Incremental sample screening method for power grid transient stability discrimination model training

Abstract

The invention discloses an incremental sample screening method for power grid transient stability discrimination model training. The method comprises the steps of obtaining a basic sample set based on a plurality of initial operation modes of a power grid; obtaining a to-be-selected sample based on the generated power grid operation mode; calculating the distance between the to-be-selected sample and each basic sample in the basic sample set; constructing a neighbor sample subset of the to-be-selected sample based on the distance; calculating a neighbor distance between the to-be-selected sample and the neighbor sample subset; judging whether the to-be-selected sample is an incremental sample based on the neighbor distance; and training the power grid transient stability discrimination model by adopting the incremental sample. The screening of the incremental samples only depends on time domain simulation and distance calculation, the identification difficulty of the incremental samples can be greatly reduced, and the data requirement of the artificial intelligence model training for the transient stability discrimination of the power grid can be effectively reduced by identifying the key samples with important value for stability discrimination.

Description

Technical Field

[0001] This invention relates to the field of power system analysis, and in particular to an incremental sample selection method for training a power grid transient stability discrimination model. Background Technology

[0002] A safe and reliable power supply is the foundation of socio-economic development. After a fault occurs, the power system may face the risk of transient instability. The activation of the stability control system may lead to generator and load shedding or even out-of-synchronization disconnection, causing the power grid to malfunction. Therefore, real-time transient stability assessment of the power grid is crucial for risk prevention and stability restoration. However, with the continuous expansion of my country's power grid, the construction and integration of large-scale renewable energy power plants, the operation of inter-regional DC systems, and the application of numerous power electronic devices, the dynamic characteristics of the system after a fault have become more complex, making accurate assessment of power grid transient stability increasingly difficult. In this context, artificial intelligence (AI) technology offers a new approach to stability assessment. However, AI training relies on massive amounts of sample data. When applied to large power grids, traversing various operating modes and faults becomes extremely time-consuming, making it difficult to directly implement the generation of transient response data under various operating modes for AI model training. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide an incremental sample selection method for training a power grid transient stability discrimination model. By identifying key samples that are of great value to stability discrimination in power grid transient stability discrimination, the amount of data relied upon in the training process of the stability discrimination artificial intelligence model is reduced, thus solving the technical problem that the training of the power grid transient stability discrimination model requires a large amount of sample data.

[0004] This invention provides an incremental sample selection method for training a power grid transient stability discrimination model, comprising:

[0005] A basic sample set is obtained based on multiple initial operating modes of the power grid. Each initial operating mode corresponds to multiple preset faults. The basic sample set includes multiple basic samples, and the basic samples are the electrical transient response trajectories of a single preset fault under the initial operating mode.

[0006] A feasible power grid operation mode is generated, and a candidate sample is obtained based on the generated power grid operation mode. The candidate sample includes the electrical transient response trajectory of a single preset fault under the generated power grid operation mode.

[0007] Calculate the distance between the candidate sample and each base sample in the base sample set;

[0008] Construct a subset of nearest neighbor samples for the candidate sample based on the distance;

[0009] Calculate the nearest neighbor distance between the candidate sample and the nearest neighbor subset;

[0010] Based on the nearest neighbor distance, determine whether the candidate sample is an incremental sample;

[0011] The incremental samples are used to train the power grid transient stability discrimination model.

[0012] Furthermore, the preset faults include anticipated faults and unanticipated faults. The anticipated faults include faults that are identified in advance and included in the safety verification scope during the planning stage of the power system, while the unanticipated faults include faults for which effective countermeasures have not been formulated during the planning stage of the power system.

[0013] Furthermore, the anticipated faults include at least one of the power grid faults N-1, N-2, and N-3, while the unintended faults include at least one of the following: multiple DC commutation failures, DC blocking, safety control device malfunction, and sudden changes in new energy output.

[0014] Furthermore, the electrical quantities of the preset fault include at least one of the following: generator active power, generator reactive power, generator speed, bus voltage amplitude, bus voltage phase angle, branch active power, and branch reactive power.

[0015] Further, calculating the distance between the candidate sample and each base sample includes:

[0016] Calculate the Euclidean distance and cosine similarity distance between the candidate sample and each of the base samples respectively;

[0017] Calculate the composite distance between the candidate sample and each base sample, wherein the composite distance includes the product of the Euclidean distance and the cosine similarity distance;

[0018] The distance is the composite distance.

[0019] Furthermore, constructing a subset of nearest neighbor samples for the candidate sample based on the distance includes:

[0020] Sort the distances by size to get N S There are N minimum distances, where N is the minimum distance. S It is a positive integer;

[0021] The base samples are grouped with N. S The base sample corresponding to the minimum distance is taken as the nearest neighbor sample subset.

[0022] Further, calculating the nearest neighbor distance between the candidate sample and the nearest neighbor subset includes:

[0023] The center-nearest neighbor distance between the candidate sample and the center point of the nearest neighbor subset is calculated using a composite distance method; or / and

[0024] Calculate the N S The average of the minimum distances, where the average is the average nearest neighbor distance.

[0025] Further, determining whether the candidate sample is an incremental sample based on the nearest neighbor distance includes:

[0026] The candidate samples whose center nearest neighbor distance is greater than the center nearest neighbor threshold are incremental samples; or

[0027] The candidate samples whose average nearest neighbor distance is greater than the average nearest neighbor threshold are incremental samples; or

[0028] The candidate samples whose center nearest neighbor distance is greater than the center nearest neighbor threshold and whose average nearest neighbor distance is greater than the average nearest neighbor threshold are incremental samples.

[0029] Further, determining whether the candidate sample is an incremental sample based on the nearest neighbor distance includes:

[0030] Calculate the composite boundary index λ of the candidate samples. C The formula is as follows:

[0031]

[0032] Where exp{·} is the exponential function, r is the integral variable, μ is the average nearest neighbor distance, R is the central nearest neighbor distance, and σ is the standard deviation of the composite distance between the candidate sample and the basic sample in the nearest neighbor sample subset;

[0033] The composite boundary index is compared with the critical threshold, and the candidate samples that are less than the critical threshold are selected as incremental samples.

[0034] Furthermore, the power grid transient stability discrimination model is an artificial intelligence model.

[0035] The beneficial effects of the embodiments of the present invention are:

[0036] This invention constructs a nearest neighbor subset of candidate samples by analyzing the distances between candidate samples and each basic sample in the basic sample set under generated power grid operation modes. Based on the nearest neighbor distance between the candidate sample and the nearest neighbor subset, it determines whether the candidate sample is an incremental sample. Therefore, it is an unsupervised screening method to identify key incremental samples for power grid transient stability judgment. The screening of incremental samples only relies on time-domain simulation and distance calculation, which greatly reduces the difficulty of identifying incremental samples. By identifying key samples that are of great value to stability judgment, this invention can effectively reduce the data requirements for training artificial intelligence models for power grid transient stability judgment. This invention further designs a composite boundary index to determine whether a candidate sample is a key incremental sample. Since the composite boundary index can simultaneously measure the overall similarity between a candidate sample and its nearest neighbor subset as well as the probability that the candidate sample belongs to the sample space of its nearest neighbor subset, the candidate samples selected by the composite boundary index as key incremental samples can provide additional effective information for the training and updating of the model. It can effectively reduce the training data requirements of the artificial intelligence model for power grid transient stability discrimination while improving the efficiency of model construction and the accuracy of discrimination results, thereby accelerating the development and deployment of artificial intelligence models and having important practical value. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart illustrating the incremental sample selection method for training the power grid transient stability discrimination model of the present invention.

[0039] Figure 2 This is a simplified network structure diagram of a main network in a certain region of China, which is a specific application of the screening method of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0041] This invention provides a method for selecting incremental samples for training a power grid transient stability discrimination model, such as... Figure 2As shown, the method of this invention uses transient simulation data under a simplified main network typical operation mode in a certain region of China as the research object. Figure 2 The network structure comprises 197 nodes in Region A and Region B, connected by single-circuit lines, double-circuit lines, or busbars. It includes multiple thermal power plants, renewable energy power stations, power stations, four DC transmission lines, and corresponding grid-connected converters (LCCs). The renewable energy power stations include wind farms and photovoltaic power plants.

[0042] Combination Figure 1 As shown, the incremental sample selection method for training the power grid transient stability discrimination model of the present invention includes the following steps:

[0043] S1. Obtain a basic sample set based on multiple initial operating modes of the power grid.

[0044] Multiple initial operating modes for different power grid conditions are set. Time-domain simulation is used to obtain the transient response trajectories of electrical quantities corresponding to each initial operating mode and various preset faults. Each initial operating mode corresponds to multiple preset faults, and each basic sample is the transient response trajectory of electrical quantities for a single preset fault under a certain initial operating mode. All basic samples constitute the basic sample set. Preset faults include anticipated faults and unanticipated faults, totaling M types. Anticipated faults include faults that are pre-identified and included in the safety verification scope during the power system planning stage, such as at least one of the following: N-1, N-2, and N-3. In the N-1 criterion, N refers to the number of related lines or components in the system, including but not limited to transmission lines, transformers, generators, etc. N-1 means that when any component in the system fails or disconnects, the remaining components should be able to continue to operate normally, ensuring stable operation and continuous power supply without causing the entire system to collapse due to the failure of a single component. If it can be guaranteed that power can still be supplied normally when two devices are decommissioned, it is called N-2. The larger the number minus N, the higher the equipment redundancy and the higher the power grid's power supply reliability. Although the power system is constrained by the N-1 safety criterion and takes into account the typical set of anticipated accidents, there is still a possibility that a combination of failure events other than the anticipated failures may be triggered as the initial failure, i.e., unintended failures. Unintended failures include failures for which effective countermeasures were not formulated during the planning stage of the power system, such as at least one of the following: multiple DC commutation failures, DC blocking, failure of safety control devices to operate, and sudden changes in the output of new energy sources.

[0045] The transient simulation software used is PSD-BPA developed by China Electric Power Research Institute Co., Ltd. Based on each given initial operating mode, various preset faults are traversed and time-domain simulations are performed separately. The numerical changes in the transient responses of electrical quantities during the simulation are continuously recorded as the transient response trajectories. The simulation duration is set to T seconds, and the simulation step size is t milliseconds. P electrical quantity types are included, such as generator active power, generator reactive power, generator speed, bus voltage amplitude, bus voltage phase angle, branch active power, and branch reactive power. The transient response trajectories obtained from the simulation are used as basic samples. Each basic sample includes P transient response trajectories corresponding to one preset fault. For L different initial operating modes, time-domain simulations are performed on all M preset faults under each initial operating mode. P transient response trajectories of electrical quantities are collected for each preset fault, thus generating a total of L*M basic samples forming a basic sample set.

[0046] S2. Generate feasible power grid operation modes, and obtain candidate samples based on the generated power grid operation modes. The candidate samples include the electrical transient response trajectory of a single preset fault under the generated power grid operation modes.

[0047] Based on the actual operating conditions of the system, the output and load of new energy sources are randomly adjusted to generate a feasible power grid operation mode. A fault is randomly set, and the transient response trajectories of electrical quantities corresponding to the generated power grid operation mode and the preset fault are obtained through time-domain simulation. The simulation duration is still set to T seconds, and the simulation step size is still set to t milliseconds. The electrical quantity types include generator active power, generator reactive power, generator speed, bus voltage amplitude, bus voltage phase angle, branch active power, and branch reactive power. That is, the electrical quantity types in step S2 are consistent with the electrical quantity types in step S1, totaling P electrical quantity types. The transient response trajectories of electrical quantities obtained from the simulation are used as candidate samples.

[0048] S3. Calculate the distance between the candidate sample and each basic sample in the basic sample set;

[0049] S3-1. Calculate the Euclidean distance and cosine similarity distance between the candidate sample and each base sample respectively.

[0050] The distance between the candidate sample and each basic sample in the basic sample set is calculated using Euclidean distance. The calculation formula is as follows:

[0051]

[0052] Where D E x is a Euclidean distance metric. d x is the candidate sample. s The basic samples in the basic sample set, where k represents the k-th type of electrical quantity, and N FN represents the total number of electrical quantity types, l represents the l-th data point of the transient response trajectory of the electrical quantity, and N represents the total number of electrical quantity types. T This represents the total number of data points for the transient response trajectory of electrical quantities in the time-domain simulation. Let l be the value of the transient response trajectory of the k-th type of electrical quantity in the candidate sample at the l-th data point. It is the value of the transient response trajectory of the k-th type of electrical quantity in the basic sample set at the l-th data point.

[0053] The distance between the candidate sample and each base sample in the base sample set is calculated using cosine similarity. The calculation formula is as follows:

[0054]

[0055] Where D S For cosine similarity measurement, x d x is the candidate sample. s The basic samples in the basic sample set, where k represents the k-th type of electrical quantity, and N F N represents the total number of electrical quantity types, l represents the l-th data point of the transient response trajectory of the electrical quantity, and N represents the total number of electrical quantity types. T This represents the total number of data points for the transient response trajectory of electrical quantities in the time-domain simulation. Let l be the value of the transient response trajectory of the k-th type of electrical quantity in the candidate sample at the l-th data point. It is the value of the transient response trajectory of the k-th type of electrical quantity in the basic sample set at the l-th data point;

[0056] S3-2. Calculate the composite distance between the candidate sample and each base sample. The composite distance includes the product of the Euclidean distance and the cosine similarity distance.

[0057] The composite distance between the candidate sample and each base sample in the base sample set is calculated using the following formula:

[0058] D C (x d ,x s ) = D E (x d ,x s )·D S (x d ,x s )

[0059] Where D C It is a composite distance metric.

[0060] S3-3. The distance between the candidate sample and each basic sample is the composite distance.

[0061] In this invention, the composite distance is used as the distance between the candidate sample and each basic sample. The composite distance combines two measures: Euclidean distance and cross-similarity. It can comprehensively reflect the differences between samples from both the numerical and shape aspects of the transient response trajectory of electrical quantities, and is more accurate than a single measure.

[0062] To simplify the distance calculation process, one of the following can be used to calculate the distance between the candidate sample and each base sample: Euclidean distance, cosine similarity, or Manhattan distance.

[0063] S4. Construct a subset of nearest neighbor samples for the candidate samples based on distance;

[0064] Sort the distances obtained in step S3 by size to get N. S There are N minimum distances, where N is the minimum distance. S It is a positive integer;

[0065] Set the base samples together with N S The base samples corresponding to the minimum distances are taken as the nearest neighbor subsets. Therefore, the nearest neighbor subsets include the base sample set and N. S N corresponding to the minimum distance S N basic samples, S A smaller positive integer should be chosen, usually N. S The set consists of 10 samples, which means selecting 10 basic samples with the smallest distance from the basic sample set to construct a subset of the nearest neighbor samples of the candidate samples.

[0066] S5. Calculate the nearest neighbor distance between the candidate sample and the nearest neighbor subset;

[0067] Nearest neighbor distance can include central nearest neighbor distance and / or average nearest neighbor distance.

[0068] S5-1. Calculate the center-nearest neighbor distance between the candidate sample and the center point of the nearest neighbor sample subset using composite distance;

[0069] The formula for calculating the centroid of the nearest neighbor subset is as follows:

[0070]

[0071] Where x o N is the center point of the nearest neighbor subset of samples. S x is the number of samples in the nearest neighbor subset. s,i It is the i-th sample in the nearest neighbor sample subset;

[0072] The formula for calculating the center-nearest neighbor distance between the candidate sample and the center point of the nearest neighbor subset is as follows:

[0073] R=D C (x d ,xo )

[0074] Where R is the center-nearest neighbor distance between the candidate sample and the center point of the nearest neighbor subset, and D is the center-nearest neighbor distance between the candidate sample and the center point of the nearest neighbor subset. C It is a composite distance metric.

[0075] S5-2, Calculate N S The average of the minimum distances is the average nearest neighbor distance.

[0076] Calculate the average value μ of the composite distance between the candidate sample and the samples in the nearest neighbor subset. The average value μ is the average nearest neighbor distance, and the formula is as follows:

[0077]

[0078] Where N S D represents the number of samples in the nearest neighbor subset. C For composite distance metric, x d x is the candidate sample. s,i It is the i-th sample in the nearest neighbor sample subset;

[0079] To simplify the distance calculation process, one of the following can be used to calculate the center nearest neighbor distance and the average nearest neighbor distance: Euclidean distance, cosine similarity, or Manhattan distance.

[0080] S6. Determine whether a candidate sample is an incremental sample based on the nearest neighbor distance;

[0081] In step S4, a subset of nearest neighbors is obtained by determining several samples that are closest to the candidate sample in the basic sample set. If the candidate sample and its nearest neighbor subset are far enough apart, the candidate sample is a key incremental sample. The candidate sample is retained as an incremental sample for training the power grid transient stability discrimination model.

[0082] Determining whether a candidate sample is an incremental sample based on nearest neighbor distance includes:

[0083] Samples whose center-nearest neighbor distance R is greater than the center-nearest neighbor threshold are considered incremental samples; or

[0084] Candidate samples whose average nearest neighbor distance μ is greater than the average nearest neighbor threshold are incremental samples; or

[0085] Candidate samples whose center nearest neighbor distance R is greater than the center nearest neighbor threshold and whose average nearest neighbor distance μ is greater than the average nearest neighbor threshold are incremental samples.

[0086] In this invention, to screen for more effective incremental samples, a composite boundary index calculation method is designed to determine whether a candidate sample is a key incremental sample. The composite boundary index λ of the candidate sample... C The calculation formula is as follows:

[0087]

[0088] Where exp{·} is the exponential function, r is the integral variable, μ is the average nearest neighbor distance, R is the central nearest neighbor distance, and σ is the standard deviation of the composite distance between the candidate sample and the basic samples in the nearest neighbor subset. The composite boundary index is defined using the Gaussian distribution law in the probability density model: since the distance from the sample to the center of the space formed by the nearest neighbor subsets follows a Gaussian distribution, this invention uses the average nearest neighbor distance μ as the mean, and the standard deviation σ of the composite distance between the candidate sample and the basic samples in the nearest neighbor subsets as the variance. Correspondingly, the defined composite boundary index characterizes the cumulative probability that the candidate sample belongs to the nearest neighbor subset. The smaller the composite boundary index, the smaller the probability that the candidate sample belongs to the nearest neighbor subset space, and the higher the uniqueness of the candidate sample compared to the basic sample set; therefore, it should be regarded as an incremental sample. The essence of using the composite boundary index to determine whether a candidate sample is a key incremental sample lies in the fact that the composite boundary index simultaneously measures the overall similarity between the candidate sample and its nearest neighbor subsets, as well as the probability that the candidate sample belongs to the sample space of the nearest neighbor subset. The smaller this probability, the smaller the overall similarity. Such candidate samples, as key incremental samples, can provide additional effective information for the training and updating of the model.

[0089] After calculating the composite boundary index, the composite boundary index is compared with a critical threshold. Candidate samples with a composite boundary index less than the critical threshold are designated as incremental samples. Preferably, the critical threshold is set to 0.1737. If the composite boundary index of a candidate sample is less than the critical threshold of 0.1737, the candidate sample is determined to be an incremental sample.

[0090] S7. Incremental sample training is used to train the power grid transient stability discrimination model.

[0091] Different operating modes and preset faults are randomly generated and iterated to generate multiple combinations of operating modes and preset faults. The previous steps are repeated to obtain different candidate samples. After using the preferred composite boundary index to determine whether the candidate sample is an incremental sample, all incremental samples are combined into an incremental sample set for training the power grid transient stability discrimination model.

[0092] The following comparative examples illustrate the technical effect of the present invention in training a power grid transient stability discrimination model using incremental samples. The present invention uses an incremental sample screening method to generate a total of 2,567 incremental samples, which are used to train the power grid transient stability discrimination model. Comparative Example 1 uses all samples, totaling 10,000 samples, to train the power grid transient stability discrimination model. Comparative Example 2 uses 2,567 randomly generated samples, the same number as those selected in the present invention, to train the power grid transient stability discrimination model.

[0093] Taking the 2567 incremental samples generated above as an example, and using this incremental sample set as the training set, the transient response trajectory of electrical quantities within the first 150ms of the time-domain simulation was used as the input feature, and the 0,1 labels of the corresponding power grid transient stability discrimination model were used as the output. An artificial intelligence model suitable for power grid transient stability discrimination was constructed and trained based on the gated recurrent unit (GRU) algorithm. Furthermore, 2000 combinations of operating modes and faults were randomly generated and iterated through different operating modes and faults. The corresponding transient response trajectories of electrical quantities were obtained through time-domain simulation as the test set to evaluate the accuracy of the artificial intelligence model. The results are shown in Table 1. The test results recorded in Table 1 show that the incremental sample screening method proposed in this invention can effectively reduce the number of samples required to train the transient stability discrimination artificial intelligence model. It only requires 2567 incremental samples for training to achieve an overall discrimination accuracy of 96.48%, which is higher than the method in Comparative Example 1 that uses all samples to directly train the model, and even higher than the method in Comparative Example 2 that uses randomly generated samples to directly train the model. The incremental sample screening method for power grid transient stability discrimination proposed in this invention can effectively improve the accuracy of transient stability discrimination and reduce the probability of misjudgment and missed judgment.

[0094] Table 1

[0095]

[0096] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for selecting incremental samples for training a power grid transient stability discrimination model, characterized in that, include: A basic sample set is obtained based on multiple initial operating modes of the power grid. Each initial operating mode corresponds to multiple preset faults. The basic sample set includes multiple basic samples, and the basic samples are the electrical transient response trajectories of a single preset fault under the initial operating mode. A feasible power grid operation mode is generated, and a candidate sample is obtained based on the generated power grid operation mode. The candidate sample includes the electrical transient response trajectory of a single preset fault under the generated power grid operation mode. Calculate the distance between the candidate sample and each base sample in the base sample set; Construct a subset of nearest neighbor samples for the candidate sample based on the distance; Calculate the nearest neighbor distance between the candidate sample and the nearest neighbor subset; Based on the nearest neighbor distance, determine whether the candidate sample is an incremental sample; The incremental samples are used to train the power grid transient stability discrimination model.

2. The method according to claim 1, characterized in that, The preset faults include anticipated faults and unanticipated faults. The anticipated faults include faults that are identified in advance and included in the safety verification scope during the planning stage of the power system, while the unanticipated faults include faults for which effective countermeasures have not been formulated during the planning stage of the power system.

3. The method according to claim 2, characterized in that, The anticipated faults include at least one of the following: grid N-1, N-2, and N-3. The unintended faults include at least one of the following: multiple DC commutation failures, DC blocking, failure of safety control devices to operate, and sudden changes in the output of new energy sources.

4. The method according to claim 1, characterized in that, The electrical quantities of the preset fault include at least one of the following: generator active power, generator reactive power, generator speed, bus voltage amplitude, bus voltage phase angle, branch active power, and branch reactive power.

5. The method according to claim 1, characterized in that, Calculating the distance between the candidate sample and each base sample includes: Calculate the Euclidean distance and cosine similarity distance between the candidate sample and each of the base samples respectively; Calculate the composite distance between the candidate sample and each base sample, wherein the composite distance includes the product of the Euclidean distance and the cosine similarity distance; The distance is the composite distance.

6. The method according to claim 5, characterized in that, Constructing a subset of nearest neighbor samples for the candidate sample based on the distance includes: Sort the distances by size to get N S There are N minimum distances, where N is the minimum distance. S It is a positive integer; The base samples are grouped with N. S The base sample corresponding to the minimum distance is taken as the nearest neighbor sample subset.

7. The method according to claim 6, characterized in that, Calculating the nearest neighbor distance between the candidate sample and the nearest neighbor subset includes: The center-nearest neighbor distance between the candidate sample and the center point of the nearest neighbor subset is calculated using a composite distance method; or / and Calculate the N S The average of the minimum distances, where the average is the average nearest neighbor distance.

8. The method according to claim 7, characterized in that, Determining whether a candidate sample is an incremental sample based on the nearest neighbor distance includes: The candidate samples whose center nearest neighbor distance is greater than the center nearest neighbor threshold are incremental samples; or The candidate samples whose average nearest neighbor distance is greater than the average nearest neighbor threshold are incremental samples; or The candidate samples whose center nearest neighbor distance is greater than the center nearest neighbor threshold and whose average nearest neighbor distance is greater than the average nearest neighbor threshold are incremental samples.

9. The method according to claim 7, characterized in that, Determining whether a candidate sample is an incremental sample based on the nearest neighbor distance includes: Calculate the composite boundary index λ of the candidate samples. C The formula is as follows: Where exp{·} is the exponential function, r is the integral variable, μ is the average nearest neighbor distance, R is the central nearest neighbor distance, and σ is the standard deviation of the composite distance between the candidate sample and the basic sample in the nearest neighbor sample subset; The composite boundary index is compared with the critical threshold, and the candidate samples that are less than the critical threshold are taken as incremental samples.

10. The method according to claim 1, characterized in that, The power grid transient stability discrimination model is an artificial intelligence model.

Images