Power system transient stability evaluation model construction method based on sample extraction method

Abstract

The invention relates to a power system transient stability evaluation model construction method based on a sample extraction method. And key samples playing an important role in model training are extracted according to the spatial distribution information of the samples and the fault severity of the samples, and redundant samples are eliminated, so that optimization of the sample set is realized, and the performance of the model is improved. The method comprises the following steps: firstly, calculating based on a support vector machine (SVM) to obtain a support vector acting on a decision boundary, and extracting a critical sample in combination with the severity of a sample fault; then, counting position information of a neighbor sample and an approximate tangent plane of the similar curved surface, and extracting an edge sample in combination with the severity of a sample fault; and finally, the extracted samples are used for constructing a new sample set for training the stability evaluation model of the power system, so that the model precision and the training efficiency are effectively improved, and the evaluation tendency of the model is improved at the same time.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and in particular relates to a method for constructing a transient stability assessment model for a power system based on a sample extraction method. Background Art

[0002] As an important field of power system analysis, the analysis of power system stability assessment models is crucial to ensuring the safe and stable operation of power systems. In recent years, with the rapid development of artificial intelligence technology, the rapid and accurate judgment of the stability state of power systems can be achieved through deep mining and analysis of massive data resources, which can effectively avoid the problems of low accuracy and efficiency in traditional power system stability analysis methods. However, these methods are based on data-driven. There are many redundant samples in the power system stability assessment sample set. These samples do not contribute much to the model. Using them to train deep learning models will greatly increase the training cost of the model and even reduce the evaluation accuracy of the model. Therefore, it is necessary to design a method that can extract key samples that can represent the spatial distribution of the original samples and use these key samples for training to improve the training efficiency and evaluation accuracy of the model, and even improve the evaluation tendency of the model. Summary of the invention

[0003] The purpose of the present invention is to provide a method for constructing a transient stability assessment model for a power system based on a sample extraction method, combining the fault severity of the sample and using the support vector of the SVM to extract critical samples, and then extracting edge samples through the position information of the surface-like neighboring samples, thereby optimizing the training samples. Finally, the optimized sample set is used to train the final power system stability assessment model, which improves the training efficiency and assessment accuracy of the model.

[0004] To achieve the above-mentioned purpose, the technical solution of the present invention is: a method for constructing a transient stability assessment model of an electric power system based on a sample extraction method. First, a support vector that acts on the decision boundary is obtained based on SVM calculation, and critical samples are extracted in combination with the severity of sample faults; then, the position information of the approximate tangent plane of the neighboring samples and the class surface is statistically analyzed, and edge samples are extracted in combination with the severity of sample faults; finally, a new sample set is constructed using the extracted samples for training the electric power system stability assessment model.

[0005] In one embodiment of the present invention, the method is specifically implemented as follows:

[0006] 1) A large number of training samples are obtained for the simulation model through different fault information; a large number of training samples are obtained through the permutations and combinations of different operating conditions, transmission lines, fault locations, and fault durations; the power angle and angular velocity deviations of each generator set in the system before, at, and after the fault is cleared are selected as the feature quantities of the samples, and the sample features are normalized by z-score to construct a sample set S = {x i ,i=1,2,...,N}, stable sample set S s ={x s,c ,c=1,2,...,N s}, and unstable sample set S u ={x u,z ,z=1,2,...,N u}, Where N is the number of samples, N s is the number of stable samples, N u is the number of unstable samples;

[0007] 2) Calculate the fault severity of each sample, as shown in formula (1):

[0008]

[0009] Among them: F i For sample x i The severity of the fault; t CCT (x i ) represents the sample x i The corresponding critical fault removal time; t i (x i ) represents the sample x i Duration of the fault;

[0010] 3) Input the sample set S, and train the SVM with the goal of solving the classification hyperplane that maximizes the geometric interval to obtain the Lagrange multiplier α of each sample i ; When the sample x i Alpha i Satisfying 0<α i ≤C and 0≤F i ≤μ c , extract the sample as the critical sample and construct the critical sample set, as shown in formula (2):

[0011] S c ={x i ,i=1,2,...,N|α i ∈(0,C],F i ∈[0,μ c ]} (2)

[0012] Where: C is the penalty coefficient of SVM; μ c is the sample low fault severity threshold coefficient;

[0013] 4) Calculate the center points of the stable sample set and the unstable sample set respectively, as shown in formula (3) and formula (4):

[0014]

[0015]

[0016] Where: v S and v U are the center points of the stable sample set and the unstable sample set respectively;

[0017] 5) The k nearest neighbor algorithm is used to calculate the k nearest neighbor samples of each sample, and the positional relationship between the k nearest neighbor samples and the tangent plane passing through the target sample is calculated. Then, by statistically analyzing these positional information, it is determined whether the target sample is at the edge of the class, as shown in equations (5) and (6):

[0018]

[0019]

[0020] Where: L s,c For a stable sample x s,c Location information indicator; L u,z For unstable samples x u,z Position information indicator; x s,c,j For a stable sample x s,c The jth nearest neighbor sample of x; u,z,j For unstable samples x u,z The jth nearest neighbor sample of ; when the input variable of the I(·) function is greater than or equal to zero, the function outputs 1, otherwise it outputs 0;

[0021] When the target sample location information and fault severity exceed the set threshold, the sample will be extracted as an edge sample and the edge sample set will be constructed, as shown in formula (7):

[0022]

[0023] Among them: F s,c For a stable sample x s,c The severity of the fault; F u,z For unstable samples x u,z The severity of the fault; γ is the local surface threshold coefficient; μ e is the high fault severity threshold coefficient;

[0024] 6) Combine the critical sample set and the marginal sample set, as shown in formula (8):

[0025] S new =S c ∪S e (8)

[0026] 7) Input the optimized sample set S new , training the power system stability assessment model.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention extracts key samples that can fully represent the original sample space according to the spatial distribution of sample data and the severity of sample faults in the power system, and eliminates redundant samples that have a low contribution to the model, thereby formulating a sample extraction method based on the sample space distribution. Existing methods usually do not consider the importance of key samples to model training during the sampling process, and are likely to eliminate key samples, thereby destroying the spatial distribution of the original sample set and resulting in unsatisfactory model performance. The present invention comprehensively considers the spatial distribution of samples and the severity of sample faults, and the optimized training set is used for training, which improves the training efficiency and evaluation accuracy of the power system stability assessment model, while also improving the evaluation tendency of the model to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 The present invention provides a specific flow chart of the offline training of the power system stability assessment model. DETAILED DESCRIPTION

[0030] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings.

[0031] The present invention provides a method for constructing a transient stability assessment model for a power system based on a sample extraction method. According to the spatial distribution information of the samples and the fault severity of the samples, key samples that play an important role in model training are extracted, and redundant samples are eliminated, thereby optimizing the sample set and improving the performance of the model. First, the support vector that acts on the decision boundary is obtained based on SVM calculation, and the critical samples are extracted in combination with the fault severity of the samples; then, the position information of the approximate tangent plane of the neighboring samples and the class surface is statistically analyzed, and the edge samples are extracted in combination with the fault severity of the samples; finally, the extracted samples are used to construct a new sample set for training the stability assessment model of the power system.

[0032] The following are specific implementation examples of the present invention.

[0033] The simulation software uses PSD-BPA, and the test system uses an IEEE-39 node system with a speed regulator and a voltage regulator. The operating state of the power system is set to five load levels: 90%, 95%, 100%, 105% and 110%, and permanent three-phase short-circuit faults are set at 10%, 50% and 90% of the distance from the head end of each fault line. The fault removal time is set to 5 to 20 cycles after the fault occurs, the longest simulation time is 4s, and the system operation information is sampled 120 times per second. A total of 15012 samples are generated on the IEEE-39 node system, including 7191 stable samples and 7821 unstable samples. Randomly extract sample data from stable samples and unstable samples to form training sets and test sets. The number of training sample sets is 9000, the number of test samples is 2000, the ratio of unstable samples to stable samples in the training set is 1:2, and the ratio of unstable samples to stable samples in the test set is 1:1.

[0034] The calculation example uses a stacked sparse autoencoder (SSAE) to establish a power system stability assessment model. The parameters of the stacked sparse autoencoder are as follows: the number of neurons in the hidden layer is 100, 50, and 25 respectively; pre-training is 100 times; fine-tuning is 400 times; L1 regularization parameter is 0.0005, and L2 regularization parameter is 0.00005. The training set is optimized according to the patented method and then used to train the SSAE model. The parameters of this method are set to k=25, γ=0.10, C=0.01, μ c =0.2, μ e =0.7.

[0035] This example is based on the SSAE classifier and compares several commonly used sample sampling methods, including oversampling methods (SMOTE, ADASYN and Borderline-SMOTE), random undersampling method (RUS) and the sample extraction method introduced in this invention;

[0036] The oversampling methods (ADASYN, SMOTE and Borderline-SMOTE) generate system unstable samples until the number is consistent with the system stable samples, and then input the samples into SSAE for training; the random undersampling method (RUS) randomly extracts samples with the same number of system unstable samples from the system stable samples, and then inputs the extracted samples into the SSAE model for training.

[0037] The stable sample prediction error rate is defined as follows.

[0038]

[0039] In the formula, F s is the stable number of samples with incorrect predictions, T sis the stable number of samples for correct prediction.

[0040] The unstable sample prediction error rate is defined as follows.

[0041]

[0042] In the formula, F u is the number of unstable samples with incorrect predictions, T u is the number of unstable samples predicted correctly.

[0043] The global accuracy of the evaluated model is shown below.

[0044]

[0045] The experimental results of each method are compared in Table 1:

[0046] Table 1 Comparison of test results of different sampling methods

[0047]

[0048]

[0049] It can be seen from Table 1 that the model training efficiency of the method of the present invention is the fastest, the overall accuracy of the model is the highest, and the evaluation tendency correction effect of the model is the best.

[0050] The above are preferred embodiments of the present invention. Any changes made according to the technical solution of the present invention, as long as the resulting functions do not exceed the scope of the technical solution of the present invention, belong to the protection scope of the present invention.

Claims

1. A method for constructing a transient stability assessment model for a power system based on a sample extraction method, characterized in that: Firstly, the support vector acting on the decision boundary is obtained based on SVM calculation, and the critical samples are extracted in combination with the severity of sample faults. Then, the position information of the approximate tangent plane between the nearest samples and the class surface is statistically analyzed, and the edge samples are extracted in combination with the severity of sample faults. Finally, the extracted samples are used to construct a new sample set for training the power system stability assessment model.

2. The method for constructing a power system transient stability assessment model based on a sample extraction method according to claim 1, characterized in that: The specific implementation of this method is as follows: 1) Obtain a large number of training samples for the simulation model through different fault information; obtain a large number of training samples through the arrangement and combination of different operating conditions, transmission lines, fault locations, and fault durations; The power angle and angular velocity deviations of each generator set in the system before, at, and after fault clearing are selected as the feature quantities of the samples, and the sample features are normalized by z-score to construct a sample set S = {x i ,i=1,2,...,N}, stable sample set S s ={x s,c ,c=1,2,...,N s }, and unstable sample set S u ={x u,z ,z=1,2,...,N u }, Where N is the number of samples, N s is the number of stable samples, N u is the number of unstable samples; 2) Calculate the fault severity of each sample, as shown in formula (1): Among them: F i For sample x i The severity of the fault; t CCT (x i ) represents the sample x i The corresponding critical fault removal time; t i (x i ) represents the sample x i Duration of the fault; 3) Input the sample set S, and train the SVM with the goal of solving the classification hyperplane that maximizes the geometric interval to obtain the Lagrange multiplier α of each sample i ; When the sample x i Alpha i Satisfying 0<α i ≤C and 0≤F i ≤μ c , extract the sample as the critical sample and construct the critical sample set, as shown in formula (2): S c ={x i ,i=1,2,...,N|α i ∈(0,C],F i ∈[0,μ c ]} (2) Where: C is the penalty coefficient of SVM; μ c is the sample low fault severity threshold coefficient; 4) Calculate the center points of the stable sample set and the unstable sample set respectively, as shown in formula (3) and formula (4): Where: v S and v U are the center points of the stable sample set and the unstable sample set respectively; 5) The k nearest neighbor algorithm is used to calculate the k nearest neighbor samples of each sample, and the positional relationship between the k nearest neighbor samples and the tangent plane passing through the target sample is calculated. Then, by statistically analyzing these positional information, it is determined whether the target sample is at the edge of the class, as shown in equations (5) and (6): Where: L s,c For a stable sample x s,c Location information indicator; L u,z For unstable samples x u,z Position information indicator; x s,c,j For a stable sample x s,c The jth nearest neighbor sample of x; u,z,j For unstable samples x u,z The jth nearest neighbor sample of ; when the input variable of the I(·) function is greater than or equal to zero, the function outputs 1, otherwise it outputs 0; When the target sample location information and fault severity exceed the set threshold, the sample will be extracted as an edge sample and the edge sample set will be constructed, as shown in formula (7): Where: S e,s is the edge sample of the stable sample; S e,u is the edge sample of the unstable sample; F s,c For a stable sample x s,c The severity of the fault; F u,z For unstable samples x u,z The severity of the fault; γ is the local surface threshold coefficient; μ e is the high fault severity threshold coefficient; 6) Combine the critical sample set and the marginal sample set, as shown in formula (8): S new =S c ∪S e (8) 7) Input the optimized sample set S new , training the power system stability assessment model.

Images