Method, device and product for selecting key features of power system transient stability

Abstract

The embodiment of the present application provides a kind of power system transient stability key feature selection method, device and product, belong to information processing technical field.In the embodiment of the present application, the operation data not carried out in power grid safety and stability evaluation is considered, and the manifold structure of operation data sample of unlabeled data is reserved by local linear embedding;In the embodiment of the present application, the forward sequence selection method based on Fisher Score is used, and unlabeled data is processed by local linear embedding, and the variance cumulative contribution rate is calculated by principal component analysis method to reduce the dimension of feature;After dimension reduction, efficient incremental feature selection is carried out, so that the key feature affecting power system transient stability is selected.The embodiment of the present application can comprehensively utilize the operation data not subjected to safety and stability evaluation generated in the process of power system operation, and improve the classification accuracy of the selected key feature on test set.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of power system transient stability analysis technology, and in particular to a method, apparatus, electronic device and storage medium for selecting key transient stability features of power systems. Background Technology

[0002] Transient stability assessment of power systems involves evaluating their safety and stability. Uncertainties in grid operating conditions and fluctuations, as well as the integration of renewable energy and new loads, all contribute to the challenges of this assessment. Commonly used transient stability analysis methods include time-domain simulation, energy function methods, or deep learning feature selection to optimize grid safety and stability.

[0003] Among the aforementioned transient stability analysis methods, the time-domain simulation method establishes accurate mathematical models of each part of the system, describing the system state and topology using differential and algebraic equations. Using the steady-state point as the initial value, it solves for numerical solutions under disturbances or faults, and judges system stability based on the rotor yaw curve. This method provides accurate solutions, but its modeling is complex, integration calculations are time-consuming, and it cannot calculate stability margins, limiting its application to offline analysis. The energy function method determines the system state by comparing the energy function values ​​at the fault clearing time and the ultimate clearing time. It offers fast calculation speed and can assess stability margins. However, constructing accurate energy functions and solving for critical points remain challenging, and current methods cannot provide accurate solutions, resulting in conservative evaluations.

[0004] Given a thorough and accurate understanding of the power grid's operational status, selecting key features highly correlated with system stability for evaluation is a crucial direction for optimizing power grid stability analysis. Forward sequence selection based on the Fisher Score is a well-performing and widely used supervised feature selection method. This method uses forward sequence selection as the search strategy and the Fisher Score as the evaluation metric, maximizing inter-class divergence while minimizing intra-class divergence, and selecting key features using a scoring threshold. However, this feature selection method has limitations: it cannot effectively utilize the large amount of operational data generated during power grid operation that has not undergone safety and stability assessment; and the computational complexity of scoring candidate features increases cubically with the number of selected features.

[0005] Therefore, there is an urgent need for a new method for selecting key features of transient stability in power systems. Summary of the Invention

[0006] This invention provides a method, apparatus, electronic device, and storage medium for selecting key transient stability features of a power system, in order to solve the problems existing in related technologies.

[0007] The first aspect of this invention provides a method for selecting key features of transient stability in a power system, the method comprising:

[0008] Add data labels to power system simulation data samples to obtain labeled simulation data samples. Use power system operation data samples as unlabeled operation data samples. Input the simulation data samples and the operation data samples as data samples into the original feature set. The data labels are based on power angle instability as the criterion.

[0009] The original feature set is subjected to initial feature screening to obtain the initial screening feature set;

[0010] The inter-class scatter matrix and the first total scatter matrix of the simulated data samples in the feature set after initial screening are calculated.

[0011] Local linear embedding is performed on the running data samples in the feature set after initial screening to obtain the local linear representation matrix and the second total divergence matrix of the running data samples.

[0012] Based on the cumulative variance contribution rate threshold, principal component analysis is performed on the second total scatter matrix to reduce its dimensionality, thus obtaining the first feature subset;

[0013] If the number of features in the first feature subset is greater than the number of features to be extracted, calculate the Fisher score corresponding to the first feature subset;

[0014] If the Fisher score reaches the output threshold, the first feature subset is output as the key feature set for transient stability of the power system.

[0015] Optionally, the method further includes:

[0016] If the number of features in the feature subset is less than or equal to the number of features to be extracted, the cumulative variance contribution rate threshold is adjusted, and principal component analysis is performed on the second total scatter matrix to reduce its dimensionality, thereby obtaining the second feature subset.

[0017] Optionally, the original feature set is subjected to initial feature screening, including:

[0018] The original feature set is initially screened using a low-variance feature filtering method. The variance of each original feature in all data samples is sorted to eliminate original features with variances below the variance threshold.

[0019] Optionally, before performing initial feature screening on the original feature set, the method further includes:

[0020] The mean of the simulated data samples is subtracted from the mean of all data samples in the original feature set.

[0021] Optionally, the original feature set includes bus phase angle, bus voltage amplitude, line power flow active power, line power flow reactive power, power flow at key sections, line current, generator power angle, and generator speed.

[0022] Optionally, the calculation method for the local linear embedding is as follows:

[0023]

[0024]

[0025] M ij =0, if

[0026] Among them, z i For data points with locally linear embedding, N i ={z i1 ,z i2 ,…,z ik} for z i The neighborhood set of z j For z i The j-th nearest neighbor, along with N i The expression within is consistent, M ij The weight coefficient for each neighboring point, M = (M ij ) is the obtained local linear representation matrix.

[0027] Optionally, the method further includes: randomly selecting 80% of the data samples as a training set and the remaining 20% ​​as a test set, and using a 1-nearest neighbor classifier to determine the classification accuracy of the key feature set.

[0028] A second aspect of this invention provides a device for selecting key transient stability features of a power system, the device comprising:

[0029] The data sample processing module is used to add data tags to power system simulation data samples to obtain labeled simulation data samples, and to take power system operation data samples as unlabeled operation data samples. The simulation data samples and the operation data samples are used as data sample inputs to the original feature set, and the data tags are based on power angle instability as the criterion.

[0030] The feature screening module is used to perform initial feature screening on the original feature set to obtain the initial screened feature set.

[0031] The first calculation module is used to calculate the simulation data samples in the feature set after initial screening, and obtain the inter-class scatter matrix and the first total scatter matrix of the simulation data samples.

[0032] The local linear embedding module is used to perform local linear embedding on the running data samples in the feature set after initial screening, to obtain the local linear representation matrix and the second total divergence matrix with the running data samples added.

[0033] The principal component analysis module is used to perform principal component analysis to reduce the dimensionality of the second total scatter matrix based on the cumulative variance contribution rate threshold, so as to obtain the first feature subset.

[0034] The second calculation module is used to calculate the Fisher score corresponding to the first feature subset when the number of features in the first feature subset is greater than the number of features to be extracted.

[0035] The output module is used to output the first feature subset as the key feature set for transient stability of the power system when the Fisher score reaches the output threshold.

[0036] Optionally, the device further includes:

[0037] The adjustment module is used to adjust the cumulative variance contribution rate threshold when the number of features in the feature subset is less than or equal to the number of features to be extracted, and to perform principal component analysis to reduce the dimensionality of the second total scatter matrix to obtain the second feature subset.

[0038] Optionally, the feature screening module is specifically used for:

[0039] A device employing low-variance feature filtering performs initial feature screening on the original feature set, sorts the variance of each original feature for all data samples, and eliminates original features with variances below a variance threshold.

[0040] Optionally, the device further includes:

[0041] The centralization processing module is used to centralize all data samples in the original feature set by subtracting the mean of the simulation data samples from all data samples in the original feature set.

[0042] Optionally, the original feature set includes: bus phase angle, bus voltage amplitude, line power flow active power, line power flow reactive power, power flow at key sections, line current, generator power angle, and generator speed.

[0043] Optionally, the calculation method for the local linear embedding is as follows:

[0044]

[0045]

[0046] M ij =0, if

[0047] Among them, z i For data points with locally linear embedding, N i ={z i1 ,zi2,…,z ik} for z i The neighborhood set of z j Let M be the j-th nearest neighbor of zi, consistent with the description in Ni. ij Let M = (Mij) be the weight coefficients corresponding to each neighboring point, and M = (Mij) be the obtained local linear representation matrix.

[0048] Optionally, the device further includes:

[0049] The verification module is used to randomly select 80% of the data samples as the training set and the remaining 20% ​​as the test set, and use a 1-nearest neighbor classifier to determine the classification accuracy of the key feature set.

[0050] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect of the present invention.

[0051] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the steps of the method described in the first aspect of the present invention.

[0052] In this embodiment of the invention, operational data from the power grid that has not undergone safety and stability assessment is considered. Local linear embedding is used to preserve the manifold structure of the unlabeled operational data samples. This embodiment employs a forward sequence selection method based on FisherScore and uses local linear embedding to process the unlabeled data. Simultaneously, principal component analysis is used to calculate the cumulative variance contribution rate to reduce the dimensionality of the features. After dimensionality reduction, efficient incremental feature selection is performed to identify key features affecting the transient stability of the power system. This embodiment of the invention can reasonably utilize the combination of simulation data and operational data. Considering the time complexity of incremental selection, the transient stability analysis results can comprehensively take into account the classification accuracy and computation time of different feature numbers and methods, making the conclusions on the selection of key transient stability features more reasonable and effective.

[0053] The embodiments of the present invention can make full use of the operational data generated during the operation of the power system that has not undergone safety and stability assessment, thereby improving the classification accuracy of the selected key features on the test set. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of a method for selecting key features of transient stability in a power system according to an embodiment of the present invention;

[0056] Figure 2 This is a flowchart of another method for selecting key features of transient stability in power systems according to an embodiment of the present invention;

[0057] Figure 3 This is a block diagram of a power system transient stability key feature selection device according to an embodiment of the present invention. Detailed Implementation

[0058] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0059] In power systems, the amount of operational data is enormous, while the amount of simulation data is extremely small. If only simulation data is used in the selection of key features, the results will undoubtedly be inaccurate. Therefore, this invention proposes to consider the influence of operational data on the selection of key features for power system transient stability.

[0060] Specifically, this invention proposes a method for selecting key transient stability features of a semi-supervised power system based on manifold regularization, referring to... Figure 1 The flowchart illustrates a method for selecting key features of transient stability in a power system according to an embodiment of the present invention. The method for selecting key features of transient stability in a power system provided by the present invention may include the following steps:

[0061] S101, add data labels to the power system simulation data samples to obtain labeled simulation data samples, take the power system operation data samples as unlabeled operation data samples, and input the simulation data samples and the operation data samples as data samples into the original feature set. The data labels are based on power angle instability as the criterion.

[0062] The power system mentioned in this embodiment of the invention can be the IEEE 39-bus system.

[0063] The IEEE 39-bus system is a well-known regional transmission system network in the field of power systems, also known as the New England 39Bus System (NE39BS). This benchmark network is located in the New England region of the United States and consists of 39 buses, including 10 generator buses and 19 load buses. It is widely used in fields such as small-signal stability research, dynamic stability analysis, power quality analysis and control.

[0064] In this embodiment of the invention, simulation software can be used to perform transient stability time-domain simulation of the power system, conduct fault analysis, obtain simulation data, and export it as a power system simulation data sample.

[0065] In this embodiment of the invention, for each simulation data sample, a label can be added based on its power angle instability. For example, a simulation data sample with power angle instability is assigned data label 0, and a simulation data sample with stable power angle is assigned data label 1.

[0066] With the rapid development of wide-area measurement systems, it has become possible to obtain sufficient operational data to characterize the operating state of power systems. If this operational data is processed into training samples, even without knowing whether the corresponding system states are stable, it can still provide information about the data distribution for model training.

[0067] Therefore, in this embodiment of the invention, power operation data samples are added to the data samples as unlabeled operation data samples.

[0068] In this embodiment of the invention, an initial feature set can be preset in advance. The initial feature set is a set of features related to the transient process, including bus phase angle, bus voltage amplitude, line power flow active power, line power flow reactive power, power flow at key sections, line current, generator power angle, generator speed, etc.

[0069] In this embodiment of the invention, after the data samples are sorted out, all data samples can be input into the original feature set according to each feature.

[0070] S102, Perform initial feature screening on the original feature set to obtain the initial feature set.

[0071] In this embodiment of the invention, in order to reduce the computational load and improve the computational efficiency of subsequent calculations, the features in the original feature set can be pre-screened.

[0072] Specifically, labeled simulation data samples can be used to pre-screen features that are not highly correlated with the data labels based on the data labels they carry.

[0073] S103, calculate the inter-class scatter matrix and the first total scatter matrix of the simulated data samples in the feature set after initial screening.

[0074] In the embodiments of the present invention, the relevant calculation methods are well known to those skilled in the art and will not be described in detail here.

[0075] In this embodiment of the invention, the simulated data samples in the feature set after initial screening can be calculated to obtain the inter-class scatter matrix and the first total scatter matrix of the simulated data samples, so as to facilitate the subsequent Fisher scoring of each feature.

[0076] S104, perform local linear embedding on the running data samples in the feature set after initial screening to obtain the local linear representation matrix and the second total scatter matrix of the running data samples.

[0077] In this embodiment of the invention, the core idea of ​​locally linear embedding is to linearly represent a sample by several other samples in its neighborhood. Therefore, in this embodiment, power system operation data can be applied to obtain a locally linear representation matrix and a second total divergence matrix incorporating the operation data samples, thereby completing the Fisher score for each feature based on the power system operation data.

[0078] S105, Based on the cumulative variance contribution rate threshold, perform principal component analysis to reduce the dimensionality of the second total scatter matrix and obtain the first feature subset.

[0079] In this embodiment of the invention, principal component analysis can be performed on the second total scatter matrix to reduce its dimensionality. Based on the cumulative variance contribution rate threshold, several features ranked at the top can be selected to form a first feature subset.

[0080] S106, If the number of features in the first feature subset is greater than the number of features to be extracted, calculate the Fisher score corresponding to the first feature subset.

[0081] In this embodiment of the invention, the Fisher score corresponding to the first feature subset can be directly calculated to determine whether it reaches the output threshold. Alternatively, efficient incremental feature selection can be performed on the first feature subset obtained after dimensionality reduction, and the corresponding Fisher score can be calculated.

[0082] Specifically, in this embodiment of the invention, the Fisher Score feature subset scoring method is as follows:

[0083] SemiFS(X) = tr{(S st )-1S sb}=tr{(X(F+λL)XT)-1XWXT}

[0084] Where X is the low-dimensional feature matrix after local linear embedding, and Ss t and S s b represents the improved semi-supervised inter-class scatter matrix and total scatter matrix. λ is the coefficient of the manifold regularization term, L = (IM) T (IM) is the Laplace matrix of the original high-dimensional data, and I is the identity matrix. C is the number of categories, n k Let l be the total number of data in the k-th class, and l be the number of labeled samples.

[0085] S107, if the Fisher score reaches the output threshold, output the corresponding feature subset as the key feature set for transient stability of the power system.

[0086] In this embodiment of the invention, the output threshold can be preset in advance by a technician. The final output feature subset has a Fisher score that reaches the output threshold and its feature count is greater than the number of features to be extracted.

[0087] In this embodiment of the invention, the final output feature subset can be either the first feature subset or a subset of the first feature subset. Therefore, this embodiment of the invention can comprehensively utilize operational data generated during power system operation that has not undergone safety and stability assessment to determine the key feature set of power system transient stability.

[0088] Reference Figure 2 The flowchart illustrates a method for selecting key features of transient stability in a power system according to an embodiment of the present invention. The method for selecting key features of transient stability in a power system provided by the present invention may include the following steps:

[0089] S201, add data labels to the power system simulation data samples to obtain labeled simulation data samples, take the power system operation data samples as unlabeled operation data samples, and input the simulation data samples and the operation data samples as data samples into the original feature set. The data labels are based on power angle instability as the criterion.

[0090] Step S201 is similar to step S101 above, and will not be described again here.

[0091] S202, perform centering processing on all data samples of the original feature set by subtracting the mean of the simulation data samples from all data samples in the original feature set.

[0092] In this embodiment of the invention, the data samples of the original feature set are centered to ensure that the data mean is 0, which can guarantee the continuity of subsequent calculations.

[0093] Specifically, in this embodiment of the invention, the mean of the simulated data samples is subtracted from all data samples in the original feature set, so that the original data samples are shifted to the origin of the coordinate system to achieve the purpose of centering.

[0094] S203, a low-variance feature filtering method is used to perform initial feature screening on the original feature set, and the variance of each original feature of all data samples is sorted to eliminate original features with variances lower than the variance threshold.

[0095] Specifically, in this embodiment of the invention, the variance threshold can be 10. -4 This eliminates variances below 10. -4 The original features are used to form a feature set after initial screening.

[0096] S204 calculates the inter-class scatter matrix and the first total scatter matrix of the simulated data samples in the feature set after initial screening.

[0097] Step S204 is similar to step S103 above, and will not be described again here.

[0098] S205, perform local linear embedding on the running data samples in the feature set after initial screening to obtain the local linear representation matrix and the second total scatter matrix of the running data samples.

[0099] In this embodiment of the invention, local linear embedding can be performed on unlabeled samples according to formula (1) to obtain the local linear embedding matrix M;

[0100]

[0101]

[0102] M ij =0, if

[0103] Among them, z i For data points with locally linear embedding, N i ={z i1 ,z i2 ,…,z ik} for z i The neighborhood set of z j Let be the j-th nearest neighbor of zi, and N i The expression within is consistent, M ij The weight coefficient for each neighboring point, M = (M ij ) is the obtained local linear representation matrix.

[0104] S206. Based on the cumulative variance contribution rate threshold, principal component analysis is performed on the second total scatter matrix to reduce its dimensionality, thus obtaining the first feature subset.

[0105] In this embodiment of the invention, principal component analysis can be performed on the second total scatter matrix to reduce its dimensionality, and several features ranked first can be selected according to the cumulative variance contribution rate threshold to form a first feature subset.

[0106] If the cumulative contribution rate of variance is greater than the threshold and the number of features after dimensionality reduction is greater than the number of features to be extracted, then proceed to step S207.

[0107] If the cumulative variance contribution rate is greater than the threshold and the number of features after dimensionality reduction is less than or equal to the number of features to be extracted, then adjust the threshold for the cumulative variance contribution rate, continue to perform principal component analysis to reduce the dimensionality of the total scatter matrix, and calculate the eigenvalues ​​and cumulative variance contribution rate of its covariance matrix so that the number of features after dimensionality reduction is greater than the number of features to be extracted.

[0108] Specifically, in this embodiment of the invention, after data is digitized, its eigenvalues ​​can be calculated by solving the covariance matrix and then calculated using SVD singular value decomposition.

[0109] By calculating the cumulative contribution rate of variance (as shown in Formula 2), and combining it with the data model, dimensionality reduction of the data can be achieved. Based on the variance contribution rate, the final dimensionality reduction can be determined.

[0110] Cumulative contribution rate of variance:

[0111]

[0112]

[0113] Where, λ k Let ∑ be the k-th eigenvalue of the covariance matrix of the total divergence matrix, ∑ be the eigenvalue matrix, and U be the matrix consisting of columns of the corresponding eigenvectors. r Let ∑ be the eigenvector matrix corresponding to the first r eigenvalues ​​sorted from largest to smallest. r The purpose of retaining the first r eigenvalues ​​is to make the subsequent eigenvalues ​​approximately equal to 0, thereby simplifying the total scatter matrix.

[0114] In this embodiment of the invention, by setting a cumulative variance contribution rate threshold, the feature dimension after dimensionality reduction is compared with the number of features to be extracted. If the feature dimension after dimensionality reduction is greater than the number of features to be extracted, the total scatter matrix is ​​simplified and features are extracted incrementally and efficiently. Otherwise, the cumulative variance contribution rate threshold is adjusted and principal component analysis is repeated until the feature dimension after dimensionality reduction is greater than the number of features to be extracted.

[0115] S207. The total scatter matrix of the incremental addition is decomposed by SVD to optimize and accelerate the calculation of the incremental total scatter matrix, and the efficient incremental semi-supervised Fisher Score based on the manifold regularization term is obtained.

[0116] In this embodiment of the invention, in order to reduce the amount of computation, when the number of features in the first feature subset is greater than the number of features to be extracted, several features can be selected first to perform SVD decomposition on the second total scatter matrix to simplify the second total scatter matrix. Subsequently, candidate features are gradually added to obtain an incrementally added total scatter matrix, and the corresponding Fisher score is calculated until the Fisher score reaches the output threshold.

[0117] Specifically, in this embodiment of the invention, step S207 includes:

[0118] 1. Simplify the second total divergence matrix

[0119] Assuming that d features have been selected, perform SVD decomposition on the second total scatter matrix.

[0120] S st =X(F+λL)X T =XGG T X T =QQ T (4)

[0121] in, Q=XG∈R d×N .

[0122] in, for SVD decomposition, for eigenvalue matrix, for The corresponding eigenvectors are matrices composed of columns, let Let Q be the SVD decomposition of Q = U∑V T Where V = Q T U∑ -1 .

[0123] 2. Determine the total divergence matrix of the acceleration increment.

[0124] When selecting the (d+1)th feature, let f be a candidate feature, and use S st =QQ T To speed up As can be seen from the feature subspace fusion method, The feature space can be derived from QQ T Feature space and vector G T f is generated.

[0125] Among them, Q T The eigenvalue decomposition of Q is known, and is Q T Q=V∑ 2 V T ,but The eigenvalue decomposition is unknown and needs to be determined, let it be denoted as .

[0126] When the number of selected features is d, the time complexity of evaluating the (d+1)th candidate feature based on the Fisher Score forward sequence feature selection method is O(min(d+1,N)). 3 ), where N is the number of samples. As d increases, the required computation time will increase. Therefore, based on the semi-supervised Fisher Score feature evaluation method proposed above, this embodiment of the invention constructs an efficient incremental semi-supervised feature selection method using feature subspace theory.

[0127]

[0128] f is a candidate feature, therefore, Q T The basis set φ of the characteristic space of Q can be derived from the columns of the orthogonal matrix V, and G T f on QQ T The residuals in the characteristic space are composed of:

[0129]

[0130] Where orth(.) is the orthogonalization function, and the basis set φ is related to the desired eigenvectors. The relationship is:

[0131]

[0132] R is the rotation matrix, which is substituted into... In the middle, we get and then

[0133]

[0134] The above formula is The eigenvalue matrix can be obtained by eigenvalue decomposition. And the eigenvector matrix R, thus Depend on Therefore, let SVD decomposition into but so The eigenvalues ​​are decomposed as follows:

[0135]

[0136] Finally, incremental data was obtained. Score:

[0137]

[0138] S208, determine whether the output threshold has been reached at this time.

[0139] If yes, proceed to step S209; otherwise, proceed to step S207.

[0140] In this embodiment of the invention, incremental data is calculated. After obtaining the score, it can be determined whether the score reaches the output threshold. If it does, the incremental data can be... The corresponding feature subset is output as the optimal feature subset.

[0141] S209 outputs the optimal feature subset and related metrics.

[0142] In this embodiment of the invention, considering the time complexity issue, the first feature subset obtained after dimensionality reduction can also be subjected to efficient incremental feature selection to finally obtain the optimal feature subset in which the Fisher score meets the output threshold and the number of features meets the required number of features. This subset serves as the key feature set for transient stability of the power system. Furthermore, the key feature set for transient stability of the power system can be further analyzed to obtain relevant indicators to evaluate the accuracy of the key feature set for transient stability of the power system.

[0143] Specifically, when the number of features in the first feature subset is greater than the number of features to be extracted, the second total divergence matrix can be simplified and features can be extracted incrementally and efficiently step by step. The incrementally added total divergence matrix is ​​decomposed by SVD to optimize and accelerate the calculation of the incremental total divergence matrix, and the corresponding Fisher score is obtained.

[0144] Therefore, in this embodiment of the invention, when the number of features in the first feature subset is greater than the number of features to be extracted, an incremental feature scoring update strategy can be adopted to gradually obtain the optimal feature subset whose number satisfies the number of features to be extracted, thereby reducing time complexity and improving computation speed.

[0145] The relevant metrics include at least: the Fisher Score of the subset, computation time, and classification accuracy on the test set.

[0146] Specifically, after determining the optimal subset, 80% of the data samples are randomly selected as the training set, and the remaining 20% ​​is used as the test set. The classification accuracy of the key feature set is determined using a 1-nearest neighbor classifier.

[0147] Based on the same inventive concept, embodiments of the present invention provide a device for selecting key features of transient stability in a power system, with reference to... Figure 3 , Figure 3 This is a schematic diagram of a power system transient stability key feature selection device provided in an embodiment of the present invention. Figure 3 As shown, the device includes:

[0148] The data sample processing module 301 is used to add data tags to power system simulation data samples to obtain labeled simulation data samples, take power system operation data samples as unlabeled operation data samples, and input the simulation data samples and the operation data samples as data samples into the original feature set. The data tags are based on power angle instability as the criterion.

[0149] The feature screening module 302 is used to perform initial feature screening on the original feature set to obtain the initial screened feature set.

[0150] The first calculation module 303 is used to calculate the simulation data samples in the feature set after initial screening, and obtain the inter-class scatter matrix and the first total scatter matrix of the simulation data samples.

[0151] The local linear embedding module 304 is used to perform local linear embedding on the running data samples in the feature set after initial screening, to obtain a local linear representation matrix and a second total divergence matrix with the running data samples added.

[0152] Principal component analysis module 305 is used to perform principal component analysis to reduce the dimensionality of the second total scatter matrix based on the cumulative variance contribution rate threshold, so as to obtain the first feature subset;

[0153] The second calculation module 306 is used to calculate the Fisher score corresponding to the first feature subset when the number of features in the first feature subset is greater than the number of features to be extracted.

[0154] The output module 307 is used to output the first feature subset as the key feature set for transient stability of the power system when the Fisher score reaches the output threshold.

[0155] Optionally, the device further includes:

[0156] The adjustment module is used to adjust the cumulative variance contribution rate threshold when the number of features in the feature subset is less than or equal to the number of features to be extracted, and to perform principal component analysis to reduce the dimensionality of the second total scatter matrix to obtain the second feature subset.

[0157] Optionally, the feature screening module 302 is specifically used for:

[0158] A device employing low-variance feature filtering performs initial feature screening on the original feature set, sorts the variance of each original feature for all data samples, and eliminates original features with variances below a variance threshold.

[0159] Optionally, the device further includes:

[0160] The centralization processing module is used to centralize all data samples in the original feature set by subtracting the mean of the simulation data samples from all data samples in the original feature set.

[0161] Optionally, the original feature set includes: bus phase angle, bus voltage amplitude, line power flow active power, line power flow reactive power, power flow at key sections, line current, generator power angle, and generator speed.

[0162] Optionally, the calculation method for the local linear embedding is as follows:

[0163]

[0164]

[0165] M ij =0, if

[0166] Among them, z i For locally linearly embedded data points, Ni = {zi1, zi2, ..., z} ik} for z i The neighborhood set of z j For z i The j-th nearest neighbor, along with N i The expression within is consistent, M ij The weight coefficient for each neighboring point, M = (M ij ) is the obtained local linear representation matrix.

[0167] Optionally, the device further includes:

[0168] The verification module is used to randomly select 80% of the data samples as the training set and the remaining 20% ​​as the test set, and use a 1-nearest neighbor classifier to determine the classification accuracy of the key feature set.

[0169] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0170] Based on the same inventive concept, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the power system transient stability key feature selection method described in any of the above embodiments.

[0171] Based on the same inventive concept, embodiments of the present invention provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps in the power system transient stability key feature selection method described in any of the above embodiments.

[0172] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0173] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0174] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0175] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0176] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0177] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

[0178] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0179] The present invention has provided a detailed description of a method, apparatus, electronic device, and storage medium for selecting key transient stability features of a power system. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for selecting a key feature of power system transient stability, characterized in that, The method comprises: adding data labels to power system simulation data samples to obtain labeled simulation data samples, taking power system operation data samples as unlabeled operation data samples, inputting the simulation data samples and the operation data samples into an original feature set as data samples, and taking power angle instability as a criterion for the data labels; performing initial screening on the original feature set to obtain a screened feature set; calculating the simulation data samples in the screened feature set to obtain an inter-class scatter matrix of the simulation data samples and a first total scatter matrix; performing local linear embedding on the operation data samples in the screened feature set to obtain a local linear representation matrix and a second total scatter matrix with the operation data samples added; performing principal component analysis dimension reduction on the second total scatter matrix according to a variance cumulative contribution rate threshold to obtain a first feature subset; in a case where the number of features in the first feature subset is greater than the number of features to be extracted, calculating a Fisher score corresponding to the first feature subset; in a case where the Fisher score reaches an output threshold, outputting a corresponding feature subset as a power system transient stability key feature set.

2. The method of claim 1, wherein, The method further comprises: in a case where the number of features in the feature subset is less than or equal to the number of features to be extracted, adjusting the variance cumulative contribution rate threshold, performing principal component analysis dimension reduction on the second total scatter matrix to obtain a second feature subset.

3. The method of claim 1, wherein, The initial screening on the original feature set comprises: performing initial screening on the original feature set by using a low-variance feature filtering method, and sorting the variances of each original feature of all data samples to eliminate original features with variances lower than a variance threshold.

4. The method of claim 1, wherein, Before the initial screening on the original feature set, the method further comprises: performing centering processing on all data samples in the original feature set by subtracting the mean value of the simulation data samples from all data samples in the original feature set.

5. The method of claim 1, wherein, The original feature set comprises bus phase angles, bus voltage amplitudes, line power flow active powers, line power flow reactive powers, key section power flows, line currents, generator power angles and generator speeds.

6. The method of claim 1, wherein, The calculation method of the local linear embedding is: M ij = 0, if where z i is the data point of local linear embedding, Ni={zi1,zi2,…,z ik i} is the neighborhood set of z i , zi is the jth neighbor of z j i, consistent with the expression in Ni, Mij is the weight coefficient corresponding to each neighbor, and M=(Mij) is the obtained local linear representation matrix.

7. The method of claim 1, wherein, The method further comprises: randomly selecting 80% of the data samples as a training set and the remaining 20% as a test set, and using a 1-nearest neighbor classifier to determine the classification accuracy of the key feature set.

8. An apparatus for selecting a critical feature of power system transient stability, characterized by, The device comprises: a data sample processing module configured to add data labels to power system simulation data samples to obtain labeled simulation data samples, take power system operation data samples as unlabeled operation data samples, input the simulation data samples and the operation data samples into an original feature set as data samples, and take power angle instability as a criterion for the data labels; an initial screening module configured to perform initial screening on the original feature set to obtain a screened feature set; a first calculation module configured to calculate the simulation data samples in the screened feature set to obtain an inter-class scatter matrix of the simulation data samples and a first total scatter matrix; The local linear embedding module is configured to perform local linear embedding on the operation data samples in the post-screening feature set to obtain a local linear representation matrix and a second total scatter matrix with the operation data samples added; The principal component analysis module is configured to perform principal component analysis dimension reduction on the second total scatter matrix according to a variance cumulative contribution rate threshold to obtain a first feature subset; The second calculation module is configured to calculate a Fisher score corresponding to the first feature subset in a case where the number of features in the first feature subset is greater than the number of features to be extracted. The output module is configured to output the first feature subset as the power system transient stability key feature set in a case where the Fisher score reaches an output threshold.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the power system transient stability key feature selection method according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps of the power system transient stability key feature selection method according to any one of claims 1-7.

Images