Key element analysis method for transient power angle stability in whole fault process of power system

Abstract

The invention discloses a key element analysis method for transient power angle stability in the whole fault process of a power system, and belongs to the technical field of power systems. Comprising the following steps: decomposing a constructed dynamic feature matrix into a low-rank global coherent component and a sparse local abnormal component, and extracting main modal features of a generated core feature sequence; performing characteristic decomposition on the main modal characteristics in the linear potential space to obtain modal parameters, and integrating the sparse local abnormal components, the core characteristic sequence and the modal parameters into a time sequence input vector; obtaining a transient power angle instability score and a judgment result of the operation state of the power system; calculating a mutual information value between the generated feature time sequence statistics and a preset system stability label, and a regression weight of the key element, and fusing the mutual information value and the regression weight to obtain a comprehensive influence index; and analyzing the key element according to the comprehensive influence index to obtain a key element analysis result. According to the method, high-dimensional data can be processed, the stable state can be evaluated, and key elements can be quantified.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method for analyzing key elements of transient power angle stability throughout the entire fault process of a power system. Background Technology

[0002] Transient power angle stability of a power system is a core foundation for ensuring the safe and reliable operation of the power grid. When the system encounters large disturbances such as short-circuit faults, large generator disconnection, or sudden load switching, key electrical quantities such as voltage, current, generator power angle, and system frequency will fluctuate and oscillate drastically within seconds or even milliseconds. Currently, the widely deployed Supervisory Control and Data Acquisition (SCADA) systems and phasor measurement units (PMUs) provide massive amounts of high-dimensional, multi-source time-series measurement data to capture these dynamic processes.

[0003] However, these data are not only highly dimensional and interconnected, but also inevitably contain measurement noise, communication errors, and outliers caused by local equipment failures (such as bus voltage collapse or individual unit loss of synchronization), posing a significant challenge to direct feature extraction and stability analysis. Traditional transient stability analysis methods heavily rely on time-domain analysis based on physical simulation, resulting in enormous computational overhead and difficulty in real-time applications. Their stability determination often depends on engineers' experience, manually setting thresholds by observing whether the power angle curve diverges, lacking objective and quantitative standards. More importantly, existing methods struggle to automatically and accurately identify the key factors causing system instability or the most sensitive factors from the data (e.g., which generators lost synchronization, which line's power angle difference is most critical, and what the dominant oscillation mode parameters are). This is crucial for a deep understanding of instability mechanisms, optimizing the accuracy of safety control strategies for generator and load shedding, and designing novel damping control devices. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a key element analysis method for transient power angle stability throughout the entire process of power system faults. This method can automatically process high-dimensional data, evaluate the stable state in real time, and quantify the key elements of transient power angle stability throughout the entire process of power system faults.

[0005] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0006] On the one hand, this invention provides a key element analysis method for transient power angle stability throughout the entire fault process of a power system, including:

[0007] Based on the acquired full-time-domain measurement information of the power system, a dynamic feature matrix of the entire fault process is constructed;

[0008] The dynamic feature matrix is ​​decomposed into low-rank global coherent components and sparse local anomaly components. A core feature sequence is generated based on the low-rank global coherent components, and the main modal features of the core feature sequence are extracted.

[0009] The main modal features are mapped to a linear latent space, and the main modal features in the linear latent space are decomposed to obtain modal parameters. The sparse local anomaly components, core feature sequences, and modal parameters are integrated into a temporal input vector.

[0010] The transient power angle instability score is obtained based on the time-series input vector, and the power system operating state is determined based on the transient power angle instability score.

[0011] The time-series input vector and transient power angle instability score are aggregated to generate characteristic time-series statistics. Based on the power system operation status discrimination results, the mutual information value between the characteristic time-series statistics and the preset system stability label, as well as the regression weight of key elements, are calculated. The mutual information value and regression weight are fused to obtain a comprehensive impact index.

[0012] Based on the comprehensive impact index, the key factors for transient power angle stability of the power system are ranked and quantified to obtain the key factor analysis results.

[0013] Optionally, the dynamic feature matrix of the entire fault process is represented as:

[0014] ;

[0015] in, A dynamic feature matrix representing the entire fault process; , , Representing time 1, time t, respectively The dynamic eigenvalue submatrix at time step; Indicates matrix transpose; This represents the magnitude of the bus voltage at time t; The phase angle of the bus voltage at time t is represented. This represents the amplitude of the line current at time t; The phase angle of the line current at time t; This represents the generator rotor power angle at time t; This represents the generator angular velocity offset at time t.

[0016] Optionally, the dynamic feature matrix can be decomposed into low-rank global coherence components and sparse local anomaly components, as shown in the following formula:

[0017] ;

[0018] The core feature sequence is represented as follows:

[0019] ;

[0020] The dominant modal feature is represented as follows:

[0021] ;

[0022] in, This represents the optimal solution for the low-rank global coherence component matrix; This represents the optimal solution for the sparse local outlier component matrix; Let L and S represent the solutions that yield the optimal values ​​for the objective function; Represents the low-rank global coherence component matrix; Represents a sparse local anomaly component matrix; Indicates the balancing weights; , , These represent the L1 norm, nuclear norm, and F norm, respectively. Indicates constraints; A dynamic feature matrix representing the entire fault process; Indicates noise tolerance; This represents the power angle difference of the generator group at time t; This represents the difference matrix used to construct the angular difference between the two generators; This represents the selection matrix used to extract the generator power angle; Let represent the low-rank global coherence component matrix at time t; This represents the phase angle difference of the bus voltage at time t; This represents the difference matrix used to construct the angle difference between two generatrices; This represents the selection matrix used to extract the bus phase angle; Indicates matrix transpose; Describes a left singular vector matrix, containing The modal direction; represents the singular value matrix, with diagonal elements indicating the importance of modes; V represents the right singular vector matrix. Represents the first r left singular vectors; This indicates taking the first r columns of U, corresponding to the r most important modes; This represents the dominant mode feature at time k.

[0023] Optionally, the principal modality features are mapped to a linear latent space, as shown in the following formula:

[0024] ;

[0025] The modal parameters are expressed as follows:

[0026] ;

[0027] The timing input vector is represented as follows:

[0028] ;

[0029] in, , Let represent the potential state vectors at time k and time k+1, respectively; Represents the dominant mode features of the original space at time k. Represents the Koopman linear evolution matrix; Indicates the noise margin at time k; The dominant mode features in the linear latent space of the reconstruction space at time k are represented. , These represent the encoder neural network and the decoder neural network, respectively. This represents the eigenvalue of the i-th modal feature; The damping factor represents the characteristic of the i-th mode; Represents the oscillation angular frequency of the i-th modal characteristic; Indicates the sampling time step; The damping ratio represents the characteristic of the i-th mode. This represents the modal amplitude of the i-th modal feature; The unit vector representing the i-th modal feature; Represents the mode matrix; Indicates the instant of fault clearing The potential state vector; Indicates matrix transpose; This represents the time-series input vector at time k; This represents the power angle difference of the generator group at time k; This represents the phase angle difference of the bus voltage at time k; Let represent the sparse local anomaly component matrix at time k; This represents the oscillation frequency of the i-th modal feature.

[0030] Optionally, obtaining the transient power angle instability score based on the time-series input vector includes:

[0031] The time-series input vector is input into the S4 time-series model, and the transient power angle instability score is output.

[0032] The transient power angle instability score is expressed as follows:

[0033] ;

[0034] in, This represents the internal hidden state at time k; This represents the time-series input vector at time k; Represents the S4 sequence model; The transient power angle instability score at time k is represented. , This represents the error weight and error bias.

[0035] 6. The key element analysis method for transient power angle stability throughout the entire fault process of a power system according to claim 1, characterized in that, the judgment result of the power system operating state is obtained based on the transient power angle instability score, including:

[0036] The critical moment closest to the stability boundary during the entire process of a power system fault is identified based on the transient power angle instability score. Based on the critical moment and the score threshold, the judgment result of the power system operating state is obtained.

[0037] If the critical moment is lower than the scoring threshold, the determination result of the power system operating state is stable;

[0038] If the critical moment is not lower than the scoring threshold, the determination result of the power system operating state is unstable.

[0039] Optionally, the critical moment is represented as:

[0040] ;

[0041] The determination result of the power system operating status is expressed as follows:

[0042] ;

[0043] in, Indicates the critical moment; Indicates to make The solution for k that yields the optimal value; The transient power angle instability score at time k is represented. The result indicates the determination of the operating status of the power system; Indicates the critical moment Transient work angle instability score; Indicates the scoring threshold; This indicates that the value within the box can be retrieved.

[0044] Optionally, it also includes jointly optimizing the transient power angle instability score and the power system operating state discrimination result, as shown in the formula:

[0045] ;

[0046] in, The parameters representing the S4 sequence model Error weights Error bias Joint minimization; The transient power angle instability score at time k is represented. The simulation reference score at time k is represented. The result indicates the determination of the operating status of the power system; This indicates historically stable annotation results; This represents the binary cross-entropy loss; , Indicates the loss weighting coefficient; This represents the L2 norm.

[0047] Optionally, the characteristic time-series statistics are expressed as:

[0048] ;

[0049] The mutual information value between the characteristic time-series statistics and the preset system stability label is represented as follows:

[0050] ;

[0051] The regression weights of the key elements are expressed as follows:

[0052] ;

[0053] The comprehensive impact index is expressed as follows:

[0054] ;

[0055] in, This represents the time series statistic of the j-th feature; express The temporal input vector at each time step; express The transient work angle instability score at time t; Represents the aggregation function for the j-th feature time series statistic; This represents the mutual information value between the j-th feature time series statistic and the preset system stability label; This represents a neural network estimator; The result indicates the determination of the operating status of the power system; Represents the parameters of the neural network estimator Find the optimal value; This indicates that the expected value is to be obtained within the box; Indicates the regression weights; This represents the regression weights before optimization that bring the objective function to its optimal value. The solution; This represents the regression weights of the j-th feature before optimization of the time series statistics; , Represents the regularity coefficient; , Let L1 norm and L2 norm be represented respectively; The comprehensive impact index represents the time series statistic of the j-th feature. Indicates the fusion weights; This represents the regression weight of the time series statistic of the j-th feature; Represents a set of key elements; Indicates all that satisfy The set consisting of the j-th characteristic time series statistics belonging to the K highest values.

[0056] On the other hand, the present invention provides a computer system comprising:

[0057] Memory, used to store computer programs / instructions;

[0058] A processor is used to execute the computer program / instructions to implement the steps of the key element analysis method for transient power angle stability during the entire process of power system faults as described in the first aspect.

[0059] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0060] This invention compresses high-dimensional data of the entire fault process into stable and interpretable dynamic master mode features, revealing the main energy transfer and phase evolution laws of the system. Through feature decomposition and integration, a comprehensive time-series input vector characterizing the multi-dimensional dynamic behavior of the system is formed, providing accurate data support for transient power angle instability scoring. On this basis, real-time and accurate mapping from measurement data to stability judgment is realized. Finally, the key factors affecting transient power angle stability are automatically quantified and sorted for analysis, providing direct and effective decision-making basis for system stability analysis, protection setting optimization and transient control strategy design, thereby improving the safe and stable operation level of the power system. Attached Figure Description

[0061] Figure 1 The diagram shown is a flowchart of one embodiment of the key element analysis method for transient power angle stability during the entire fault process of the power system according to the present invention. Detailed Implementation

[0062] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0063] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0064] Example 1

[0065] like Figure 1 As shown in the figure, this embodiment introduces a key element analysis method for transient power angle stability during the entire fault process of a power system, including the following steps:

[0066] Step 1: Dynamic feature extraction of the entire power system fault process by integrating robust principal component analysis and singular value decomposition, specifically as follows:

[0067] After a power system fault occurs, such as a line short circuit, generator disconnection, or sudden load change, key electrical quantities such as voltage, current, power angle, and frequency will fluctuate drastically within a short period of time. To comprehensively characterize the dynamic characteristics of the system throughout the entire fault process, firstly, full-time domain measurement information of the power system is obtained from the Supervisory Control and Data Acquisition (SCADA) system and the Phasor Measurement Unit (PMU). Based on this full-time domain measurement information, a dynamic characteristic matrix covering the entire fault process—before the fault occurs, during fault clearing, and during transient recovery—is constructed, represented as follows:

[0068] ;

[0069] in, A dynamic feature matrix representing the entire fault process; , , Representing time 1, time t, respectively The dynamic eigenvalue submatrix at time step; Indicates matrix transpose; This represents the magnitude of the bus voltage at time t; The phase angle of the bus voltage at time t is represented. This represents the amplitude of the line current at time t; The phase angle of the line current at time t; This represents the generator rotor power angle at time t; This represents the generator angular velocity offset at time t.

[0070] Then, to address noise interference and local anomalies in the data, such as local bus voltage collapse or individual unit out-of-synchronization, the entire time series data is decomposed into two parts based on the Robust Principal Component Analysis (Robust PCA) algorithm: first, a low-rank global coherent component, reflecting the overall synchronous oscillation characteristics of the system; and second, a sparse local anomaly component, characterizing the abnormal response of individual regions or units. The formula is as follows:

[0071] ;

[0072] in, This represents the optimal solution for the low-rank global coherence component matrix; This represents the optimal solution for the sparse local outlier component matrix; Let L and S represent the solutions that yield the optimal values ​​for the objective function; Represents the low-rank global coherence component matrix; Represents a sparse local anomaly component matrix; Indicates the balancing weights; , , Let L1 norm (sparse local anomaly), kernel norm (restricted system order), and F norm be represented respectively; Indicates constraints; A dynamic feature matrix representing the entire fault process; Indicates noise tolerance.

[0073] Subsequently, the generator power angle difference and bus voltage phase angle difference are extracted from the low-rank global coherent components to form a core feature sequence reflecting the system's phase consistency and power angle dynamics, represented as:

[0074] ;

[0075] in, This represents the power angle difference of the generator group at time t; This represents the difference matrix used to construct the angular difference between the two generators; This represents the selection matrix used to extract the generator power angle; Let represent the low-rank global coherence component matrix at time t; This represents the phase angle difference of the bus voltage at time t; This represents the difference matrix used to construct the angle difference between two generatrices; This represents the selection matrix used to extract the bus phase angle; This indicates the matrix transpose.

[0076] Finally, based on Singular Value Decomposition (SVD), the dominant modal features of the core feature sequence are further extracted to provide high-quality input for subsequent dynamic modeling. The dominant modal features are represented as follows:

[0077] ;

[0078] in, Describes a left singular vector matrix, containing The modal direction; represents the singular value matrix, with diagonal elements indicating the importance of modes; V represents the right singular vector matrix. Represents the first r left singular vectors; This indicates taking the first r columns of U, corresponding to the r most important modes; This represents the dominant mode feature at time k.

[0079] This step compresses the high-dimensional, multi-source measurement information of the power system into stable, interpretable dynamic master mode characteristics, reflecting the main energy transfer and phase evolution processes throughout the fault process.

[0080] Step 2: Transient mode identification based on deep Koopman operators, specifically:

[0081] After acquiring the main modal characteristics of the entire process, in order to reveal the inherent oscillation characteristics and stability mechanism of the power system in the transient stage after a fault, the transient modes are identified based on the deep Koopman operator learning algorithm. First, the nonlinear dynamic main modal characteristics of the system are mapped to a latent linear space through a nonlinear neural network, making the complex power angle oscillation process approximately linear in this space, as shown in the formula:

[0082] ;

[0083] in, , Let represent the potential state vectors at time k and time k+1, respectively; Represents the dominant mode features of the original space at time k. Represents the Koopman linear evolution matrix; Indicates the noise margin at time k; The dominant mode features in the linear latent space of the reconstruction space at time k are represented. , These represent the encoder neural network and the decoder neural network, respectively.

[0084] Then, the learned Koopman linear operator is used to perform eigenvalue decomposition on the principal mode features in the linear latent space, thereby extracting transient mode parameters such as eigenvalues, oscillation frequencies, damping ratios, and initial amplitudes of each mode. The mode parameters are expressed as follows:

[0085] ;

[0086] in, This represents the eigenvalue of the i-th modal feature; The damping factor represents the characteristic of the i-th mode; Represents the oscillation angular frequency of the i-th modal characteristic; Indicates the sampling time step; The damping ratio represents the characteristic of the i-th mode. The modal amplitude represents the i-th modal characteristic, characterizing the energy intensity of each mode in the early stage of the fault; The unit vector representing the i-th modal feature; Represents the mode matrix; Indicates the instant of fault clearing The potential state vector; This indicates the matrix transpose.

[0087] The above modal parameters reflect the synchronous oscillation relationship and energy decay rate among the various units of the system after a fault. When the damping ratio of a certain mode is too low or the real part of the eigenvalue is positive, it means that there is a risk of instability in the power angle of the system.

[0088] Finally, the above modal parameters are fused with the sparse local anomaly components extracted in step one, as well as core features such as power angle difference and phase angle difference, to form a comprehensive time-series input vector characterizing the multidimensional dynamic behavior of the system. This provides accurate data support for subsequent transient stability scoring. The time-series input vector is represented as follows:

[0089] ;

[0090] in, This represents the time-series input vector at time k; This represents the power angle difference of the generator group at time k; This represents the phase angle difference of the bus voltage at time k; Let represent the sparse local anomaly component matrix at time k; This represents the oscillation frequency of the i-th modal feature.

[0091] The timing input vector comprehensively represents three types of information in the power grid during the entire fault process: energy flow, phase angle synchronization, and oscillation characteristics.

[0092] Step 3: Transient power angle stability scoring and discrimination based on the S4 sequence model, specifically as follows:

[0093] Intelligent modeling of time-series input vectors based on the Structured State Space Sequence (S4) model can effectively capture the spatiotemporal correlation characteristics of power systems on a scale of several seconds to tens of seconds, and capture the system state change patterns over long time spans. It is particularly suitable for non-stationary oscillation sequences in the transient phase after a fault. First, the S4 sequence model takes the time-series input vector generated in step three as input, gradually learns the temporal evolution pattern of the system state, and outputs the transient power angle instability score corresponding to each moment, expressed as:

[0094] ;

[0095] in, This represents the internal hidden state at time k; This represents the time-series input vector at time k; Represents the S4 sequence model; The transient power angle instability score at time k is represented. , This represents the error weight and error bias.

[0096] A higher transient power angle instability score indicates that the system has poorer power angle synchronization and a higher risk of instability at that moment.

[0097] Then, by analyzing the transient power angle instability score sequence, the critical moment when the system is closest to the stability boundary during the entire fault process can be identified, and based on this critical moment, it can be determined whether the system has crossed the transient stability threshold, that is, whether the system operating state judgment result is stable.

[0098] If the critical moment is below the scoring threshold, the power system's operating state is determined to be stable; if the critical moment is not below the scoring threshold, the power system's operating state is determined to be unstable. This is equivalent to automatically identifying the "critical stability point," eliminating the need for manual setting of the power angle limit. The critical moment is expressed as:

[0099] ;

[0100] The result of determining the operating status of the power system is expressed as follows:

[0101] ;

[0102] in, Indicates the critical moment; Indicates to make The solution that yields the optimal value; The transient power angle instability score at time k is represented. The result indicates the operating status of the power system, with 1 indicating stability and 0 indicating instability. Indicates the critical moment Transient work angle instability score; Indicates the scoring threshold; This indicates that the value inside the box is 1 if the condition inside the box is met, and 0 if the condition inside the box is not met.

[0103] The regression error (the difference between the predicted transient power angle instability score and the target transient power angle instability score) and the classification error (the stability of the discrimination result) are jointly optimized to achieve end-to-end training. This step completes the closed loop from data to transient stability score to discrimination decision, and the formula is:

[0104] ;

[0105] in, The parameters representing the S4 sequence model Error weights Error bias Joint minimization; The transient power angle instability score at time k is represented. The simulation reference score at time k is represented. The result indicates the determination of the operating status of the power system; This indicates historically stable annotation results; This represents the binary cross-entropy loss; , Indicates the loss weighting coefficient; This represents the L2 norm.

[0106] This step enables intelligent mapping from measurement data to stability assessment, allowing the power system to identify the power angle stability state in real time during the dynamic evolution of faults, thus providing a basis for decision-making in defense control.

[0107] Step 4: Analysis of key transient stability factors based on MINE and elastic mesh, specifically:

[0108] To further analyze the core factors affecting transient power angle stability, based on the aforementioned stability score results, a method combining Mutual Information Neural Estimator (MINE) and elastic network regression was adopted to explore the correspondence between key features and system stability.

[0109] First, the time-series input vector and transient power angle instability score are statistically aggregated to generate characteristic time-series statistics, such as maximum power angle difference, frequency fluctuation rate, and oscillation energy, which provide a basis for subsequent feature importance assessment. The characteristic time-series statistics are expressed as follows:

[0110] ;

[0111] in, This represents the time series statistic of the j-th feature; express The temporal input vector at each time step; express The transient work angle instability score at time t; This represents the aggregation function for the j-th feature time series statistic.

[0112] Then, the mutual information value between the feature time-series statistics and the preset system stability label is calculated using the MINE model to measure the influence of the feature on the stability results. The larger the mutual information value, the more significant the influence of the feature on the transient power angle stability determination. The mutual information value is expressed as:

[0113] ;

[0114] in, This represents the mutual information value between the j-th feature time series statistic and the preset system stability label; This represents a neural network estimator; The result indicates the determination of the operating status of the power system; Represents the parameters of the neural network estimator Find the optimal values ​​and optimal neural network parameters; This indicates that the desired value is to be obtained within the box.

[0115] Next, feature sparsity screening is achieved through elastic network regression to automatically identify the core factors most sensitive to transient power angle stability throughout the fault process. The results can be directly correlated with physical elements such as critical lines and critical units. The regression weights of the key elements are expressed as follows:

[0116] ;

[0117] in, Indicates the regression weights; This represents the regression weights before optimization that bring the objective function to its optimal value. The solution; This represents the regression weights of the j-th feature before optimization of the time series statistics; , Represents the regularity coefficient; , Let L1 and L2 be the norms, respectively.

[0118] Finally, the mutual information value and regression weights are combined to form a comprehensive impact index, which is expressed as:

[0119] ;

[0120] in, The comprehensive impact index represents the time series statistic of the j-th feature. Indicates the fusion weights; This represents the regression weight of the time series statistic of the j-th feature; Represents a set of key elements; Indicates all that satisfy The set consists of the j-th characteristic time-series statistics that belong to the K highest values. Based on the comprehensive impact index, the key elements of transient power angle stability of the power system (such as key generator pairs, key line power angle differences, main oscillation mode parameters, etc.) are ranked and quantified to obtain the key element analysis results. These results can be directly used for system stability analysis, protection setting optimization, and transient control strategy design.

[0121] Example 2

[0122] This embodiment describes a computer system, including:

[0123] Memory, used to store computer programs / instructions;

[0124] A processor is used to execute the computer program / instructions to implement the steps of the key element analysis method for transient power angle stability during the entire process of power system faults as described in Embodiment 1.

[0125] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0126] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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 apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function 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.

[0128] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable 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.

[0129] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method of analyzing key elements of power system fault full process transient angle stability, characterized in that, The method comprises the following steps: According to the acquired power system full-time domain measurement information, a dynamic characteristic matrix of the whole process of the fault is constructed; the dynamic characteristic matrix is decomposed into a low-rank global coherent component and a sparse local abnormal component, a core feature sequence is generated according to the low-rank global coherent component, and a principal modal feature of the core feature sequence is extracted; The principal modal feature is mapped to a linear latent space, and the principal modal feature in the linear latent space is subjected to feature decomposition to obtain modal parameters; the sparse local abnormal component, the core feature sequence and the modal parameters are integrated into a time sequence input vector; A transient power angle instability score is obtained according to the time sequence input vector, and a discrimination result of the power system operating state is obtained according to the transient power angle instability score; The time sequence input vector and the transient power angle instability score are aggregated to generate a characteristic time sequence statistic, the mutual information value between the characteristic time sequence statistic and a preset system stability label and the regression weight of a key element are calculated according to the discrimination result of the power system operating state, and a comprehensive influence index is obtained by fusing the mutual information value and the regression weight; The key elements of the transient power angle stability of the power system are sorted and quantified according to the comprehensive influence index, and a key element analysis result is obtained.

2. The method of claim 1, wherein the key factor analysis of the power system fault full process transient angle stability is characterized by, The dynamic characteristic matrix of the whole process of the fault is represented as: ； wherein, represents the dynamic characteristic matrix of the whole fault process; , , respectively represent the dynamic characteristic sub-matrix at time 1, time t, time t; represents the matrix transpose; represents the bus voltage amplitude at time t; represents the bus voltage phase angle at time t; represents the line current amplitude at time t; represents the line current phase angle at time t; represents the generator rotor power angle at time t; represents the generator angular velocity offset at time t.

3. The method of claim 1, wherein the key factor analysis of the power system fault full process transient angle stability is characterized by, The dynamic characteristic matrix is decomposed into a low-rank global coherent component and a sparse local abnormal component, and the formula is: ； The core feature sequence is represented as: ； The principal modal feature is represented as: ； wherein, represents the optimal solution of the low-rank global coherent component matrix; represents the optimal solution of the sparse local abnormal component matrix; represents the solution of L and S when the objective function reaches the optimal value; represents the low-rank global coherent component matrix; represents the sparse local abnormal component matrix; represents the balance weight; , , respectively represent the L1 norm, the kernel norm, and the F norm; represents the constraint; represents the dynamic characteristic matrix of the whole process of the fault; represents the noise tolerance; represents the generator group power angle difference at time t; represents the difference matrix for constructing the angle difference between two generators; represents the selection matrix for extracting the generator power angle; represents the low-rank global coherent component matrix at time t; represents the bus voltage phase angle difference at time t; represents the difference matrix for constructing the angle difference between two buses; represents the selection matrix for extracting the bus phase angle; represents the matrix transpose; represents the left singular vector matrix, containing the modal direction; represents the singular value matrix, and the diagonal elements represent the importance of the modal; represents the right singular vector matrix; represents the first r left singular vectors; represents the first columns of , corresponding to the most important modes; represents the main modal feature at time k.

4. The method of claim 1, wherein the key factor analysis of the power system fault full- process transient angle stability is characterized by, The principal modal feature is mapped to a linear latent space, and the formula is: ； The modal parameters are represented as: ； The time sequence input vector is represented as: ； wherein, , respectively represent the latent state vector at time k, k+1; represent the dominant modal feature in the original space at time k; represent the Koopman linear evolution matrix; represent the noise tolerance at time k; represent the dominant modal feature in the linear latent space of the reconstruction space at time k; , respectively represent the encoder neural network, the decoder neural network; represent the eigenvalue of the i-th modal feature; represent the damping factor of the i-th modal feature; represent the oscillation angular frequency of the i-th modal feature; represent the sampling time step; represent the damping ratio of the i-th modal feature; represent the modal amplitude of the i-th modal feature; represent the unit vector of the i-th modal feature; represent the modal matrix; represent the latent state vector at the fault clearing instant ; represent the matrix transpose; represent the time series input vector at time k; represent the generator group power angle difference at time k; represent the bus voltage phase angle difference at time k; represent the sparse local anomaly component matrix at time k; represent the oscillation frequency of the i-th modal feature.

5. The method of claim 1, wherein the key factor analysis of power system fault full- process transient angle stability is characterized by, The transient power angle instability score is obtained according to the time sequence input vector, which comprises: The time sequence input vector is input into an S4 time sequence model to output the transient power angle instability score; The transient power angle instability score is represented as: ； wherein, represents the internal hidden state at time k; represents the time series input vector at time k; represents the S4 sequence model; represents the transient power angle instability score at time k; , represents the error weight, error bias.

6. The method of claim 1, wherein the key factor analysis of the power system fault full- process transient angle stability is characterized by, The discrimination result of the power system operating state is obtained according to the transient power angle instability score, which comprises: The critical time closest to the stability boundary in the whole process of the power system fault is identified according to the transient power angle instability score, and the discrimination result of the power system operating state is obtained according to the critical time and a score threshold; If the critical time is lower than the score threshold, the discrimination result of the power system operating state is stable; If the critical time is not lower than the score threshold, the discrimination result of the power system operating state is unstable.

7. The method of claim 6, wherein the key factor analysis of the power system fault full- process transient angle stability is characterized by, The critical time is represented as: ； The discrimination result of the power system operating state is represented as: ； wherein, denotes a critical time instant; denotes making the solution of k for which the optimum is attained; denotes the transient power angle instability score at k; denotes the result of the discrimination of the operating state of the power system; denotes the transient power angle instability score at the critical time instant ; denotes a score threshold value; denotes taking the value within the box.

8. The method of claim 6, wherein the key factor analysis of the power system fault full- process transient angle stability is characterized by, The transient power angle instability score and the discrimination result of the power system operating state are jointly optimized, and the formula is: ； wherein, parameters of the S4 sequence model error weight error bias jointly minimizing; transient power angle instability score at time k; simulation reference score at time k; discrimination result of the power system operating state; historical labeled stability result; binary cross-entropy loss; , loss weighting coefficient; L2 norm.

9. The method of claim 1, wherein the key factor analysis of power system fault full- process transient angle stability is characterized by, The characteristic time sequence statistic is represented as: ； The mutual information value between the characteristic time sequence statistic and the preset system stability label is represented as: ； The regression weight of the key element is represented as: ； The comprehensive influence index is represented as: ； wherein, denotes the jth feature time series statistics; denotes the time series input vector at time t; denotes the transient angle instability score at time t; denotes the aggregation function of the jth feature time series statistics; denotes the mutual information value between the jth feature time series statistics and the preset system stability label; denotes the neural network estimator; denotes the discrimination result of the power system operating state; denotes the parameter of the neural network estimator; denotes the expectation value in the box; denotes the regression weight; denotes the solution of the regression weight before optimization when the objective function reaches the optimal value; denotes the regression weight before optimization of the jth feature time series statistics; , denotes the regularization coefficient; , denote the L1 norm and the L2 norm, respectively; denotes the comprehensive influence index of the jth feature time series statistics; denotes the fusion weight; denotes the regression weight of the jth feature time series statistics; denotes the key element set; denotes the set of the jth feature time series statistics belonging to the highest K values.

10. A computer system, characterized by The method comprises the following steps: A memory is used to store computer programs / instructions; A processor is used to execute the computer programs / instructions to realize the steps of the key element analysis method of the transient power angle stability of the whole process of the power system fault according to any one of claims 1-9.

Images