Operating scene change-oriented discriminant domain adaptation transient stability evaluation method

Abstract

The invention discloses an operation scene change-oriented discriminant domain adaptation transient stability evaluation method, and relates to the technical field of power systems, and the method comprises the steps: obtaining the operation data of a power system, and constructing a source domain task model; updating the source domain task model based on the change of the operation scene to obtain a target domain task model; and evaluating the power system based on the operation data and the target domain task model, and constructing a discriminant domain adaptive transient stability evaluation framework to realize evaluation of the transient stability of the power system. According to the invention, the problem that the accurate prediction of the stability of the power system is influenced because the operation state of the power grid is frequently changed and the power grid in a new operation scene cannot be accurately evaluated by a model after offline training for the change of the new operation scene of the power system can be solved; and the effect of accurately evaluating the stability of the current power system is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a discriminant domain adaptive transient stability assessment method oriented to operation scenario changes. Background Art

[0002] With the continuous access of large-scale renewable energy and power electronic equipment, the safety and stability mechanism of the power system is becoming increasingly complex. Ensuring the safe and stable operation of the power system has become a severe technical challenge. Accurate assessment of the transient stability of the system is particularly important. Therefore, it is urgent to establish a reliable and fast transient stability assessment TSA model to timely identify potential transient stability risks to ensure the safe and stable operation of the power grid under the new situation. At present, traditional transient stability assessment methods mainly include time domain simulation method and direct method. Among them, although the time domain simulation method can obtain more accurate assessment results, it takes a long time to calculate and it is difficult to meet the needs of real-time assessment. The direct method has a fast calculation speed, but the accuracy and reliability of the assessment results are poor, and its scope of application is limited. In recent years, with the development of artificial intelligence technology, transient stability assessment methods based on deep learning have gradually become a research hotspot.

[0003] Although deep learning models can show excellent performance in TSA, it depends on the establishment of an implicit assumption, that is, the training data of the model and the data in actual application are independent and identically distributed. However, this assumption is not always true, because the power system may encounter unpredictable dynamic disturbances during operation, such as unplanned outages caused by sudden failures of key equipment or sharp fluctuations in load demand, which in turn leads to the emergence of different data feature distributions, affecting model performance and even causing erroneous prediction results; in addition, when dealing with the problem of transient stability assessment of power systems, existing deep learning models often ignore the impact of changes in system operation scenarios on model performance. When the system operation scenario changes, the generalization ability of the model will decrease significantly, and it will not be able to accurately assess the transient stability in the new scenario; at the same time, the existing methods lack an effective domain adaptation mechanism and cannot make full use of historical data and empirical knowledge to improve the performance of the model in new scenarios. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to provide a discriminant domain adaptive transient stability assessment TSA-DDA framework for operating scenario changes, update the source domain task model according to the current power system operating scenario, and then use the updated target domain task model to evaluate the transient stability of the current power system, so as to achieve the effect of accurately evaluating the stability of the current power system.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides a discriminant domain adaptive transient stability assessment method for operating scenario changes, which includes acquiring operating data of the power system and constructing a source domain task model; updating the source domain task model based on changes in the operating scenario to obtain a target domain task model; evaluating the power system based on the operating data and the target domain task model, constructing a discriminant domain adaptive transient stability assessment framework, and realizing the assessment of transient stability of the power system.

[0008] As a preferred solution of the discriminant domain adaptive transient stability assessment method for operation scenario changes described in the present invention, the construction of the source domain task model includes the following steps: based on BPA and DSP computing and analysis software, a sample set is generated according to the data of the power system; feature extraction is performed on the sample set to obtain N voltage trajectory cluster features and divide them according to a specific ratio to obtain a training set and a test set; the training set is input into the source domain task model and model training is performed to obtain a trained source domain task model.

[0009] As an optimal solution of the discriminant domain adaptive transient stability assessment method for operating scenario changes described in the present invention, wherein: the source domain task model is updated based on the change of the operating scenario to obtain the target domain task model, which includes the following steps: judging the operating scenario of the power system, if the operating scenario of the power system changes, starting TSA-DDA migration, updating the source domain task model according to the operating scenario of the power system, obtaining the target domain task model and obtaining the current operating data of the power system, and executing the next step; evaluating the transient stability of the power system based on the current operating data of the power system and the target domain task model; if the operating scenario of the power system has not changed, the source domain task model is not updated, the current operating data of the power system is obtained, and the next step is executed; evaluating the transient stability of the power system based on the current operating data of the power system and the source domain task model.

[0010] As a preferred solution of the discriminant domain adaptive transient stability assessment method for operation scenario changes described in the present invention, wherein: updating the source domain task model according to the operation scenario of the power system includes the following steps: based on the multi-dimensional migration regularization term, comprehensive MMD and related alignment, constructing a multi-dimensional migration regularization term MC to measure the data distribution differences between different domains; based on the edge probability distribution alignment mechanism, measuring and continuously reducing the edge probability distribution of the source domain and the target domain, aligning the edge probability distribution of the two domains; based on the conditional probability distribution alignment mechanism, while realizing the alignment of the edge probability distribution between domains, using the Bayesian theorem to further align related sub-domains, and realizing the refined alignment of data distribution; based on the parameter-adjustable discriminant Softmax function, forcing the distance between samples of the same category to be compact, and samples of different categories to be far away from each other, helping the model learn more highly discriminative transient stability features.

[0011] As a preferred solution of the discriminant domain adaptive transient stability assessment method for operation scenario changes described in the present invention, wherein: the construction of the multidimensional migration regularization term MC refers to combining the MMD measurement method and the CORAL measurement method, starting from the two perspectives of mean and covariance, to obtain the multidimensional migration regularization term MC, and the specific formula of the multidimensional migration regularization term MC is as follows:

[0012] MC(X S ,X T )=MMD(X S ,X T )+CORAL(X S ,X T )

[0013] Among them, MC(·) is the multi-dimensional migration regularization term MC; X S is the sample feature matrix of the source domain; X T is the sample feature matrix of the target domain; MMD is the maximum mean difference; CORAL(·) is the correlation alignment.

[0014] As a preferred solution of the discriminant domain adaptive transient stability assessment method for operation scenario changes described in the present invention, the discriminant domain adaptive transient stability assessment framework includes an inter-domain dual distribution adaptive method, a discriminant Softmax function and an adaptive assessment process; the edge probability distribution alignment mechanism and the conditional probability distribution alignment mechanism are integrated to obtain the inter-domain dual distribution adaptive method, and the specific formula of the loss function of the inter-domain dual distribution adaptive method is as follows:

[0015]

[0016] Among them, L IDA is the loss function of the inter-domain dual distribution adaptive method; G(·) is the feature extraction function; E(·) is the mathematical expectation function; ||·||MC is the multi-dimensional migration regularization term MC; X S is the sample feature matrix of the source domain; X T is the sample feature matrix of the target domain; Y S is the sample feature matrix X of the source domain S The corresponding category label; Y T is the sample feature matrix X of the target domain T The corresponding category label; P(Y S =c) is the prior probability that the source domain belongs to category c; P(Y T =c) is the prior probability that the target domain belongs to category c; R is the number of categories.

[0017] As a preferred solution of the discriminant domain adaptive transient stability assessment method for operation scenario changes described in the present invention, the discriminant Softmax function is obtained by introducing adjustable parameters m and k and performing weight normalization. The specific formula of the discriminant Softmax function is as follows:

[0018]

[0019] Among them, D-Softmax is the discriminant Softmax function; is the result of normalizing the r-category column vector in the weight matrix; z is the eigenvector of the stable sample; m and k are additional parameters; is the result of normalizing the j-category column vector in the weight matrix; r is 0 or 1; the adaptive evaluation process includes source domain offline training, target domain migration update and online application.

[0020] On the second aspect, in order to further solve the safety problems existing in the power system, the embodiment of the present invention provides a discriminant domain adaptive transient stability assessment system for operating scenario changes, which includes: a source domain task module, used to obtain the operating data of the power system and construct a source domain task model; a model update module, used to update the source domain task model when the operating scenario changes to obtain a target domain task model; an assessment module, used to assess the transient stability of the power system based on the operating data of the power system, the source domain task model and the target domain task model.

[0021] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of a discriminant domain adaptive transient stability assessment method for operating scenario changes as described in the first aspect of the present invention.

[0022] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of a discriminant domain adaptive transient stability assessment method for operating scenario changes as described in the first aspect of the present invention.

[0023] Beneficial effects of the invention: The present invention proposes a discriminant domain adaptive transient stability assessment TSA-DDA framework for operating scenario changes, which can update the source domain task model according to the changes in the current power system operating scenario, and then use the updated target domain model to evaluate the transient stability of the current power system. It solves the problem in the prior art that due to the frequent changes in the operating status of the power grid, the model completed by offline training cannot accurately evaluate the power grid in the new operating scenario for the changes in the new operating scenario of the power system, which affects the accurate prediction of the stability of the power system, and achieves the effect of accurately evaluating the stability of the current power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0025] Figure 1 This is an overall flow chart of a discriminant domain adaptive transient stability assessment method for operation scenario changes in Example 1.

[0026] Figure 2 Schematic diagram of the discriminant domain adaptation framework in Example 1.

[0027] Figure 3 This is a comparison chart of the effects of Softmax and D-Softmax in Example 1.

[0028] Figure 4 This is a schematic diagram of the structure of the computer device in Example 3.

[0029] Figure 5 Schematic diagram of the adaptability results of the TSA-DDA framework in Example 4.

[0030] Figure 6 Schematic diagram of verification of sustainable learning capability of the TSA-DDA framework in Example 4.

[0031] Figure 7 Schematic diagram of the effects of different migration schemes in Example 4. DETAILED DESCRIPTION

[0032] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0033] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0034] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0035] Example 1

[0036] Reference Figure 1 to Figure 3 , which is the first embodiment of the present invention, and provides a discriminant domain adaptive transient stability assessment method for operation scenario changes.

[0037] The existing power system stability assessment methods mainly have the following problems: First, although deep learning models can show excellent performance in TSA, it depends on the establishment of an implicit assumption, that is, the training data of the model and the data in actual applications are independent and identically distributed. However, this assumption is not always true, because the power system may encounter unpredictable dynamic disturbances during operation, such as unplanned outages caused by sudden failures of key equipment or sharp fluctuations in load demand, which in turn leads to the emergence of different data feature distributions, affecting model performance and even causing erroneous prediction results; in addition, when dealing with the problem of transient stability assessment of power systems, existing deep learning models often ignore the impact of changes in system operation scenarios on model performance. When the system operation scenario changes, the generalization ability of the model will significantly decrease, and it will not be able to accurately assess the transient stability in the new scenario; at the same time, the existing methods lack an effective domain adaptation mechanism and cannot make full use of historical data and empirical knowledge to improve the performance of the model in new scenarios.

[0038] The present application provides an effective solution to the above-mentioned problems. Next, multiple embodiments will be combined to explain in detail how to implement the discriminant domain adaptive transient stability assessment method for operating scenario changes.

[0039] Figure 1 The overall flow chart of a discriminant domain adaptive transient stability assessment method for operation scenario changes is shown, including:

[0040] S1: Obtain the operation data of the power system and build the source domain task model.

[0041] Preferably, constructing the source domain task model includes the following steps: generating a sample set based on the data of the power system based on BPA and DSP computing and analysis software.

[0042] Feature extraction is performed on the sample set to obtain N voltage trajectory cluster features and divide them according to a specific ratio to obtain a training set and a test set.

[0043] The training set is input into the source domain task model and the model is trained to obtain the trained source domain task model.

[0044] It should be noted that the present invention uses BPA and DSP computing and analysis software for data simulation. For example, the total simulation time is set to 3 seconds, the simulation step is 0.04s, and the fault duration is from 1 cycle to 18 cycles, with a step of 1 cycle; under different fault durations, the fault is set to be located at 1% to 99% of each line, with a step of 1%, and the fault type is a three-phase short circuit fault; under the above conditions, 17,500 samples are randomly generated, of which 9,000 are stable samples and 8,500 are unstable samples.

[0045] Preferably, the present invention uses BPA and DSP computing and analysis software to generate a sample set. By setting refined simulation parameters, such as a total duration of 3 seconds, a step size of 0.04s, or a fault duration of 1-18 cycles, the dynamic characteristics of the power system under different fault conditions can be fully simulated. Compared with traditional methods, the fine-grained simulation used in the present invention can obtain richer transient characteristic information; by setting fault points at 1% to 99% of each line, combined with the three-phase short-circuit fault type, it can systematically cover the possible fault scenarios in the power grid, thereby improving the representativeness and completeness of the training data.

[0046] S2: Update the source domain task model based on changes in the operating scenario to obtain the target domain task model.

[0047] Preferably, the source domain task model is updated based on the change of the operating scenario to obtain the target domain task model, which includes the following steps: judging the operating scenario of the power system, if the operating scenario of the power system changes, starting the TSA-DDA migration, updating the source domain task model according to the operating scenario of the power system, obtaining the target domain task model and acquiring the current operating data of the power system, and executing the next step.

[0048] The transient stability of the power system is evaluated based on the current operating data of the power system and the target domain task model.

[0049] If the operation scenario of the power system has not changed, the source domain task model is not updated, the current operation data of the power system is obtained, and the next step is executed.

[0050] The transient stability of the power system is evaluated based on the current operating data of the power system and the source domain task model.

[0051] Specifically, updating the source domain task model according to the operation scenario of the power system includes the following steps:

[0052] Based on the multi-dimensional migration regularization term, MMD and related alignment are integrated to construct the multi-dimensional migration regularization term MC, which measures the data distribution differences between different domains from multiple perspectives and angles, thereby further improving the domain adaptability of the model.

[0053] Based on the marginal probability distribution alignment mechanism, the marginal probability distributions of the source domain and the target domain are measured and continuously reduced to align the marginal probability distributions of the two domains.

[0054] Based on the conditional probability distribution alignment mechanism, while realizing the alignment of edge probability distribution between domains, the Bayesian theorem is used to further align related sub-domains to achieve refined alignment of data distribution and reduce the difficulty of model migration.

[0055] To address the problem of insufficient attention paid to the discriminative learning of features, a discriminative Softmax function with adjustable parameters is used to force samples of the same category to be compact and samples of different categories to be far away from each other, helping the model learn more highly discriminative transient stable features, thereby improving the performance of the transfer framework in the TSA task.

[0056] It should be noted that the present invention combines the discriminant domain adaptation method when updating the source domain task model.

[0057] Furthermore, the multi-dimensional transfer regularization term means that in domain adaptation, it is crucial to select an appropriate transfer regularization term to measure the data distribution difference between the source domain and the target domain. Among them, the MMD measurement method and the CORAL measurement method are the two most widely used measurement methods.

[0058] Specifically, the MMD measurement method refers to measuring the similarity or difference between data distributions in different domains. The specific formula of the MMD measurement method is as follows:

[0059]

[0060] Where MMD is the maximum mean difference; n S is the number of source domain samples; is the i-th sample in the source domain; D S is the source domain; φ(·) is the mapping function from sample data to the reproducing kernel Hilbert space; n T is the number of samples in the target domain; is the i-th sample in the target domain; D T is the source domain task; H is the reproducing kernel Hilbert space; k(·) is the Gaussian kernel function, where the Gaussian kernel is selected as the kernel function of the MMD metric; r is the width of the Gaussian filter.

[0061] Furthermore, the specific formula for unbiased estimation of the MMD metric is as follows:

[0062]

[0063] Where MMD is the maximum mean difference; n S is the number of source domain samples; is the i-th sample in the source domain; is the jth sample in the source domain; D S is the source domain; φ(·) is the mapping function from sample data to the reproducing kernel Hilbert space; n T is the number of samples in the target domain; is the i-th sample in the target domain; is the jth sample in the target domain; D T is the target domain; H is the target domain task; k(·) is the Gaussian kernel function, where the Gaussian kernel is selected as the kernel function of the MMD metric.

[0064] It should be noted that the MMD metric measures the distribution difference between the source domain and the target domain data by comparing their distances in the feature space. By minimizing the distance, the model is guided to reduce the difference in feature distribution between domains.

[0065] Specifically, the CORAL metric method refers to matching the second-order covariance statistics of the two domains, corresponding to the MMD metric method aligning the mean of the source domain and the target domain in the reproducing kernel Hilbert space. The specific formula is as follows:

[0066]

[0067] Among them, CORAL(·) is Correlation Alignment for Deep Domain Adaptation; X S is the sample feature matrix of the source domain; X T is the sample feature matrix of the target domain; d is the dimension of the feature, which is used for normalization; O S is the covariance matrix of the source domain; O T is the covariance matrix of the target domain; is the Frobenius norm of the characteristic matrix; n S is the number of source domain samples; is the transpose of the source domain sample feature matrix; v is a column vector with all elements equal to 1; nT is the number of samples in the target domain; is the transpose of the target domain sample feature matrix.

[0068] It should be noted that the CORAL metric achieves domain adaptation by utilizing the similarity of the covariance matrix. By minimizing the difference in the covariance of source domain and target domain data features, the CORAL metric encourages the model to align and match features of different domains.

[0069] Furthermore, since the MMD measurement method focuses on the mean difference of different data distributions, and the CORAL measurement method focuses on the covariance similarity of data distributions, a new multi-dimensional migration regularization term MC is designed by combining the two statistical characteristics from the perspectives of mean and covariance. The specific formula of the multi-dimensional migration regularization term MC is as follows:

[0070] MC(X S ,X T )=MMD(X S ,X T )+CORAL(X S ,X T )

[0071] Among them, MC(·) is the multi-dimensional migration regularization term MC; X S is the sample feature matrix of the source domain; X T is the sample feature matrix of the target domain; MMD is the maximum mean difference; CORAL(·) is the correlation alignment (Correlation Alignment for Deep Domain Adaptation).

[0072] Preferably, the present invention innovatively combines the two metrics of MMD and CORAL to achieve accurate measurement of the data distribution difference between the source domain and the target domain in the reproducing kernel Hilbert space; selects the Gaussian kernel as the kernel function to provide nonlinear mapping capability; and effectively reduces the feature distribution difference between domains by minimizing the MMD distance; at the same time, by focusing on the mean difference and covariance similarity of the data distribution, the feature alignment is made more comprehensive and accurate, which can more accurately characterize the distribution difference between the source domain and the target domain, thereby improving the domain adaptability of the model.

[0073] S3: Evaluate the power system based on operating data and target domain task model to realize the transient stability evaluation of the power system.

[0074] Preferably, Figure 2The figure shows a schematic diagram of a discriminant domain adaptation framework. The discriminant domain adaptation transient stability assessment framework proposed in the present invention, namely, the TSA-DDA framework, includes an inter-domain dual distribution adaptation method, a discriminant Softmax function and an adaptive evaluation process, and is composed of two feature extractors with shared parameters. While extracting transient stability features, it can quantify and reduce the differences in marginal probability distributions and conditional probability distributions between different domains through a loss function, thereby aligning the feature distributions of the two domains and achieving refined domain adaptation. Among them, the classifier adopts a discriminant Softmax function, which enhances the recognition of discriminative features by maximizing the inter-class distance and minimizing the intra-class distance, and captures more highly discriminative transient stability information, thereby improving the evaluation performance of the model. The feature extractor is composed of a 1D-CNN with a flexible structure and a fully connected 1D convolutional neural network.

[0075] Furthermore, the inter-domain dual distribution adaptive method refers to the P S (X S )≠P T (X T ) and P S (Y S |X S )≠P T (Y T |X T ) case, where the sample feature matrix X of the source domain S The corresponding category label is Y S , the sample feature matrix X of the target domain T The corresponding category label is Y T , the loss function is constructed using the multi-dimensional migration regularization term MC to measure the differences between the source domain and the target domain in the marginal probability distribution and conditional probability distribution, and continuously forces the feature extractor to narrow these differences, achieve effective alignment of features between the two domains, improve the classification performance of the target domain, and reduce dependence on labeled samples in the target domain.

[0076] Specifically, the inter-domain dual distribution adaptive method includes the following steps: Since the marginal probability distribution directly affects the overall distribution characteristics of the sample, the marginal probability distribution alignment mechanism (Marginal Probability Distribution Alignment, MPDA) is introduced. The specific formula of the loss function of the marginal probability distribution alignment mechanism is as follows:

[0077]

[0078] Among them, L MPDA is the loss function of the edge probability distribution alignment mechanism; G(·) is the feature extraction function; E(·) is the mathematical expectation function; ||·|| MC is the multi-dimensional migration regularization term MC; X Sis the sample feature matrix of the source domain; X T is the sample feature matrix of the target domain; the loss function L MPDA It is used to measure the marginal probability distribution distance between domains, aiming to ensure the consistency of the overall data distribution between the source domain and the target domain, and maintain the similarity of the data even without considering the category label, thereby improving the generalization and robustness of the model.

[0079] It should be noted that the emphasis on marginal distribution alignment often leads to the neglect of conditional probability distribution differences, which may be significant between the source domain and the target domain. The inconsistency of the conditional probability distribution may cause confusion between samples of different categories in the target domain, thus posing a challenge to the accuracy of the classification task. Although some studies have begun to try to adjust the conditional probability distribution through the relationship between categories in order to achieve better domain adaptability, these studies usually adopt a simplified approximation strategy, that is, treating the category conditional probability distribution as a substitute for the conditional probability distribution. The limitation of this strategy is that there is not always a one-to-one correspondence between the category conditional probability distribution and the conditional probability distribution. In the framework of domain adaptation, this approximation may make it impossible to fully eliminate the difference in conditional probability distribution between the source domain and the target domain, thereby affecting the overall performance of the domain adaptation algorithm. Therefore, the present invention proposes to use Bayes' theorem to transform the conditional probability distribution into a quasi-conditional probability distribution.

[0080] Specifically, the specific formula for converting the conditional probability distribution into the form of quasi-conditional probability distribution using Bayes' theorem is as follows:

[0081]

[0082]

[0083] Among them, P(X S |Y S =c) is the category conditional probability of category c in the source domain, that is, the confidence of the deep learning model in classifying the sample; P(X T |Y T =c) is the category conditional probability of category c in the target domain, that is, the confidence of the deep learning model in classifying the sample; X S is the sample feature matrix of the source domain; X T is the sample feature matrix of the target domain; Y S is the sample feature matrix X of the source domain S The corresponding category label; Y T is the sample feature matrix X of the target domain T The corresponding category label; P(Y S =c) is the prior probability that the source domain belongs to category c; P(Y T=c) is the prior probability that the target domain belongs to category c; R is the number of categories; n S c is the number of samples of category c in the source domain; is the number of samples of category c in the target domain; n S is the number of source domain samples; n T is the number of samples in the target domain.

[0084] Furthermore, based on the Bayesian theorem, the conditional probability distribution is converted into a quasi-conditional probability distribution form. The present invention designs a new conditional probability distribution alignment mechanism (Conditional Probability Distribution Alignment, CPDA). The specific formula of the loss function of the conditional probability distribution alignment mechanism is as follows:

[0085]

[0086] Among them, L CPDA is the loss function of the conditional probability distribution alignment mechanism, which is used to measure the conditional probability distribution distance between domains, so that the model can focus more on adjusting the data distribution under a given category in the source domain and the target domain, thereby reducing confusion between categories, ensuring the consistency of data representation under a specific task, and improving the classification performance of the model on the target domain; in addition, in the loss function L of the conditional probability distribution alignment mechanism CPDA In the proposed method, a key strategy is to use the true conditional probability distribution to align the data separately, rather than simply using approximate substitution. This refined alignment strategy helps the model to more accurately capture the similarities and differences between the source and target domains, thereby effectively improving the ability of domain confusion.

[0087] Furthermore, the edge probability distribution alignment mechanism and the conditional probability distribution alignment mechanism are integrated to obtain the inter-domain dual distribution adaptive method. The specific formula of the loss function of the inter-domain dual distribution adaptive method is as follows:

[0088]

[0089] Among them, L IDA is the loss function of the inter-domain dual distribution adaptive method, where the inter-domain dual distribution adaptive method can comprehensively consider the differences between the marginal probability distribution and the conditional probability distribution of the two domains, take a suitable migration regularization term MC to measure the difference, continuously reduce the difference and obtain a highly generalized feature representation, thereby promoting the cross-domain migration and expression of knowledge; G(·) is the feature extraction function; E(·) is the mathematical expectation function; ||·|| MC is the multi-dimensional migration regularization term MC; X S is the sample feature matrix of the source domain; X T is the sample feature matrix of the target domain; YS is the sample feature matrix X of the source domain S The corresponding category label; Y T is the sample feature matrix X of the target domain T The corresponding category label; P(Y S =c) is the prior probability that the source domain belongs to category c; P(Y T =c) is the prior probability that the target domain belongs to category c; R is the number of categories.

[0090] Preferably, the present invention converts the conditional probability distribution into the class conditional probability distribution by using Bayes' theorem, thus overcoming the limitation of the traditional method that only considers the class conditional probability, achieving the refined alignment of the marginal probability distribution and the conditional probability distribution, and significantly improving the classification performance of the model in the target domain.

[0091] It should be noted that the Softmax function plays a key role in the multi-category classification task of neural networks. By converting the network output into a probability distribution, it provides an intuitive explanation for the classification results. However, experimental observations have revealed the performance limitations of the Softmax function under specific conditions, especially in the sample feature space. If there is inconsistency in the distribution of samples within a class or fuzzy boundaries between samples between classes, the performance of the Softmax function may be limited; specifically, when the distribution of samples of the same category in the feature space is relatively scattered, the Softmax function may be difficult to accurately distinguish due to the lack of sufficient intra-class compactness, which directly affects the classification accuracy of the model; in addition, if samples of different categories are too close in the feature space, that is, the inter-class distinction is not high, the Softmax function also finds it difficult to distinguish these categories, further weakening the classification effect. Therefore, the present invention adopts a novel discriminative Softmax function (Discriminative Softmax, D-Softmax) to learn transient stability features with more high discrimination to improve the migration effect.

[0092] Furthermore, constructing the discriminant Softmax function includes the following steps: Since the TSA problem is a binary classification problem, assuming that z is the feature vector of the stable sample, if the prediction of the feature vector z of the stable sample is correct, the Softmax function satisfies the following formula:

[0093]

[0094] Among them, z is the eigenvector of the stable sample; w0 is the column vector corresponding to the stable category in the weight matrix; w1 is the column vector corresponding to the unstable category in the weight matrix.

[0095] Due to the characteristics of the Softmax function, it is difficult to ensure a strict decision boundary, and thus it is impossible to maximize the distance between classes and minimize the distance within a class. In order to obtain a stricter decision boundary and achieve correct stability evaluation, additional parameters m and k are introduced to obtain an improved decision boundary and D-Softmax function. The specific formula of the improved decision boundary is as follows:

[0096]

[0097] Among them, z is the feature vector of stable samples; w0 is the column vector corresponding to the stable category in the weight matrix; w1 is the column vector corresponding to the unstable category in the weight matrix; m and k are additional parameters. When m ≥ 1 and k > 0, the interval between categories can be increased, so that the classification task has a stricter decision boundary.

[0098] The specific formula of the D-Softmax function is as follows:

[0099]

[0100] Among them, D-Softmax is the discriminant Softmax function; is the transpose of the r category column vector in the weight matrix; is the transpose of the j-category column vector in the weight matrix; z is the eigenvector of the stable sample; m and k are additional parameters; r is 0 or 1.

[0101] Furthermore, the D-Softmax function is further described from a geometric perspective to present a clearer D-Softmax function, where the inner product concept is used to and Converted into vector angle form, the specific formula is as follows:

[0102]

[0103] Among them, θ0 is the angle between the column vector w0 corresponding to the stable category in the weight matrix and the eigenvector z of the stable sample; θ1 is the angle between the column vector w1 corresponding to the unstable category in the weight matrix and the eigenvector z of the stable sample.

[0104] It should be noted that, assuming that z is a stable sample feature, the larger the value of ‖w0||||z||cos(θ0), the better the classification effect. When ||w0||=||w1||, the classification situation depends entirely on the difference between the angles, such as Figure 3 As shown in Figure 2, only when ‖w0||||z||cos(θ0)>||w1||||z||cos(θ1)(θ0<θ1) is satisfied, can the classification be correct and there is only one decision boundary. For the D-Softmax function, when For example, the following formula is satisfied:

[0105]

[0106] Among them, θ0 is the angle between the column vector w0 corresponding to the stable category in the weight matrix and the eigenvector z of the stable sample; θ1 is the angle between the column vector w1 corresponding to the unstable category in the weight matrix and the eigenvector z of the stable sample; is the equivalent edge angle,

[0107] Obviously, the D-Softmax function satisfies Moreover, stable samples and unstable samples have different decision boundaries. This design makes the classification more rigorous, making samples of the same category closer to each other, while samples of different categories are relatively distant, thereby providing the model with more discriminative feature learning.

[0108] Furthermore, the weights of the D-Softmax function are normalized to further promote the learning of more discriminative features. The specific formula for normalizing the weight vector using the L2 norm is as follows:

[0109]

[0110] in, is the result of normalizing the r category column vector in the weight matrix; w r is the r category column vector in the weight matrix.

[0111] Specifically, by introducing additional adjustable parameters m and k and performing weight normalization, the final D-Softmax function is obtained, thereby maximizing the inter-class distance and minimizing the intra-class distance, thereby enhancing the model's learning of discriminative features and improving the performance of the transfer framework in the TSA task. The specific formula of the D-Softmax function is as follows:

[0112]

[0113] Among them, D-Softmax is the discriminant Softmax function; is the result of normalizing the r-category column vector in the weight matrix; z is the eigenvector of the stable sample; m and k are additional parameters; is the result of normalizing the j-category column vector in the weight matrix; r is 0 or 1.

[0114] Preferably, the present invention achieves stricter decision boundary division by introducing adjustable parameters m and k and combining weight normalization technology, which can effectively increase the distance between classes, reduce the distance within classes, and enhance the model's ability to identify transient stability characteristics.

[0115] Furthermore, the adaptive evaluation process includes offline training in the source domain, migration update in the target domain, and online application.

[0116] Specifically, source domain offline training refers to using time domain simulation technology to generate source domain data required for offline model training, extracting 29 voltage trajectory cluster features from the source domain data, and dividing them into training sets and test sets according to specific proportions for training source domain task models.

[0117] Specifically, the target domain migration update means starting the TSA-DDA migration and updating the source domain task model when a change in the power grid operation scenario is detected.

[0118] Specifically, online application refers to obtaining real-time operation data of the power system and performing data processing, and realizing fast and accurate transient stability assessment TSA by using the updated model.

[0119] In summary, the present invention proposes a discriminant domain adaptive transient stability assessment TSA-DDA framework for operating scenario changes. It can update the source domain task model according to the changes in the current power system operating scenario, and then use the updated target domain model to evaluate the transient stability of the current power system. It solves the problem in the prior art that due to the frequent changes in the operating state of the power grid, the model completed by offline training cannot accurately evaluate the power grid in the new operating scenario for the changes in the new operating scenario of the power system, which affects the accurate prediction of the stability of the power system, and achieves the effect of accurately evaluating the stability of the current power system.

[0120] Embodiment 2 is an embodiment of the present invention, which provides a discriminant domain adaptive transient stability assessment system for operation scenario changes, including: a source domain task module, used to obtain the operation data of the power system and construct a source domain task model; a model update module, used to update the source domain task model when the operation scenario changes to obtain a target domain task model; an assessment module, used to assess the transient stability of the power system based on the operation data of the power system, the source domain task model and the target domain task model.

[0121] Embodiment 3 is an embodiment of the present invention, which is different from the previous embodiment in that:

[0122] like Figure 4As shown, if the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0123] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0124] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0125] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0126] Example 4 is an embodiment of the present invention, which provides a discriminant domain adaptive transient stability assessment method for operation scenario changes. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0127] This example uses the IEEE 39-bus system as an example for analysis. The New England 10-machine 39-bus power system includes 39 buses, 10 generators, and 46 lines. The total simulation time is set to 3 seconds, the simulation step is 0.04s, and the fault duration ranges from 1 cycle to 18 cycles with a step of 1 cycle. Under different fault durations, the fault is set at 1% to 99% of each line with a step of 1%, and the fault type is a three-phase short-circuit fault. Under the above conditions, 17,500 samples are randomly generated, of which 9,000 are stable samples and 8,500 are unstable samples.

[0128] First, in order to verify the performance of 1D-CNN for TSA tasks, this example selects four machine learning models for comparison, including SVM, DT, RF and KNN; the kernel function of SVM adopts radial basis kernel function, DT adopts C5.0 algorithm, the number of base classifiers of RF is 20, the number of nearest neighbors in KNN is 10, the learning rate of 1D-CNN is 0.001, and the cross entropy loss function and Adam optimization algorithm are adopted; the performance test results of the above five models are shown in Table 1. It can be seen from the table that SVM, DT, RF and KNN, as shallow ML models, have limited feature expression capabilities and are difficult to fully mine deep features. Therefore, they perform worse than 1D-CNN in various evaluation indicators. The 1D-CNN model, based on its powerful nonlinear feature aggregation, makes each indicator reach more than 98%.

[0129] Table 1 Average performance evaluation results of each indicator of five models

[0130] Model <![CDATA[A cc / %]]> <![CDATA[P rec / %]]> <![CDATA[R ec / %]]> <![CDATA[f1 / %]]> 1D-CNN 98.25 98.55 98.07 98.31 SVM 92.74 92.93 92.00 92.46 DT 96.93 96.99 96.67 96.83 RF 96.94 98.62 95.24 96.77 KNN 94.84 99.26 90.04 94.41

[0131] Secondly, in order to simulate the situation where the grid operation scenario changes significantly during online application, it is necessary to assume that a node in the system and its transmission line are out of operation. By removing the No. 4 load node and the three related transmission lines, the active output of the No. 32 generator is adjusted from 300MW to 200MW, and the changes in the power system flow and operation scenario are adjusted. Under this condition, a total of 1500 target domain samples are simulated and generated, and the training set and test set are divided into a ratio of 1:4. The test results of the source domain task model in the new operation scenario are shown in Table 2.

[0132] Table 2 Test results of source domain task model in new scenario

[0133] Sample Set <![CDATA[A cc / %]]> <![CDATA[P rec / %]]> <![CDATA[R ec / %]]> <![CDATA[f1 / %]]> Training set 89.62 90.31 89.66 89.98 Test Set 90.25 91.01 90.14 90.57

[0134] As can be seen from Table 2, by changing the operation mode of the power grid, the prediction accuracy of the source domain task model has dropped significantly and can no longer meet the actual needs. Therefore, the TSA-DDA framework is needed to improve the performance of the model in the new operation scenario; 50, 100, 150 and 200 samples are randomly selected from the target domain training set, and they are merged with the 3500 randomly selected source domain samples to construct mixed sample sets S1, S2, S3 and S4; the parameters of the TSA-DDA framework are set as follows: the sample batch size is 130 (the number of source domain samples and target domain samples are set to 100 and 30 respectively), the number of iterations of model training is 200, the learning rate is 0.001, and the parameters of the D-Softmax function are m=1 and k=0.32; considering the training of different mixed sample sets, the adaptability results of the TSA-DDA framework are as follows Figure 5 As shown in the figure, it can be seen that when using the mixed sample S1 for training, due to the limited number of target samples it contains, the model fails to fully learn and understand the distribution of the data and its potential patterns, so the evaluation effect is poor, and the accuracy rate is only 95.13%. However, with the increase of the number of target domain samples in the mixed sample set, the evaluation accuracy gradually increases; when the number of target domain samples in the mixed sample set increases to 200, the prediction accuracy of the model increases from 89.62% to 98.18%, indicating that the TSA-DDA framework can effectively restore the evaluation performance of the model.

[0135] In addition, in order to illustrate the effectiveness of each module in the TSA-DDA framework, four other incomplete TSA-DDA frameworks are set for ablation experiments, in which ablation model 1 does not contain the CPDA mechanism, only aligns the marginal probability distributions of the source domain and the target domain, and uses the original Softmax function for activation. Ablation model 2 does not contain the MPDA mechanism, only aligns the conditional probability distributions of the source domain and the target domain, and uses the original Softmax function for activation. Ablation model 3 does not contain the CPDA mechanism, only aligns the marginal probability distributions of the source domain and the target domain, and uses the D-Softmax function for activation. Ablation model 4 does not contain the MPDA mechanism, only aligns the conditional probability distributions of the source domain and the target domain, and uses the D-Softmax function for activation. The test results are shown in Tables 3 and 4. Figure 6 shown.

[0136] Table 3 Average results of ablation experiments on various indicators of four models

[0137] Model <![CDATA[A cc / %]]> <![CDATA[P rec / %]]> <![CDATA[R ec / %]]> <![CDATA[f1 / %]]> Ablation model 1 97.37 99.25 95.67 97.42 Ablation Model 2 97.13 99.24 95.19 97.18 Ablation Model 3 97.62 99.74 95.69 97.67 Ablation Model 4 97.75 99.27 96.39 97.80

[0138] From Table 3 and Figure 6 It can be seen that when ablation model 1 and ablation model 2 only align the marginal probability distribution or the conditional probability distribution, the model performance decreases, and the prediction accuracy does not reach 97.50%; on the basis of aligning the feature distribution, ablation model 3 and ablation model 4 enhance the discriminative feature learning of the model by introducing the D-Softmax function, and the prediction accuracy is improved by 0.25% and 0.62% respectively, indicating that the D-Softmax function enhances the generalization performance of the model to a certain extent, making it better adapt to the target domain tasks.

[0139] Finally, in order to further verify the effectiveness of the TSA-DDA framework, the migration effects of the framework are compared with those of some mainstream transfer learning schemes, and two schemes without transfer learning are included. Scheme 1 does not use transfer learning and only retrains with the target domain samples in the mixed sample set S5. Scheme 2 does not use transfer learning and uses the mixed sample set S5 as the training set for retraining. Scheme 3 uses model migration to migrate the parameters of the source domain model to the target domain, and fine-tunes the entire network using the target domain samples in the mixed sample set S5. Scheme 4 uses model migration to migrate the parameters of the source domain model to the target domain, freezes the feature extractor layer parameters, and initializes the feature extractor layer. The classifier layer parameters are only locally fine-tuned using the target domain samples in the mixed sample set S5. Scheme 5 combines sample migration and model migration to migrate the parameters of the source domain model to the target domain. The mixed sample set S5 is used as the training set to fine-tune the entire network. Scheme 6 adopts an unsupervised domain adversarial migration method. By adding a domain classifier and a gradient reversal layer, the feature extractor is forced to find it difficult to distinguish the source of the samples. The mixed sample set S5 is used as the training set, and the target domain samples do not participate in the training of the classifier. Scheme 7 adopts a supervised domain adversarial migration method. The model structure and training set are consistent with Scheme 6, but the target domain samples participate in the training of the classifier. Scheme 8 is the TSA-DDA framework.

[0140] like Figure 7The figure shows the experimental results of the above 8 schemes. In the new running scenario, the performance of Scheme 1 is significantly worse. After training with 200 target domain samples, its accuracy failed to reach 95%; Scheme 2 makes full use of a large number of source domain samples while utilizing a small number of target domains, which improves the performance of the model under limited new scene samples. Compared with Scheme 1, this strategy of utilizing source domain samples significantly improves the prediction accuracy of the model; Compared with the first two schemes, Schemes 3, 4 and 5 share the structure and parameters of the source domain task model to provide a learning starting point with good basic features for building a new target domain model. This model transfer-based method enables the model prediction accuracy to reach 87.25% when initially applied to a new scenario. It is worth noting that Scheme 5 combines model transfer and sample transfer techniques to make full use of source domain information, effectively improving the evaluation performance of the model in the stage of scarce labeled samples in new scenarios. After 00 rounds of training, the model prediction accuracy reached 96.98%; Scheme 6 adopts an unsupervised domain adversarial migration method, which can only rely on source domain data and domain adversarial loss for domain adaptation. The lack of participation of labeled samples in the target domain makes it difficult for the model to effectively learn the feature representation of the target domain, and therefore performs poorly in domain adaptation; Scheme 7 introduces a small number of labeled target domain samples to achieve better prediction results than Scheme 6. However, the alignment of the conditional probability distribution within the domain is ignored during the processing, and the accuracy reaches 96.38%, but subsequent training cannot further improve the model performance; Scheme 8, i.e. the method proposed in the present invention, significantly improves the prediction accuracy of the model to 98.18% by finely aligning the marginal probability distribution and conditional probability distribution between different domains, and introducing the D-Softmax function to promote the model to learn more transient stable features with high discrimination. The experimental results confirm that its performance is better than the aforementioned schemes.

[0141] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A discriminant domain adaptive transient stability assessment method for operation scenario changes, characterized by: include: Obtain the operation data of the power system and build the source domain task model; The source domain task model is updated based on the change of the operation scenario to obtain a target domain task model; The power system is evaluated based on the operating data and the target domain task model, a discriminant domain adaptive transient stability evaluation framework is constructed, and the transient stability evaluation of the power system is realized.

2. The discriminant domain adaptive transient stability assessment method for operation scenario changes according to claim 1 is characterized by: The construction of the source domain task model comprises the following steps: Generate sample sets based on power system data based on BPA and DSP computing and analysis software; Extracting features from the sample set to obtain N voltage trajectory cluster features and dividing the features according to a specific ratio to obtain a training set and a test set; The training set is input into the source domain task model and model training is performed to obtain a trained source domain task model.

3. The discriminant domain adaptive transient stability assessment method for operation scenario changes according to claim 2 is characterized by: Updating the source domain task model based on the change of the operation scenario to obtain the target domain task model includes the following steps: The operation scenario of the power system is judged. If the operation scenario of the power system changes, TSA-DDA migration is started, the source domain task model is updated according to the operation scenario of the power system, the target domain task model is obtained, the current operation data of the power system is obtained, and the next step is executed; Evaluating the transient stability of the power system based on the current operating data of the power system and the target domain task model; If the operation scenario of the power system has not changed, the source domain task model is not updated, the current operation data of the power system is obtained, and the next step is executed; The transient stability of the power system is evaluated based on the current operating data of the power system and the source domain task model.

4. The discriminant domain adaptive transient stability assessment method for operation scenario changes according to claim 3 is characterized by: Updating the source domain task model according to the operation scenario of the power system includes the following steps: Based on the multi-dimensional migration regularization term, MMD and related alignment are integrated to construct the multi-dimensional migration regularization term MC to measure the data distribution differences between different domains; Based on the marginal probability distribution alignment mechanism, the marginal probability distributions of the source domain and the target domain are measured and continuously reduced to align the marginal probability distributions of the two domains. Based on the conditional probability distribution alignment mechanism, while realizing the alignment of the edge probability distribution between domains, the Bayesian theorem is used to further align the related sub-domains to achieve the refined alignment of data distribution; The discriminant Softmax function with adjustable parameters forces samples of the same category to be close to each other and samples of different categories to be far away from each other, helping the model learn more highly discriminative transient stable features.

5. The discriminant domain adaptive transient stability assessment method for operation scenario changes according to claim 4 is characterized in that: The construction of the multi-dimensional migration regularization term MC refers to combining the MMD measurement method and the CORAL measurement method, starting from the two perspectives of mean and covariance, to obtain the multi-dimensional migration regularization term MC. The specific formula of the multi-dimensional migration regularization term MC is as follows: MC(X S ,X T )=MMD(X S ,X T )+CORAL(X S ,X T ) Among them, MC(·) is the multi-dimensional migration regularization term MC; X S is the sample feature matrix of the source domain; X T is the sample feature matrix of the target domain; MMD is the maximum mean difference; CORAL(·) is the correlation alignment.

6. The discriminant domain adaptive transient stability assessment method for operation scenario changes according to claim 5, characterized in that: The discriminant domain adaptive transient stability assessment framework includes an inter-domain dual distribution adaptive method, a discriminant Softmax function and an adaptive assessment process; By combining the edge probability distribution alignment mechanism and the conditional probability distribution alignment mechanism, we obtain an inter-domain dual distribution adaptive method. The specific formula of the loss function of the inter-domain dual distribution adaptive method is as follows: Among them, L IDA is the loss function of the inter-domain dual distribution adaptive method; G(·) is the feature extraction function; E(·) is the mathematical expectation function; ||·|| MC is the multi-dimensional migration regularization term MC; X S is the sample feature matrix of the source domain; X T is the sample feature matrix of the target domain; Y S is the sample feature matrix X of the source domain S The corresponding category label; Y T is the sample feature matrix X of the target domain T The corresponding category label; P(Y S =c) is the prior probability that the source domain belongs to category c; P(Y T =c) is the prior probability that the target domain belongs to category c; R is the number of categories.

7. The discriminant domain adaptive transient stability assessment method for operation scenario changes according to claim 6, characterized in that: By introducing adjustable parameters m and k and performing weight normalization, a discriminant Softmax function is obtained. The specific formula of the discriminant Softmax function is as follows: Among them, D-Softmax is the discriminant Softmax function; is the result of normalizing the r-category column vector in the weight matrix; z is the eigenvector of the stable sample; m and k are additional parameters; is the result of normalizing the j-category column vector in the weight matrix; r is 0 or 1; The adaptive evaluation process includes offline training in the source domain, migration update in the target domain and online application.

8. A discriminant domain adaptive transient stability assessment system for operation scenario changes, based on a discriminant domain adaptive transient stability assessment method for operation scenario changes according to any one of claims 1 to 7, characterized in that: include, The source domain task module is used to obtain the operation data of the power system and build the source domain task model; The model updating module is used to update the source domain task model when the operation scenario changes to obtain the target domain task model; The evaluation module is used to evaluate the transient stability of the power system based on the operation data of the power system, the source domain task model and the target domain task model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the processor implements the steps of a discriminant domain adaptive transient stability assessment method for operation scenario changes according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a discriminant domain adaptive transient stability assessment method for operation scenario changes according to any one of claims 1 to 7 are implemented.

Images