A zero leakage judgment method and system for power grid transient voltage stability and electronic equipment

Abstract

This invention discloses a zero-leakage detection method, system, and electronic equipment for transient voltage stability of a power grid. The method includes: performing simulation calculations on the power grid for preset faults to obtain voltage time-series data of the bus under different faults; calculating two-dimensional orthogonal feature vectors based on the voltage time-series data to obtain two-dimensional orthogonal feature vectors corresponding to different faults; determining two segmentation curves that divide the two-dimensional feature space into an unstable region, a fuzzy region, and a stable region, respectively serving as a baseline instability discrimination boundary and an extended instability discrimination boundary; and determining the stable state of the power grid under different faults based on the positional relationship between the two-dimensional orthogonal feature vectors, the baseline instability discrimination boundary, and the extended instability discrimination boundary. This invention uses wavelet orthogonal feature vectors with a clear mechanism as input, forming a complete and efficient technical chain from feature extraction to security discrimination. It has low computational complexity and can meet the engineering requirements for rapid batch screening of massive amounts of anticipated fault scenarios.

Description

Technical Field

[0001] This invention relates to the field of power system safety and stability analysis technology, and more specifically, to a method, system, and electronic equipment for zero-leakage judgment of power grid transient voltage stability. Background Technology

[0002] With a high proportion of new energy sources being connected to the grid via power electronic converters, the power system is evolving into a power electronic grid, and the underlying mechanisms and analytical needs for transient voltage stability problems are becoming increasingly complex. Novel instability mechanisms, such as multi-balance point switching and fast-slow dynamic decoupling based on the control dynamics of power electronic equipment, have been widely recognized. Against this backdrop, transient voltage stability assessment of the power grid, especially the rapid screening of numerous anticipated faults, has become a crucial link in ensuring the safe operation of the power grid. However, existing assessment methods still face significant challenges in addressing the core safety principle of absolutely avoiding missed faults.

[0003] Effective stability determination relies on accurate characterization of system dynamics. In recent years, research has focused on constructing feature spaces with clear mechanisms to better characterize the aforementioned instability mechanisms, such as using wavelet transforms to extract feature quantities with explicit physical meaning. However, after obtaining high-quality feature representations, designing matching discrimination strategies that strictly meet the zero-miss requirement in engineering practice and possess good interpretability remains an urgent problem to be solved.

[0004] In existing technologies, the main stability discrimination strategies can be divided into two categories: The first category is discrimination methods based on deterministic physical rules or thresholds. These methods typically set fixed thresholds for one or more characteristic quantities and draw conclusions through simple logical comparisons (e.g., instability occurs if the value exceeds the threshold). Their advantages are transparent rules and fast computation. However, in the complex and ever-changing scenarios of electronic power grids, a single fixed threshold is insufficient to maintain reasonable discrimination accuracy while ensuring zero missed detections. It often falls into a dilemma: raising the threshold to prevent missed detections leads to a surge in false alarms, while lowering the threshold to reduce false alarms increases the risk of missed detections. There is a lack of a mechanism for finely balancing and adjusting between safety and economy.

[0005] The second category is data-driven intelligent discrimination models. Models such as deep learning and support vector machines can learn complex decision boundaries from historical data and possess strong nonlinear fitting capabilities. However, the decision-making process of these models is usually a black box, with its internal logic difficult to interpret, and the discrimination conclusions cannot be directly linked to specific physical characteristics or mechanisms. In the operation of power systems where safety is paramount, this lack of interpretability seriously affects the operators' trust in the model results. Furthermore, their performance is highly dependent on the completeness and quality of the training data. When faced with extreme or novel scenarios not fully covered by the training set, their reliability is questionable, making it difficult to assume the absolute safety responsibility of zero missed judgments.

[0006] Therefore, a transient voltage stability zero-leakage detection method is needed for power electronic power grids. Summary of the Invention

[0007] This invention proposes a zero-leakage judgment method, system, and electronic equipment for power grid transient voltage stability, in order to solve the problem of how to accurately judge the stability of power grid transient voltage.

[0008] To address the aforementioned problems, according to one aspect of the present invention, a method for determining zero leakage in power grid transient voltage stability is provided, the method comprising: Simulation calculations of preset faults are performed on the power grid to obtain voltage timing data of the bus under different fault conditions; Two-dimensional orthogonal feature vectors are calculated based on the voltage time series data to obtain two-dimensional orthogonal feature vectors corresponding to different faults; Two segmentation curves are determined to divide the two-dimensional feature space into an unstable region, a fuzzy region, and a stable region. These two segmentation curves are used as the baseline instability discrimination boundary and the extended instability discrimination boundary, respectively. Based on the positional relationship between the two-dimensional orthogonal eigenvectors, the benchmark instability discrimination boundary, and the extended instability discrimination boundary, the stability state of the power grid under different faults is determined.

[0009] Preferably, the calculation of two-dimensional orthogonal feature vectors based on the voltage time-series data to obtain two-dimensional orthogonal feature vectors corresponding to different faults includes: Discrete wavelet transform is performed on the voltage time series data of the bus under each fault to decompose the voltage time series data into mutually orthogonal approximate subspace and detail subspace coefficient sequences; wherein, the approximate subspace corresponds to the low frequency component and the detail subspace corresponds to the high frequency component. Construct a two-dimensional orthogonal eigenvector [F1, F2] based on the mutually orthogonal approximate subspace and detail subspace coefficient sequences to obtain two-dimensional orthogonal eigenvectors corresponding to different faults; where F1 is the first characteristic quantity and F2 is the second characteristic quantity. F1 is used to characterize the high-frequency transient impact energy and corresponds to the energy characteristics extracted from the detail subspace; F2 is used to characterize the dynamic energy structure relationship across time scales and corresponds to the proportional relationship between the energies of the detail subspace and the approximate subspace.

[0010] Preferably, the discrete wavelet transform processing is performed on the voltage time series data of the bus under each fault to decompose the voltage time series data into mutually orthogonal approximate subspace and detail subspace coefficient sequences, including: For any fault, when performing discrete wavelet transform processing on the voltage time series data of the bus under the any fault based on the orthogonal wavelet basis function, through multi-resolution analysis, the voltage time series data of the bus under the any fault is decomposed into multi-layer mutually orthogonal approximate subspace and detail subspace coefficient sequences.

[0011] Preferably, the method further includes: Construct an energy aggregation function based on the detail subspace coefficient sequences from the first layer to the Jth layer to determine the first characteristic quantity; where J≥2 and is a positive integer, and the energy aggregation function is calculated based on the p-norm of the coefficient sequences.

[0012] Preferably, the method further includes: Construct an energy ratio function based on the selected approximate subspace coefficient sequences of different frequency bands to determine the second characteristic quantity, including: F2 = Ψ(cD_m) / Ψ(cA_n), where cD_m and cA_n respectively represent the coefficients of a high-frequency detail subspace and a low-frequency approximate subspace, m < n, and Ψ is an energy calculation operator.

[0013] Preferably, determining two segmentation curves that divide the two-dimensional feature space into an instability region, a fuzzy region, and a stable region, and using the two segmentation curves as the reference instability discrimination boundary and the extended instability discrimination boundary respectively, includes: Determine the cost function C with the goal of minimizing the discrimination risk: C = W miss , miss , fp , miss , fp , fp , , fp ,

[0013] , miss , , miss , fp *N miss +W fp *N fp , where N miss is the number of instability samples misjudged, N fp is the number of stable samples misreported, W miss and W fp are penalty weights, satisfying W miss ≥h*W fp , h is a preset multiple; Using a historical simulation sample set containing a large number of known stable state labels, with the absolute safety constraint that all unstable sample points must be located on the same side of the segmentation curve, the segmentation curve L1 that satisfies the minimum cost function is determined in the F1-F2 two-dimensional feature space, and the segmentation curve L1 is used as the benchmark instability discrimination boundary. Using a historical simulation sample set containing a large number of known stable state labels, and with the absolute safety constraint that a predetermined number of unstable sample points must be located on the same side of the segmentation curve, the segmentation curve L2 that satisfies the minimum cost function is determined in the two-dimensional feature space F1-F2, and the segmentation curve L2 is used as the extended instability discrimination boundary; where F1 is the first feature quantity and F2 is the second feature quantity.

[0014] Preferably, the determination of the power grid stability state under different faults based on the positional relationship between the two-dimensional orthogonal eigenvectors, the baseline instability discrimination boundary, and the extended instability discrimination boundary includes: For any two-dimensional orthogonal eigenvector, if the two-dimensional orthogonal eigenvector is located in the instability region defined by the baseline instability discrimination boundary, then the power grid is determined to be instable; if the two-dimensional orthogonal eigenvector is located in the stable region defined by the extended instability discrimination boundary, then the power grid is determined to be stable; if the two-dimensional orthogonal eigenvector is located in the fuzzy region formed by the extended instability discrimination boundary and the baseline instability discrimination boundary, then the power grid is determined to be instable.

[0015] Preferably, the method further includes: The benchmark instability discrimination boundary is divided into multiple segments based on the angles between the benchmark instability discrimination boundary and the horizontal and vertical axes in the feature space; wherein the horizontal and vertical axes are determined based on the first feature quantity F1 and the second feature quantity F2. For any boundary segment, calculate the angle between the boundary segment and the horizontal and vertical axes respectively. If the angle between the boundary segment and any coordinate axis is less than or equal to a preset angle threshold, then the boundary segment is determined to be dominated by the feature quantity corresponding to the coordinate axis. If the angle between the boundary segment and both coordinate axes is greater than the preset angle threshold, then the boundary segment is determined to be dominated by the mixture of the two feature quantities. For any two-dimensional orthogonal eigenvector whose stable state is unstable, calculate the signed distance from the two-dimensional orthogonal eigenvector whose stable state is unstable to each boundary segment; The boundary segment with the largest signed distance is selected as the target boundary segment, and the dominant feature quantity corresponding to the target boundary segment is used as the stability criterion identifier for the two-dimensional orthogonal feature vector of any stable state being unstable. According to another aspect of the present invention, a zero-leakage judgment system for power grid transient voltage stability is provided, the system comprising: The data acquisition unit is used to perform simulation calculations on the power grid for preset faults and acquire voltage timing data of the bus under different faults. The feature calculation unit is used to calculate a two-dimensional orthogonal feature vector based on the voltage time series data to obtain two-dimensional orthogonal feature vectors corresponding to different faults. The instability discrimination boundary determination unit is used to determine two segmentation curves that divide the two-dimensional feature space into an unstable region, a fuzzy region, and a stable region, and to use the two segmentation curves as the baseline instability discrimination boundary and the extended instability discrimination boundary, respectively. The judgment unit is used to determine the stability state of the power grid under different faults based on the positional relationship between the two-dimensional orthogonal feature vector, the benchmark instability discrimination boundary, and the extended instability discrimination boundary.

[0016] Preferably, the feature calculation unit calculates a two-dimensional orthogonal feature vector based on the voltage time-series data to obtain two-dimensional orthogonal feature vectors corresponding to different faults, including: Discrete wavelet transform is performed on the voltage time series data of the bus under each fault to decompose the voltage time series data into mutually orthogonal approximate subspace and detail subspace coefficient sequences; wherein, the approximate subspace corresponds to the low frequency component and the detail subspace corresponds to the high frequency component. Two-dimensional orthogonal feature vectors [F1, F2] are constructed based on the coefficient sequences of mutually orthogonal approximate subspaces and detail subspaces to obtain two-dimensional orthogonal feature vectors corresponding to different faults. Among them, F1 is the first feature quantity and F2 is the second feature quantity. F1 is used to characterize the high-frequency transient impact energy, corresponding to the energy feature extracted from the detail subspace. F2 is used to characterize the dynamic energy structure relationship across time scales, corresponding to the ratio of energy between the detail subspace and the approximate subspace.

[0017] Preferably, the feature calculation unit performs discrete wavelet transform processing on the voltage time series data of the bus under each fault to decompose the voltage time series data into mutually orthogonal approximate subspace and detail subspace coefficient sequences, including: For any fault, when performing discrete wavelet transform on the voltage time series data of the bus under any fault based on orthogonal wavelet basis functions, multi-resolution analysis is used to decompose the voltage time series data of the bus under any fault into multi-layered mutually orthogonal approximate subspace and detail subspace coefficient sequences.

[0018] Preferably, the feature quantity calculation unit is further configured to: An energy aggregation function is constructed based on the detail subspace coefficient sequence from the first layer to the Jth layer to determine the first characteristic quantity; wherein J≥2 and is a positive integer, and the energy aggregation function is calculated based on the p-order norm of the coefficient sequence.

[0019] Preferably, the feature quantity calculation unit is further configured to: Construct an energy ratio function based on the selected approximate subspace coefficient sequences of different frequency bands to determine a second feature quantity, including: F2 = Ψ(cD_m) / Ψ(cA_n), where cD_m and cA_n respectively represent the coefficients of a high-frequency detail subspace and a low-frequency approximation subspace, m < n, and Ψ is an energy calculation operator.

[0020] Preferably, the instability discrimination boundary determination unit determines two segmentation curves that divide the two-dimensional feature space into an instability region, a fuzzy region, and a stable region, and uses the two segmentation curves as the reference instability discrimination boundary and the extended instability discrimination boundary respectively, including: Determine a cost function C with the goal of minimizing the discrimination risk: C = W miss *N miss +W fp *N fp , where N miss is the number of instability samples misjudged, N fp is the number of stable samples with false alarms, W miss and W fp are penalty weights, satisfying W miss ≥h*W fp , h is a preset multiple; Use a historical预想故障仿真样本集 (simulated sample set of预想故障) containing a large number of known stable state labels. With the absolute safety constraint that all instability sample points must be on the same side of the segmentation curve, on the F1-F2 two-dimensional feature space, determine the segmentation curve L1 that satisfies the minimum of the cost function, and use the segmentation curve L1 as the reference instability discrimination boundary; Use a historical预想故障仿真样本集 (simulated sample set of预想故障) containing a large number of known stable state labels. With the absolute safety constraint that a preset number of instability sample points must be on the same side of the segmentation curve, on the F1-F2 two-dimensional feature space, determine the segmentation curve L2 that satisfies the minimum of the cost function, and use the segmentation curve L2 as the extended instability discrimination boundary; where F1 is the first feature quantity and F2 is the second feature quantity. <​​​For any two-dimensional orthogonal eigenvector, if the two-dimensional orthogonal eigenvector is located in the instability region defined by the baseline instability discrimination boundary, then the power grid is determined to be instable; if the two-dimensional orthogonal eigenvector is located in the stable region defined by the extended instability discrimination boundary, then the power grid is determined to be stable; if the two-dimensional orthogonal eigenvector is located in the fuzzy region formed by the extended instability discrimination boundary and the baseline instability discrimination boundary, then the power grid is determined to be instable.

[0022] Preferably, the system further includes: The identification unit divides the benchmark instability discrimination boundary into multiple segments based on the angle between the benchmark instability discrimination boundary and the horizontal and vertical axes in the feature space; wherein the horizontal and vertical axes are determined based on the first feature quantity F1 and the second feature quantity F2. For any boundary segment, calculate the angle between the boundary segment and the horizontal and vertical axes respectively. If the angle between the boundary segment and any coordinate axis is less than or equal to a preset angle threshold, then the boundary segment is determined to be dominated by the feature quantity corresponding to the coordinate axis. If the angle between the boundary segment and both coordinate axes is greater than the preset angle threshold, then the boundary segment is determined to be dominated by the mixture of the two feature quantities. For any two-dimensional orthogonal eigenvector whose stable state is unstable, calculate the signed distance from the two-dimensional orthogonal eigenvector whose stable state is unstable to each boundary segment; The boundary segment with the largest signed distance is selected as the target boundary segment, and the dominant feature quantity corresponding to the target boundary segment is used as the stability criterion identifier for the two-dimensional orthogonal feature vector whose stable state is unstable. Based on another aspect of the present invention, the present invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of any one of the zero-leakage determination methods for power grid transient voltage stability. Based on another aspect of the present invention, the present invention provides an electronic device, comprising: The aforementioned computer-readable storage medium; and One or more processors for executing a program in the computer-readable storage medium.

[0023] This invention provides a method, system, and electronic device for zero-leakage judgment of transient voltage stability in a power grid. The method includes: performing simulation calculations of the power grid under preset faults to obtain voltage time-series data of the bus under different faults; calculating two-dimensional orthogonal feature vectors based on the voltage time-series data to obtain two-dimensional orthogonal feature vectors corresponding to different faults; determining two segmentation curves that divide the two-dimensional feature space into an unstable region, a fuzzy region, and a stable region, and using the two segmentation curves as the baseline instability discrimination boundary and the extended instability discrimination boundary, respectively; and determining the stable state of the power grid under different faults based on the positional relationship between the two-dimensional orthogonal feature vectors, the baseline instability discrimination boundary, and the extended instability discrimination boundary. This invention establishes a security discrimination mechanism with zero-leakage judgment as the highest priority, using wavelet orthogonal feature vectors with a clear mechanism as input, forming a complete and efficient technical chain from feature extraction to security discrimination. It has low computational complexity, can meet the engineering requirements for rapid batch screening of massive anticipated fault scenarios, and can integrate visualization and trend analysis functions, directly serving the online security assessment and offline planning analysis of the power grid. Attached Figure Description

[0024] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures: Figure 1 This is a flowchart of a zero-leakage detection method 100 for power grid transient voltage stability according to an embodiment of the present invention; Figure 2 The following is a Python analysis diagram of the SMIB instability time-domain simulation results according to an embodiment of the present invention; Figure 3 This is a two-dimensional feature map of the bus voltage wavelet Ehf-Rhl corresponding to SMIB instability according to an embodiment of the present invention; Figure 4 The distribution diagram of the expected fault calculation examples of multiple lines of CSEE-VS according to the embodiments of the present invention in the wavelet two-dimensional feature Ehf-Rhl space; Figure 5 This is a schematic diagram of the zero-leakage detection system 500500 for power grid transient voltage stabilization according to an embodiment of the present invention. Detailed Implementation

[0025] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0026] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0027] Figure 1 This is a flowchart of a zero-leakage detection method 100 for grid transient voltage stability according to an embodiment of the present invention. Figure 1 As shown, the zero-leakage judgment method for power grid transient voltage stability provided by this invention establishes a security discrimination mechanism with zero-leakage judgment as the highest priority. It uses wavelet orthogonal feature vectors with a clear mechanism as input, forming a complete and efficient technical chain from feature extraction to security discrimination. It has low computational complexity, meets the engineering requirements for rapid batch screening of massive anticipated fault scenarios, and can integrate visualization and trend analysis functions, directly serving online security assessment and offline planning analysis of the power grid. The zero-leakage judgment method 100 for power grid transient voltage stability provided by this invention starts from step 101. In step 101, a simulation calculation of preset faults is performed on the power grid to obtain the voltage time series data of the bus under different faults. In this invention, during the data preparation and input stage, the voltage time series data of the bus for each fault scenario is obtained after simulating the anticipated fault set in a power electronic power grid under a given topology and operating mode. The data sources include offline simulation software or the online security assessment function program of the dispatch center. The data sampling rate is usually not less than 100Hz, which can characterize electromechanical transient dynamics. This step clarifies that the data object processed by the method is a known, complete simulation waveform.

[0028] In step 102, a two-dimensional orthogonal feature vector is calculated based on the voltage timing data to obtain the two-dimensional orthogonal feature vectors corresponding to different faults.

[0029] Preferably, the calculation of two-dimensional orthogonal feature vectors based on the voltage time-series data to obtain two-dimensional orthogonal feature vectors corresponding to different faults includes: Discrete wavelet transform is performed on the voltage time series data of the bus under each fault to decompose the voltage time series data into mutually orthogonal approximate subspace and detail subspace coefficient sequences; wherein, the approximate subspace corresponds to the low frequency component and the detail subspace corresponds to the high frequency component. Construct a two-dimensional orthogonal eigenvector [F1, F2] based on the coefficient sequences of mutually orthogonal approximation subspace and detail subspace to obtain two-dimensional orthogonal eigenvectors corresponding to different faults. Among them, F1 is the first feature quantity, F2 is the second feature quantity. F1 is used to characterize the high-frequency transient impact energy, corresponding to the energy feature extracted from the detail subspace; F2 is used to characterize the dynamic energy structure relationship across time scales, corresponding to the proportional relationship between the energies of the detail subspace and the approximation subspace.

[0030] Preferably, the voltage time series data of the bus under each fault is processed by discrete wavelet transform to decompose the voltage time series data into coefficient sequences of mutually orthogonal approximation subspace and detail subspace, including: For any fault, when performing discrete wavelet transform on the voltage time series data of the bus under the fault based on the orthogonal wavelet basis function, through multi-resolution analysis, the voltage time series data of the bus under the fault is decomposed into multi-layer coefficient sequences of mutually orthogonal approximation subspace and detail subspace.

[0031] Preferably, the method further includes: Construct an energy aggregation function based on the detail subspace coefficient sequences from the first layer to the Jth layer to determine the first feature quantity. Among them, J≥2 and is a positive integer, and the energy aggregation function is calculated based on the p-norm of the coefficient sequence.

[0032] Preferably, the method further includes: Construct an energy ratio function based on the selected approximation subspace coefficient sequences of different frequency bands to determine the second feature quantity, including: F2 = Ψ(cD_m) / Ψ(cA_n), where cD_m and cA_n respectively represent the coefficients of a high-frequency detail subspace and a low-frequency approximation subspace, m < n, and Ψ is an energy calculation operator.

[0033] In the present invention, in the orthogonal feature quantity calculation stage, a two-dimensional orthogonal eigenvector [F1, F2] of each预想故障场景 (anticipated fault scenario) generated based on the voltage time series data is obtained. Among them, F1 is a feature quantity characterizing the high-frequency transient impact energy (such as the normalized high-frequency detail energy), and its physical essence is related to the multi-equilibrium point switching mechanism; F2 is a feature quantity characterizing the proportional relationship between the fast and slow dynamic energies of the system (such as the normalized high-low frequency energy ratio), and its physical essence is related to the fast-slow dynamic decoupling mechanism. Obtaining these features with orthogonality and clear physical meanings is the premise for the discriminant method of the present invention to achieve high reliability and interpretability.

[0034] In the present invention, discrete wavelet transform processing is performed on the voltage time-series data of the bus under each fault to decompose the voltage time-series data into mutually orthogonal approximation subspace and detail subspace coefficient sequences, and then a two-dimensional orthogonal eigenvector [F1, F2] is constructed based on the mutually orthogonal approximation subspace and detail subspace coefficient sequences to obtain two-dimensional orthogonal eigenvectors corresponding to different faults.

[0035] Among them, the coefficient sequences are determined based on the multi-resolution decomposition of the orthogonal wavelet basis. Specifically, an orthogonal wavelet basis function with compact support and regularity (such as the Daubechies wavelet family, the Symlets wavelet family, or the Coiflets wavelet family) is used to perform discrete wavelet transform (DWT) on the transient voltage time-series signal. Through multi-resolution analysis (MRA), this transform decomposes the input signal into a set of mutually orthogonal approximation subspaces (corresponding to low-frequency components) and detail subspace (corresponding to high-frequency components) coefficient sequences.

[0036] In the present invention, based on the mapping design, a two-dimensional orthogonal eigenvector [F1, F2] is constructed according to the obtained subspace coefficient sequences, where F1 is the first eigenquantity and F2 is the second eigenquantity.

[0037] Among them, the first eigenquantity F1 is an energy aggregation function based on one or more selected high-frequency detail subspace coefficients; among them, the energy aggregation function is constructed based on the detail subspace coefficient sequences from the first layer to the Jth layer; where J≥2 and is a positive integer, and the energy aggregation function is calculated based on the p-norm of the coefficient sequences. For example, F1 can be constructed based on the detail subspace coefficients from the first layer to the Jth layer (J≥2), and its functional form can be calculated based on the p-(order) norm of the coefficient sequences (such as the sum of absolute values or the square root of the sum of squares) to quantify the intensity of high-frequency transient impacts.

[0038] Among them, the second eigenquantity F2 is an energy ratio function based on the approximate subspace coefficients selected from different frequency bands; among them, the energy ratio function is constructed based on the approximate subspace coefficient sequences of the selected different frequency bands. F2 can be constructed as F2 = Ψ(cD_m) / Ψ(cA_n), where cD_m and cA_n respectively represent the coefficients of a high-frequency detail subspace and a low-frequency approximation subspace (m < n), and Ψ is an energy calculation operator. This eigenquantity is used to quantify the energy structure relationship between the fast and slow dynamics of the system.

[0039] In an embodiment of the present invention, when the orthogonal wavelet basis function is selected as the Daubechies 5th-order wavelet (db5), the decomposition level is 5 layers, the energy aggregation function F1 is specifically the average value of the sum of squares of the detail subspace coefficients from the first to the third layer, and the energy ratio function F2 is specifically the ratio of F1 to the sum of the sum of squares of the detail subspace coefficients of the fourth and fifth layers, a specific feature space is obtained.

[0040] In step 103, two segmentation curves are determined to divide the two-dimensional feature space into an unstable region, a fuzzy region, and a stable region, and the two segmentation curves are used as the baseline instability discrimination boundary and the extended instability discrimination boundary, respectively.

[0041] Preferably, the method involves determining two segmentation curves that divide the two-dimensional feature space into an unstable region, a fuzzy region, and a stable region, and using these two segmentation curves as the baseline instability discrimination boundary and the extended instability discrimination boundary, respectively, including: Determine the cost function C with the objective of minimizing the discrimination risk: C=W miss *N miss +W fp *N fp , where N miss N represents the number of unstable samples that were missed in detection. fp W represents the stable number of false positives. miss and W fp For the penalty weight, satisfy W miss ≥h*W fp h is a preset multiple; Using a historical simulation sample set containing a large number of known stable state labels, with the absolute safety constraint that all unstable sample points must be located on the same side of the segmentation curve, the segmentation curve L1 that satisfies the minimum cost function is determined in the F1-F2 two-dimensional feature space, and the segmentation curve L1 is used as the benchmark instability discrimination boundary. Using a historical simulation sample set containing a large number of known stable state labels, and with the absolute safety constraint that a predetermined number of unstable sample points must be located on the same side of the segmentation curve, the segmentation curve L2 that satisfies the minimum cost function is determined in the two-dimensional feature space F1-F2, and the segmentation curve L2 is used as the extended instability discrimination boundary; where F1 is the first feature quantity and F2 is the second feature quantity.

[0042] The core of this step is to establish a safety-oriented approach to the discrimination process from a fundamental perspective. To achieve the goal of "zero missed detections," a set of weight parameters needs to be configured for the discrimination logic. The core principle is that the penalty weight assigned to "missed detections" (i.e., misclassifying an actual unstable fault as stable) must be significantly higher than the penalty weight assigned to "false alarms" (i.e., misclassifying an actual stable fault as unstable). For example, by setting W_ miss ≥2*W_ fp The relationship between these factors mathematically makes the discriminative logic naturally inclined to make a conservative judgment that is "unstable" when the evidence is ambiguous, thereby systematically suppressing the risk of missed judgments.

[0043] Specifically, the concept of zero missed detections is transformed into a specific, conservative discrimination boundary in the feature space [F1,F2]. This boundary defines the dividing line between the unstable and stable regions, and its generation process strictly follows the principle of safety first.

[0044] Definition of safety cost function: First, define a cost function C that aims to minimize the discrimination risk: C = W miss *N miss +W fp *N fp , where N miss N represents the number of unstable samples that were missed in detection. fp Let W be the stable number of false positives. To achieve zero false negatives, a penalty weight is set to satisfy W. miss >>W fp (e.g. W) miss ≥10*W fp This makes the optimization process naturally inclined to increase N. fp (False alarm), and resolutely reduce Nmiss to zero.

[0045] Boundary search based on sample set: Utilizing a historical simulated fault sample set S={([F1 i F2 i ],label i On the F1-F2 two-dimensional plane, search for a (or a set of) dividing curves L such that: Absolute safety constraint: All sample points with labeli=instability must be located on the same side of curve L (defined as the instability side).

[0046] Optimality objective: Under the premise of satisfying the above absolute safety constraints, minimize the cost function C by adjusting the shape and position of curve L. Since W miss When the value is extremely large, the optimization process will prioritize squeezing L towards the stable sample region to ensure that all unstable points are included, even if it means surrounding some stable points.

[0047] Mathematical representation of the discrimination boundary: The discrimination boundary L is not limited to a simple straight line, but is a piecewise curve or a complex closed region boundary determined by a safe optimization process in a two-dimensional feature space. For example, it can be a conservative outward boundary of a convex hull formed by connecting multiple line segments end to end, or a nonlinear decision hyperplane. The final mathematical form of the boundary L (such as parametric equations or a system of inequalities) constitutes the core of the discrimination rule.

[0048] In step 104, based on the positional relationship between the two-dimensional orthogonal eigenvectors, the benchmark instability discrimination boundary, and the extended instability discrimination boundary, the stability state of the power grid under different faults is determined.

[0049] Preferably, the determination of the power grid stability state under different faults based on the positional relationship between the two-dimensional orthogonal eigenvectors, the baseline instability discrimination boundary, and the extended instability discrimination boundary includes: For any two-dimensional orthogonal eigenvector, if the two-dimensional orthogonal eigenvector is located in the instability region defined by the baseline instability discrimination boundary, then the power grid is determined to be instable; if the two-dimensional orthogonal eigenvector is located in the stable region defined by the extended instability discrimination boundary, then the power grid is determined to be stable; if the two-dimensional orthogonal eigenvector is located in the fuzzy region formed by the extended instability discrimination boundary and the baseline instability discrimination boundary, then the power grid is determined to be instable.

[0050] In this invention, this step transforms security weights into specific discrimination operations. The received feature vectors [F1, F2] are input into a predefined set of instability discrimination rules. Based on the weight configuration in step S2, this rule set defines an instability region in the F1-F2 feature space that prioritizes "zero false negatives." This rule set can employ mechanisms such as weighted voting and can integrate comprehensive risk indicators derived from F1 and F2 for decision-making. The rule set includes a long-term fixed baseline instability discrimination boundary, which serves as the absolute benchmark for security discrimination; simultaneously, it supports setting an extended peripheral instability discrimination region to attempt to reduce false alarms and optimize overall discrimination accuracy without violating the zero false negative commitment.

[0051] Specifically, this invention inputs the received feature vector [F1, F2] into a predefined instability core discrimination rule. Based on the weight configuration in step S2, this rule defines an instability region in the F1-F2 feature space that prioritizes zero false positives. The rule set includes a long-term fixed baseline instability boundary, which serves as the absolute benchmark for discrimination safety; simultaneously, it supports setting an extended peripheral instability boundary to attempt to reduce false alarms and optimize overall discrimination accuracy without violating the zero false positive commitment.

[0052] Core discrimination rule: The core of the discrimination rule set is the baseline instability discrimination boundary Lcore determined in step 103. For any input feature point P=[F1,F2], the core discrimination rule is: IFP is located on the instability side defined by Lcore, THEN conclusion = "instability". This rule is the core principle of discrimination safety, because the generation of Lcore has guaranteed that all historical instability samples meet this condition, thus theoretically promising zero missed detections.

[0053] Determining the baseline boundary Lcore: Lcore is obtained by applying a conservative envelope to the historical instability sample point set Sunstable. Specific methods include, but are not limited to: Convex hull expansion method: Calculate the convex hull of the Sunstable, and then translate it outward along the convex hull normal direction by a safety margin δ to form the Lcore. δ can be set according to sample uncertainty or measurement error.

[0054] The minimum closed region method: Find a closed region (such as a rectangle or ellipse) with a simple shape that can contain all Sunstable points; its boundary is the Lcore. To ensure conservatism, the region range is usually slightly larger than the actual distribution range of the sample points.

[0055] Extended Discriminant Region and Rules: To further optimize performance, an extended instability discriminant region can be defined around Lcore, with its outer boundary being Lext. Lext is determined based on the distribution analysis of the stable sample set Ssta. For example, Lext can be set as the boundary (e.g., a confidence ellipse) containing the vast majority (e.g., 99%) of stable samples. This forms the extended discriminant rules: IFP is located on the unstable side defined by Lcore -> Conclusion = "Instability" (high confidence); ELSEIFP is located in the region between Lcore and Lext -> Conclusion = "Instability" (conservative judgment) or triggers "manual review"; ELSE (P is located in a stable region outside the Lex) -> Conclusion = "Stable". This hierarchical rule, while adhering to the absolute safety red line of Lcore, defines the ambiguity region through Lex to reduce false alarms. In this invention, the final decision is made based on the position of the feature vector in the feature space. If the feature vector falls into the unstable region (including within the baseline unstable discrimination boundary and the extended unstable discrimination region), the conclusion "Instability" is output; otherwise, the conclusion "Stable" is output.

[0056] Preferably, the method further includes: The benchmark instability discrimination boundary is divided into multiple segments based on the angles between the benchmark instability discrimination boundary and the horizontal and vertical axes in the feature space; wherein the horizontal and vertical axes are determined based on the first feature quantity F1 and the second feature quantity F2. For any boundary segment, calculate the angle between the boundary segment and the horizontal and vertical axes respectively. If the angle between the boundary segment and any coordinate axis is less than or equal to a preset angle threshold, then the boundary segment is determined to be dominated by the feature quantity corresponding to the coordinate axis. If the angle between the boundary segment and both coordinate axes is greater than the preset angle threshold, then the boundary segment is determined to be dominated by the mixture of the two feature quantities. For any two-dimensional orthogonal eigenvector whose stable state is unstable, calculate the signed distance from this two-dimensional orthogonal eigenvector to each boundary segment; select the boundary segment with the largest signed distance as the target boundary segment, and use the dominant feature quantity corresponding to the target boundary segment as the stability criterion identifier for this two-dimensional orthogonal eigenvector whose stable state is unstable. In this invention, a discrimination identifier can also be output, indicating which feature quantity (F1 or F2) dominates the current discrimination conclusion, thereby directly linking the higher-order "stable / unstable" state with the underlying physical mechanism features, enhancing the interpretability of the results. The generation logic of this identifier is directly related to the physical meaning of the quasi-instability discrimination boundary Lcore determined in step 103, and is specifically implemented as follows: Boundary segment association mechanism: When defining each boundary segment of Lcore, a dominant feature identifier is pre-associated with it. For example: If a boundary segment is approximately perpendicular to the F1 axis (i.e., primarily sensitive to the F1 value), it is identified as an F1-dominant boundary segment. This boundary segment primarily responds to the exceeding of high-frequency transient impact energy (F1).

[0057] If a boundary segment is approximately perpendicular to the F2 axis (i.e., primarily sensitive to the F2 value), it is identified as an F2-dominant boundary segment. This boundary segment primarily responds to exceeding the limit of the fast-slow dynamic energy ratio (F2).

[0058] If it is a diagonal line segment, the relative strength of its sensitivity to F1 and F2 can be judged based on its slope, and it can be identified as a mixed dominant boundary segment of F1-F2.

[0059] Criterion identification and determination rules: For a feature point P=[F1,F2] that is judged to be unstable, the system determines its position relative to the Lcore and executes the following logic to determine the dominant criterion: Calculate the violation distance: Calculate the signed distance from point P to each boundary line of Lcore (positive for points on the unstable side). Find the most critical boundary segment that makes point P instability, i.e., the segment where point P has the largest violation distance (largest positive value) to that boundary.

[0060] Output identifier: The dominant feature identifier of the most critical boundary segment pre-associated is output as the criterion identifier for this discrimination.

[0061] Special case: If point P seriously violates multiple boundaries at the same time (such as being far from both the F1 dominant segment and the F2 dominant segment at the same time), the mixed mechanism identifier can be output, or the contribution of the two feature quantities can be output at the same time.

[0062] Suppose a feature point is defined as [E_hf, R_hl] = [0.85, 1.6], which is deemed unstable because it exceeds the vertical segment of L_core. The system calculations reveal that the violation distance of this point to this vertical segment is significantly greater than its distance to other segments. Since this vertical segment is identified as E_hf dominant, the system outputs a criterion label along with the instability conclusion: "Dominant Criterion: E_hf (High-Frequency Impact Energy Exceeds Limit)".

[0063] This design directly links high-order stable / instability states with underlying physical mechanisms. Operators not only know whether the system is unstable, but can also immediately understand whether the likely cause of instability is excessive high-frequency impact or severe dynamic decoupling. This provides direct, mechanistic guidance for subsequent scheduling control or in-depth analysis, significantly enhancing the engineering value of the judgment results.

[0064] This invention also records and visualizes the analysis results, aiming to form a complete analytical loop and support long-term monitoring. After an analysis cycle (such as the completion of a batch of anticipated fault identifications), the system can automatically save all data related to this identification (feature vectors, rule set parameters, conclusions) and generate a visual analysis chart. This chart intuitively displays the distribution of all fault samples in the F1-F2 feature space and overlays the used identification boundaries. When the method of this invention is integrated into an online system and executed periodically, by comparing the historical and current cycle visualization charts, operators can gain insight into the evolution trend of the power grid's transient stability characteristics.

[0065] The discrimination method of this invention is a logical framework that is applied in conjunction with a specific feature space. As a specific embodiment, when the feature vectors [F1, F2] are concretized as [E_hf, R_hl] obtained using the db5 wavelet basis and 5-level decomposition, and the discrimination rule set adopts a weighted voting mechanism, and the highest decision weight is assigned to the comprehensive index WMTVSI, a complete and implementable zero-miss discrimination scheme is constituted. The benchmark discrimination boundary and weight parameters used to ensure zero-miss discrimination can be comprehensively determined based on statistical analysis and mechanistic understanding of historical typical stable / instability scenario simulation data; the entire process does not rely on data-driven "black box" training.

[0066] The overall approach of this invention is as follows: Based on the well-defined mechanism of wavelet orthogonal feature space, and addressing the core engineering requirement of "zero missed detections" in anticipated fault safety screening, this invention constructs an efficient, transparent, and traceable stable discrimination process by designing safety-priority discrimination rules and weight configurations. This invention aims to solve the problems of missed detection risks and opaque decision-making logic faced by existing stable discrimination methods in engineering safety scenarios. Specific objectives include: A security discrimination mechanism with zero missed detections as the highest priority is established. By configuring weight parameters that significantly favor penalizing missed detections in the discrimination process, and defining a discrimination region in a feature space with a clear mechanism to ensure zero missed detections, the risk of missed detections is systematically suppressed at the algorithm design level, providing more reliable discrimination conclusions for power grid security defense.

[0067] This achieves transparency in the discrimination process and traceability of the conclusions. By utilizing feature inputs with clear physical meaning, the discrimination conclusions can be directly linked to the specific feature quantities that triggered the discrimination. By combining the visualized records of feature space distribution, it provides operators with direct evidence for understanding, verifying, and tracing each discrimination decision, overcoming the trust problem of black box models.

[0068] It provides long-term stability and asymptotic adjustability of the discrimination strategy. By setting a fixed benchmark instability discrimination boundary as a long-term evaluation baseline, it ensures the consistency and comparability of the discrimination criteria. At the same time, it allows for the setting of an extended discrimination region under the premise of rigorously verifying security, providing a feasible technical approach to optimize the overall discrimination accuracy without sacrificing the commitment to zero false negatives.

[0069] A closed-loop discrimination mechanism is constructed that is closely integrated with the upstream feature space construction method and is practical for engineering applications. Using wavelet orthogonal feature vectors with a clear mechanism as input, a complete and efficient technical chain is formed from feature extraction to safety discrimination. This method has low computational complexity, meets the engineering requirements for rapid batch screening of massive amounts of anticipated fault scenarios, and can integrate visualization and trend analysis functions to directly serve the online safety assessment and offline planning analysis of the power grid.

[0070] Compared with the prior art, the transient voltage stability zero leakage detection method and system provided by the present invention have the following beneficial effects: 1. In principle, it systematically reduces the risk of missed detections.

[0071] By assigning significantly higher penalty weights to "missed detections" and defining a discrimination region in the feature space that prioritizes zero missed detections, the discrimination mechanism itself is made to have a safety-first characteristic.

[0072] 2. Make the criteria for judgment transparent and the results traceable.

[0073] The judgment conclusion can be linked to the specific feature quantity that triggered the judgment, and combined with the visual record of the feature space distribution, it provides operators with a direct basis for understanding and verifying the judgment results.

[0074] 3. It balances the long-term consistency of the judgment criteria with the adjustability of the strategy.

[0075] By establishing a fixed benchmark discrimination boundary, the stability of the evaluation standard is ensured, while allowing for the setting of extended regions under strict verification, providing a feasible path for optimizing discrimination accuracy.

[0076] 4. Capable of efficiently handling stable screening tasks for massive amounts of anticipated faults.

[0077] This method directly discriminates the feature vectors extracted from the simulation waveform. The calculation process is simple and clear, meeting the engineering requirements for rapid and batch analysis of large-scale anticipated fault sets.

[0078] 5. It forms a technical closed loop with the preceding feature space construction method.

[0079] This invention uses the well-defined wavelet orthogonal features provided by the superior patent as input to realize a coherent process from feature extraction to security discrimination, thereby enhancing the integrity and practicality of the technical system.

[0080] The following specific examples illustrate the embodiments of the present invention. In embodiments of the present invention, voltage stability is determined based on a pre-constructed specific wavelet orthogonal feature space. The specific process includes: 1. Instantiation of the feature space This embodiment first acquires the feature quantities. Specifically, the Daubechies 5th order wavelet (db5) is selected as the orthogonal wavelet basis, and the voltage time series signal after the fault is decomposed into 5 levels of discrete wavelet decomposition. Based on the coefficients obtained from the decomposition, the first feature quantity F1 is constructed as the average of the sum of squares of the coefficients of the first to third level detail subspaces (i.e., the high-frequency detail energy E_hf), and the second feature quantity F2 is constructed as the ratio of F1 to the sum of the sums of squares of the coefficients of the fourth and fifth level detail subspaces (i.e., the high-low frequency energy ratio R_hl). This set of features [E_hf, R_hl] constitutes the two-dimensional feature vector input received by the discrimination method of this embodiment.

[0081] For example, let's take the simulated bus voltage V(t) after a specific anticipated fault (such as a three-phase short circuit on a critical transmission line) as an example. Figure 2 As shown, the instability voltage curve is a drop-down curve, illustrating the characteristic calculation process: Input: V(t), sampling frequency fs=10kHz, observation window after fault t=0~2s.

[0082] Wavelet decomposition: A 5-level DWT was performed using the db5 wavelet. The detail coefficients D1~D5 and the approximation coefficient cA5 for each level were calculated.

[0083] Characteristic quantity calculation: Calculate the high-frequency detail energy E_hf: E_hf = (sum(D1) / (D2)) 2 )+sum(D22 )+sum(D3 2 )) / 3. Assume the normalized result is E_hf=0.85.

[0084] Calculate the low-frequency reference energy E_lf: E_lf = sum(D4) 2 )+sum(D5 2 ).

[0085] Calculate the high-frequency / low-frequency energy ratio R_hl: R_hl = E_hf / E_lf. Assume the normalized result is R_hl = 2.1.

[0086] Output: The feature vector of this fault is [F1,F2]=[0.85,2.1].

[0087] By calculating hundreds of such anticipated faults, a sample feature set can be obtained for training the discrimination boundary.

[0088] 2. Determining the discrimination rule set and boundaries After obtaining a sufficient number of sample feature points, their distribution on the E_hf-R_hl plane is as follows: Figure 3 As shown, stable samples (green dots) are mostly clustered in the lower left region, while unstable samples (red dots) are concentrated in the upper right region, but the boundary between the two is not a straight line. Based on the aforementioned feature space, by statistically analyzing a large number of simulated failure samples with known labels (stable / unstable), and combining this with the engineering goal of zero missed detections, the discrimination rule set of this embodiment can be determined. The core of this rule set lies in defining two key geometric boundaries in the feature space: the baseline instability discrimination boundary (Lcore) and the extended instability discrimination region boundary (Lext). Figure 3 As shown, stable samples (green dots) and unstable samples (red dots) exhibit good inter-class separation in the E_hf-R_hl feature space. The unstable sample points are mainly clustered in the upper right region of the coordinate system.

[0089] 1. Determination of the Benchmark Instability Detection Boundary (Lcore): The benchmark instability detection boundary Lcore is a piecewise linear boundary (polyline) that divides the stable and unstable regions in the feature space. Its location is determined strictly according to the highest safety criterion of "zero missed detections," and the specific method is as follows: Safety Envelope Principle: Lcore's path is designed to completely enclose all historical unstable sample points (red dots) on one side (defined as the unstable side or the inner side of the boundary). To ensure absolute safety, the boundary line is usually not right next to the outermost unstable sample point, but rather leaves a preset safety margin to cope with sample uncertainties and future critical instability scenarios with weaker features.

[0090] Piecewise linear implementation: such as Figure 3As shown by the bold black polyline, the Lcore consists of several straight line segments connected end to end. For example, it can be composed of a vertical segment, a horizontal segment, and a diagonal segment. This piecewise linear design has significant advantages: clear physical meaning, with each boundary segment associated with a different dominant instability mechanism. For example, the vertical segment may primarily respond to E_hf (high-frequency shock) exceedances, the horizontal segment primarily responds to R_hl (dynamic imbalance) exceedances, and the diagonal segment corresponds to a combination of the two mechanisms. It is engineering-friendly, with a simple mathematical expression for the polyline (defined by a series of inequalities a*E_hf + b*R_hl > c), high computational efficiency, and easy integration and configuration in online systems. It is conservative and controllable; by adjusting the coordinates of the polyline's inflection points, the degree of conservatism of the boundary can be easily controlled, ensuring complete envelopment of the instability sample points.

[0091] 2. Determination of the Extended Instability Detection Region Boundary (Lext) The extended instability detection region boundary (Lext) is another boundary located outside the Lcore (e.g., Figure 3 As shown by the dashed line in the middle, it, together with Lcore, defines a ring-shaped "extended discrimination region".

[0092] The determination of Lext is primarily based on the distribution analysis of the stable sample set. The goal is to define a region such that the vast majority (e.g., over 95%) of the historical stable sample points (green dots) are located outside this boundary.

[0093] The region between Lcore and Lext is the extended discrimination region. Sample points falling into this region, although not touching the absolutely safe Lcore boundary, are considered to have a high risk of instability because they are far from the stable sample density area and relatively close to the unstable sample area. The system can adopt a conservative strategy (directly judge instability) or trigger further analysis for such points.

[0094] 3. The discrimination rule set is constructed based on the above two boundaries, forming the following three-level discrimination rules: Rule 1 (Determining Instability): If the feature point [E_hf, R_hl] falls inside the L_core boundary (i.e.) Figure 3 If the area is shaded in the upper right corner of the L_core, it is directly considered unstable. This conclusion is based on an absolute safety boundary and has the highest reliability.

[0095] Rule 2 (Conservative Judgment or Verification): If a feature point falls within the extended discriminant region between Lcore and Lext ( Figure 3 If the area is a shaded area in the middle grid, it can be conservatively judged as unstable or marked as high risk according to the system's preset security policy - manual review is recommended.

[0096] Rule 3 (Determining Stability): If the feature point is located outside the Lex boundary ( Figure 3If the area marked by the green dot in the lower left corner of the z-axis is considered stable, then it is determined to be stable.

[0097] This set of rules rigorously ensures a zero-miss rate commitment mathematically through the design of piecewise linear boundaries and a two-layer discrimination region (Rule 1). At the same time, by introducing an extended region (Rule 2), it provides a flexible engineering means to reduce false alarms and optimize overall discrimination accuracy. Furthermore, the clear geometric division makes the discrimination process highly transparent and interpretable.

[0098] In specific judgment, for a new anticipated fault scenario to be evaluated, the system executes the following process: Calculate its eigenvalues ​​[E_hf,R_hl].

[0099] The feature point is compared with the predefined Lcore boundary.

[0100] Conclusion: If the point falls inside the Lcore, the output becomes unstable; if it is outside the Lext, the output becomes stable; if it is in the extended region, the output becomes unstable according to the strategy or a review is recommended.

[0101] Criterion identifier generation (for instability conclusion): The system internally calculates the violation distance of the instability point relative to the boundaries of each segment of Lcore. Assume Lcore consists of three segments: a vertical segment (dominantly E_hf), a horizontal segment (dominantly R_hl), and a diagonal segment (mixed). If the calculation finds that the violation distance of this point is the largest relative to the vertical segment, the system automatically adds the output criterion identifier: "Dominant Mechanism: High-Frequency Impact Energy (E_hf) Exceeds Limit". If the violation distance is the largest relative to the horizontal segment, it outputs "Dominant Mechanism: Fast-Slow Dynamic Imbalance (R_hl) Exceeds Limit".

[0102] For example, for a fault with a characteristic point [0.85, 2.1], the system determines that it falls inside the Lcore, resulting in output instability. Simultaneously, because this point primarily violates the diagonal segment of the Lcore (designed to be sensitive to high R_hl), the system outputs the criterion label: "Dominant Criterion: R_hl (Abnormal Fast-Slow Dynamic Energy Ratio)". This suggests to operators that the instability may be related to insufficient system damping and dynamic decoupling, rather than a simple instantaneous power surge.

[0103] 4. Implementation of visualization and recording functions The discrimination system in this embodiment can realize the result recording and visualization functions as described in the claims. The system can automatically generate similar results. Figure 3 and Figure 4 The visualization analysis chart not only displays the feature points of all faults to be identified, but also clearly overlays the benchmark instability identification boundary. For online periodic analysis, the system can save such images generated for each analysis, and compare multiple images from different dates (e.g., ...). Figure 4As shown in the distribution map, operators can intuitively observe whether the overall distribution of the feature point cloud has drifted, or whether new fault points are approaching or crossing the discrimination boundary, thereby achieving dynamic and visual tracking of the transient voltage stability risk of the power grid.

[0104] In summary, this embodiment, by referencing actual simulation cases and result figures, specifically illustrates how to apply the zero-missing discrimination method described in this invention to a specific wavelet orthogonal feature space that has been verified to be effective, thereby forming a complete and implementable engineering solution from theory, features to discrimination and visualization.

[0105] Compared with the prior art, the transient voltage stability zero leakage detection method and system provided by the present invention have the following beneficial effects: 1. In principle, it systematically reduces the risk of missed detections.

[0106] By assigning significantly higher penalty weights to "missed detections" and defining a discrimination region in the feature space that prioritizes zero missed detections, the discrimination mechanism itself is made to have a safety-first characteristic.

[0107] 2. Make the criteria for judgment transparent and the results traceable.

[0108] The judgment conclusion can be linked to the specific feature quantity that triggered the judgment, and combined with the visual record of the feature space distribution, it provides operators with a direct basis for understanding and verifying the judgment results.

[0109] 3. It balances the long-term consistency of the judgment criteria with the adjustability of the strategy.

[0110] By establishing a fixed benchmark discrimination boundary, the stability of the evaluation standard is ensured, while allowing for the setting of extended regions under strict verification, providing a feasible path for optimizing discrimination accuracy.

[0111] 4. Capable of efficiently handling stable screening tasks for massive amounts of anticipated faults.

[0112] This method directly discriminates the feature vectors extracted from the simulation waveform. The calculation process is simple and clear, meeting the engineering requirements for rapid and batch analysis of large-scale anticipated fault sets.

[0113] 5. It forms a technical closed loop with the preceding feature space construction method.

[0114] This invention uses the well-defined wavelet orthogonal features provided by the superior patent as input to realize a coherent process from feature extraction to security discrimination, thereby enhancing the integrity and practicality of the technical system.

[0115] Figure 5 This is a schematic diagram of the zero-leakage detection system 500500 for power grid transient voltage stabilization according to an embodiment of the present invention. Figure 5 As shown, the zero-leakage judgment system 500 for power grid transient voltage stability provided by the present invention includes: a data acquisition unit 601, a feature quantity calculation unit 602, an instability judgment boundary determination unit 603, and a judgment unit 604.

[0116] Preferably, the data acquisition unit 601 is used to perform simulation calculations of preset faults in the power grid and acquire voltage timing data of the bus under different faults.

[0117] Preferably, the feature quantity calculation unit 602 is used to calculate a two-dimensional orthogonal feature vector based on the voltage timing data to obtain two-dimensional orthogonal feature vectors corresponding to different faults.

[0118] Preferably, the feature calculation unit 602 calculates a two-dimensional orthogonal feature vector based on the voltage time series data to obtain two-dimensional orthogonal feature vectors corresponding to different faults, including: Discrete wavelet transform is performed on the voltage time series data of the bus under each fault to decompose the voltage time series data into mutually orthogonal approximate subspace and detail subspace coefficient sequences; wherein, the approximate subspace corresponds to the low frequency component and the detail subspace corresponds to the high frequency component. Two-dimensional orthogonal feature vectors [F1, F2] are constructed based on the coefficient sequences of mutually orthogonal approximate subspaces and detail subspaces to obtain two-dimensional orthogonal feature vectors corresponding to different faults. Among them, F1 is the first feature quantity and F2 is the second feature quantity. F1 is used to characterize the high-frequency transient impact energy, corresponding to the energy feature extracted from the detail subspace. F2 is used to characterize the dynamic energy structure relationship across time scales, corresponding to the ratio of energy between the detail subspace and the approximate subspace.

[0119] Preferably, the feature calculation unit 602 performs discrete wavelet transform processing on the voltage time series data of the bus under each fault to decompose the voltage time series data into mutually orthogonal approximate subspace and detail subspace coefficient sequences, including: For any fault, when performing discrete wavelet transform on the voltage time series data of the bus under any fault based on orthogonal wavelet basis functions, multi-resolution analysis is used to decompose the voltage time series data of the bus under any fault into multi-layered mutually orthogonal approximate subspace and detail subspace coefficient sequences.

[0120] Preferably, the feature quantity calculation unit 602 is further configured to: An energy aggregation function is constructed based on the detail subspace coefficient sequence from the first layer to the Jth layer to determine the first characteristic quantity; wherein J≥2 and is a positive integer, and the energy aggregation function is calculated based on the p-order norm of the coefficient sequence.

[0121] Preferably, the feature quantity calculation unit 602 is further configured to: Construct an energy ratio function based on the approximate subspace coefficient sequences of selected different frequency bands to determine a second feature quantity, including: F2 = Ψ(cD_m) / Ψ(cA_n), where cD_m and cA_n respectively represent the coefficients of a high-frequency detail subspace and a low-frequency approximation subspace, m < n, and Ψ is an energy calculation operator.

[0122] Preferably, the instability discrimination boundary determination unit 603 is configured to determine two segmentation curves that divide the two-dimensional feature space into an instability region, a fuzzy region, and a stable region, and use the two segmentation curves as the reference instability discrimination boundary and the extended instability discrimination boundary, respectively.

[0123] Preferably, the instability discrimination boundary determination unit 603 determines two segmentation curves that divide the two-dimensional feature space into an instability region, a fuzzy region, and a stable region, and uses the two segmentation curves as the reference instability discrimination boundary and the extended instability discrimination boundary, respectively, including: Determine a cost function C with the goal of minimizing the discrimination risk: C = W miss *N miss +W fp *N fp , where N miss is the number of misjudged unstable samples, N [[ID=​​​​​​​​​​​​​​​​​​Preferably, the judgment unit 604 determines the stability state of the power grid under different faults based on the positional relationship between the two-dimensional orthogonal feature vector, the reference instability discrimination boundary, and the extended instability discrimination boundary, including: For any two-dimensional orthogonal eigenvector, if the two-dimensional orthogonal eigenvector is located in the instability region defined by the baseline instability discrimination boundary, then the power grid is determined to be instable; if the two-dimensional orthogonal eigenvector is located in the stable region defined by the extended instability discrimination boundary, then the power grid is determined to be stable; if the two-dimensional orthogonal eigenvector is located in the fuzzy region formed by the extended instability discrimination boundary and the baseline instability discrimination boundary, then the power grid is determined to be instable.

[0126] Preferably, the system further includes: The identification unit is used to divide the benchmark instability discrimination boundary into multiple segments according to the angle between the benchmark instability discrimination boundary and the horizontal and vertical axes in the feature space; wherein the horizontal and vertical axes are determined based on the first feature quantity F1 and the second feature quantity F2. For any boundary segment, calculate the angle between the boundary segment and the horizontal and vertical axes respectively. If the angle between the boundary segment and any coordinate axis is less than or equal to a preset angle threshold, then the boundary segment is determined to be dominated by the feature quantity corresponding to the coordinate axis. If the angle between the boundary segment and both coordinate axes is greater than the preset angle threshold, then the boundary segment is determined to be dominated by the mixture of the two feature quantities. For any two-dimensional orthogonal eigenvector whose stable state is unstable, calculate the signed distance from the two-dimensional orthogonal eigenvector whose stable state is unstable to each boundary segment; The boundary segment with the largest signed distance is selected as the target boundary segment, and the dominant feature quantity corresponding to the target boundary segment is used as the stability criterion identifier for any two-dimensional orthogonal feature vector whose stable state is unstable.

[0127] The zero-leakage detection system 500 for grid transient voltage stability in this embodiment of the invention corresponds to the zero-leakage detection method 100 for grid transient voltage stability in another embodiment of the invention, and will not be described again here.

[0128] According to another aspect of the present invention, 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 any one of the zero-leakage detection methods for power grid transient voltage stability.

[0129] According to another aspect of the present invention, the present invention provides an electronic device, comprising: The aforementioned computer-readable storage medium; and One or more processors for executing a program in the computer-readable storage medium.

[0130] The present invention has been described with reference to a few embodiments. However, it will be apparent to those skilled in the art that other embodiments besides those disclosed above fall equivalently within the scope of the present invention.

[0131] Generally, all terms used in this invention are interpreted according to their ordinary meaning in the art, unless otherwise expressly defined herein. All references to “a / the / the [device, component, etc.]” ​​are openly interpreted as at least one instance of said device, component, etc., unless otherwise expressly stated. The steps of any method disclosed herein need not be performed in the exact order disclosed, unless explicitly stated otherwise.

[0132] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention 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.

[0133] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (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 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 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.

[0134] 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.

[0135] 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.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for determining zero leakage in the transient voltage stability of a power grid, characterized in that, The method includes: Performing simulation calculations of preset faults on the power grid to obtain voltage time-series data of the busbars under different faults; Calculating two-dimensional orthogonal eigenvectors based on the voltage time-series data to obtain two-dimensional orthogonal eigenvectors corresponding to different faults; Determining two segmentation curves that divide the two-dimensional feature space into an unstable region, a fuzzy region, and a stable region, and using the two segmentation curves as the reference instability discrimination boundary and the extended instability discrimination boundary respectively; Determining the stable state of the power grid under different faults based on the positional relationship between the two-dimensional orthogonal eigenvectors, the reference instability discrimination boundary, and the extended instability discrimination boundary.

2. The method according to claim 1, characterized in that, Calculating two-dimensional orthogonal eigenvectors based on the voltage time-series data to obtain two-dimensional orthogonal eigenvectors corresponding to different faults, including: Performing discrete wavelet transform processing on the voltage time-series data of the busbars under each fault to decompose the voltage time-series data into mutually orthogonal approximation subspace and detail subspace coefficient sequences; where the approximation subspace corresponds to the low-frequency component, and the detail subspace corresponds to the high-frequency component; Constructing two-dimensional orthogonal eigenvectors [F1, F2] based on the mutually orthogonal approximation subspace and detail subspace coefficient sequences to obtain two-dimensional orthogonal eigenvectors corresponding to different faults; where F1 is the first characteristic quantity, F2 is the second characteristic quantity, F1 is used to characterize the high-frequency transient impact energy and corresponds to the energy feature extracted from the detail subspace; F2 is used to characterize the dynamic energy structure relationship across time scales and corresponds to the proportional relationship between the energies of the detail subspace and the approximation subspace.

3. The method according to claim 2, characterized in that, Performing discrete wavelet transform processing on the voltage time-series data of the busbars under each fault to decompose the voltage time-series data into mutually orthogonal approximation subspace and detail subspace coefficient sequences, including: For any one fault, when performing discrete wavelet transform processing on the voltage time-series data of the busbars under that fault based on the orthogonal wavelet basis function, through multi-resolution analysis, the voltage time-series data of the busbars under that fault is decomposed into multiple layers of mutually orthogonal approximation subspace and detail subspace coefficient sequences.

4. The method according to claim 2, characterized in that, The method further includes: Constructing an energy aggregation function based on the detail subspace coefficient sequences from the first layer to the Jth layer to determine the first characteristic quantity; where J≥2 and is a positive integer, and the energy aggregation function is calculated based on the p-norm of the coefficient sequences.

5. The method according to claim 2, characterized in that, The method further includes: Constructing an energy ratio function based on the selected approximation subspace coefficient sequences of different frequency bands to determine the second characteristic quantity, including: F2 = Ψ(cD_m) / Ψ(cA_n), where cD_m and cA_n respectively represent the coefficients of a high-frequency detail subspace and a low-frequency approximation subspace, m < n, and Ψ is an energy calculation operator.

6. The method according to claim 1, characterized in that, Determining two segmentation curves that divide the two-dimensional feature space into an unstable region, a fuzzy region, and a stable region, and using the two segmentation curves as the reference instability discrimination boundary and the extended instability discrimination boundary respectively, including: Determine the cost function C with the objective of minimizing the discrimination risk: C=W miss *N miss +W fp *N fp , where N miss N represents the number of unstable samples that were missed in detection. fp W represents the stable number of false positives. miss and W fp For the penalty weight, satisfy W miss ≥h*W fp h is a preset multiple; Using a historical simulation sample set containing a large number of known stable state labels, with the absolute safety constraint that all unstable sample points must be located on the same side of the segmentation curve, the segmentation curve L1 that satisfies the minimum cost function is determined in the F1-F2 two-dimensional feature space, and the segmentation curve L1 is used as the benchmark instability discrimination boundary. Using a historical simulation sample set containing a large number of known stable state labels, and with the absolute safety constraint that a predetermined number of unstable sample points must be located on the same side of the segmentation curve, the segmentation curve L2 that satisfies the minimum cost function is determined in the two-dimensional feature space F1-F2, and the segmentation curve L2 is used as the extended instability discrimination boundary; where F1 is the first feature quantity and F2 is the second feature quantity.

7. The method according to claim 1, characterized in that, Based on the positional relationship between the two-dimensional orthogonal eigenvectors, the baseline instability discrimination boundary, and the extended instability discrimination boundary, the stability state of the power grid under different faults is determined, including: For any two-dimensional orthogonal eigenvector, if the two-dimensional orthogonal eigenvector is located in the instability region defined by the baseline instability discrimination boundary, then the power grid is determined to be instable; if the two-dimensional orthogonal eigenvector is located in the stable region defined by the extended instability discrimination boundary, then the power grid is determined to be stable; if the two-dimensional orthogonal eigenvector is located in the fuzzy region formed by the extended instability discrimination boundary and the baseline instability discrimination boundary, then the power grid is determined to be instable.

8. The method according to claim 1, characterized in that, The method further includes: The benchmark instability discrimination boundary is divided into multiple segments based on the angles between the benchmark instability discrimination boundary and the horizontal and vertical axes in the feature space; wherein the horizontal and vertical axes are determined based on the first feature quantity F1 and the second feature quantity F2. For any boundary segment, calculate the angle between the boundary segment and the horizontal and vertical axes respectively. If the angle between the boundary segment and any coordinate axis is less than or equal to a preset angle threshold, then the boundary segment is determined to be dominated by the feature quantity corresponding to the coordinate axis. If the angle between the boundary segment and both coordinate axes is greater than the preset angle threshold, then the boundary segment is determined to be dominated by the mixture of the two feature quantities. For any two-dimensional orthogonal eigenvector whose stable state is unstable, calculate the signed distance from the two-dimensional orthogonal eigenvector whose stable state is unstable to each boundary segment; The boundary segment with the largest signed distance is selected as the target boundary segment, and the dominant feature quantity corresponding to the target boundary segment is used as the stability criterion identifier for any two-dimensional orthogonal feature vector whose stable state is unstable.

9. A zero-leakage detection system for transient voltage stability of a power grid, characterized in that, The system includes: The data acquisition unit is used to perform simulation calculations of preset faults in the power grid and acquire voltage timing data of the bus under different faults. The feature calculation unit is used to calculate a two-dimensional orthogonal feature vector based on the voltage time series data to obtain two-dimensional orthogonal feature vectors corresponding to different faults. The instability discrimination boundary determination unit is used to determine two segmentation curves that divide the two-dimensional feature space into an unstable region, a fuzzy region, and a stable region, and to use the two segmentation curves as the baseline instability discrimination boundary and the extended instability discrimination boundary, respectively. The judgment unit is used to determine the stability state of the power grid under different faults based on the positional relationship between the two-dimensional orthogonal feature vector, the benchmark instability discrimination boundary, and the extended instability discrimination boundary.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-8.

11. An electronic device, characterized in that, include: The computer-readable storage medium as described in claim 10; as well as One or more processors for executing a program in the computer-readable storage medium.

Images