Power system transient stability adaptive evaluation method and system oriented to operation scene switching

Abstract

The invention discloses an adaptive evaluation method and system for transient stability of a power system for operation scene switching. Comprising the steps of 1, generating small sample data of a new scene; step 2, performing feature decoupling on the evaluation classifier; 3, adaptively updating the transient stability evaluation model; step 3.1, obtaining known and unknown spatial feature projections; step 3.2, calculating the divergence of the category score inconsistency; step 3.3, adaptively updating the evaluation model; step 4, performing online evaluation; step 4.1, outputting a transient stability online evaluation result; and step 4.2, updating and judging the evaluation model. According to the method, the known feature space of the source domain of the original operation scene and the unknown feature space of the target domain after the operation scene is switched are distinguished, and the loss function of the category score inconsistency divergence is introduced, so that the transient stability evaluation model is adaptive to different domain features in updating; therefore, the evaluation precision and reliability of the transient stability evaluation model of the power system after the operation scene switching are ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system automation, and particularly relates to a power system transient stability adaptive evaluation method and system for operation scenario switching. Background Art

[0002] China's power system is gradually showing the characteristics of high proportion of new energy and high proportion of power electronic transpositions. Therefore, the dynamic characteristics of the power system become more complex, greatly increasing the risk of power system transient instability. In order to prevent major losses in all aspects caused by transient instability, it is necessary to conduct transient stability assessment of the power system.

[0003] With the installation of a large number of phasor measurement units (PMUs) in the wide-area measurement system and the development of artificial intelligence technology, data-driven power system transient stability assessment methods have received extensive attention. However, in an actual power system, the original topological structure and operating characteristics may change due to the maintenance of generators or lines, or the addition of generators or lines. This will greatly reduce the evaluation performance of the historical model trained with the original topological power grid data. Collecting a large amount of new topological power grid data and retraining an available model often takes a lot of time and cannot adapt to the frequent operation scenario switching of the power grid. Therefore, a power system transient stability adaptive evaluation method for operation scenario switching is proposed, which is characterized by updating the evaluation model only using small sample data to adapt to the transient stability assessment in the new scenario and improving the evaluation accuracy in the new scenario. Therefore, this method has certain value. Summary of the Invention

[0004] The present invention is made to solve the above problems, and aims to provide a power system transient stability adaptive evaluation method and system for operation scenario switching, which can use small samples to adaptively update the transient stability evaluation model when the power system undergoes operation scenario switching, thereby improving the accuracy of power system transient stability evaluation after the operation scenario switching of the power system.

[0005] To achieve the above objectives, the present invention adopts the following solutions:

[0006] In the first aspect, the present invention provides a power system transient stability adaptive evaluation method for operation scenario switching, and the method includes:

[0007] Step 1. According to the network topology of the power system after operation scenario switching, use time-domain simulation to generate a small number of time series samples with electrical characteristics and a sample number of N T , match them with the corresponding transient stability states, and normalize the electrical characteristics in the time series samples to set them as small sample data in the target domain;

[0008] Step 2. Obtain the source model for evaluating the transient stability state of the original operation scenario of the power system, and based on the source model, obtain the electrical characteristics related to the known scenarios in the source domain and the electrical characteristics related to the unknown scenarios in the target domain;

[0009] Step 3. Input the small sample data in the target domain in Step 1 into the source model, obtain the mapping of the electrical characteristic space related to the known scenarios in the source domain and the mapping of the electrical characteristic space related to the unknown scenarios in the target domain, and adaptively update the overall transient stability assessment model;

[0010] Step 4. Online obtain the measurement data of the power system after the operation scenario is switched, input it into the overall transient stability assessment model after the adaptive update is completed, and obtain the transient stability assessment result applicable to the new scenario. If the operation scenario of the power system changes again, repeat Steps 1 to 4.

[0011] Further, Step 1 includes the following sub-steps:

[0012] Step 1.1, according to the network topology of the power system after the operation scenario is switched, use the time-domain simulation method to obtain a small number of simulation samples, set the number of samples generated by the simulation to N T , record the active power and reactive power of the generator, the voltage and phase angle of the bus, the active power and reactive power transmitted by the line, and the active power and reactive power of the load at the moment one sampling period before the fault occurs, at the moment of the fault, and T end sampling periods after the fault is cleared, construct a feature vector x containing time series * , and then set the transient stability state y of the power system to 0 or 1, representing the transient stability and transient instability of the system respectively, and associate the transient stability state y in each operating state with the feature vector x * correspondingly;

[0013] Step 1.2, perform normalization processing on the feature vector x * to obtain the normalized feature vector x:

[0014] Step 1.3, based on the normalized feature vector x and the transient stability state y of the power system, obtain the sample set {x, y} after the feature phasor is normalized, and set this sample set as the small sample data in the target domain.

[0015] Further, the source model includes a feature extractor network and an evaluation classifier network.

[0016] Further, obtaining the electrical characteristics related to the known scenarios in the source domain and the electrical characteristics related to the unknown scenarios in the target domain includes:

[0017] Obtain the weights of the evaluation classifier network in the source model; perform orthogonal decoupling of the features of the weights of the evaluation classifier network to obtain electrical features related to known scenarios in the source domain and electrical features related to unknown scenarios in the target domain.

[0018] Further, the source model is:

[0019] g θ (x) = f θ (h θ (x))

[0020] where θ represents network parameters, h θ (x) represents the feature extractor network of the source model, f θ () represents the evaluation classifier network of the source model, g θ (x) represents the source model for evaluating the transient stability state of the original scenario;

[0021] The orthogonal decoupling of the features of the weights of the evaluation classifier network is as follows:

[0022] W cls = UΣV T ,

[0023] F knw = span{v n | n＝1,...,C}

[0024] F unk = span{v n | n＝C+1,...,D}

[0025] where, W cls ∈R C×D represents the weights of the evaluation classifier network f θ (), ∑∈R C×D represents a diagonal matrix, U∈R C×C and V∈R D×D both represent orthogonal unit matrices, F knw represents the known feature space in the source domain, F unk represents the positive orthogonal complement space of F knw , that is, the unknown feature space, span represents the vector space spanned by a series of feature vectors, represents the column vectors of the orthogonal unit matrix V, C represents the dimension of the known feature space in the source domain, and D represents the total dimension of the known feature space and the unknown feature space in the source domain.

[0026] Further, the step 3 of adaptively updating the overall transient stability evaluation model includes:

[0027] Based on the electrical characteristics related to the known scenarios in the source domain and the electrical characteristics related to the unknown scenarios in the target domain obtained by orthogonal decoupling in Step 2, obtain the feature space mapping of the known scenarios in the source domain and the feature space mapping of the unknown scenarios in the target domain; then add two scoring auxiliary classifier networks to the source model, calculate the class scoring inconsistency divergence between the known scenarios in the source domain and the unknown scenarios in the target domain, so as to adaptively update the overall transient stability assessment model.

[0028] Further, Step 3 includes the following sub-steps:

[0029] Step 3.1, using the small sample dataset {x, y} of the target domain, input the feature extractor network h θ (x), and set the feature mapping obtained from the small sample data of the target domain as Expressed as:

[0030]

[0031]

[0032]

[0033]

[0034] where i represents the i-th small sample of the target domain, and represent the projections on the known and unknown spaces in the source domain respectively, and l n ∈R represents the weight on the basis vector v n ;

[0035] Step 3.2, add two scoring auxiliary classifier networks to the source model, denoted as f1 and f2 respectively, for calculating the class scoring inconsistency divergence, specifically expressed as:

[0036]

[0037] where, represents the class scoring inconsistency divergence between the known scenarios in the source domain and the unknown scenarios in the target domain, represents the scoring function, and SD represents the inconsistency between the two scoring functions f1 and f2 in the same domain,

[0038] Step 3.3, execute the loop from t = 1 until t = t max , iteratively train the feature extractor network and the scoring auxiliary classifier network of the transient stability assessment model to reduce the class scoring inconsistency divergence between the known scenarios in the source domain and the unknown scenarios in the target domain; the overall objective function of the transient stability assessment model is expressed as:

[0039]

[0040] In the formula represents the classification loss function of the target domain small sample data;

[0041] The process of updating the parameters of the feature extractor network and the scoring auxiliary classifier network is expressed as:

[0042]

[0043] In the formula, θ h represents the parameters of the feature extractor network, and θ a represents the parameters of the scoring auxiliary classifier network, represents the classification loss function of the target domain small sample data, represents the divergence loss function of class score inconsistency, λ represents the variable gradient reversal balance coefficient, and η represents the exponential decay learning rate;

[0044] Judge whether the iterative loop of the transient stability assessment model training is completed. If so, end the training iteration process to obtain a transient stability assessment model that can be used for the power system operation scenario after switching.

[0045] Furthermore, SD is expressed as:

[0046]

[0047]

[0048]

[0049]

[0050] In the formula, K represents the number of sample categories, O represents the known or unknown scenario of the source domain, and i and j respectively represent the row vector index and column vector index of M (ρ) The row vector index and column vector index of, Φ ρ (x) represents the ramp loss function, ρ represents the threshold parameter of the ramp loss function, and μ k (f(x), y) represents the absolute margin function, and y represents the set of class label spaces.

[0051] Furthermore, step 4 includes the following sub-steps:

[0052] Step 4.1, use the phasor measurement unit PMU in the wide area measurement system to obtain the measurement data after the power system operation scenario is switched, and input it into the transient stability assessment model obtained in step 3 for online assessment;

[0053] Step 4.2: Determine whether the current operating scenario of the power system has switched. If there is no switching of the operating scenario, continue to execute Step 4. If there is a switching of the operating scenario, loop through Steps 1 to 4.

[0054] In a second aspect, the present invention provides a transient stability adaptive evaluation system for a power system facing operating scenario switching. The system includes:

[0055] A small sample data generation module for the new scenario of the power system, which is used to generate a small number of time series samples with various electrical characteristics according to the network topology of the power system after the operating scenario switching, with the number of samples being N T using time domain simulation. Then, match them with the corresponding transient stability states and normalize the electrical characteristics in the time series samples, and set them as the small sample data for the target domain.

[0056] A feature decoupling module for the transient stability evaluation classifier of the power system, which is used to obtain the source model for evaluating the transient stability state of the original operating scenario of the power system, and based on the source model, obtain the electrical characteristics related to the known scenarios in the source domain and the electrical characteristics related to the unknown scenarios in the target domain.

[0057] An adaptive update module for the transient stability evaluation model of the power system. Input the small sample data for the target domain into the source model, obtain the electrical characteristic space mapping related to the known scenarios in the source domain and the electrical characteristic space mapping related to the unknown scenarios in the target domain, and adaptively update the overall transient stability evaluation model.

[0058] An online transient stability evaluation module for the power system, which is used to online obtain the measurement data of the power system after the operating scenario switching, input it into the overall transient stability evaluation model after the adaptive update is completed, and obtain the transient stability evaluation result applicable to the new scenario. If the operating scenario of the power system changes again, start the above modules again.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The power system transient stability adaptive evaluation method, system and storage medium for operation scenario switching according to the present invention decouple the features of the evaluation classifier weights of the model, thereby distinguishing the source domain known feature space of the original operation scenario and the target domain unknown feature space after the operation scenario switching, and determining the initial difference of data in different domains; further, a calculation method of class score inconsistency divergence is introduced, which can more effectively measure the inconsistency between data in different domains, thereby prompting the transient stability evaluation model to adaptively update during the iterative training process, continuously shortening the distance between the target domain unknown feature distribution and the source domain known feature distribution, and achieving the goal of consistency of features in different domains; finally, the feature extractor network of the updated transient stability evaluation model can extract common features in different domains, that is, different operation scenarios, and the classifier network gives accurate prediction results for the target domain data during the transient stability evaluation, ensuring the evaluation accuracy and reliability of the power system transient stability evaluation model after the operation scenario switching. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the embodiments or the description of the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to these drawings.

[0062] Figure 1 It is a flowchart of the power system transient stability adaptive evaluation method for operation scenario switching according to the embodiment of the present invention;

[0063] Figure 2 It is a network structure diagram of the transient stability evaluation model according to the embodiment of the present invention.

[0064] Figure 3 It is a feature mapping change diagram of the target domain unknown scenario and the source domain known scenario according to the embodiment of the present invention.

[0065] Figure 4 It is a continuous scenario switching evaluation result change diagram according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] In order to make the above objects, features and advantages of the present application more obvious and understandable, the following will make a detailed description of the specific embodiments of the present application with reference to the drawings. Many specific details are set forth in the following description in order to fully understand the present application. However, the present application can be implemented in many other ways different from those described herein. Those skilled in the art can make similar improvements without departing from the connotation of the present application. Therefore, the present application is not limited by the specific embodiments disclosed below.

[0067] Embodiment 1

[0068] As Figure 1 shown, the power system transient stability adaptive evaluation method for operation scenario switching provided in this embodiment includes the following steps:

[0069] Step 1: Generation of small-sample data for the new scenario. According to the network topology after the operation scenario switching of the power system, time-domain simulation is used to generate a small number of samples of time series containing various electrical characteristics with the sample number of N T , and then they are matched with the corresponding transient stability states and the characteristics are normalized, which are set as the small-sample data of the target domain.

[0070] Step 1.1, in this embodiment, the IEEE-39 bus system in which the synchronous generators of busbars No. 33, 34, and 36 are replaced by three wind farms with equal capacity and the synchronous generators of busbars No. 37 and 38 are replaced by two photovoltaic solar farms with equal capacity is selected as the test power system. It is set that the network topology change after the operation scenario switching of the power system is to remove 2 generators, namely generator G31 of busbar No. 31 and generator G35 of busbar No. 34, and 3 loads, namely load L16 of busbar No. 16, load L25 of busbar No. 25, and load L29 of busbar No. 29. The time-domain simulation method is used to obtain a small number of simulation samples, and the number of samples generated by simulation is set as N T = 400. The active power and reactive power of the generator, the voltage and phase angle of the busbar, the active power and reactive power transmitted by the line, and the active power and reactive power of the load at one moment before the fault occurrence, the moment of fault occurrence, and T end = 20 sampling periods after the fault clearance are recorded to construct a feature vector x * containing time series, and then the transient stability state y of the power system is set to 0 or 1, which respectively represent the transient stability and transient instability of the system. The transient stability state y in each operating state is corresponding to the feature vector x * .

[0071] Step 1.2, the feature vector x * is normalized to change the data range to between [-1, 1] to obtain the normalized feature vector x:

[0072]

[0073] In the formula: represents the jth feature of the ith sample, represents the maximum value of the features of the ith sample, represents the minimum value of the features of the ith sample, represents the mean value of the features of the ith sample. The sample set {x, y} after the feature vector normalization is obtained, and this sample set is set as the small-sample data of the target domain.

[0074] Step 2: Evaluate the decoupling of classifier features. Obtain the source model for evaluating the transient stability state of the original operation scenario of the power system in this embodiment, including a feature extractor module and an evaluation classifier module. Then, obtain the weights of the evaluation classifier network layer, perform orthogonal decoupling of its features, and obtain the features related to the known scenarios in the source domain and the features related to the unknown scenarios in the source domain.

[0075] In this embodiment, the transient stability evaluation model for evaluating the original operation scenario of the power system is obtained and set as the source model, which is composed of a feature extractor network and an evaluation classifier network, and can be expressed as:

[0076] g θ (x) = f θ (h θ (x))

[0077] where θ represents the network parameters, h θ represents the feature extractor network of the source model, f θ represents the evaluation classifier network of the source model, and g θ represents the source model for evaluating the transient stability state of the original scenario;

[0078] Define the weight of the evaluation classifier network f θ as W cls ∈ R C×D . To achieve the goal of distinguishing between known and unknown data in the source domain, perform orthogonal decoupling of W cls , and decompose the sample features into two independent parts as follows:

[0079] W cls = UΣV T ,

[0080] F knw = span{v n | n＝1,...,C},

[0081] F unk = span{v n | n＝C+1,...,D}

[0082] where ∑ ∈ R C×D represents a diagonal matrix, U ∈ R C×C and V ∈ R D×D both represent orthogonal unit matrices, F knw represents the known feature space in the source domain, F unk represents the positive orthogonal complement space of F knw , that is, the unknown feature space, and span represents the vector space spanned by a series of feature vectors. Denote the column vectors of the orthogonal unit matrix V, C represents the dimension of the known feature space in the source domain, and D represents the total dimension of the known feature space and the unknown feature space in the source domain.

[0083] Step 3: Adaptive update of the transient stability assessment model. In this embodiment, the small sample data of the target domain in Step 1 is input into the feature extractor, and the mapping of the known scenario feature space in the source domain and the mapping of the unknown scenario feature space in the target domain are obtained by combining the decoupled features in Step 2. Then, two scoring auxiliary classifiers are added to the source model to calculate the divergence of the class score inconsistency between the known scenario in the source domain and the unknown scenario in the target domain, thereby adaptively updating the overall transient stability assessment model so that the model can be used for transient stability assessment after the operation scenario is switched.

[0084] Step 3.1, Obtain the projections of the known and unknown space features.

[0085] In this embodiment, the small sample data set {x, y} of the target domain obtained in Step 1 is used as the input to the feature extraction network h θ , and the feature mapping obtained from the small sample data of the target domain is set as which can be obtained by the weighted sum of the two orthogonal basis vectors F unk and F knw , specifically expressed as:

[0086]

[0087]

[0088]

[0089]

[0090] where i represents the i-th small sample of the target domain, and represent the projections on the known and unknown spaces in the source domain respectively. l n ∈R represents the weight on the basis vector v n .

[0091] Step 3.2, Calculate the divergence of the class score inconsistency.

[0092] In this embodiment, two scoring auxiliary classifiers are added to the source model, denoted as f1 and f2 respectively, to calculate the divergence of the class score inconsistency, specifically expressed as:

[0093]

[0094] where, Indicates the divergence of class scores between the known scenarios in the source domain and the unknown scenarios in the target domain. Denotes the scoring function. SD represents the inconsistency between two scoring functions f1 and f2 in the same domain, and SD can be specifically expressed as:

[0095]

[0096]

[0097]

[0098]

[0099] In the formula, O represents the known or unknown scenarios in the source domain, i and j respectively represent the row vector index and column vector index of M (ρ) of, Φ ρ (x) represents the ramp loss function, ρ represents the threshold parameter of the ramp loss function, μ k (f(x), y) represents the absolute margin function, and y ∈ [0, 1] represents the set of class label spaces.

[0100] Step 3.3, adaptively update the evaluation model.

[0101] In this embodiment, the loop is executed from t = 1 to t = 100 to iteratively train the feature extractor network and the scoring auxiliary classifier network of the transient stability evaluation model, reducing the divergence of class scores between the known scenarios in the source domain and the unknown scenarios in the target domain, increasing the consistency between the two domains, thereby promoting the feature mappings of the feature extractor network in the unknown scenarios of the target domain and in the known scenarios of the source domain to tend to be consistent, achieving domain adaptation. The overall objective function of the transient stability evaluation model can be expressed as:

[0102]

[0103] In the formula represents the classification loss function of the small sample data in the target domain, and the cross-entropy loss function is adopted;

[0104] The process of updating the parameters of the feature extractor network and the scoring auxiliary classifier network can be expressed as:

[0105]

[0106] In the formula, θ h represents the parameters of the feature extractor network, θ a represents the parameters of the scoring auxiliary classifier network, represents the classification loss function of the small sample data in the target domain, It represents the category scoring inconsistency divergence loss function, λ represents the variable gradient flip balance coefficient, which is used to flexibly control the training levels of the scoring auxiliary classifier and the feature extractor, η represents the exponential decay learning rate, and the specific representations of λ and η are as follows:

[0107]

[0108]

[0109] In the formula, p is the ratio of the current training iteration number to the total iteration number of 100, representing the training process from 0 to 1.

[0110] Judge whether the iterative loop of the transient stability assessment model training is completed. If so, end the training iteration process to obtain a transient stability assessment model that can be used after the power system operation scenario is switched.

[0111] During the training process, the two-dimensional dimensionality reduction distribution of the feature mapping of the feature extractor network in the unknown scenario of the target domain and the known scenario of the source domain is as Figure 3 shown Figure 3 (a) to 3(d) are the distribution results when training 0 times, 5 times, 50 times, and 100 times respectively. Before the training starts, the feature distributions of the source domain and the target domain are quite different. After multiple rounds of training, the feature distributions show a trend of gradually approaching and merging, indicating that the feature extractor can capture more common features between different scenarios in the feature space, enabling the classifier network to better adapt to different scenarios.

[0112] In this embodiment, the accuracy rate λ a , recall rate λ r and Gmeans value λ G are selected to reflect the performance of the model evaluation:

[0113]

[0114]

[0115]

[0116] In the formula, N TS , N FS , N FU , N TU represent the number of correct predictions of stability, the number of incorrect predictions of stability, the number of incorrect predictions of instability, and the number of correct predictions of instability respectively.

[0117] Table 1 Scenario switching test results

[0118]

[0119] The test results of whether to update the model using the method of the present invention are compared as shown in Table 1. It can be seen that after the model is updated, the impact on the evaluation effect of the original scenario is small, and a relatively high evaluation accuracy can still be maintained. After the model is updated, it can adapt to the switching of the power system operation scenario, improve the evaluation accuracy of the scenario after the model evaluation switches, and achieve the adaptive evaluation of the operation scenario switching.

[0120] Step 4: Online evaluation. The PMU online obtains the measurement data of the power system after the operation scenario switches, and inputs it into the transient stability evaluation model that has been adaptively updated in Step 3 to obtain the transient stability evaluation result applicable to the new scenario. If the power system operation scenario changes again, repeat the above Steps 1 to 4.

[0121] Step 4.1, Output of the transient stability online evaluation result.

[0122] In this embodiment, the phasor measurement unit PMU in the wide area measurement system is used to obtain the measurement data after the power system operation scenario switches, and input it into the transient stability evaluation model obtained in Step 3 for online evaluation. This process can be expressed as:

[0123] y new =f θ (h θ ′(x new ))

[0124] In the formula, x new represents the measurement data after the power system operation scenario switches, h θ ′ represents the feature extractor network after the model is updated, and y new represents the online transient stability evaluation result.

[0125] Step 4.2, Judgment on the update of the evaluation model.

[0126] In this embodiment, it is judged whether the current power system operation scenario switches. If the operation scenario does not switch, continue to execute Step 4. If the operation scenario switches, loop to execute Steps 1 to 4.

[0127] In this embodiment, it is set that the operation scenario switches again, the removed loads and generators are reconnected to the grid, and at the same time, two new lines are added, connecting bus 14 and bus 17, bus 18 and bus 25 respectively, and a synchronous generator is connected to bus 2. The method of this embodiment is used to update the model again for evaluation. The evaluation results from the original scenario to the second switching scenario are as Figure 4 shown. In Figure 4 , the existing scenario migration algorithm DANN and the fine-tuning algorithm are compared. Figure 4 (a) to 4(c) are the accuracy rates λ a , recall rates λ rand the Gmeans value λ G As for the variation, it can be seen that the accuracy rate, recall rate, and Gmeans value of the proposed method are all the highest, indicating that for the changes in the continuous operation scenario, the evaluation model has strong adaptability and good generalization performance, and can effectively update the model to adapt to the changes in the power system scenario.

[0128] Embodiment 2

[0129] This embodiment provides a transient stability adaptive evaluation system for a power system facing operation scenario switching that can automatically implement the above-mentioned method of the present invention, including a small sample data generation module for new scenarios of the power system, a feature decoupling module for a transient stability evaluation classifier of the power system, an adaptive update module for a transient stability evaluation model of the power system, and an online transient stability evaluation module for the power system.

[0130] The small sample data generation module for new scenarios of the power system is used to execute the content described in step 1 above to generate small sample data for new scenarios. This module is used to obtain the time series feature vector after the operation scenario of the power system is switched, match it with the corresponding transient stability state, and then perform normalization processing on the feature vector to obtain a normalized feature vector, and construct small sample data for the target domain.

[0131] The feature decoupling module for a transient stability evaluation classifier of the power system is used to execute the content described in step 2 above to obtain the features related to the known scenarios in the source domain and the features related to the unknown scenarios. This module extracts the weights of the evaluation classifier network layer by using the evaluation classifier module in the source model, and then performs orthogonal decoupling of the features to obtain the known feature space and unknown feature space in the source domain.

[0132] The adaptive update module for a transient stability evaluation model of the power system is used to execute the content described in step 3 above to update the transient stability evaluation model for use in the transient stability evaluation of the power system after the scenario is switched. This module includes a unit for obtaining projections of known and unknown space features, a unit for calculating the category score inconsistency divergence, and a unit for adaptively updating the evaluation model. The unit for obtaining projections of known and unknown space features is used to execute the content described in step 3.1 above to obtain the projections on the known and unknown spaces in the source domain. The unit for calculating the category score inconsistency divergence is used to execute the content described in step 3.2 above to calculate the category score inconsistency divergence of the projections of the known and unknown space features by using two scoring auxiliary classifiers. The unit for adaptively updating the evaluation model is used to execute the content described in step 3.3 above to obtain a transient stability evaluation model that can be used after the operation scenario of the power system is switched.

[0133] The on-line evaluation module of the power system is used to execute the content described in step 4 above, and on-line evaluate the transient stability state after the operation scenario of the power system is switched. This module includes a transient stability on-line evaluation result output unit and an evaluation model update determination unit. The transient stability on-line evaluation result output unit is used to execute the content described in step 4.1 above, and obtain the on-line transient stability evaluation result after the operation scenario of the power system is switched. The evaluation model update determination unit is used to execute the content described in step 4.2 above, and determine whether to enable the aforementioned module again according to whether the operation scenario of the power system is switched again.

[0134] The input display module is used to display the input, output and intermediate processing data of the corresponding part in the form of text, list, or two-dimensional / three-dimensional static / dynamic graph according to the operation instructions input by the operator.

[0135] The control module is communicatively connected to the small sample data generation module for the new scenario of the power system, the feature decoupling module for the transient stability evaluation classifier of the power system, the adaptive update module for the transient stability evaluation model of the power system, the on-line transient stability evaluation module of the power system, and the input display module, and controls their operations.

[0136] It should be understood that the parts not elaborated in detail in this specification all belong to the prior art.

[0137] It should be understood that the above description of the preferred embodiment is relatively detailed, and it should not be considered as a limitation to the protection scope of the present invention. Under the inspiration of the present invention, those of ordinary skill in the art can also make substitutions or deformations without departing from the protection scope defined by the claims of the present invention, and all fall within the protection scope of the present invention. The scope of protection claimed by the present invention shall be subject to the appended claims.

Claims

1. A transient stability adaptive assessment method for power systems facing operation scenario switching, characterized in that, The method includes: Step 1. According to the network topology of the power system after the operation scenario is switched, use time-domain simulation to generate a small number of time series samples with N T electrical characteristics. Then match them with the corresponding transient stable states and normalize the electrical characteristics in the time series samples, which are set as the small sample data in the target domain; Step 2. Obtain a source model for evaluating the transient stability state of the original operation scenario of the power system, and based on the source model, obtain electrical characteristics related to known scenarios in the source domain and electrical characteristics related to unknown scenarios in the target domain; Step 3. Input the small sample data in the target domain in Step 1 into the source model, obtain the mapping of the electrical characteristic space related to known scenarios in the source domain and the mapping of the electrical characteristic space related to unknown scenarios in the target domain, and adaptively update the overall transient stability evaluation model; Step 4. Online obtain the measurement data of the power system after the operation scenario is switched, input it into the overall transient stability evaluation model that has completed adaptive update, and obtain the transient stability evaluation result applicable to the new scenario. If the operation scenario of the power system changes again, repeat Steps 1 to 4.

2. The transient stability adaptive assessment method for a power system facing operation scenario switching according to claim 1, characterized in that, The said Step 1 includes the following sub-steps: Step 1.1, according to the network topology of the power system after the operation scenario is switched, obtain a small number of simulation samples by using the time-domain simulation method, and set the number of samples generated by the simulation to N T , record the active power and reactive power of the generator, the voltage and phase angle of the bus, the active power and reactive power transmitted by the line, and the active power and reactive power of the load at the moment before the fault occurs, the moment when the fault occurs, and T end sampling periods after the fault is cleared, and construct a feature vector x containing time series * , then set the transient stability state y of the power system to 0 or 1, which represent the transient stability and transient instability of the system respectively, and associate the transient stability state y with the feature vector x * correspondingly; Step 1.

2. Normalize the feature vector x * to obtain the normalized feature vector x: Step 1.3, based on the normalized feature vector x and the transient stability state y of the power system, obtain the sample set {x, y} after the feature phasor is normalized, and set this sample set as the small sample data in the target domain.

3. The transient stability adaptive evaluation method for a power system facing operating scenario switching according to claim 1, characterized in that The said source model includes a feature extractor network and an evaluation classifier network.

4. The transient stability adaptive evaluation method for a power system facing operation scenario switching according to claim 3, characterized in that Obtaining the electrical characteristics related to known scenarios in the source domain and the electrical characteristics related to unknown scenarios in the target domain includes: Obtain the weights of the evaluation classifier network in the source model; perform orthogonal decoupling of the features of the weights of the evaluation classifier network to obtain the electrical characteristics related to known scenarios in the source domain and the electrical characteristics related to unknown scenarios in the target domain.

5. The transient stability adaptive evaluation method for a power system facing operation scenario switching according to claim 4, characterized in that The said source model is: g θ (x) = f θ (h θ (x)) where θ represents the network parameter, h θ (x) represents the feature extractor network of the source model, f θ () represents the evaluation classifier network of the source model, g θ (x) represents the source model for evaluating the transient stability state of the original scenario; The orthogonal decoupling of the features of the weights of the evaluation classifier network is as follows: W cls = UΣV T , F knw = span{v n | n＝1,...,C}, F unk = span{v n | n＝C+1,...,D} where, W cls ∈R C×D represents the weights of the evaluation classifier network f θ (), ∑∈R C×D represents a diagonal matrix, U∈R C×C and V∈R D×D both represent orthogonal unit matrices, F knw represents the known feature space of the source domain, F unk represents the knw positive orthogonal complement space of F , that is, the unknown feature space, span represents the vector space spanned by a series of feature vectors, represents the column vectors of the orthogonal unit matrix V, C represents the dimension of the known feature space of the source domain, and D represents the total dimension of the known feature space and the unknown feature space of the source domain.

6. The transient stability adaptive evaluation method for a power system facing operation scenario switching according to claim 5, characterized in that , The adaptive update of the overall transient stability evaluation model in Step 3 includes: Based on the electrical characteristics related to known scenarios in the source domain and the electrical characteristics related to unknown scenarios in the target domain obtained by orthogonal decoupling in Step 2, obtain the mapping of the feature space of known scenarios in the source domain and the mapping of the feature space of unknown scenarios in the target domain; then add two scoring auxiliary classifier networks to the source model, and calculate the divergence of the category scores of known scenarios in the source domain and unknown scenarios in the target domain, so as to adaptively update the overall transient stability evaluation model.

7. The transient stability adaptive evaluation method for a power system facing operation scenario switching according to claim 6, characterized in that Step 3 includes the following sub-steps: Step 3.1, using the target domain small sample dataset {x, y}, input the feature extractor network h θ (x), and set the feature map obtained from the target domain small sample data as Expressed as: where \(i\) represents the \(i\)-th small sample in the target domain, and respectively represent the projections on the known and unknown spaces of the source domain, \(l\) n \(\in\mathbb{R}\) represents the weight on the basis vector \(\mathbf{v}\) n ; Step 3.2, add two scoring auxiliary classifier networks to the source model, denoted as f1 and f2 respectively, for calculating the divergence of category scores, specifically expressed as: In the formula, represents the divergence of the class scores between the known scenario in the source domain and the unknown scenario in the target domain, represents the scoring function, and SD represents the inconsistency between two scoring functions f1 and f2 in the same domain. Step 3.3, execute the loop t = 1 until t = t max , iteratively train the feature extractor network and the scoring auxiliary classifier network of the transient stability assessment model to reduce the divergence of the class scores between the known scenarios in the source domain and the unknown scenarios in the target domain; the overall objective function of the transient stability assessment model is expressed as: where represents the classification loss function of the small sample data in the target domain; The process of updating the parameters of the feature extractor network and the scoring auxiliary classifier network is expressed as: where θ h represents the network parameters of the feature extractor, and θ a represents the network parameters of the scoring auxiliary classifier, represents the classification loss function of the target domain few-shot data, represents the divergence loss function of the class score inconsistency, λ represents the variable gradient reversal balance coefficient, and η represents the exponential decay learning rate; Judge whether the iterative loop of the transient stability evaluation model training is completed. If so, end the training iteration process and obtain a transient stability evaluation model that can be used for the power system after the operation scenario is switched.

8. The transient stability adaptive evaluation method for a power system facing operation scenario switching according to claim 7, characterized in that SD is expressed as: Wherein, K represents the number of sample categories, O represents the known or unknown scenarios in the source domain, and i and j respectively represent the row vector index and column vector index of M (ρ) The ρ (x) represents the ramp loss function, ρ represents the threshold parameter of the ramp loss function, and μ k (f(x), y) represents the absolute margin function, and y represents the set of category label spaces.

9. The transient stability adaptive evaluation method for power systems facing operation scenario switching according to claim 1, characterized in that The said Step 4 includes the following sub-steps: Step 4.1, use the phasor measurement unit PMU in the wide area measurement system to obtain the measurement data of the power system after the operation scenario is switched, and input it into the transient stability evaluation model obtained in Step 3 for online evaluation; Step 4.2, determine whether the current operation scenario of the power system has been switched. If the operation scenario is not switched, continue to execute Step 4. If the operation scenario is switched, loop to execute Steps 1 to 4.

10. A transient stability adaptive evaluation system for power systems facing operation scenario switching, characterized in that The said system includes: A small-sample data generation module for new scenarios of power systems, which is used to generate a small number of time-series samples with various electrical characteristics according to the network topology of the power system after the operation scenario is switched, with the number of samples being N T Then, match them with the corresponding transient stability states and normalize the electrical characteristics in the time-series samples, which are set as the small-sample data in the target domain; The transient stability assessment classifier feature decoupling module of the power system is used to obtain the source model for evaluating the transient stability state of the original operation scenario of the power system, and based on the source model, obtain the electrical features related to the known scenarios in the source domain and the electrical features related to the unknown scenarios in the target domain; The adaptive update module of the power system transient stability assessment model inputs the small sample data in the target domain into the source model, obtains the electrical feature space mapping related to the known scenarios in the source domain and the electrical feature space mapping related to the unknown scenarios in the target domain, and adaptively updates the overall transient stability assessment model; The online transient stability assessment module of the power system is used to obtain the measurement data of the power system after the operation scenario is switched online, input it into the overall transient stability assessment model after adaptive update, and obtain the transient stability assessment result applicable to the new scenario. If the operation scenario of the power system changes again, the above module is started again; The power system transient stability adaptive assessment system for operation scenario switching is used to execute the steps in the power system transient stability adaptive assessment method for operation scenario switching described in any one of claims 1-9.

Images