Method and device for evaluating transient stability of power system

Abstract

The invention relates to the technical field of power system transient stability evaluation, and discloses a power system transient stability evaluation method and device, and the method comprises the steps: obtaining initial samples of a power system in different operation scenes, and the initial samples comprise steady-state features, fault positions and transient stability discrimination results; extracting a training set and a verification set from the initial sample; screening important features from the steady-state features to obtain a candidate feature set; training based on the training set and the verification set to obtain an initial transient stability evaluation neural network; calculating uncertainty indexes of all samples in the training set based on multiple forward propagation results of a Dropout layer of the initial transient stability evaluation neural network, and binning the samples according to the uncertainty indexes; constructing a conditional generative adversarial network model, and training the conditional generative adversarial network model based on the samples after binning to obtain a transient stability evaluation model; and performing transient stability evaluation on the power grid according to the transient stability evaluation model.

Description

Technical Field

[0001] This application relates to the field of power system transient stability assessment technology, specifically to a power system transient stability assessment method and apparatus. Background Technology

[0002] Transient stability of a power system refers to the ability of synchronous generators to maintain synchronous operation and transition to a new or restored stable operating mode after a large disturbance. Transient instability can lead to widespread power outages, causing enormous economic and social losses. Therefore, rapid and accurate assessment of transient stability is a core aspect of power system security and defense. With the increase in new energy sources and the continuous expansion of the power grid, the difficulty of power system transient stability analysis is constantly increasing, making the use of artificial intelligence methods for transient stability assessment a key focus.

[0003] In related technologies, a transient stability assessment method for power systems based on Conditional Generative Adversarial Networks (CGANs) and Graph Convolutional Networks (GCNs) has been proposed. This method uses a CGAN generator to generate unstable samples in a targeted manner, which, together with the original training set, constitute a new augmented dataset. A GCN model is then trained based on this augmented dataset. However, this method does not pay attention to critical samples near the transient stability boundary, resulting in low model reliability. Another approach proposes a transient stability assessment method for power systems based on Bayesian convolutional neural networks. Although this method can obtain the transient stability prediction results of the power system and estimate the uncertainty of the prediction results, it does not consider improving the model's performance based on the uncertainty estimation results, which also affects the model's reliability. Summary of the Invention

[0004] This application aims to at least address the technical problem in the related technologies where the reliability of power system transient stability assessment models is low, affecting the accuracy of the assessment.

[0005] To address the aforementioned technical problems, embodiments of this application provide a method for assessing the transient stability of a power system, comprising:

[0006] Initial samples are obtained under different operating scenarios of the power system, wherein the initial samples include steady-state characteristics, fault locations and transient stability discrimination results;

[0007] A training set and a validation set are extracted from the initial samples;

[0008] Important features are selected from the steady-state features to obtain a candidate feature set;

[0009] An initial transient stability evaluation neural network is trained based on the training set and validation set, wherein the initial transient stability evaluation neural network is a deep neural network model containing a Dropout layer, and the input of the initial transient stability evaluation neural network includes the candidate feature set;

[0010] Based on the results of multiple forward propagations of the Dropout layer of the initial transient stability evaluation neural network, the uncertainty index of all samples in the training set is calculated, and the samples are binned according to the uncertainty index.

[0011] A conditional generative adversarial network (GAN) model is constructed, and the GAN model is trained based on binned samples to obtain a transient stability evaluation model.

[0012] The power grid is subjected to transient stability assessment based on the transient stability assessment model.

[0013] In some embodiments, obtaining initial samples under different operating scenarios of the power system includes:

[0014] Based on the historical operation of the power system and load forecasting, obtain the first number of operation scenarios;

[0015] Based on the historical operating conditions of the power system, load forecasts, and possible topology changes, a second number of new operating scenarios are obtained;

[0016] An initial feature vector is obtained from the first number of operating scenarios and the second number of new operating scenarios, wherein the initial feature vector includes a steady-state feature vector and a preset fault vector;

[0017] Based on power flow calculation and transient stability simulation calculation, the generator power angle curves under the first number of operating scenarios and the second number of new operating scenarios are obtained;

[0018] Based on the generator power angle curve, transient stability quantitative indicators under different operating scenarios are calculated.

[0019] The transient stability determination result is obtained based on the aforementioned transient stability quantification index.

[0020] In some embodiments, important features are screened from the steady-state features to obtain a candidate feature set, including:

[0021] The steady-state features in the initial features are preprocessed to obtain normalized steady-state features;

[0022] Based on the Pearson correlation coefficients between each steady-state feature and the transient stability quantification index in the training set, the importance index of the steady-state feature is determined.

[0023] The importance threshold is determined based on the descending order of the importance index values ​​of each steady-state feature.

[0024] The steady-state features whose importance index values ​​are greater than or equal to the importance threshold are determined as the candidate feature set.

[0025] In some embodiments, an initial transient stability evaluation neural network is trained based on the training set and the validation set, including:

[0026] The model structure of the neural network is determined. The input of the neural network is all steady-state features of the candidate feature set and one-hot encoding of the preset fault vector. The output of the neural network is the transient stability discrimination result. The neural network contains at least one Dropout layer.

[0027] Based on the samples in the training set and the gradient descent algorithm based on adaptive moment estimation, the parameters to be determined for the neural network are iteratively calculated.

[0028] Based on the validation set and different neural network model parameters, the validation set accuracy of models with different parameters is determined;

[0029] The initial transient stability evaluation neural network is determined based on the parameters of the neural network model with the highest accuracy on the validation set.

[0030] In some embodiments, based on the results of multiple forward propagations of the Dropout layer of the initial transient stability evaluation neural network, an uncertainty index for all samples in the training set is calculated, and the samples are binned according to the uncertainty index, including:

[0031] Using the trained initial transient stability evaluation neural network, perform T forward propagations on all samples in the training set;

[0032] Calculate the uncertainty index for all samples;

[0033] Based on the uncertainty index results of all samples, calculate the maximum and minimum values ​​of the uncertainty index;

[0034] Set the binning threshold based on the maximum and minimum values;

[0035] All samples in the training set are binned according to the binning threshold.

[0036] In some embodiments, a conditional generative adversarial network (GAN) model is constructed, and the GAN model is trained based on binned samples to obtain a transient stability evaluation model, including:

[0037] The model structure of the generator of the conditional generative adversarial network is determined, wherein the input of the generator is a noise vector and a one-hot encoding of the fault location, the intermediate layer of the generator includes multiple stacked neural network layers, and the output of the generator is all features of the candidate feature set;

[0038] The model structure of the discriminator of the conditional generative adversarial network is determined, wherein the input of the discriminator is a real sample or a synthetic sample, and the output is the probability distribution of whether the sample is real or fake;

[0039] The conditional generative adversarial network is trained based on the binned high-uncertainty samples to obtain a generator model for generating high-uncertainty samples.

[0040] In some embodiments, the method further includes:

[0041] Based on the generator model, samples are continuously generated, and the transient stability evaluation model is fine-tuned to obtain the final transient stability evaluation model.

[0042] In some embodiments, performing a transient stability assessment of the power grid based on the transient stability assessment model includes:

[0043] Acquire actual power grid operation data and preset faults;

[0044] The transient stability assessment model is used to determine the transient stability of the power grid after a preset fault occurs, and the transient stability assessment result is obtained.

[0045] In some embodiments, the actual operating data of the power grid includes at least one of the active power of the generator, the generator terminal voltage, the active power of the load, and the reactive power of the load.

[0046] This application also provides a power system transient stability assessment device, including:

[0047] The acquisition module is configured to acquire initial samples under different operating scenarios of the power system, wherein the initial samples include steady-state characteristics, fault locations, and transient stability discrimination results.

[0048] The extraction module is configured to extract a training set and a validation set from the initial samples;

[0049] The feature filtering module is configured to filter important features from the steady-state features to obtain a candidate feature set;

[0050] The first model construction module is configured to train an initial transient stability evaluation neural network based on the training set and the validation set, wherein the initial transient stability evaluation neural network is a deep neural network model containing a Dropout layer, and the input of the initial transient stability evaluation neural network includes the candidate feature set;

[0051] The sample binning module is configured to calculate the uncertainty index of all samples in the training set based on the results of multiple forward propagations of the Dropout layer of the initial transient stability evaluation neural network, and to bin the samples according to the uncertainty index.

[0052] The second model building module is configured to build a conditional generative adversarial network model and train the conditional generative adversarial network model based on the binned samples to obtain a transient stability evaluation model.

[0053] The transient stability assessment module is configured to perform transient stability assessment on the power grid based on the transient stability assessment model.

[0054] This application also provides an electronic device, which includes at least a processor and a memory. The memory stores a computer program, and the processor implements the above-described power system transient stability assessment method when executing the computer program in the memory.

[0055] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described power system transient stability assessment method.

[0056] The power system transient stability assessment method and apparatus provided in this application acquire initial samples under different operating scenarios of the power system, wherein the initial samples include steady-state characteristics, fault locations, and transient stability discrimination results; extract training and validation sets from the initial samples; screen important features from the steady-state characteristics to obtain a candidate feature set; train an initial transient stability assessment neural network based on the training and validation sets, wherein the initial transient stability assessment neural network is a deep neural network model containing Dropout layers, and the input of the initial transient stability assessment neural network includes the candidate feature set; and train the Dropout layers of the initial transient stability assessment neural network multiple times. The forward propagation results are used to calculate the uncertainty index of all samples in the training set, and the samples are binned according to the uncertainty index. A conditional generative adversarial network (GAN) model is constructed, and the GAN model is trained based on the binned samples to obtain a transient stability assessment model. The transient stability assessment model is used to assess the power grid transient stability. Dropout can be used to calculate the sample uncertainty index through multiple forward propagations, identify sample regions with low confidence in the model judgment, and bin the samples according to uncertainty. This provides target guidance for the subsequent generation of high uncertainty samples, enhances the model's assessment performance in sample uncertainty regions, and obtains more accurate power system transient stability assessment results. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart of the power system transient stability assessment method according to an embodiment of this application;

[0059] Figure 2 This is another flowchart of the power system transient stability assessment method according to an embodiment of this application;

[0060] Figure 3 This is a schematic diagram of the power system transient stability assessment device according to an embodiment of this application. Detailed Implementation

[0061] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and 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.

[0062] It should be understood that various modifications can be made to the embodiments described herein. Therefore, the specification should not be considered limiting, but merely as an example of embodiments. Other modifications within the scope and spirit of this application will be apparent to those skilled in the art.

[0063] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0064] These and other features of this application will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0065] It should also be understood that although this application has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of this application, which have the features described in the claims and are therefore all within the scope of protection defined herein.

[0066] The above and other aspects, features and advantages of this application will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0067] Specific embodiments of this application are described thereafter with reference to the accompanying drawings; however, it should be understood that the claimed embodiments are merely examples of this application, which can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the application. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely serve as the basis and representative basis for the claims to teach those skilled in the art to use this application in a variety of substantially any suitable detailed structures.

[0068] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in other embodiments,” all of which may refer to one or more of the same or different embodiments according to this application.

[0069] Example 1

[0070] Figure 1 A flowchart illustrating a power system transient stability assessment method according to an embodiment of this application is shown. Figure 1 As shown in the figure, this application provides a method for assessing the transient stability of a power system, including:

[0071] S101: Obtain initial samples under different operating scenarios of the power system, wherein the initial samples include steady-state characteristics, fault locations and transient stability discrimination results.

[0072] Step S101 specifically includes the following steps:

[0073] S1011: Based on the historical operation of the power system and load forecasting, obtain the first number of operation scenarios;

[0074] S1012: Based on the historical operation of the power system, load forecasting, and possible topology changes, obtain a second number of new operation scenarios;

[0075] S1013: Obtain an initial feature vector from the first number of operating scenarios and the second number of new operating scenarios, wherein the initial feature vector includes a steady-state feature vector and a preset fault vector;

[0076] S1014: Based on power flow calculation and transient stability simulation calculation, the generator power angle curves under the first number of operating scenarios and the second number of new operating scenarios are obtained;

[0077] S1015: Based on the generator power angle curve, calculate the transient stability quantitative index under different operating scenarios;

[0078] S1016: Obtain the transient stability discrimination result based on the transient stability quantification index.

[0079] In step S1011, for a power system with N generators and M loads, K operating scenarios are randomly set based on the historical operating conditions and load forecasts of the power system, and the K operating scenarios are randomly ordered. For example, in this embodiment, the power system adopts a 10-generator 39-node system, with N=10 generators, M=30 loads, and K set to 20000.

[0080] In step S1011, based on the historical operation of the power system, load forecasting, and possible topology changes, e new operation scenarios are set up and merged with the K operation scenarios to obtain K+e operation scenarios.

[0081] In step S1013, initial feature vectors for K+e operating scenarios are obtained. These initial feature vectors include steady-state feature vectors and preset fault vectors. The steady-state feature vectors contain 2(N+M) features, including the active power and terminal voltage of N generators, and the active and reactive power of M loads. The preset fault vectors employ one-hot encoding. For example, in one embodiment of this application, there are 80 initial features and 34 preset faults. One-hot encoding is a technique for converting categorical variables into binary vectors.

[0082] In step S1014, the generator power angle curves under K+e operating scenarios are obtained based on power flow calculation and transient stability simulation calculation.

[0083] In step S1015, the maximum relative power angle difference between generators during the transient stability simulation is determined based on the generator power angle curve. The transient stability quantification index is then calculated based on this maximum relative power angle difference during the simulation. Specifically, the transient stability quantification index η for the k-th operating scenario is... k Represented as:

[0084] ;

[0085] Where, |Δδ| max_k Let $e$ represent the maximum absolute value of the power angle difference between any two generators in the $k$-th operating scenario, where $k = 1, ..., $K+e$.

[0086] In step S1016, based on the transient stability quantification index η k The transient stability determination result y is obtained. k If η k If the value is greater than or equal to 0, then the power system can maintain transient stability in the k-th operating scenario. Let the transient stability determination result be y. k =0, if η k If the value is less than 0, then the power system is transiently unstable in the k-th operating scenario. Let the transient stability determination result be y. k =1.

[0087] S102: Extract training and validation sets from the initial samples.

[0088] After obtaining the initial samples, the steady-state feature vectors in the initial samples are sequentially subjected to min-max normalization to obtain the normalized steady-state feature vectors. Then, the normalized steady-state features, preset fault vectors, and transient stability discrimination results of the first 3×e operating scenarios are extracted from K operating scenarios to form the training set D. train Wherein, the i-th normalized steady-state feature in the k-th running scenario in the training set is denoted as x i k The transient stability determination result is y k The normalized steady-state features, preset fault vectors, and transient stability discrimination results for the remaining K-2e operating scenarios constitute the validation set D. verify .

[0089] S103: Select important features from the steady-state features to obtain a candidate feature set.

[0090] Step S103 specifically includes the following steps:

[0091] S1031: Preprocess the steady-state features in the initial features to obtain normalized steady-state features;

[0092] S1032: Based on the Pearson correlation coefficient between each steady-state feature and the transient stability quantification index in the training set, determine the importance index of the steady-state feature;

[0093] S1033: Determine the importance threshold based on the descending order of the importance index values ​​of each steady-state feature;

[0094] S1034: The steady-state features whose importance index values ​​are greater than or equal to the importance threshold are determined as the candidate feature set.

[0095] In step S1031, the normalized steady-state feature can be the normalized steady-state feature vector in step S102.

[0096] In step S1032, the training set is calculated sequentially. i Normalized steady-state characteristics x i =[ x i 1 ,…, x i k ,…, xi e and transient stability quantitative indicators or =[ or 1,…, or k ,…, or e The absolute value of the Pearson correlation coefficient between [ ] is used as an indicator of importance. :

[0097] ;

[0098] in, It is the first i Normalized steady-state characteristics x i =[ x i 1 ,…, x i k ,…, x i e and transient stability quantitative indicators or =[ or 1,…, or k ,…, or e The covariance of ] yes x i variance yes or The variance.

[0099] Then, in step S1033, the importance index values ​​of all normalized steady-state features are sorted in descending order, and the importance threshold is determined according to the following formula. i :

[0100] ;

[0101] in, After sorting the importance index values ​​of the normalized steady-state characteristics in descending order, the value is ranked as follows: t The importance of the normalized steady-state characteristics of the position, and t The value satisfies t ≥ 0.2×(2× N +2× M ) .

[0102] Then, the criteria are selected through step S1034. ≥ iThe steady-state features are used to obtain the candidate feature set C. The steady-state features that are not selected into the candidate feature set are irrelevant features.

[0103] S104: An initial transient stability evaluation neural network is trained based on the training set and validation set, wherein the initial transient stability evaluation neural network is a deep neural network model containing a Dropout layer, and the input of the initial transient stability evaluation neural network includes the candidate feature set.

[0104] In this step, a deep neural network model with Dropout layers is trained based on the training and validation sets to serve as the initial transiently stable evaluation neural network. The Dropout layer can suppress overfitting of the neural network and maintain the stability of the overall expected output values.

[0105] like Figure 2 As shown, step S104 specifically includes the following steps:

[0106] S1041: Determine the model structure of the neural network, wherein the input of the neural network is all steady-state features of the candidate feature set and one-hot encoding of the preset fault vector, the output of the neural network is the transient stability discrimination result, and the neural network contains at least one Dropout layer;

[0107] S1042: Based on the samples in the training set and the gradient descent algorithm based on adaptive moment estimation, iteratively calculate the parameters to be determined for the neural network;

[0108] S1043: Determine the validation set accuracy of different parameter models based on the validation set and different neural network model parameters;

[0109] S1044: Determine the initial transient stability evaluation neural network based on the parameters of the neural network model with the highest accuracy on the validation set.

[0110] First, determine the model structure of the neural network, with the input of the model being the candidate feature set. C All steady-state features and preset fault vectors are one-hot encoded, and the output is the transient stability discrimination result. The model contains at least one Dropout layer. Then, based on the training set samples and the gradient descent algorithm based on adaptive moment estimation, the parameters to be determined in the neural network model structure are iteratively calculated. Based on the validation set and different neural network model parameters, the validation set accuracy corresponding to different parameter models is obtained. The neural network model parameters with the highest validation set accuracy are selected to form the initial neural network model. M 0. The initial neural network model M 0 represents the initial transient stability evaluation neural network.

[0111] S105: Based on the results of multiple forward propagations of the Dropout layer of the initial transient stability evaluation neural network, calculate the uncertainty index of all samples in the training set, and bin the samples according to the uncertainty index.

[0112] Step S105 specifically includes the following steps:

[0113] S1051: Using the trained initial transient stability evaluation neural network, perform T forward propagations on all samples in the training set;

[0114] S1052: Calculate the uncertainty index for all samples;

[0115] S1053: Based on the uncertainty index results of all samples, calculate the maximum and minimum values ​​of the uncertainty index;

[0116] S1054: Set the binning threshold based on the maximum and minimum values;

[0117] S1055: Bin all samples in the training set according to the binning threshold.

[0118] First, using the pre-trained initial neural network model M 0, for all training sets e The sample undergoes T forward propagation iterations, with Dropout enabled during each iteration, resulting in T predictions, where the th... k The first sample t The prediction result is: The preferred value for T is... T ≥30, for example, T is 50.

[0119] Then, calculate all of the training set. e The uncertainty index of the nth sample, where the nth k Uncertainty index of a sample It is the variance of all T predictions. Then, based on all the training set... e Given the uncertainty index results for each sample, calculate the maximum value of the uncertainty index, max{ } and minimum value min{ }. Then, according to max{ } and min{ Set a binning threshold to bin all training sets. e The samples are divided into 3 bins, where the samples in bin 1 satisfy the following: , indicating a low-uncertainty sample; the samples in bin 2 satisfy: , representing intermediate samples; the samples in bin 3 satisfy: , representing samples with high uncertainty, and the binning threshold satisfies the following formula:

[0120] ;

[0121] .

[0122] In this embodiment, the sample is divided into three bins, representing low-uncertainty samples, intermediate samples, and high-uncertainty samples, respectively. In specific implementations, more or fewer sample bins can be divided as needed (e.g., only low-uncertainty samples and high-uncertainty samples).

[0123] S106: Construct a conditional generative adversarial network model, and train the conditional generative adversarial network model based on the binned samples to obtain a transient stability evaluation model.

[0124] In this step, the generator model for generating transiently stable high-uncertainty samples is mainly obtained by training the conditional generative adversarial network model.

[0125] Step S106 specifically includes the following steps:

[0126] S1061: Determine the model structure of the generator of the conditional generative adversarial network, wherein the input of the generator is a noise vector and a one-hot encoding of the fault location, the intermediate layer of the generator includes multiple stacked neural network layers, and the output of the generator is all features of the candidate feature set;

[0127] S1062: Determine the model structure of the discriminator of the conditional generative adversarial network, wherein the input of the discriminator is a real sample or a synthetic sample, and the output is the probability distribution of whether the sample is real or fake;

[0128] S1063: Train the conditional generative adversarial network based on the binned high-uncertainty samples to obtain a generator model for generating high-uncertainty samples. M G .

[0129] First, the generator model structure of the Conditional Generative Adversarial Network (CGAN) is set up. The generator's input consists of a noise vector and a one-hot encoded fault location, containing multiple stacked neural network layers. The generator's output is all features from the candidate feature set C. Next, the discriminator model structure of the CGAN is set up. The discriminator's input consists of either real or synthetic samples, and its output is the probability distribution of whether a sample is real or fake. Then, the CGAN model is trained using all samples from bin 3 (high-uncertainty samples) to obtain the generator and discriminator. The generator model is the one subsequently used to generate high-uncertainty samples. M G The generator model MG This is the transient stability assessment model (also known as the transient stability assessment neural network model).

[0130] S107: Perform transient stability assessment on the power grid according to the transient stability assessment model.

[0131] After constructing the transient stability assessment model, the actual operating data of the power grid can be input into the model to conduct a transient stability assessment of the power grid and obtain the transient stability assessment results.

[0132] The power system transient stability assessment method provided in this application utilizes Dropout to calculate sample uncertainty indices through multiple forward propagations, identifies sample regions with low model confidence, bins samples according to uncertainty, provides target guidance for the subsequent generation of high-uncertainty samples, enhances the model's assessment performance in sample uncertainty regions, and obtains more accurate power system transient stability assessment results.

[0133] This application constructs a transient stability assessment model that includes a Dropout layer, which can quantify uncertainty to identify weak areas in the model's cognition, and actively generate samples in these areas based on a generative adversarial network to strengthen the model, ultimately resulting in a transient stability assessment model with excellent performance and credibility assessment.

[0134] In some embodiments, the method further includes:

[0135] S108: Based on the generator model, samples are continuously generated, and the transient stability evaluation model is fine-tuned to obtain the final transient stability evaluation model.

[0136] The generator model is obtained through training in step S106. M G Subsequently, the transient stability assessment model can be continuously fine-tuned by generating samples through the generator model, gradually optimizing the model's discrimination ability in regions of high uncertainty, and obtaining the final transient stability assessment model. M T The final transient stability assessment model has stronger generalization ability and higher assessment accuracy.

[0137] Step S108 specifically includes the following steps:

[0138] Step S1081: Set the maximum number of iterations L and the preset threshold d for validation set accuracy. set Let the number of iterations l=0, and let the final transient stability evaluation model... M T := M 0 ;

[0139] Step S1082: Let l:=l+1, set the preset fault type and the number of transient stable samples to be generated W0;

[0140] Step S1083: Perform one-hot encoding on the preset fault type to obtain the preset fault vector, and input W0 Gaussian noises into the generator model. Given a preset fault vector, the generator outputs W0 candidate feature vectors under different preset faults;

[0141] Step S1084: Input the W0 candidate feature vectors output by the generator and the preset fault vector as W0 samples into the transient stability evaluation neural network M. T In this process, T forward propagations are performed on all W0 samples (in one embodiment of this patent, T is set to 50, but T ≥ 30 is recommended). Dropout is enabled during each forward propagation, resulting in T predictions. The a-th prediction result for the e-th sample is... ;

[0142] Step S1085: Calculate the uncertainty index for all W0 samples, where the uncertainty index for the e-th sample is... It is the variance of all T prediction results, and the uncertainty index of all W0 samples. Compare with the binning threshold, if If the value is 0, it indicates that the sample is a high-uncertainty sample; otherwise, the sample is not a high-uncertainty sample, and finally W1 samples are obtained.

[0143] Step S1086: Inverse normalize the candidate feature vectors of W1 samples and randomly select irrelevant features from W1 samples to obtain 2(N+M) complete initial steady-state features. Perform power flow calculation based on the initial steady-state features. If the power flow does not converge, the transient simulation will not be performed. If the power flow calculation converges, the transient simulation will continue to obtain the transient stability quantification index, and finally W2 transient stability simulation samples will be obtained.

[0144] Step S1087: Merge the newly generated W2 transient stable simulation samples with the initial training set samples to obtain a new training set;

[0145] Step S1088: Evaluate the transient stability evaluation neural network M based on the new training set. T The parameters are fine-tuned, and the accuracy d of the transient stability evaluation neural network is obtained based on the validation set.

[0146] Step S1089: Compare the accuracy d of the transient stability evaluation neural network on the validation set with a preset threshold d. set Compare, if d≥d set If d <d setThen repeat steps S1082 to S1089 until d ≥ d set .

[0147] In some embodiments, step S107, performing a transient stability assessment of the power grid based on the transient stability assessment model, includes:

[0148] S1071: Obtain actual power grid operation data and preset faults;

[0149] S1072: Based on the transient stability assessment model, the transient stability of the power grid after a preset fault is determined, and the transient stability assessment result is obtained.

[0150] The actual operating data of the power grid includes at least one of the following: generator active power, generator terminal voltage, load active power, and load reactive power.

[0151] When performing transient stability assessment of the power grid, the active power and terminal voltage of all generators in the actual power grid, as well as the active power and reactive power of all loads, are obtained. The active power and terminal voltage of all generators and the active power and reactive power of all loads in the actual power grid are normalized to obtain a steady-state feature vector. Preset faults are then encoded using one-hot encoding to obtain a preset fault vector. Finally, the steady-state feature vector and the preset fault vector are input into the transient stability assessment model. M T In this process, transient stability assessment results were obtained.

[0152] Example 2

[0153] Figure 3 This is a schematic diagram of the power system transient stability assessment device according to an embodiment of this application. Figure 3 As shown in the figure, this application provides a power system transient stability assessment device, including:

[0154] The acquisition module 10 is configured to acquire initial samples under different operating scenarios of the power system, wherein the initial samples include steady-state characteristics, fault locations and transient stability discrimination results;

[0155] Extraction module 20 is configured to extract training set and validation set from the initial sample;

[0156] Feature filtering module 30 is configured to filter important features from the steady-state features to obtain a candidate feature set;

[0157] The first model construction module 40 is configured to train an initial transient stability evaluation neural network based on the training set and the validation set, wherein the initial transient stability evaluation neural network is a deep neural network model containing a Dropout layer, and the input of the initial transient stability evaluation neural network includes the candidate feature set;

[0158] The sample binning module 50 is configured to calculate the uncertainty index of all samples in the training set based on the result of multiple forward propagations of the Dropout layer of the initial transient stability evaluation neural network, and to bin the samples according to the uncertainty index.

[0159] The second model building module 60 is configured to build a conditional generative adversarial network model and train the conditional generative adversarial network model based on the binned samples to obtain a transient stability evaluation model.

[0160] The transient stability assessment module 70 is configured to perform transient stability assessment on the power grid based on the transient stability assessment model.

[0161] In some embodiments, the acquisition module 10 is further configured to:

[0162] Based on the historical operation of the power system and load forecasting, obtain the first number of operation scenarios;

[0163] Based on the historical operating conditions of the power system, load forecasts, and possible topology changes, a second number of new operating scenarios are obtained;

[0164] An initial feature vector is obtained from the first number of operating scenarios and the second number of new operating scenarios, wherein the initial feature vector includes a steady-state feature vector and a preset fault vector;

[0165] Based on power flow calculation and transient stability simulation calculation, the generator power angle curves under the first number of operating scenarios and the second number of new operating scenarios are obtained;

[0166] Based on the generator power angle curve, transient stability quantitative indicators under different operating scenarios are calculated.

[0167] The transient stability determination result is obtained based on the aforementioned transient stability quantification index.

[0168] In some embodiments, the feature filtering module 30 is further configured to:

[0169] The steady-state features in the initial features are preprocessed to obtain normalized steady-state features;

[0170] Based on the Pearson correlation coefficients between each steady-state feature and the transient stability quantification index in the training set, the importance index of the steady-state feature is determined.

[0171] The importance threshold is determined based on the descending order of the importance index values ​​of each steady-state feature.

[0172] The steady-state features whose importance index values ​​are greater than or equal to the importance threshold are determined as the candidate feature set.

[0173] In some embodiments, the first model building module 40 is further configured to:

[0174] The model structure of the neural network is determined. The input of the neural network is all steady-state features of the candidate feature set and one-hot encoding of the preset fault vector. The output of the neural network is the transient stability discrimination result. The neural network contains at least one Dropout layer.

[0175] Based on the samples in the training set and the gradient descent algorithm based on adaptive moment estimation, the parameters to be determined for the neural network are iteratively calculated.

[0176] Based on the validation set and different neural network model parameters, the validation set accuracy of models with different parameters is determined;

[0177] The initial transient stability evaluation neural network is determined based on the parameters of the neural network model with the highest accuracy on the validation set.

[0178] In some embodiments, the sample binning module 50 is further configured to:

[0179] Using the trained initial transient stability evaluation neural network, perform T forward propagations on all samples in the training set;

[0180] Calculate the uncertainty index for all samples;

[0181] Based on the uncertainty index results of all samples, calculate the maximum and minimum values ​​of the uncertainty index;

[0182] Set the binning threshold based on the maximum and minimum values;

[0183] All samples in the training set are binned according to the binning threshold.

[0184] In some embodiments, the second model building module 60 is further configured to:

[0185] The model structure of the generator of the conditional generative adversarial network is determined, wherein the input of the generator is a noise vector and a one-hot encoding of the fault location, the intermediate layer of the generator includes multiple stacked neural network layers, and the output of the generator is all features of the candidate feature set;

[0186] The model structure of the discriminator of the conditional generative adversarial network is determined, wherein the input of the discriminator is a real sample or a synthetic sample, and the output is the probability distribution of whether the sample is real or fake;

[0187] The conditional generative adversarial network is trained based on the binned high-uncertainty samples to obtain a generator model for generating high-uncertainty samples.

[0188] In some embodiments, the power system transient stability assessment device further includes a correction module configured to:

[0189] Based on the generator model, samples are continuously generated, and the transient stability evaluation model is fine-tuned to obtain the final transient stability evaluation model.

[0190] In some embodiments, the transient stability assessment module 70 is further configured to:

[0191] Acquire actual power grid operation data and preset faults;

[0192] The transient stability assessment model is used to determine the transient stability of the power grid after a preset fault occurs, and the transient stability assessment result is obtained.

[0193] In some embodiments, the actual operating data of the power grid includes at least one of the active power of the generator, the generator terminal voltage, the active power of the load, and the reactive power of the load.

[0194] The power system transient stability assessment device provided in this application corresponds to the power system transient stability assessment method in the above embodiments. Any optional options in the embodiments of the power system transient stability assessment method are also applicable to the embodiments of the power system transient stability assessment device, and will not be repeated here.

[0195] Example 3

[0196] This application also provides an electronic device, including at least a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-described power system transient stability assessment method when executing the computer program in the memory.

[0197] In some embodiments, the processor executing a computer program may be a processing device that includes one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), etc. More specifically, the processor may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor that runs other instruction sets, or a processor that runs a combination of instruction sets. The processor may also be one or more special-purpose processing devices, such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), system-on-a-chip (SoCs), etc.

[0198] The memory may be a read-only memory (ROM), random access memory (RAM), phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), electrically erasable programmable read-only memory (EEPROM), other types of random access memory (RAM), flash drives or other forms of flash memory, cache, registers, static memory, optical disc read-only memory (CD-ROM), digital versatile optical disc (DVD) or other optical storage, magnetic tape cassette or other magnetic storage devices, or any other possible non-transitory medium used to store information or instructions that can be accessed by computer equipment.

[0199] The electronic devices in this application may include, but are not limited to, fixed terminal devices such as servers, desktop computers, and digital TVs, as well as mobile terminal devices such as in-vehicle devices (e.g., head-up displays), handheld devices (e.g., mobile phones, tablets), and wearable devices (e.g., smartwatches, smart bracelets).

[0200] Example 4

[0201] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described power system transient stability assessment method.

[0202] The computer-readable storage medium in this application embodiment can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. In this application embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device; for example, it can be the aforementioned memory.

[0203] The computer programs of embodiments of this application can be organized into one or more computer-executable components or modules. Various aspects of this application can be implemented with any number and combination of such components or modules. For example, aspects of this application are not limited to the specific computer-executable instructions or specific components or modules shown in the drawings and described herein. Other embodiments may include different computer-executable instructions or components having more or fewer functions than those shown and described herein.

[0204] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for assessing the transient stability of a power system, characterized in that, include: Initial samples are obtained under different operating scenarios of the power system, wherein the initial samples include steady-state characteristics, fault locations and transient stability discrimination results; A training set and a validation set are extracted from the initial samples; Important features are selected from the steady-state features to obtain a candidate feature set; An initial transient stability evaluation neural network is trained based on the training set and validation set, wherein the initial transient stability evaluation neural network is a deep neural network model containing a Dropout layer, and the input of the initial transient stability evaluation neural network includes the candidate feature set; Based on the results of multiple forward propagations of the Dropout layer of the initial transient stability evaluation neural network, the uncertainty index of all samples in the training set is calculated, and the samples are binned according to the uncertainty index. A conditional generative adversarial network (GAN) model is constructed, and the GAN model is trained based on binned samples to obtain a transient stability evaluation model. The power grid is subjected to transient stability assessment based on the transient stability assessment model.

2. The power system transient stability assessment method according to claim 1, characterized in that, Obtain initial samples under different operating scenarios of the power system, including: Based on the historical operation of the power system and load forecasting, obtain the first number of operation scenarios; Based on the historical operating conditions of the power system, load forecasts, and possible topology changes, a second number of new operating scenarios are obtained; An initial feature vector is obtained from the first number of operating scenarios and the second number of new operating scenarios, wherein the initial feature vector includes a steady-state feature vector and a preset fault vector; Based on power flow calculation and transient stability simulation calculation, the generator power angle curves under the first number of operating scenarios and the second number of new operating scenarios are obtained; Based on the generator power angle curve, transient stability quantitative indicators under different operating scenarios are calculated. The transient stability determination result is obtained based on the aforementioned transient stability quantification index.

3. The power system transient stability assessment method according to claim 2, characterized in that, Important features are selected from the steady-state features to obtain a candidate feature set, including: The steady-state features in the initial features are preprocessed to obtain normalized steady-state features; Based on the Pearson correlation coefficients between each steady-state feature and the transient stability quantification index in the training set, the importance index of the steady-state feature is determined. The importance threshold is determined based on the descending order of the importance index values ​​of each steady-state feature. The steady-state features whose importance index values ​​are greater than or equal to the importance threshold are determined as the candidate feature set.

4. The power system transient stability assessment method according to claim 1, characterized in that, An initial transient stability evaluation neural network is trained based on the training set and validation set, including: The model structure of the neural network is determined. The input of the neural network is all steady-state features of the candidate feature set and one-hot encoding of the preset fault vector. The output of the neural network is the transient stability discrimination result. The neural network contains at least one Dropout layer. Based on the samples in the training set and the gradient descent algorithm based on adaptive moment estimation, the parameters to be determined for the neural network are iteratively calculated. Based on the validation set and different neural network model parameters, the validation set accuracy of models with different parameters is determined; The initial transient stability evaluation neural network is determined based on the parameters of the neural network model with the highest accuracy on the validation set.

5. The power system transient stability assessment method according to claim 1, characterized in that, Based on the results of multiple forward propagations of the Dropout layer of the initial transient stability evaluation neural network, the uncertainty index of all samples in the training set is calculated, and the samples are binned according to the uncertainty index, including: Using the trained initial transient stability evaluation neural network, perform T forward propagations on all samples in the training set; Calculate the uncertainty index for all samples; Based on the uncertainty index results of all samples, calculate the maximum and minimum values ​​of the uncertainty index; Set the binning threshold based on the maximum and minimum values; All samples in the training set are binned according to the binning threshold.

6. The power system transient stability assessment method according to claim 1, characterized in that, A conditional generative adversarial network (GAN) model is constructed, and the GAN model is trained based on binned samples to obtain a transient stability evaluation model, including: The model structure of the generator of the conditional generative adversarial network is determined, wherein the input of the generator is a noise vector and a one-hot encoding of the fault location, the intermediate layer of the generator includes multiple stacked neural network layers, and the output of the generator is all features of the candidate feature set; The model structure of the discriminator of the conditional generative adversarial network is determined, wherein the input of the discriminator is a real sample or a synthetic sample, and the output is the probability distribution of whether the sample is real or fake; The conditional generative adversarial network is trained based on the binned high-uncertainty samples to obtain a generator model for generating high-uncertainty samples.

7. The power system transient stability assessment method according to claim 6, characterized in that, The method further includes: Based on the generator model, samples are continuously generated, and the transient stability evaluation model is fine-tuned to obtain the final transient stability evaluation model.

8. The power system transient stability assessment method according to claim 1, characterized in that, The transient stability assessment of the power grid is performed based on the aforementioned transient stability assessment model, including: Acquire actual power grid operation data and preset faults; The transient stability assessment model is used to determine the transient stability of the power grid after a preset fault occurs, and the transient stability assessment result is obtained.

9. The power system transient stability assessment method according to claim 8, characterized in that, The actual operating data of the power grid includes at least one of the following: generator active power, generator terminal voltage, load active power, and load reactive power.

10. A power system transient stability assessment device, characterized in that, include: The acquisition module is configured to acquire initial samples under different operating scenarios of the power system, wherein the initial samples include steady-state characteristics, fault locations and transient stability discrimination results; The extraction module is configured to extract a training set and a validation set from the initial samples; The feature filtering module is configured to filter important features from the steady-state features to obtain a candidate feature set; The first model construction module is configured to train an initial transient stability evaluation neural network based on the training set and the validation set, wherein the initial transient stability evaluation neural network is a deep neural network model containing a Dropout layer, and the input of the initial transient stability evaluation neural network includes the candidate feature set; The sample binning module is configured to calculate the uncertainty index of all samples in the training set based on the results of multiple forward propagations of the Dropout layer of the initial transient stability evaluation neural network, and to bin the samples according to the uncertainty index. The second model building module is configured to build a conditional generative adversarial network model and train the conditional generative adversarial network model based on the binned samples to obtain a transient stability evaluation model. The transient stability assessment module is configured to perform transient stability assessment on the power grid based on the transient stability assessment model.

Images