; g (1:g-1,:) is Г g The first g-1 rows of Г g (g+1:2g,:) is Г g The last g-1 row.
[0077] The calculation process of the system's input matrix B and feedforward matrix D is as follows:
[0078]
[0079] Among them, [L 1 L 2 L 3 ···L g ]=U 2 ; I is the unit matrix; n y is the order of the output sequence y(k).
[0080] After the system matrices A, B, C and D are obtained by using the MOESP algorithm, the state space model M 0 As the initial value input to the PEM algorithm for iteration.
[0081] Step 3: Use the PEM algorithm to obtain the global optimal system state space model M*
[0082] The PEM algorithm is used for multiple iterations to obtain the global optimal state space model M*. The algorithm flow is as follows:
[0083] First, the system parameters are fully parameterized using the parameter vector θ. A(θ) is the parameterized system matrix, B(θ) is the parameterized input matrix, C(θ) is the parameterized output matrix, D(θ) is the parameterized feedforward matrix, K(θ) is the parameterized noise matrix, and the state space equation is transformed into M(θ):
[0084]
[0085] in, is the estimated value of the system state at the kth sampling time, x(k+1) is the system state at the k+1th sampling time, e(k) is the system noise at the kth sampling time, and the parameter vector θ is a priori unknown. Each θ p (p=1,2…n) all represent a state space model M(θ p ); n is the number of iterations of the algorithm.
[0086] Secondly, each state space model M(θ p ) can output a set of predicted output sequences Comparing it with the measured output sequence y(k), we can derive the model M(θ p ) of the prediction error ε(k,θ p ) and the corresponding mean square error V N (θ p ):
[0087]
[0088] The optimization process of the state space model can be expressed as:
[0089]
[0090] Here, the optimization algorithm used in the solution process is Gauss-Newton line search, and its prediction error criterion is the mean square error V N (θ p ), the p+1th iteration value is as follows:
[0091]
[0092] Among them, the Jacobian matrix J(θ p ) is V N (θ p )The first-order derivative with respect to θ is p The value of .
[0093] Finally, repeat the above iterative process until Δθ p infinitely approaches 0. At this point, the obtained state space model is the global optimal solution M*.
[0094] Step 4: Extract system parameters
[0095] 4.1 Convert the global optimal system state space model M* into a discrete transfer function G(z):
[0096]
[0097] Among them, z represents the discrete domain, Y(z) is the output sequence in the discrete domain, and U(z) is the input sequence in the discrete domain.
[0098] 4.2 Use the zero-order holder to convert G(z) into G(s):
[0099]
[0100] Where s represents the Laplace operator, Y(s) is the time domain function of the output sequence, U(s) is the time domain function of the input sequence, and k 0 ,k 1 ,……,k h-1 ,k h is the coefficient of the time domain function of the output sequence, l 0 ,l 1 ,……,l h-1 ,l h are the coefficients of the time domain function of the input sequence.
[0101] 4.3 The transfer function G'(s) obtained by reducing the order of G(s) can still describe the system characteristics at the node well:
[0102]
[0103] Among them, H is the required node inertia, and D is the damping coefficient of the system.
[0104] The specific process of the improved node inertia online evaluation of the power system of the present invention is as follows:
[0105] exist Figure 1 In the simple power system shown, Bus2 is selected as the node to be evaluated, which is equivalent to an equivalent synchronous generator, and the rest is regarded as an external power system connected to it. The node data is measured by a synchronous phasor measurement unit and the frequency and active power data are fed back for preprocessing.
[0106] Normalization:
[0107]
[0108] Among them, P 0,bus2 The active power of bus2 in steady state operation is 163MW; S gn,bus2 The apparent power of the No. 2 synchronous unit is 192 kVA; f 0 The power frequency is 50Hz; f bus 2 and P bus 2 are the frequency and active power of bus2 at time i respectively.
[0109] After detrending, we can get Figure 3 The input and output sequence shown is:
[0110]
[0111] Among them, the input quantity u(k) is the processed active power data; the output quantity y(k) is the processed frequency.
[0112] will sequence As input, run the composite algorithm described above and get the following discrete system parameters:
[0113] A=[1],B=[0.017],C=[-0.0585],D=[0],K=[-13.88]
[0114] Convert the system state space model into a discrete transfer function G(z):
[0115]
[0116] In order to verify the fit between the obtained G(z) and the actual system, the data within 10-60s is used for verification. Figure 4 This is the fitting result. The fitting degree is as high as 98.83%. Obviously, the calculated state space model meets the requirements, and a zero-order holder is used to convert G(z) into G(s):
[0117]
[0118] Since the identification result is exactly the first-order transfer function, there is no need to reduce the order. The required information can be directly extracted from the transfer function, that is, the node inertia H = 5.0277s at bus2 and the damping coefficient D = -0.0231. Compared with the true values listed in Table 1, the error is small.
[0119] In order to avoid large deviations in the inertia evaluation results due to interference from random events, a 10s sliding window is used to continuously evaluate the node inertia. The deviation of the continuous online inertia evaluation results within 0-500s is as follows: Figure 5 shown.
[0120] Application examples:
[0121] The following takes a simple three-machine nine-node power system as an example and combines simulation results to verify the application effect of the technical solution of the present invention.
[0122] The simulation parameters are as follows:
[0123] Table 1: Main parameters of generator
[0124]
[0125] Table 2: Main parameters of the line
[0126] serial number Positive sequence resistance R / Ω Positive sequence reactance X / Ω Voltage level / kV Line 4-5 5.29 44.964 230 Lines 4-6 8.993 48.668 230 Line 5-7 16.928 85.169 230 Lines 6-9 20.631 89.93 230 Lines 7-9 4.4965 38.088 230 Lines 8-9 6.2951 53.3232 230
[0127] Table 3: Main parameters of load
[0128]
[0129]
[0130] Figure 1 It is a three-machine nine-node power system, and its main parameters are shown in Tables 1, 2, and 3. Figure 5 and Figure 6 They are the relative estimation results of the inertia time constant of each generator in the system and the evaluation results of the damping coefficient. It can be seen from the figure that the improved node inertia online evaluation method proposed in the present invention can perform a relatively accurate continuous online evaluation of the node inertia of the power system.
[0131] In summary, the improved node inertia online evaluation method suitable for power systems of the present invention can realize more accurate online monitoring of power system node inertia, and the proposed method is also more robust.
[0132] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.
Claims
1. An improved node inertia online evaluation method suitable for power system, characterized in that: The steps include: Step 1: collect active power changes and frequency changes at the nodes to be evaluated in the power system, and pre-process the collected data; Step 2: Using random subspace algorithm to obtain the local optimal system state space model; Step 3, using the local optimal system state space model parameters obtained in step 2 as initial values, adopting a prediction error minimization algorithm to iteratively optimize the model parameters to obtain a global optimal system state space model; Step 4: Convert the global optimal system state space model obtained in step 3 into a discrete transfer function, perform continuous processing on the discrete transfer function, perform order reduction processing on the time domain transfer function obtained by continuous processing, and extract the inertia constant and system damping coefficient at the node to be evaluated from the time domain function after order reduction processing.
2. According to claim 1, an improved node inertia online evaluation method suitable for power system, characterized in that: In the step one, the collected data is preprocessed including standardization and detrending.
3. The improved node inertia online evaluation method applicable to the power system according to claim 2, characterized in that: The method for normalizing the collected data in step 1 is: Calculate the per-unit value of the active power change and the per-unit value of the frequency change at the node to be evaluated, where: is the per-unit value of the frequency change at the nth node to be evaluated, busn; f busn is the measured value of the frequency at the nth node to be evaluated, busn; f0 is the rated frequency of the system; is the per unit value of the active power change at the nth node to be evaluated busn; busn is the measured value of active power at the nth node to be evaluated, busn; P 0,busn is the active power of the nth node to be evaluated busn in steady state operation; S gn,busn is the apparent power of the synchronous generator connected to the nth node busn to be evaluated.
4. The improved node inertia online evaluation method applicable to the power system according to claim 3 is characterized in that: The method for detrending the collected data in step 1 is: The time series of the active power change at the node to be evaluated after the per-unit processing or the time series of the frequency change at the node to be evaluated after the per-unit processing is x(t), t = 1, 2, ..., N, and the average value of x(t) is calculated. The cumulative outlier value x(t) of N is the length of the time series; Divide x′(t) into m non-overlapping intervals, each of which has a length of N / m, and use the least squares method to fit the local trend x of each subsequence. v (t), x v (t) = a0 + a1t + a2t 2 +…+a α t α v=1,2,…,m,α0,α1,α2,…,α α are polynomial coefficients, α is the order of polynomial fitting; Remove the local trend of each subsequence from the time series x(t) to obtain the final time series x*(t). The input sequence u(k) is constructed according to the per-unit value of the active power change at the node to be evaluated after detrending, and the output sequence y(k) is constructed according to the per-unit value of the frequency change at the node to be evaluated after detrending, where k=1,2,…,N.
5. The improved node inertia online evaluation method applicable to the power system according to claim 4, characterized in that: The specific method of obtaining the local optimal system state space model by using the random subspace algorithm in step 2 is: According to the output sequence and the input sequence, an output Hankel matrix Y and an input Hankel matrix U are constructed respectively. With the current sampling time g as the limit, the output Hankel matrix Y is divided into the past output Hankel matrix Y p and the future output Hankel matrix Y f , divide the input Hankel matrix U into the past input Hankel matrix U p and the future input Hankel matrix U f ; The output Hankel matrix and the input Hankel matrix are compressed by QR decomposition to obtain U f The row space of Y f The orthogonal projection matrix O of the row space g ; For the orthogonal projection matrix O g Perform singular value decomposition and use the generalized matrix Г g Denotes the orthogonal projection matrix O g , Calculate the generalized energy observable matrix Г g , Using the generalized observable matrix Г g Calculate the system matrix A and the output matrix C, Calculate the input matrix B and the feedforward matrix D, Among them, S is O g The diagonal matrix composed of the eigenvalues of U, V T O g The singular vectors composed of the eigenvectors of , S1 and S2 are sub-matrices of S, U1 and U2 are sub-matrices of U, V T The subarray of is the state sequence W g The non-stationary Kalman filter estimate of are the Kalman filter estimates of the system state at the g-th sampling moment, the g+1-th sampling moment, and the g+j-1-th sampling moment, respectively. is the pseudo-inverse operation of the Moore-Penrose matrix, Г g (1,:) is Г g The first line of g (1:g-1,:) is Г g The first g-1 rows of Г g (g+1:2g,:) is Г g The last g-1 row, [L1 L2L3···L g ]=U2, is the identity matrix, n y is the order of the output sequence y(k).
6. The improved node inertia online evaluation method applicable to the power system according to claim 5, characterized in that: The specific method of obtaining the global optimal system state space model in step 3 is: The parameter vector θ is used to parameterize the local optimal system state space model parameters obtained in step 2. The system state space model is expressed as: in, is the estimated value of the system state at the kth sampling moment, x(k+1) is the system state at the k+1th sampling moment, e(k) is the system noise at the kth sampling moment, A(θ) is the parameterized system matrix, B(θ) is the parameterized input matrix, C(θ) is the parameterized output matrix, D(θ) is the parameterized feedforward matrix, and K(θ) is the parameterized noise matrix; With the goal of minimizing the mean square error of the system spatial state model, the parameter vector θ is iteratively solved. According to Δθ p θ when it approaches 0 infinitely p Get the best estimate of the parameter vector θ Among them, θ p+1 is the p+1th iteration value of the parameter vector θ, Δθ p is the increment of the parameter vector θ at the pth iteration, J(θ p ) is θ p The Jacobian matrix at V N (θ p ) is the pth iteration value θ based on the parameter vector θ p The obtained mean square error of the system spatial state model is ε(k,θ p ) is the pth iteration value θ based on the parameter vector θ p Predict output sequence of error.
7. The improved node inertia online evaluation method applicable to the power system according to claim 6, characterized in that: The fourth step converts the global optimal system state space model obtained in the third step into a discrete transfer function G(z). Among them, z represents the discrete domain, Y(z) is the output sequence in the discrete domain, and U(z) is the input sequence in the discrete domain.
8. The improved node inertia online evaluation method applicable to the power system according to claim 7, characterized in that: The fourth step uses a zero-order holder to perform continuous processing on the discrete transfer function, and continuously converts the discrete transfer function G(z) into a time domain transfer function G(s). Where s represents the Laplace operator, Y(s) is the time domain function of the output sequence, U(s) is the time domain function of the input sequence, k0, k1, ..., k h-1 ,k h are the coefficients of the time domain function of the output sequence, l0,l1,…,l h-1 ,l h are the coefficients of the time domain function of the input sequence.
9. The improved node inertia online evaluation method applicable to the power system according to claim 8, characterized in that: In step 4, the time domain transfer function obtained by the continuous processing is reduced in order. The time domain function after the reduced order processing is: H is the inertia constant at the node to be evaluated, and D is the damping coefficient at the system.
10. A computer-readable storage medium having a computer program stored thereon, wherein the computer program executes the steps of the improved node inertia online evaluation method according to claim 1 when the computer program is executed.
Citation Information
Cited By
Engine inertia measurement method based on EMD algorithm
CN119437548A
