Feature data screening method and device for power grid fault assessment

Abstract

The invention discloses a feature data screening method and device for power grid fault assessment, and belongs to the field of power systems, and the method comprises the steps: firstly obtaining a local bus set of a power grid and a candidate feature set corresponding to the set; performing a fault simulation experiment on the power grid according to a preset power system simulation platform to obtain characteristic quantity data corresponding to each characteristic quantity and form a characteristic quantity data set; according to a Mahalanobis distance principle and a Spearman level correlation coefficient calculation method, sorting each characteristic quantity data in the characteristic quantity data set to obtain an intermediate candidate characteristic set; screening each intermediate candidate feature in the intermediate candidate feature set by adopting an incremental feature subset strategy to obtain a plurality of final candidate features; and finally, performing fault assessment on the power grid according to each final candidate feature and the local bus set, and outputting a fault assessment result, so that the accuracy of power grid fault assessment can be improved through implementation of the method.

Description

Technical Field

[0001] This invention relates to the field of power systems, and more particularly to a method and apparatus for screening characteristic data for power grid fault assessment. Background Technology

[0002] In modern power systems, a mainstream approach to achieve rapid and accurate assessment of grid transient stability, especially voltage stability, is based on data-driven models. These models heavily rely on selecting key features from system operating data that effectively distinguish between stable and unstable states. Therefore, feature data selection methods for grid fault assessment are a core component in constructing a reliable assessment system.

[0003] In existing technologies, the screening of feature data related to power grid fault assessment largely relies on traditional power grid transient stability analysis methods. This is mainly achieved indirectly through time-domain simulation or direct methods. Time-domain simulation requires precise numerical integration of the power system's differential-algebraic equations to obtain full time-series data before feature selection. Direct methods extract a limited number of features based on simplified system models for assessment. However, the numerical integration process of time-domain simulation is computationally intensive and time-consuming, making it difficult to quickly screen key features to meet the needs of rapid online assessment after power grid faults. Direct methods, on the other hand, have poor adaptability to complex power grid models and reliability issues, often failing to accurately handle feature selection related to voltage transient stability in large-scale systems. This results in extracted features that cannot fully reflect the actual transient process, leading to problems such as low feature selection efficiency, weak adaptability to complex power grids, and an inability to provide high-quality feature input for rapid assessment. Summary of the Invention

[0004] This invention provides a feature data screening method and apparatus for power grid fault assessment, which can solve the problem in the prior art that it is difficult to improve the accuracy of power grid fault assessment while ensuring the validity of feature data and the controllability of input dimensions.

[0005] In a first aspect, embodiments of the present invention provide a method for filtering feature data for power grid fault assessment, comprising: Obtain the local bus set of the power grid and the candidate feature set corresponding to the local bus set; wherein, the candidate feature set includes several feature quantities; The power grid is subjected to fault simulation experiments based on a preset power system simulation platform to obtain characteristic quantity data corresponding to each characteristic quantity, and all characteristic quantity data are combined into a characteristic quantity data set. Based on the Mahalanobis distance principle and the Spearman rank correlation coefficient calculation method, the feature data in the feature data set are sorted to obtain the intermediate candidate feature set; An incremental feature subset strategy is used to filter each intermediate candidate feature in the intermediate candidate feature set to obtain several final candidate features; The power grid is assessed for faults based on the final candidate features and the local bus set, and the fault assessment results of the power grid are output.

[0006] This application first obtains a set of local power grid buses and corresponding candidate feature sets to avoid data redundancy across the entire power grid and reduce subsequent processing load. Then, reliable feature data is obtained through fault simulation experiments, providing a high-quality data source for screening. Next, this application combines the Mahalanobis distance principle with Spearman's rank correlation coefficient sorting to ensure the features' ability to discriminate fault states while reducing feature redundancy. Subsequently, an incremental feature subset strategy is used to screen the final candidate features, preventing excessive features from complicating the model. Finally, the fault is evaluated based on the final features and local buses. This overall process solves the problems of low screening efficiency, weak adaptability to complex power grids, and insufficient feature quality in traditional methods. While ensuring feature validity and controllable input dimensions, it significantly improves the accuracy and efficiency of power grid fault assessment, adapting to the fault assessment needs of large-scale complex power grids and providing efficient support for post-fault transient stability judgment.

[0007] As a preferred example of the first aspect, obtaining the local bus set of the power grid and the candidate feature set corresponding to the local bus set specifically involves: The fault information and node impedance matrix of the power grid are obtained, and each bus in the power grid is screened according to the fault information and the node impedance matrix to obtain a local bus set and a candidate feature set corresponding to the local bus set.

[0008] In this preferred example, local buses and candidate feature sets are screened using power grid fault information and node impedance matrices, overcoming the limitations of traditional methods that lack targeted bus selection and lead to data redundancy. Fault information can accurately locate the fault's impact range, and the node impedance matrix reflects the electrical connections between buses. Combining these two methods effectively screens out local buses closely related to the fault, eliminating interference from irrelevant buses and reducing redundancy in subsequent data processing. Simultaneously, the candidate feature set corresponds to local buses, ensuring a strong correlation between features and fault states, preventing irrelevant features from interfering with subsequent screening processes. This targeted screening lays a high-quality data foundation for subsequent fault simulation, feature ranking, and other stages, further improving overall feature screening efficiency and solving the problems of heavy computational burden and poor feature targeting caused by traditional full-scale processing of bus data.

[0009] As a preferred example of the first aspect, the step of sorting the feature data in the feature data set according to the Mahalanobis distance principle and the Spearman rank correlation coefficient calculation method to obtain an intermediate candidate feature set is specifically as follows: The IMDF value corresponding to each feature data in the feature data set is calculated according to the Mahalanobis distance principle, and the feature data is sorted according to the IMDF value corresponding to each feature data to obtain an initial candidate feature set. According to the Spearman rank correlation coefficient calculation method, the Spearman rank correlation coefficient corresponding to each initial candidate feature in the initial candidate feature set is calculated, and the intermediate candidate feature set is obtained by sorting according to the Spearman rank correlation coefficient corresponding to each candidate feature using the inter-group redundancy calculation method.

[0010] In this preferred example, the IMDF values ​​are calculated and ranked using the Mahalanobis distance principle. The IMDF value quantifies the ability of features to distinguish between stable and unstable states of the power grid, prioritizing features with strong discriminative power to form a high-quality initial candidate feature set, thus addressing the problem of insufficient discriminative power in traditional methods. Furthermore, the ranking is combined with Spearman's rank correlation coefficient and inter-group redundancy calculation. The Spearman's rank correlation coefficient accurately measures the correlation between features, while the inter-group redundancy calculation effectively eliminates redundant features with significant overlap, avoiding evaluation errors caused by feature interference. This two-step ranking approach ensures both strong discriminative power and low redundancy in the intermediate candidate feature set, overcoming the shortcomings of traditional single ranking methods that struggle to balance feature effectiveness and redundancy control. This provides high-quality intermediate features for subsequent incremental screening, further improving the accuracy of the final fault assessment.

[0011] As a preferred example of the first aspect, the incremental feature subset strategy is used to filter each intermediate candidate feature in the intermediate candidate feature set to obtain several final candidate features, specifically as follows: Establish a feature set, and add each intermediate candidate feature to the feature set in sequence according to the order of the intermediate candidate features in the intermediate candidate feature set; Each time an intermediate candidate feature is added, the model is trained using the time series data corresponding to the current feature set to obtain the current fault assessment model and the classification accuracy corresponding to the current fault assessment model. If the classification accuracy of the current fault assessment model reaches its maximum value and the classification accuracy of the next fault assessment model trained after adding intermediate candidate features is less than or equal to the classification accuracy of the current fault assessment model, then each final candidate feature is determined based on the intermediate candidate features contained in the current feature set.

[0012] In this preferred example, by sequentially adding intermediate candidate features, training the model, and monitoring classification accuracy, the actual contribution of each feature to the fault assessment model can be dynamically verified. The selection process stops when the classification accuracy reaches its maximum and subsequent feature additions no longer improve accuracy. This avoids low assessment accuracy due to insufficient features and prevents redundant features from increasing model complexity and reducing assessment efficiency. This dynamic adjustment method achieves precise control over feature dimensions, ensuring the final candidate feature set is both concise and effective, solving the problem of traditional methods that require one-time feature selection and struggle to balance feature quantity and model performance. Furthermore, the selection criteria based on classification accuracy are directly linked to the fault assessment objective, ensuring that the final features maximize assessment accuracy, improve model operating efficiency, and meet the timeliness requirements of rapid assessment after power grid faults.

[0013] As a preferred example of the first aspect, the step of performing fault assessment on the power grid based on each final candidate feature and the local bus set, and outputting the fault assessment result of the power grid, specifically includes: Based on each of the final candidate features and the local bus set, extract the time-series data corresponding to each of the final candidate features corresponding to the local bus set in the power grid; The power grid is assessed for faults based on the time-series data, and the fault assessment results are output.

[0014] In this preferred example, time-series data is extracted based on final candidate features and local bus data. The final candidate features, after multiple rounds of screening, possess strong specificity and effectiveness, while the local bus data focuses on the fault correlation range. The combination of these two methods accurately extracts key time-series data reflecting the transient process of power grid faults, eliminating irrelevant data interference, reducing data processing volume, and improving assessment speed. Compared to traditional methods that use full or irrelevant data for assessment, resulting in low efficiency and poor accuracy, this approach allows time-series data to directly serve the core needs of fault assessment, ensuring that the assessment results truly reflect the power grid fault state. Simultaneously, the time-series data can fully present the dynamic changes of the power grid after a fault, providing a comprehensive basis for transient stability judgment. This solves the problems of insufficient specificity and inaccurate reflection of fault impact in traditional assessment data, achieving rapid and accurate assessment of power grid faults.

[0015] In a second aspect, the present invention provides a feature data screening device for power grid fault assessment, comprising: a data acquisition module, a simulation experiment module, a data sorting module, a data screening module, and a fault assessment module. The data acquisition module is used to acquire the local bus set of the power grid and the candidate feature set corresponding to the local bus set; wherein, the candidate feature set includes several feature quantities; The simulation experiment module is used to conduct fault simulation experiments on the power grid according to the preset power system simulation platform, obtain the characteristic quantity data corresponding to each characteristic quantity, and form a characteristic quantity data set by combining all the characteristic quantity data. The data sorting module is used to sort the feature data in the feature data set according to the Mahalanobis distance principle and the Spearman rank correlation coefficient calculation method to obtain an intermediate candidate feature set. The data filtering module is used to filter each intermediate candidate feature in the intermediate candidate feature set using an incremental feature subset strategy to obtain several final candidate features. The fault assessment module is used to assess the power grid based on each final candidate feature and the local bus set, and output the fault assessment result of the power grid.

[0016] As a preferred example of the second aspect, the data acquisition module includes a data acquisition unit; The data acquisition unit is used to acquire fault information and node impedance matrix of the power grid, and to filter each bus in the power grid according to the fault information and node impedance matrix to obtain a local bus set and a candidate feature set corresponding to the local bus set.

[0017] As a preferred example of the second aspect, the data sorting module includes a first sorting unit and a second sorting unit; The first sorting unit is used to calculate the IMDF value corresponding to each feature data in the feature data set according to the Mahalanobis distance principle, and sort them according to the IMDF value corresponding to each feature data to obtain an initial candidate feature set. The second sorting unit is used to calculate the Spearman rank correlation coefficient corresponding to each initial candidate feature in the initial candidate feature set according to the Spearman rank correlation coefficient calculation method, and sort them according to the Spearman rank correlation coefficient corresponding to each candidate feature using the inter-group redundancy calculation method to obtain an intermediate candidate feature set.

[0018] As a preferred example of the second aspect, the data filtering module includes a first filtering unit, a second filtering unit, and a third filtering unit; The first filtering unit is used to establish a feature set and add each intermediate candidate feature to the feature set in sequence according to the arrangement order of each intermediate candidate feature in the intermediate candidate feature set; The second filtering unit is used to train the model using the time series data corresponding to the current feature set after each addition of intermediate candidate features, to obtain the current fault assessment model and the classification accuracy corresponding to the current fault assessment model. The third screening unit is used to determine each final candidate feature based on the intermediate candidate features contained in the current feature set if the classification accuracy of the current fault assessment model reaches the highest value and the classification accuracy of the next round of fault assessment model trained after adding intermediate candidate features is less than or equal to the classification accuracy of the current fault assessment model.

[0019] As a preferred example of the second aspect, the fault assessment module includes a first assessment unit and a second assessment unit; The first evaluation unit is used to extract time-series data corresponding to each of the final candidate features in the power grid based on each of the final candidate features and the local bus set; The second evaluation unit is used to perform fault evaluation on the power grid based on various time-series data and output the fault evaluation result of the power grid.

[0020] In summary, this application's embodiments first obtain the local bus set and corresponding candidate feature set of the power grid, avoiding data redundancy across the entire power grid and reducing subsequent processing workload. Then, reliable feature data is obtained through fault simulation experiments, providing a high-quality data source for screening. Next, this application combines the Mahalanobis distance principle with Spearman's rank correlation coefficient sorting to ensure the features' ability to discriminate fault states while reducing feature redundancy. Subsequently, an incremental feature subset strategy is used to screen the final candidate features, preventing excessive features from leading to model complexity. Finally, faults are evaluated based on the final features and local buses. This overall process solves the problems of low screening efficiency, weak adaptability to complex power grids, and insufficient feature quality in traditional methods. While ensuring feature validity and controllable input dimensions, it significantly improves the accuracy and efficiency of power grid fault assessment, adapting to the fault assessment needs of large-scale complex power grids and providing efficient support for post-fault transient stability judgment.

[0021] Another embodiment of the present invention provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the steps of the feature data screening method for power grid fault assessment of the present invention.

[0022] Another embodiment of the present invention provides a computer-readable storage medium item, including: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform the steps of the feature data screening method for power grid fault assessment of the present invention. Attached Figure Description

[0023] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating an embodiment of a feature data screening method for power grid fault assessment provided by the present invention. Figure 2 A topology diagram of an embodiment of a feature data screening method for power grid fault assessment provided by the present invention; Figure 3 This invention provides a monitorable feature category map of an embodiment of a feature data screening method for power grid fault assessment. Figure 4 A heatmap of the normalized matrix of electrical coupling strength between faults and affected buses, which is an embodiment of a feature data screening method for power grid fault assessment provided by the present invention; Figure 5 A feature group sorting diagram is provided as an embodiment of a feature data screening method for power grid fault assessment according to the present invention. Figure 6 A correlation analysis diagram of an embodiment of a feature data screening method for power grid fault assessment provided by the present invention; Figure 7 A classification accuracy ranking chart of an embodiment of a feature data screening method for power grid fault assessment provided by the present invention; Figure 8 This is a module structure diagram of an embodiment of a feature data screening device for power grid fault assessment provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0027] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0030] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0031] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0032] Example 1 See Figure 1To address the problem in existing technologies of improving the accuracy of power grid fault assessment while ensuring the validity of feature data and controllable input dimensions, an embodiment of the present invention provides a feature data screening method for power grid fault assessment, comprising: S1. Obtain the local bus set of the power grid and the candidate feature set corresponding to the local bus set; wherein, the candidate feature set includes several feature quantities.

[0033] In some embodiments of this application, obtaining the local bus set of the power grid and the candidate feature set corresponding to the local bus set specifically involves: The fault information and node impedance matrix of the power grid are obtained, and each bus in the power grid is screened according to the fault information and the node impedance matrix to obtain a local bus set and a candidate feature set corresponding to the local bus set.

[0034] Specifically, obtaining the local bus set of the power grid and the candidate feature set corresponding to the local bus set can be implemented in the following preferred manner: ① First, the network topology and line parameters of the power grid are analyzed, and the node impedance matrix Z-bus of the power grid is calculated and stored. Due to the effect of the excitation regulator, the steady-state voltage of the generator node generally does not change after a disturbance, so the nodes connected to the generator are not considered. This matrix is ​​a... A complex matrix (N is the total number of buses in the system minus the number of generators), whose any off-diagonal elements The impact of disturbances at node j on the voltage of node i was precisely quantified, reflecting the tightness of electrical coupling between nodes. This step is a one-time offline preparation. ② When a stability assessment of a specific transient process is required, the fault information of that disturbance must first be obtained, including at least the specific location of the fault. This information serves as the trigger signal for initiating the "transient impact subsystem" screening. ③ Based on the fault location information obtained in step ②, extract the corresponding column vector from the Z-bus matrix stored in step ①. Taking the fault occurring at bus F as an example, extract the F-th column of the Z-bus matrix. The modulus of each element in this column vector. This represents the electrical coupling strength between the faulty bus F and any other bus k in the system. An electrical distance threshold is set. The range for defining "strong influence" can be between 0.1 and 0.3. Iterate through all bus k in the system, and if the following conditions are met...

[0035] Then busbar k is determined to be the busbar most severely affected by the fault. The self-impedance of the faulty busbar represents the maximum impact benchmark. All buses that meet this condition together constitute the core busbar set of the "transient impact subsystem". ④ Based on the core bus set identified in step ③, the system automatically and dynamically collects and extracts physical quantities directly related to these buses to form a candidate feature set corresponding to the local bus set specific to this fault.

[0036] S2. Conduct fault simulation experiments on the power grid according to the preset power system simulation platform, obtain the characteristic quantity data corresponding to each characteristic quantity, and form a characteristic quantity data set by combining all characteristic quantity data.

[0037] Specifically, the step of conducting fault simulation experiments on the power grid based on a preset power system simulation platform to obtain characteristic quantity data corresponding to each characteristic quantity can be implemented in the following preferred manner: ① Construct a standard simulation case containing new energy equipment in a power system simulation platform, considering the following disturbance variables: the system load level increases by 5% in increments from 80% to 130% of the baseline value; the fault type is set as a three-phase short circuit; the fault location includes all buses and main lines, with multiple location points along the entire line length; the fault duration is set within the range of 0.1 to 0.2 seconds. Multiple operating conditions are generated through the above disturbance combinations. Under each operating condition, based on the candidate feature set corresponding to the local bus set, the simulation output of typical characteristic quantities such as voltage, active power, frequency, and current from the candidate feature set corresponding to the local bus set is collected.

[0038] ② Centered on the fault clearing time, a time window of 0.5 seconds is formed by truncating 5 power frequency cycles forward and 20 power frequency cycles backward. The following time-series feature tensor is constructed for each set of sample data: Where K is the time step, This represents the number of candidate features under the current fault. All samples constitute a three-dimensional input tensor: Where N is the number of perturbation samples.

[0039] ③ The final running result of each simulation sample is judged. This invention directly determines the voltage transient stability by monitoring the voltage trajectory of a preset set of key buses. If, after fault clearance, the voltage of any key bus fails to recover to above 0.85 pu for more than 2.0 seconds, the sample is judged as "voltage instability" and labeled "1". Conversely, if the voltage of all key buses can recover to above the threshold within a specified time, it is judged as "stable" and labeled "0".

[0040] ④ To eliminate the impact of inconsistent feature dimensions on subsequent distance calculations, each feature is standardized with zero mean to obtain a unified standard input tensor. Based on the unified standard input tensor, the feature data corresponding to each feature is obtained for the next stage of feature screening and analysis.

[0041] S3. Based on the Mahalanobis distance principle and the Spearman rank correlation coefficient calculation method, sort the feature data in the feature data set to obtain the intermediate candidate feature set.

[0042] In some embodiments of this application, the step of sorting the feature data in the feature data set according to the Mahalanobis distance principle and the Spearman rank correlation coefficient calculation method to obtain an intermediate candidate feature set specifically involves: The IMDF value corresponding to each feature data in the feature data set is calculated according to the Mahalanobis distance principle, and the feature data is sorted according to the IMDF value corresponding to each feature data to obtain an initial candidate feature set. According to the Spearman rank correlation coefficient calculation method, the Spearman rank correlation coefficient corresponding to each initial candidate feature in the initial candidate feature set is calculated, and the intermediate candidate feature set is obtained by sorting according to the Spearman rank correlation coefficient corresponding to each candidate feature using the inter-group redundancy calculation method.

[0043] S4. The incremental feature subset strategy is used to filter each intermediate candidate feature in the intermediate candidate feature set to obtain several final candidate features.

[0044] In some embodiments of this application, the step of using an incremental feature subset strategy to filter each intermediate candidate feature in the intermediate candidate feature set to obtain several final candidate features specifically involves: Establish a feature set, and add each intermediate candidate feature to the feature set in sequence according to the order of the intermediate candidate features in the intermediate candidate feature set; Each time an intermediate candidate feature is added, the model is trained using the time series data corresponding to the current feature set to obtain the current fault assessment model and the classification accuracy corresponding to the current fault assessment model. If the classification accuracy of the current fault assessment model reaches its maximum value and the classification accuracy of the next fault assessment model trained after adding intermediate candidate features is less than or equal to the classification accuracy of the current fault assessment model, then each final candidate feature is determined based on the intermediate candidate features contained in the current feature set.

[0045] Specifically, an incremental feature subset strategy is used to filter each intermediate candidate feature in the intermediate candidate feature set to obtain several final candidate features. This can be implemented in the following preferred manner: An incremental feature subset strategy is adopted, adding feature groups sequentially in the order of arrangement. When the model classification accuracy reaches its best and converges to a stable state, the current feature set is determined as the optimal set. The busbars, loads, generators (including synchronous generators, photovoltaic power plants, doubly-fed induction generators, etc.), and lines from the "transient subsystem" in S1-4 will be combined with the optimal set, and time-series data will be extracted according to the following rules: For each busbar in the core busbar set, the time-series data of its voltage amplitude (F4), voltage phase angle (F5), and frequency (F6) will be extracted; For all generator sets (including synchronous machines, wind turbines, and photovoltaic power plants) directly connected to the core busbar, the time-series data of their active power output (F2, F7, F9), reactive power output (F3, F8), and power angle (F1) will be extracted (note that since photovoltaic power plants mainly output active power, their reactive power is relatively small and can be ignored in the modeling process); For all loads directly connected to the core busbar, the time-series data of their active power (F10) and reactive power (F11) will be extracted; For all lines whose two ends are connected to the core busbar set, the time-series data of their line current (F12) will be extracted. All the time-series data extracted above will be used as the final input features for subsequent modeling work.

[0046] S5. Perform fault assessment on the power grid based on each final candidate feature and the local bus set, and output the fault assessment result of the power grid.

[0047] In some embodiments of this application, the step of performing fault assessment on the power grid based on each final candidate feature and the local bus set, and outputting the fault assessment result of the power grid, specifically includes: Based on each of the final candidate features and the local bus set, extract the time-series data corresponding to each of the final candidate features corresponding to the local bus set in the power grid; The power grid is assessed for faults based on the time-series data, and the fault assessment results are output.

[0048] Specifically, to fully explain the above steps, the following scheme will be used as an example: ①This embodiment uses an improved IEEE-39 node system with new energy equipment for testing and analysis, and its topology is as follows: Figure 2 As shown. The system can monitor the following feature categories: Figure 3 As shown. According to the present invention, we do not... Figure 3 Instead of using all data from all categories as initial input, candidate feature sets are dynamically constructed for specific faults. Taking a three-phase short-circuit fault occurring on bus 20 as an example: firstly, based on... Figure 4Based on the normalized matrix heatmap of the fault-affected bus electrical coupling strength calculated using the Z-bus matrix, it was determined that the fault would primarily affect bus 20 itself, electrically adjacent buses 34 and 41, downstream critical load 20, and the power supply generator G05 and the DFIG2 wind turbine providing the main voltage support for this area. Therefore, regarding this fault, from... Figure 3 The dynamically extracted "candidate feature set" from the general categories will only include feature sets of buses, loads, and generators related to the fault. Through this step, the number of features in the initial analysis has been drastically reduced from hundreds across the entire system to a dozen or so, and each one is highly relevant to the current fault.

[0049] ② Construct a simulation sample set covering multiple operating conditions and disturbances, extract time-series data of key features before and after the disturbances, complete sample label determination and normalization processing, and provide standardized input data for feature selection. This embodiment uses an improved IEEE-39 node system with renewable energy equipment for testing and analysis, its topology as follows: Figure 2 As shown, the system comprises 10 synchronous turbine generators, 4 wind farms, and 2 photovoltaic farms. The load adopts a comprehensive load model (constant impedance + induction motor), with a base capacity of 100 MVA and a rated frequency of 50 Hz. Transient sample data is generated by building the system in the DIgSILENT platform and setting various fault scenarios. To cover diverse transient conditions, the sample generation in this embodiment comprehensively considers factors such as load level, fault location, and duration. The system load increases by 5% within the range of 80% to 130% of the base value, constructing a total of 11 operating modes, and adjusting the generator output to maintain power flow convergence. Three-phase short-circuit faults are randomly applied to 39 buses and 34 lines. The fault location along the entire line length varies from 0% to 90% in 10% increments, and the fault duration is set between 0.1 and 0.2 s, with a simulation duration of 6 s. Considering that faults in actual power systems often manifest as gradual electrical disturbances with difficult-to-define start times, and that fault clearance is typically triggered precisely by protection actions, and to meet the timeliness requirements of TSA (Transmission Control Alert), for each simulation, the method described in S1 is first used to determine the candidate feature set. Then, based on the fault clearance time, 5 power frequency cycles are extracted forward and 20 power frequency cycles (i.e., 0.4s) are extracted to capture the dynamic processes of all features within the candidate feature set. All time-series data are zero-mean standardized and used as input for model selection.

[0050] ③ Based on the Mahalanobis distance principle, the response differences of various time-series features in the candidate feature set under "stable" and "unstable" states are quantified. Feature groups are ranked according to their average discriminative power, with priority given to retaining feature groups with significant discriminative power. The IMDF values ​​of each feature group in the candidate feature set between stable and unstable samples are calculated, and the results are as follows: Figure 5As shown, the larger the IMDF value, the more important it is for subsequent classification tasks.

[0051] ④ Redundancy between feature groups is obtained by calculating the Spearman rank correlation coefficient. Feature groups with strong discriminative power are retained based on the redundancy threshold. An incremental feature subset strategy is used to determine the optimal feature group set, and time-series data of relevant components of the "transient subsystem" are extracted according to rules for modeling. After evaluating the importance of candidate feature groups, redundancy analysis is further performed based on the Spearman correlation coefficient. Figure 6 The heatmap shown identifies highly correlated feature pairs, providing a basis for subsequent screening. As can be seen from the figure, the correlation coefficients of F1 and F5, F3 and F4, F4 and F9, and F10 and F11 are all greater than 0.7, indicating strong redundancy among them. This high correlation between feature pairs largely stems from the coupling relationships of physical variables within the power system, such as the phase synchronization mechanism between power angle and voltage phase angle, and the adjustment relationship between reactive power and voltage amplitude. Retaining such redundant features may lead to an expansion of the feature space dimension, increasing the model's learning burden and increasing the risk of overfitting. According to... Figure 5 The results show that the 12 feature groups in the candidate set are sorted from high to low according to their IMDF values, forming an initial priority sequence S(1)=[F4,F11,F9,F7,F3,F10,F8,F2,F12,F1,F5,F6]. Based on this, combined with... Figure 6 Correlation analysis was performed to retain the more discriminative feature pairs from highly correlated feature pairs, while redundant and less discriminative feature pairs were moved to the end of the feature set, forming an optimized ranking S(2)=[F4,F11,F7,F8,F2,F12,F1,F6,F9,F3,F10,F5]. Subsequently, a stepwise accumulation strategy was adopted to add feature pairs sequentially to the sample feature set, train the voltage transient model, and evaluate the classification accuracy under different numbers of feature pairs. Figure 7 As shown, only the first four sets are needed to achieve the highest accuracy. Ultimately, based on... Figure 3 Based on the original features in the model, and after two-stage time-series feature selection, this paper selects the node voltage amplitude, load reactive power, wind power output active and reactive power, namely [F4, F11, F7, F8] as the inputs for subsequent analysis of the voltage transient stability model performance.

[0052] ⑤ Finally, the results of the initial physical selection and the final statistical selection are compared to filter out the reactive power of load 20, the active and reactive power of DFIG2 wind turbines, and the time-series data of the voltage amplitude of the initially selected bus. These characteristic data are standardized to eliminate dimensional differences and ensure data consistency. Then, an appropriate evaluation model is selected, and a threshold judgment method is used. If the bus voltage amplitude recovers to above 0.85 pu within 2 seconds after fault clearance, it is considered stable; otherwise, it is considered unstable. The model is trained based on the selected characteristic data to predict stability. After model training, the accuracy of the model can be verified by dividing it into training and test sets to ensure the reliability of the evaluation results. In practical applications, the system will input the current power grid characteristic data in real time, perform stability evaluation based on the model, and trigger an early warning mechanism in a timely manner if the system becomes unstable, adjusting the power grid operation to prevent further instability. Furthermore, the evaluation results can be fed back to the system, and the model and feature set can be optimized through incremental learning to improve the efficiency and accuracy of the evaluation, thereby providing decision support for power grid dispatching and fault handling.

[0053] In summary, this application's embodiments first obtain the local bus set and corresponding candidate feature set of the power grid, avoiding data redundancy across the entire power grid and reducing subsequent processing workload. Then, reliable feature data is obtained through fault simulation experiments, providing a high-quality data source for screening. Next, this application combines the Mahalanobis distance principle with Spearman's rank correlation coefficient sorting to ensure the features' ability to discriminate fault states while reducing feature redundancy. Subsequently, an incremental feature subset strategy is used to screen the final candidate features, preventing excessive features from leading to model complexity. Finally, faults are evaluated based on the final features and local buses. This overall process solves the problems of low screening efficiency, weak adaptability to complex power grids, and insufficient feature quality in traditional methods. While ensuring feature validity and controllable input dimensions, it significantly improves the accuracy and efficiency of power grid fault assessment, adapting to the fault assessment needs of large-scale complex power grids and providing efficient support for post-fault transient stability judgment.

[0054] Example 2 like Figure 8 As shown, based on the above method embodiments, corresponding device embodiments are provided; An embodiment of the present invention provides a feature data screening device for power grid fault assessment, comprising: a data acquisition module 81, a simulation experiment module 82, a data sorting module 83, a data screening module 84, and a fault assessment module 85; In some embodiments of this application, the data acquisition module 81 includes a data acquisition unit; The data acquisition unit is used to acquire fault information and node impedance matrix of the power grid, and to filter each bus in the power grid according to the fault information and node impedance matrix to obtain a local bus set and a candidate feature set corresponding to the local bus set.

[0055] In some embodiments of this application, the data sorting module 83 includes a first sorting unit and a second sorting unit; The first sorting unit is used to calculate the IMDF value corresponding to each feature data in the feature data set according to the Mahalanobis distance principle, and sort them according to the IMDF value corresponding to each feature data to obtain an initial candidate feature set. The second sorting unit is used to calculate the Spearman rank correlation coefficient corresponding to each initial candidate feature in the initial candidate feature set according to the Spearman rank correlation coefficient calculation method, and sort them according to the Spearman rank correlation coefficient corresponding to each candidate feature using the inter-group redundancy calculation method to obtain an intermediate candidate feature set.

[0056] In some embodiments of this application, the data filtering module 84 includes a first filtering unit, a second filtering unit, and a third filtering unit; The first filtering unit is used to establish a feature set and add each intermediate candidate feature to the feature set in sequence according to the arrangement order of each intermediate candidate feature in the intermediate candidate feature set; The second filtering unit is used to train the model using the time series data corresponding to the current feature set after each addition of intermediate candidate features, to obtain the current fault assessment model and the classification accuracy corresponding to the current fault assessment model. The third screening unit is used to determine each final candidate feature based on the intermediate candidate features contained in the current feature set if the classification accuracy of the current fault assessment model reaches the highest value and the classification accuracy of the next round of fault assessment model trained after adding intermediate candidate features is less than or equal to the classification accuracy of the current fault assessment model.

[0057] In some embodiments of this application, the fault assessment module 85 includes a first assessment unit and a second assessment unit; For more detailed steps and working principles of this embodiment, please refer to the relevant description in Embodiment 1, but not limited to these descriptions.

[0058] In summary, this application's embodiments first obtain the local bus set and corresponding candidate feature set of the power grid, avoiding data redundancy across the entire power grid and reducing subsequent processing workload. Then, reliable feature data is obtained through fault simulation experiments, providing a high-quality data source for screening. Next, this application combines the Mahalanobis distance principle with Spearman's rank correlation coefficient sorting to ensure the features' ability to discriminate fault states while reducing feature redundancy. Subsequently, an incremental feature subset strategy is used to screen the final candidate features, preventing excessive features from leading to model complexity. Finally, faults are evaluated based on the final features and local buses. This overall process solves the problems of low screening efficiency, weak adaptability to complex power grids, and insufficient feature quality in traditional methods. While ensuring feature validity and controllable input dimensions, it significantly improves the accuracy and efficiency of power grid fault assessment, adapting to the fault assessment needs of large-scale complex power grids and providing efficient support for post-fault transient stability judgment.

[0059] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the feature data screening method for power grid fault assessment provided by any of the above-described method embodiments of the present invention.

[0060] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0061] Example 3 Based on the above embodiments of the feature data screening method for power grid fault assessment, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the feature data screening method for power grid fault assessment of any embodiment of the present invention.

[0062] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0063] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0064] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0065] Example 4 Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the feature data screening method for power grid fault assessment described in any of the above-described method embodiments of the present invention.

[0066] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0067] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for screening feature data for power grid fault assessment, characterized in that, include: Obtain the local bus set of the power grid and the candidate feature set corresponding to the local bus set; wherein, the candidate feature set includes several feature quantities; The power grid is subjected to fault simulation experiments based on a preset power system simulation platform to obtain characteristic quantity data corresponding to each characteristic quantity, and all characteristic quantity data are combined into a characteristic quantity data set. Based on the Mahalanobis distance principle and the Spearman rank correlation coefficient calculation method, the feature data in the feature data set are sorted to obtain the intermediate candidate feature set; An incremental feature subset strategy is used to filter each intermediate candidate feature in the intermediate candidate feature set to obtain several final candidate features; The power grid is assessed for faults based on the final candidate features and the local bus set, and the fault assessment results of the power grid are output.

2. The feature data screening method for power grid fault assessment as described in claim 1, characterized in that, The specific steps for obtaining the local bus set of the power grid and the candidate feature set corresponding to the local bus set are as follows: The fault information and node impedance matrix of the power grid are obtained, and each bus in the power grid is screened according to the fault information and the node impedance matrix to obtain a local bus set and a candidate feature set corresponding to the local bus set.

3. The feature data screening method for power grid fault assessment as described in claim 1, characterized in that, The method of sorting the feature data in the feature data set according to the Mahalanobis distance principle and the Spearman rank correlation coefficient calculation method to obtain the intermediate candidate feature set is as follows: The IMDF value corresponding to each feature data in the feature data set is calculated according to the Mahalanobis distance principle, and the feature data is sorted according to the IMDF value corresponding to each feature data to obtain an initial candidate feature set. According to the Spearman rank correlation coefficient calculation method, the Spearman rank correlation coefficient corresponding to each initial candidate feature in the initial candidate feature set is calculated, and the intermediate candidate feature set is obtained by sorting according to the Spearman rank correlation coefficient corresponding to each candidate feature using the inter-group redundancy calculation method.

4. The feature data screening method for power grid fault assessment as described in claim 1, characterized in that, The incremental feature subset strategy is used to filter the intermediate candidate features in the intermediate candidate feature set to obtain several final candidate features, specifically as follows: Establish a feature set, and add each intermediate candidate feature to the feature set in sequence according to the order of the intermediate candidate features in the intermediate candidate feature set; Each time an intermediate candidate feature is added, the model is trained using the time series data corresponding to the current feature set to obtain the current fault assessment model and the classification accuracy corresponding to the current fault assessment model. If the classification accuracy of the current fault assessment model reaches its maximum value and the classification accuracy of the next fault assessment model trained after adding intermediate candidate features is less than or equal to the classification accuracy of the current fault assessment model, then each final candidate feature is determined based on the intermediate candidate features contained in the current feature set.

5. The feature data screening method for power grid fault assessment as described in claim 1, characterized in that, The process of performing fault assessment on the power grid based on each final candidate feature and the local bus set, and outputting the fault assessment result of the power grid, specifically includes: Based on each of the final candidate features and the local bus set, extract the time-series data corresponding to each of the final candidate features corresponding to the local bus set in the power grid; The power grid is assessed for faults based on the time-series data, and the fault assessment results are output.

6. A feature data screening device for power grid fault assessment, characterized in that, include: The system includes a data acquisition module, a simulation experiment module, a data sorting module, a data filtering module, and a fault assessment module. The data acquisition module is used to acquire the local bus set of the power grid and the candidate feature set corresponding to the local bus set; wherein, the candidate feature set includes several feature quantities; The simulation experiment module is used to conduct fault simulation experiments on the power grid according to the preset power system simulation platform, obtain the characteristic quantity data corresponding to each characteristic quantity, and form a characteristic quantity data set by combining all the characteristic quantity data. The data sorting module is used to sort the feature data in the feature data set according to the Mahalanobis distance principle and the Spearman rank correlation coefficient calculation method to obtain an intermediate candidate feature set. The data filtering module is used to filter each intermediate candidate feature in the intermediate candidate feature set using an incremental feature subset strategy to obtain several final candidate features. The fault assessment module is used to assess the power grid based on each final candidate feature and the local bus set, and output the fault assessment result of the power grid.

7. The feature data screening device for power grid fault assessment as described in claim 6, characterized in that, The data acquisition module includes a data acquisition unit; The data acquisition unit is used to acquire fault information and node impedance matrix of the power grid, and to filter each bus in the power grid according to the fault information and node impedance matrix to obtain a local bus set and a candidate feature set corresponding to the local bus set.

8. The feature data screening device for power grid fault assessment as described in claim 6, characterized in that, The data sorting module includes a first sorting unit and a second sorting unit; The first sorting unit is used to calculate the IMDF value corresponding to each feature data in the feature data set according to the Mahalanobis distance principle, and sort them according to the IMDF value corresponding to each feature data to obtain an initial candidate feature set. The second sorting unit is used to calculate the Spearman rank correlation coefficient corresponding to each initial candidate feature in the initial candidate feature set according to the Spearman rank correlation coefficient calculation method, and sort them according to the Spearman rank correlation coefficient corresponding to each candidate feature using the inter-group redundancy calculation method to obtain an intermediate candidate feature set.

9. A feature data screening device for power grid fault assessment as described in claim 6, characterized in that, The data filtering module includes a first filtering unit, a second filtering unit, and a third filtering unit; The first filtering unit is used to establish a feature set and add each intermediate candidate feature to the feature set in sequence according to the arrangement order of each intermediate candidate feature in the intermediate candidate feature set; The second filtering unit is used to train the model using the time series data corresponding to the current feature set after each addition of intermediate candidate features, to obtain the current fault assessment model and the classification accuracy corresponding to the current fault assessment model. The third screening unit is used to determine each final candidate feature based on the intermediate candidate features contained in the current feature set if the classification accuracy of the current fault assessment model reaches the highest value and the classification accuracy of the next round of fault assessment model trained after adding intermediate candidate features is less than or equal to the classification accuracy of the current fault assessment model.

10. The feature data screening device for power grid fault assessment as described in claim 6, characterized in that, The fault assessment module includes a first assessment unit and a second assessment unit; The first evaluation unit is used to extract time-series data corresponding to each of the final candidate features in the power grid based on each of the final candidate features and the local bus set; The second evaluation unit is used to perform fault evaluation on the power grid based on various time-series data and output the fault evaluation result of the power grid.

Images