Variance-guided domain-adaptive power system transient stability adaptive evaluation method and system

Abstract

The invention discloses a variance-guided domain-adaptive power system transient stability adaptive evaluation method and system, and the method comprises the steps: obtaining voltage track cluster features of a power system in different operation scenes to construct source domain data, training the source domain data through a neural network, and constructing a source domain task model; obtaining response data when the operation scene of the power system changes, constructing target domain data, and merging the source domain data and the target domain data to construct a mixed sample set; updating the source domain task model by using an improved transient stability evaluation method and the mixed sample set, and constructing a target domain task model; and inputting the real-time operation data of the power system into the target domain task model, and evaluating the transient stability of the power system. According to the method, the source domain task model is updated in real time to adapt to the dynamic change of a power grid operation scene, the problem of inaccurate evaluation caused by variable working conditions of a traditional deep learning model in a new energy power system is solved, and the precision and robustness of transient stability evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system evaluation, and in particular to a variance-guided domain-adaptive power system transient stability adaptive evaluation method and system. Background Art

[0002] With the transformation of the global energy structure, new energy sources such as wind power and photovoltaics have been connected to the power grid on a large scale, which has greatly promoted the intelligent and modern development of the power system. However, the intermittent and unpredictable nature of new energy sources has brought unprecedented challenges to the stable operation of the power grid. As an important indicator to measure the ability of the power grid to cope with sudden failures, the rapid and accurate evaluation of transient stability is crucial to ensure the safety of the power grid. In this context, the transient stability assessment (TSA) method based on deep learning came into being, aiming to realize the intelligent evaluation of transient stability of the power system by mining the complex relationship between electrical measurement data and transient stability.

[0003] Although the TSA method based on deep learning has powerful data processing and pattern recognition capabilities in theory, it still faces many challenges in practical applications. First, this method is highly dependent on the quality and diversity of training data, while the operating conditions of new energy power systems are complex and changeable. Traditional training sets often cannot fully cover all possible operating scenarios, resulting in insufficient generalization capabilities of the model when facing new conditions. Secondly, existing methods mostly use traditional distribution distance metrics such as Gaussian kernel functions to evaluate the differences between the source domain and the target domain, but when processing transient stability data with long-tail effects, these metrics often cannot accurately capture the extreme volatility and discreteness of the data, thereby affecting the accuracy and reliability of the evaluation results. Therefore, how to improve the generalization ability and adaptability of the TSA model to better cope with the complexity and uncertainty of the new energy power system has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a variance-guided domain-adaptive power system transient stability adaptive assessment method to solve the problem that the traditional deep learning-based transient stability assessment model is difficult to adapt to and accurately assess the transient stability of the power system due to the frequent changes in the power grid operation state under the widespread deployment of new energy and power electronic equipment.

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

[0007] In a first aspect, the present invention provides a variance-guided domain-adaptive power system transient stability adaptive assessment method, comprising:

[0008] Obtain transient stability data of power systems under different operating scenarios;

[0009] Extracting voltage trajectory cluster features from the transient stability data to construct source domain data, training the source domain data using a neural network, and building a source domain task model;

[0010] Acquire transient stability data when the power system operation scenario changes, construct target domain data, merge the source domain data with the target domain data, and construct a mixed sample set;

[0011] The source domain task model is updated by using the improved transient stability assessment method and the mixed sample set to construct a target domain task model;

[0012] The real-time operation data of the power system is input into the target domain task model to evaluate the transient stability of the power system.

[0013] As a preferred solution of the variance-guided domain-adaptive power system transient stability adaptive assessment method described in the present invention, the improved transient stability assessment method includes:

[0014] The Gaussian kernel function in the transient stability assessment method is replaced by an enhanced kernel function, and the enhanced kernel function is used to improve the tolerance of the source domain task model to extreme values;

[0015] Based on the enhanced kernel function, a variance-guided domain distribution alignment mechanism is introduced to handle the distribution alignment problem between the source domain data and the target domain data.

[0016] As a preferred solution of the variance-guided domain-adaptive power system transient stability adaptive assessment method described in the present invention, wherein: the variance-guided domain distribution alignment mechanism includes:

[0017] Based on the enhanced kernel function, the variance-guided domain distribution alignment mechanism constructs a set of bases that can express the distribution variance characteristics and reconstructs the reproducing kernel Hilbert space;

[0018] A variance-guided domain distribution difference measurement method is used in the reproducing kernel Hilbert space to quantify and reduce the difference between the source domain data and the target domain data distribution.

[0019] As a preferred solution of the variance-guided domain-adaptive power system transient stability adaptive assessment method described in the present invention, the enhanced kernel function includes:

[0020] The enhanced kernel function is used to process the extreme value problem caused by the long-tail distribution characteristics and mutation of the mixed sample set. The enhanced kernel function is expressed as:

[0021]

[0022] Where k′(x, y) is the enhanced kernel function, x is the sample data in the source domain data, y is the sample data in the target domain data, d is the degree of freedom, and G(·) is the Gamma function.

[0023] As a preferred solution of the variance-guided domain-adaptive power system transient stability adaptive evaluation method described in the present invention, wherein: the variance-guided domain distribution difference measurement method includes:

[0024] The variance-guided domain distribution difference measurement method uses biased estimation and unbiased estimation to measure the difference;

[0025] When the target domain data volume reaches a preset value, biased estimation is adopted; when the target domain data volume does not reach the preset value, unbiased estimation is adopted.

[0026] As a preferred solution of the variance-guided domain-adaptive power system transient stability adaptive assessment method described in the present invention, it also includes:

[0027] Obtain error bounds for biased estimates based on McDiarmid's theorem and the symmetry theorem;

[0028] Obtain error bounds for unbiased estimates using large deviation bounds based on the U statistic.

[0029] As a preferred solution of the variance-guided domain-adaptive power system transient stability adaptive assessment method of the present invention, wherein: reducing the difference between the distribution of the source domain data and the target domain data includes:

[0030] Based on the variance-guided domain distribution difference measurement method, a loss function of the variance-guided domain distribution alignment mechanism is established to reduce the difference in distribution between the source domain data and the target domain data. The loss function of the variance-guided domain distribution alignment mechanism is expressed as:

[0031]

[0032] Among them, τ(x,·) is the basis function defined by the enhanced kernel function, x is the sample data in the source domain data, y is the sample data in the target domain data, E x～p (·) is the mathematical expectation function of data distribution p, E y～q(·) is the mathematical expectation function of the data distribution q, h is the nonlinear continuous function in H, and H' is the reconstructed Hilbert space.

[0033] In a second aspect, the present invention provides a system for adaptive assessment of transient stability of a power system guided by a variance-based domain adaptation method, comprising:

[0034] A data acquisition module is used to obtain transient stability data of the power system under different operation scenarios;

[0035] A source domain task model construction module is used to extract voltage trajectory cluster features in the transient stability data to construct source domain data, train the source domain data using a neural network, and construct a source domain task model;

[0036] A hybrid sample acquisition module is used to acquire transient stability data when the power system operation scenario changes, construct target domain data, merge the source domain data with the target domain data, and construct a hybrid sample set;

[0037] A target domain task model construction module is used to update the source domain task model using the improved transient stability assessment method and the mixed sample set to construct a target domain task model;

[0038] The evaluation module is used to input the real-time operation data of the power system into the target domain task model to evaluate the transient stability of the power system.

[0039] In a third aspect, the present invention provides a computing device, comprising:

[0040] Memory and processor;

[0041] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the variance-guided domain adaptive power system transient stability adaptive assessment method are implemented.

[0042] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the variance-guided domain-adaptive power system transient stability adaptive assessment method.

[0043] Compared with the prior art, the present invention has the following beneficial effects: the present invention proposes an adaptive assessment framework for transient stability of power systems based on variance-guided domain adaptation, which solves the problem of inaccurate assessment of traditional deep learning models in new energy power systems due to changeable operating conditions by updating the source domain task model in real time to adapt to the dynamic changes of power grid operation scenarios. The present invention not only improves the accuracy and robustness of transient stability assessment, but also ensures accurate prediction of power system stability under different operating scenarios, providing strong support for the safe and stable operation of power grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] 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 labor. Among them:

[0045] Figure 1 It is a schematic diagram of the overall process of a variance-guided domain-adaptive power system transient stability adaptive assessment method according to an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of probability density distribution of the mean and variance of the Gaussian function of the variance-guided domain-adaptive power system transient stability adaptive assessment method according to an embodiment of the present invention and the enhanced kernel function of the present invention;

[0047] Figure 3 A schematic diagram of the TSA-VGA framework structure of a variance-guided domain-adaptive power system transient stability adaptive assessment method according to an embodiment of the present invention;

[0048] Figure 4 A schematic diagram of the domain adaptation effect of the variance-guided domain-adaptive power system transient stability adaptive assessment method according to an embodiment of the present invention;

[0049] Figure 5 A schematic diagram of ablation experiment results of a variance-guided domain-adaptive power system transient stability adaptive assessment method according to an embodiment of the present invention;

[0050] Figure 6 It is a schematic diagram of the effects of different migration schemes of the variance-guided domain-adaptive power system transient stability adaptive assessment method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purposes, 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 drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0052] 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.

[0053] 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.

[0054] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0055] At the same time, in the description of the present invention, it should be noted that the orientations or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the system or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0056] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0057] Example 1

[0058] Reference Figure 1-3, which is an embodiment of the present invention, provides a variance-guided domain-adaptive power system transient stability adaptive assessment method, comprising:

[0059] Before introducing the embodiments of the present application in detail, for the sake of clarity, some concepts related to this document are first explained.

[0060] The deep learning-based transient stability assessment method (TSA) aims to explore the intrinsic relationship between electrical measurement data and transient stability, and characterize it through mathematical models to achieve rapid real-time assessment of transient stability of power systems.

[0061] However, this mathematical model relies on a fixed training set and specific system operating conditions, which makes it difficult to adapt to the dynamic changes in the operating conditions of the renewable energy power system. To solve this problem, a domain adaptation method based on distance metrics is applied. Its core principle is to select a suitable distribution distance metric to quantify the distribution difference between the source domain and the target domain. During the model training process, the distribution of the two domains is aligned by minimizing the distribution distance metric, thereby enhancing the generalization ability and adaptability of the model.

[0062] MMD has become one of the most widely used distribution distance metrics due to its effectiveness and intuitiveness. Define H as the reproducing kernel Hilbert space, MMD[H,D s ,D t ] is the source domain D s and the target domain D t The MMD value between

[0063]

[0064] Among them, x and y are the source domain D s and the target domain D t , p, q are the true distributions of the source domain and the target domain, E is the expectation operator, h is the nonlinear continuous function in H, ||h|| H ≤1 is used to ensure the existence of a supremum, <·,·> H is the inner product in H, k(·,·) is the kernel function;

[0065] The kernel function usually chooses the Gaussian kernel function, which is expressed as:

[0066]

[0067] Among them, σ is the width of the kernel function, x, y are the source domain D s and the target domain D t of sample;

[0068] Specifically, MMD maps x and y to the feature space of H through the Gaussian kernel function k(x,y), and then calculates the mean difference between x and y in the feature space.

[0069] However, in transient stability assessment, when the power system is subjected to strong disturbances such as three-phase short circuit or load sudden change, its dynamic elements such as generators and transformers will show nonlinear and complex responses.

[0070] Therefore, the present invention proposes an improved transient stability assessment method, which mainly improves the kernel function and introduces the variance-guided domain distribution alignment (VGA) mechanism, namely, the TSA-VGA framework. Figure 2 shown.

[0071] S100: Obtain transient stability data of the power system under different operation scenarios;

[0072] In the embodiment of the present application, in order to obtain transient stability data of the power system under different operating scenarios, the present invention adopts a method based on time domain simulation technology, by constructing an accurate power system model and setting a variety of different operating scenarios, such as load mutation, generator failure, three-phase short circuit, etc., and using professional power system simulation software to perform time domain simulation analysis;

[0073] During the simulation process, the system can simulate the dynamic behavior of power components such as generators, transformers, transmission lines, etc. in different scenarios, and output corresponding electrical measurement data such as voltage, current, power, etc. These transient stability data are then collected and recorded for subsequent data preprocessing and analysis.

[0074] The preprocessing stage mainly includes steps such as data cleaning, data normalization or standardization, and feature extraction to ensure the quality and consistency of the data and provide reliable data support for the subsequent construction and training of the transient stability assessment model.

[0075] S102: extracting voltage trajectory cluster features from transient stability data to construct source domain data, using a neural network to train the source domain data, and building a source domain task model;

[0076] In the embodiment of the present application, a 1-dimensional convolutional neural network is selected to train the source domain data;

[0077] Specifically, from the complex simulation results, the present invention mainly focuses on the dynamic trajectory of voltage changes over time. These trajectories show unique patterns or sets under different operating scenarios, namely, voltage trajectory cluster characteristics. These characteristics not only reveal key information such as the amplitude, frequency and duration of voltage fluctuations, but also deeply reflect the operating characteristics of the power system under different states.

[0078] In order to effectively utilize the voltage trajectory cluster features for model training, the extracted voltage trajectory cluster feature data is first carefully preprocessed. After preprocessing, the data is divided. Data division includes reasonably dividing the sorted data set into training set, validation set and test set, laying the foundation for subsequent model training and evaluation. These preprocessing steps significantly improve the data availability and model training efficiency.

[0079] The processed voltage trajectory cluster feature data is further organized into a source domain data set, which becomes the core resource for subsequent neural network training. Based on the source domain data set, the present invention adopts a one-dimensional convolutional neural network for model training. During the training process, the neural network automatically captures and extracts the inherent laws and patterns through in-depth learning of the voltage trajectory cluster features in the source domain data. With the increase in the number of iterations, the neural network gradually builds up the ability to accurately evaluate the dynamic response of the power system, and finally forms a source domain task model.

[0080] It should be noted that the 1D convolutional neural network has advantages such as simple structure and fast training speed. The computational efficiency of the 1D convolutional neural network is usually higher than that of LSTM, etc., because it can process data in parallel, while LSTM and other cyclic structure models usually need to process data sequentially. The 1D convolutional neural network can process a large amount of transient stability data faster during the training and evaluation stages. Therefore, in order to more effectively extract the time series features of the power grid operation, the present invention selects a 1D convolutional neural network with powerful feature extraction capabilities and flexible structure to capture the long-term dependencies between different time steps and effectively solve the gradient disappearance and gradient explosion problems.

[0081] S104: Obtain transient stability data when the power system operation scenario changes, construct target domain data, merge the source domain data and the target domain data, and construct a mixed sample set;

[0082] Specifically, advanced monitoring systems and data analysis technologies are used to collect various operating parameters of the power system, such as voltage, current, frequency, and load distribution, in real time. Through comprehensive analysis of these data, the current operating status and characteristics of the power grid can be identified.

[0083] When significant changes are detected in the operation scenario of the power grid, such as the addition of large-capacity power sources, the transfer of important loads, or major adjustments to the network structure, relevant data is recorded, and high-precision power system simulation software is used to perform simulation analysis on the new operation scenario. Various possible operation conditions of the power system under the new scenario are simulated, including normal operation, fault occurrence and recovery process, etc., and detailed transient stability data is recorded.

[0084] By using signal processing and feature engineering techniques, a small number of labeled samples closely related to the target domain are extracted from the simulation results. The labeled samples not only contain the specific values ​​of electrical quantities such as voltage and current, but also come with corresponding operating scenario labels, providing valuable data support for subsequent model training.

[0085] These newly acquired target domain annotated samples are merged with the original source domain data to construct a hybrid sample set. This hybrid sample set not only contains data on the power system under various historical operating scenarios, but also incorporates dynamic response information under new scenarios, providing rich data resources for training more comprehensive and accurate power system assessment models.

[0086] S106: Using the improved transient stability assessment method and the mixed sample set to update the source domain task model, and construct the target domain task model;

[0087] Preferably, the Gaussian kernel function in the transient stability assessment method is replaced by an enhanced kernel function, and the enhanced kernel function is used to improve the tolerance of the source domain task model to extreme values;

[0088] Preferably, an enhanced kernel function is used to process the extreme value problem caused by the long-tail distribution characteristics and mutation of the mixed sample set. The enhanced kernel function is expressed as:

[0089]

[0090] Where k′(x,y) is the enhanced kernel function, x is the sample data in the source domain data, y is the sample data in the target domain data, d is the degree of freedom, and G(·) is the Gamma function;

[0091] It should be noted that the Gaussian kernel function lacks sufficient robustness when facing transient stability data with long-tail effects. This is because the tail of the Gaussian kernel function decays quickly, resulting in its high sensitivity to extreme values ​​or outliers. In contrast, the t distribution has a thicker tail, which means that it is more robust to extreme values ​​or outliers. The t distribution can more effectively accommodate these extreme values ​​and reduce their impact on the overall results, thereby performing more robustly when processing transient stability data. The enhanced kernel function of the present invention combines the heavy-tailed characteristics of the t distribution, making it more superior in processing transient stability assessments with long-tail effects. The enhanced kernel function proposed in the present invention is inspired by the t distribution; using the characteristics of the t distribution, the kernel function improves the model's tolerance to extreme values, thereby enhancing the reliability and robustness of the model when facing complex data distributions.

[0092] In the embodiment of the present application, in order to prove that the enhanced kernel function k'(x, y) of the present invention has the properties of a kernel function, that is, when d is a positive integer, it is a kernel function, two key properties of the kernel function are used to prove it indirectly;

[0093] Property 1: If z(x,y) is a kernel function, then for any constant r>0, r·z(x,y) is still a kernel function.

[0094] Property 2: If z(x,y) is a kernel function, then for any positive integer w, z(x,y) w It is still a kernel function.

[0095] First, k'(x,y) can be rewritten as:

[0096]

[0097] set up The above formula can be rewritten as:

[0098]

[0099] Apparently, and is the inverse multivariate quadratic kernel function.

[0100] Let w = d + 1, According to property 2, we can get is the kernel function.

[0101] Finally, let According to property 1, we can get It is a kernel function, that is, the enhanced kernel function k'(x, y) proposed in the present invention satisfies the properties of the kernel function and can be used to improve the robustness of the domain adaptation framework.

[0102] Preferably, based on the enhanced kernel function, a variance-guided domain distribution alignment mechanism is introduced to handle the distribution alignment problem between the source domain data and the target domain data;

[0103] Preferably, based on the enhanced kernel function, the variance-guided domain distribution alignment mechanism constructs a set of bases that can express the distribution variance characteristics, and reconstructs the reproducing kernel Hilbert space;

[0104] In the embodiment of the present application, the construction of the reproducing kernel Hilbert space is based on the basis function τ(x,·) defined by the enhanced kernel function k'(x,·), τ(x,·) is expressed as:

[0105]

[0106] It should be noted that the purpose of designing the basis function τ(x,·) is to capture and express the variance information of the source domain and the target domain. The variance information can reflect the overall characteristics of the data distribution, including the discreteness, dispersion range and volatility of the data.

[0107] Preferably, a variance-guided domain distribution difference measurement method is used in Hilbert space to quantify and reduce the difference between the source domain data and the target domain data distribution;

[0108] Preferably, the variance-guided domain distribution difference measurement method uses biased estimation and unbiased estimation to measure the difference;

[0109] Preferably, when the target domain data volume reaches a preset value, biased estimation is adopted, and when the target domain data volume does not reach the preset value, unbiased estimation is adopted;

[0110] It should be noted that the setting of the preset value should be determined comprehensively based on the requirements of the specific task, the characteristics of the target domain data, and the robustness of the domain adaptation algorithm; generally speaking, the preset value aims to find a balance point, which can ensure that when the sample size of the target domain is sufficient, the biased estimation using all the data can provide a sufficiently stable and representative measurement of the inter-domain distribution difference, and can switch to the unbiased estimation in time when the sample size is insufficient, so as to reduce the empirical error caused by the scarcity of samples;

[0111] In a possible embodiment, this value can be determined through cross-validation, domain knowledge or experimental exploration; for example, a threshold can be set so that in most experiments or application scenarios, when the target domain sample size exceeds the threshold, the performance of the domain adaptation algorithm is significantly improved and tends to be stable. At this time, the threshold can be used as a preset value; therefore, the setting of the preset value needs to be flexibly adjusted to adapt to the data volume and algorithm requirements in different scenarios.

[0112] Preferably, according to McDiarmid theorem and symmetry theorem, the error bound of the biased estimate is obtained;

[0113] Preferably, a large deviation bound based on the U statistic is used to obtain an error bound for the unbiased estimate.

[0114] Preferably, based on the variance-guided domain distribution difference measurement method, a loss function of the variance-guided domain distribution alignment mechanism is established to reduce the difference in distribution between the source domain data and the target domain data. The loss function of the variance-guided domain distribution alignment mechanism is expressed as:

[0115]

[0116] Among them, τ(x,·) is the basis function defined by the enhanced kernel function, x is the sample data in the source domain data, y is the sample data in the target domain data, E x～p (·) is the mathematical expectation function of data distribution p, E y～q (·) is the mathematical expectation function of the data distribution q, h is the nonlinear continuous function in H, and H' is the reconstructed Hilbert space;

[0117] In the embodiment of the present application, in order to simplify mathematical operations, L VGA Square it and get The definition is as follows:

[0118]

[0119] therefore, Biased estimate of It can be defined as:

[0120]

[0121] Among them, x i ,y i are the i-th samples in the source domain and the target domain respectively, m and n are the number of samples in the source domain and the target domain respectively, and H' is the reconstructed Hilbert space

[0122] τ(·,·) is defined as follows:

[0123]

[0124] An unbiased estimate of It can be defined as:

[0125]

[0126] In the embodiment of the present application, according to the McDiarmid theorem and the symmetry theorem, a biased estimate can be obtained: The error margin for:

[0127]

[0128] Among them, x is the sample data in the source domain data, y is the sample data in the target domain data, and δ is the confidence level.

[0129] Using the large deviation bound based on the U statistic, when m = n, the unbiased estimate The error margin It can be expressed as:

[0130]

[0131] Among them, x is the sample data in the source domain data, y is the sample data in the target domain data, is the floor function.

[0132] From the above description, we can see that when the sample sizes m and n of the source domain and the target domain increase, the error bound will decrease with the increase of the sample size; at the same time, the increase of the confidence level δ will lead to a stricter error bound. This result provides a reliable theoretical basis for the subsequent domain adaptation process.

[0133] It should be noted that the variance-guided domain distribution alignment mechanism overcomes the problem of insufficient robustness of the traditional Gaussian kernel function in processing transient stable data with long-tail distribution characteristics through the enhanced kernel function proposed in Section A, effectively reducing the difficulty of domain adaptation of the model. At the same time, the variance-guided domain distribution difference measurement is used to enhance the model's ability to understand the distribution differences between domains, thereby promoting the cross-domain migration and expression of knowledge and improving the model's adaptive ability.

[0134] S108: inputting the real-time operation data of the power system into the target domain task model to evaluate the transient stability of the power system;

[0135] The real-time operation data of the power system is input into the target domain task model to evaluate the transient stability of the power system. The evaluation result is stable or unstable.

[0136] Specifically, the target domain task model transformed by the present invention to improve the transient stability assessment method is as follows: Figure 3 As shown in the figure, take the New England 10-machine 39-node system and the target system 1 and target system 2 after the operation mode is changed as an example. Figure 3 The probability density distribution of the mean (Figure A) and variance (Figure B) of the transient stability data of these three systems is shown.

[0137] Depend on Figure 3 It can be clearly seen that the distributions of multiple systems have significant extensions on the right, that is, the long tail phenomenon appears. The long tail phenomenon means that under strong disturbances, the frequency of extreme values ​​in transient stability data increases, which is in sharp contrast to the characteristics of Gaussian distribution.

[0138] In a Gaussian distribution, the tail is lighter and the probability density at both ends of the data distribution is lower. This characteristic makes the traditional Gaussian kernel function perform well when processing light-tailed distribution data; however, when faced with transient stable data with long-tail effects, the Gaussian kernel function is no longer applicable because under the long-tail effect, the frequency of occurrence of extreme values ​​is much higher than expected in the Gaussian distribution, which makes the Gaussian kernel function less robust when processing such data.

[0139] Secondly, the dynamic response of the power system after being disturbed is usually manifested as the volatility of electrical measurement data, which is a key factor in analyzing the stability of the system. The mean in Figure A can show the central trend of the system as a whole, but it is not enough to fully reflect the volatility and discreteness of the system under disturbance.

[0140] In contrast, the variance distribution in Figure B can more accurately reveal the dynamic response characteristics of the system under strong disturbances because it can more clearly show the discreteness of the data.

[0141] The above is a schematic scheme of a variance-guided domain adaptive power system transient stability adaptive assessment method of this embodiment. It should be noted that the technical scheme of the variance-guided domain adaptive power system transient stability adaptive assessment system and the technical scheme of the above-mentioned variance-guided domain adaptive power system transient stability adaptive assessment method belong to the same concept. The details of the technical scheme of the variance-guided domain adaptive power system transient stability adaptive assessment system that are not described in detail in this embodiment can be found in the description of the technical scheme of the above-mentioned variance-guided domain adaptive power system transient stability adaptive assessment method.

[0142] The variance-guided domain-adaptive power system transient stability adaptive assessment system in this embodiment includes:

[0143] A data acquisition module is used to obtain transient stability data of the power system under different operation scenarios;

[0144] The source domain task model building module is used to extract the voltage trajectory cluster features in the transient stability data to construct the source domain data, use the neural network to train the source domain data, and build the source domain task model;

[0145] The hybrid sample acquisition module is used to obtain transient stability data when the power system operation scenario changes, construct target domain data, merge the source domain data with the target domain data, and construct a hybrid sample set;

[0146] The target domain task model construction module is used to update the source domain task model using the improved transient stability assessment method and the mixed sample set to construct the target domain task model;

[0147] The evaluation module is used to input the real-time operation data of the power system into the target domain task model to evaluate the transient stability of the power system.

[0148] This embodiment further provides a computing device, which is applicable to the case of variance-guided domain adaptive power system transient stability adaptive assessment, including:

[0149] A memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the variance-guided domain-adaptive power system transient stability adaptive assessment method proposed in the above embodiment.

[0150] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the variance-guided domain-adaptive power system transient stability adaptive assessment method proposed in the above embodiment is implemented.

[0151] The storage medium proposed in this embodiment and the method for implementing variance-guided domain-adaptive power system transient stability adaptive assessment proposed in the above embodiment belong to the same inventive concept. The technical details not described in detail in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0152] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0153] Example 2

[0154] Reference Figure 4-6 Table 1-3 is an embodiment of the present invention, which provides a variance-guided domain-adaptive power system transient stability adaptive assessment method. In order to verify its beneficial effects, comparison results of two schemes are provided.

[0155] This embodiment uses the benchmark operation mode of the New England 10-machine 39-node system as the source domain system, the simulation step size is set to 0.04 seconds, and the total simulation time is 3 seconds;

[0156] The fault type is a three-phase short circuit fault, the fault duration ranges from 1 cycle to 17 cycles, the step length is 1 cycle, and the fault location covers 1% to 99% of each line, the step length is 1%. In the above method, a total of 18,000 simulation samples are generated, including 10,000 stable samples and 8,000 unstable samples, and the training set and test set are randomly divided according to the ratio of 5:1.

[0157] Exemplarily, the adaptability of the source domain model under the new operation scenario can be tested by the following steps: In order to simulate the significant changes in the operation scenario of the real power system, based on the source domain, two new power grid operation scenarios are designed considering the deactivation and addition of load nodes and line equipment, and are used as two new target domain systems, including:

[0158] Target system 1: Due to a fault or maintenance, node 4 and the three transmission lines connected to it are shut down, and the output of the generator is adjusted to ensure the convergence of the power flow calculation.

[0159] Target system 2: Due to changes in electricity demand, node 40 is added and connected to generator 31. At the same time, the output of the generator is adjusted to ensure the convergence of the power flow calculation.

[0160] According to the above conditions, 2000 samples are randomly simulated for target system 1 and target system 2 respectively. Among them, target system 1 contains 1175 stable samples and 825 unstable samples; target system 2 contains 1245 stable samples and 755 unstable samples. The generated samples are divided into training set and test set according to the ratio of 1:5. The 1D-CNN network is trained offline with the training set of the source domain to obtain the source domain task model, and then the performance of the model is tested on the source domain test set and the training and test sets of target systems 1 and 2. The test results are shown in Table 1.

[0161] Table 1 Test results of source domain task models in different scenarios

[0162]

[0163] The test results show that the 1D-CNN model has a significant performance advantage due to its superior feature extraction capability, and the prediction accuracy has reached 98.62%, indicating that the 1D-CNN model is suitable for transient assessment tasks in the source domain. However, when the model is applied to target systems 1 and 2, the evaluation performance of the source domain task model drops sharply due to changes in the target system power grid operation scenario, and the prediction accuracy is less than 90%, which can no longer meet the requirements of online applications. Therefore, it is necessary to use the TSA-VGA framework to improve the applicability of the model in new scenarios.

[0164] The applicability of TSA-VGA can be verified by the following steps: 50, 100, 150, and 200 labeled samples are randomly selected from the training sets of target system 1 and target system 2 as target domain samples, and they are combined with 3700 randomly selected source domain samples to construct mixed sample sets S11, S12, S13, S14 and S21, S22, S23, and S24 of target system 1 and target system 2, respectively. The input batch size of the training model is set to 150, of which the batch size of source domain samples is 100 and the batch sample size of target domain samples is 50. The total iteration rounds of model training are set to 200, and the learning rate is 0.001. The domain adaptation effect of the TSA-VGA framework under the training of different mixed sample sets is shown in Table 2.

[0165] Table 2 Applicability results of TSA-VGA

[0166]

[0167]

[0168] The results in Table 2 show that when the source domain model is supplemented with mixed sample sets S11 and S21 containing a small number of target domain samples, the modelability evaluation performance is significantly improved, but still does not exceed 95%. This is mainly because the insufficient number of target domain samples limits the model's comprehensive learning of the target domain feature distribution and potential patterns. As the number of target domain samples increases, the model is able to more extensively access the diversity and complexity of the target domain, thereby learning the target domain information more deeply. Specifically, when the number of samples increases to 200, the prediction accuracy of the model on target systems 1 and 2 is significantly improved, with the prediction accuracy of target system 1 increasing from 87.37% to 98.25%, and the prediction accuracy of target system 2 increasing from 88.87% to 98.12%. This shows that the proposed TSA-VGA framework can significantly improve the domain adaptation effect of the model, and the TSA-VGA framework can effectively improve the evaluation performance of the model under the new operating mode using only a small number of samples.

[0169] In order to more intuitively illustrate the domain adaptation effect of the TSA-VGA framework, the feature extraction results of the target systems 1 and 2 before and after domain adaptation are visualized and mapped to a two-dimensional plane. The specific results are as follows: Figure 4 As shown;

[0170] Figure 4 It can be clearly seen that after domain adaptation, the data features of target systems 1 and 2 are closer to the feature distribution of the source domain, the inter-domain differences are significantly reduced, and the stable and unstable samples have a more obvious classification boundary, which shows that the TSA-VGA framework has achieved good results in cross-domain feature alignment.

[0171] Exemplarily, an ablation experiment can be performed through the following steps: In order to verify the effectiveness and reliability of the enhanced kernel function and variance-guided domain distribution alignment mechanism in the TSA-VGA framework, two incomplete TSA-VGA frameworks are set up for ablation experiments.

[0172] Ablation framework 1: In this framework, the enhanced kernel function is removed and the Gaussian kernel widely used in MMD is used as the kernel function for domain adaptation. Other settings remain consistent with the TSA-VGA framework.

[0173] Ablation framework 2: In this framework, the variance-guided domain adaptation mechanism is not adopted, that is, the variance feature difference between the source domain and the target domain is not included in the final distribution difference measurement. Instead, only the mean difference of MMD is used for measurement, and other settings are consistent with the TSA-VGA framework.

[0174] Figure 5 For the ablation experiment results, the ablation framework 1 uses the Gaussian kernel function instead of the enhanced kernel function. At this time, the prediction accuracy of the target system 1 and 2 failed to break through the 98% threshold, which shows that the Gaussian kernel function is not robust enough in processing transient stable data with long-tail distribution characteristics, which affects the generalization ability of the model. ;

[0175] Ablation framework 2 only considers the mean difference when measuring the data distribution difference between the source domain and the target domain, ignoring the discreteness and volatility of transient stable data, and fails to fully reflect the differences between data distributions. The limitations of this method result in its prediction accuracy on target systems 1 and 2 failing to reach 98%.

[0176] In contrast, the TSA-VGA framework proposed in the present invention better fits the transient stable data distribution through the enhanced kernel function, improves the generalization of the model to long-tail characteristic data, and uses the variance-guided domain distribution alignment mechanism to map the source domain and target domain data to the Hilbert space that can reflect the variance characteristics, thereby accurately quantifying the difference in inter-domain distribution and achieving a greater degree of inter-domain distribution difference alignment. Finally, the proposed TSA-VGA framework achieved a prediction accuracy of more than 98% in both target systems 1 and 2.

[0177] The results of the above ablation experiments verify the effectiveness of the TSA-VGA framework and confirm the importance of the enhanced kernel function and variance-guided domain adaptation mechanism for the TSA-VGA framework.

[0178] The sample requirements of the TSA-VGA framework can be verified by following the steps below:

[0179] During online operation, it is usually difficult to collect a large amount of transient stability data for new scenarios in a short period of time in the face of sudden changes in the operation mode of the power grid. Therefore, it is necessary to use as few target domain samples as possible to update the model in the shortest time and achieve the best prediction effect during online operation. This section compares and analyzes the sample requirements of the three models of retraining, fine-tuning algorithm and TSA-VGA framework, and examines the sample size, sample generation time and model training time required for each model to achieve an evaluation accuracy of 96.5% in the new scenario. The specific experimental results are shown in Table 3.

[0180] Table 3 TSA-VGA sample requirement verification results

[0181]

[0182]

[0183] Compared with retraining the model, the fine-tuning algorithm effectively retains the knowledge acquired by the model in the source domain by sharing the structure and parameters of the source domain task model, reducing the time for sample generation and model update;

[0184] However, the algorithm still requires 700 target domain samples to achieve a prediction accuracy of 96.5%, and its sample generation time is as long as 1575s. In contrast, the TSA-VGA framework proposed in the present invention only requires 150 target domain samples to achieve the same prediction accuracy, and the sample generation time is also shortened to 337s, and the model training time is only about 6s. Therefore, the TSA-VGA framework proposed in the present invention significantly reduces the time for sample generation and model training while maintaining high accuracy.

[0185] The present invention can also conduct comparative analysis on different migration schemes through the following steps: In order to verify the performance advantage of the TSA-VGA framework in power system transient stability assessment, the TSA-VGA framework is compared and analyzed with the existing mainstream migration learning strategies. The main comparison schemes include:

[0186] Strategy 1: Without using transfer learning, the target domain samples in the mixed sample set S11 are used as model input to retrain it.

[0187] Strategy 2: Using the sample migration method, the source domain and target domain samples in the mixed sample set S11 are used as model inputs and jointly retrained.

[0188] Strategy 3: Adopt an adversarial unsupervised feature transfer method, use the domain adversarial mechanism and gradient reversal layer to make the model unable to distinguish the source of samples, thereby achieving the purpose of domain confusion. Among them, the weight value in the gradient reversal layer is set to 1. The source domain and target domain samples in the mixed sample set S11 are used as model input, but the target domain samples do not participate in the training of the task classifier.

[0189] Strategy 4: Adopt an adversarial supervised feature transfer method, and the specific settings are consistent with Strategy 3. The main difference from Strategy 3 is that the target domain samples participate in the training of the task classifier.

[0190] Strategy 5: Use parameter transfer to transfer the parameter knowledge acquired by the source domain task model to the target domain, and use the target domain samples in the mixed sample set S11 as the model input to fine-tune the entire model.

[0191] Strategy 6: Joint parameter transfer and sample transfer, transfer the parameter knowledge acquired by the source domain task model to the target domain, and use the source domain and target domain samples of the mixed sample set S11 as model input to jointly fine-tune the entire model.

[0192] Strategy 7: Adopt the domain adaptation method based on maximum mean difference, use MMD to measure the difference in data distribution between the source domain and the target domain, and find common features between the two domains to improve the adaptability of the model. The source domain and target domain samples of the mixed sample set S11 are used as model input.

[0193] Strategy 8: TSA-VGA framework. The source domain and target domain samples of the mixed sample set S11 are used as model input.

[0194] The test results of the above 8 strategies on target system 1 are as follows: Figure 6 shown.

[0195] Figure 6 The results show that due to the insufficient number of samples in the target domain, Strategy 1 fails to fully capture the characteristic information of the target domain, resulting in its accuracy rate failing to exceed 95%, and its performance is poor;

[0196] Strategy 2 uses a small number of target domain samples while fusing a large number of source domain samples, so that the model can be exposed to more features and patterns during the training process. Compared with Strategy 1, this method effectively enhances the generalization ability of the model and improves the evaluation performance of the model.

[0197] Strategy 3 adopts an adversarial unsupervised feature transfer method, which aims to mine common features that can be transferred between the source domain and the target domain. However, due to the lack of sufficient target domain samples and target domain label information, the domain adaptation effect is not good.

[0198] On this basis, Strategy 4 improves the model evaluation performance by introducing a small amount of target domain samples with labeled information, making the prediction accuracy reach 96.72%;

[0199] Strategy 5 and Strategy 6 use the parameter migration method to migrate the parameter knowledge acquired by the source domain task model to the target domain, providing a good learning starting point for the new model. This parameter migration-based method enables the model prediction accuracy to reach 87.37% when initially applied to the target domain. At the same time, Strategy 6 also combines sample migration technology to introduce a large amount of source domain information into the target domain task, effectively improving the evaluation performance of the model in the sample-scarce stage, and the model prediction accuracy reaches 96.98%;

[0200] Strategy 7 uses MMD to align the feature distribution between the source domain and the target domain, with an accuracy of 97.74%. Strategy 8, i.e., the method of the present invention, improves the robustness of the model to long-tail distribution data by using an enhanced kernel function, and uses the difference in variance features between the two domains as a metric for the final distribution difference, effectively improving the domain adaptability of the model, and ultimately achieving an accuracy of 98.25%, significantly better than the previous 7 strategies.

[0201] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. 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 variance-guided domain-adaptive power system transient stability adaptive assessment method, characterized in that: include: Obtain transient stability data of power systems under different operating scenarios; Extracting voltage trajectory cluster features from the transient stability data to construct source domain data, training the source domain data using a neural network, and building a source domain task model; Acquire transient stability data when the power system operation scenario changes, construct target domain data, merge the source domain data with the target domain data, and construct a mixed sample set; The source domain task model is updated by using the improved transient stability assessment method and the mixed sample set to construct a target domain task model; The real-time operation data of the power system is input into the target domain task model to evaluate the transient stability of the power system.

2. The variance-guided domain-adaptive power system transient stability adaptive assessment method according to claim 1, characterized in that: Improved transient stability assessment methods include: The Gaussian kernel function in the transient stability assessment method is replaced by an enhanced kernel function, and the enhanced kernel function is used to improve the tolerance of the source domain task model to extreme values; Based on the enhanced kernel function, a variance-guided domain distribution alignment mechanism is introduced to handle the distribution alignment problem between the source domain data and the target domain data.

3. The variance-guided domain-adaptive power system transient stability adaptive assessment method according to claim 2, characterized in that: The variance-guided domain distribution alignment mechanism includes, Based on the enhanced kernel function, the variance-guided domain distribution alignment mechanism constructs a set of bases that can express the distribution variance characteristics and reconstructs the reproducing kernel Hilbert space; A variance-guided domain distribution difference measurement method is used in the reproducing kernel Hilbert space to quantify and reduce the difference between the source domain data and the target domain data distribution.

4. The variance-guided domain-adaptive power system transient stability adaptive assessment method according to claim 3, characterized in that: The enhanced kernel functions include: The enhanced kernel function is used to process the extreme value problem caused by the long-tail distribution characteristics and mutation of the mixed sample set. The enhanced kernel function is expressed as: Where k′(x, y) is the enhanced kernel function, x is the sample data in the source domain data, y is the sample data in the target domain data, d is the degree of freedom, and G(·) is the Gamma function.

5. The variance-guided domain-adaptive power system transient stability adaptive assessment method according to claim 4, characterized in that: Variance-guided domain distribution difference measurement methods include: The variance-guided domain distribution difference measurement method uses biased estimation and unbiased estimation to measure the difference; When the target domain data volume reaches a preset value, biased estimation is adopted; when the target domain data volume does not reach the preset value, unbiased estimation is adopted.

6. The variance-guided domain-adaptive power system transient stability adaptive assessment method according to claim 5, characterized in that: Also includes, Obtain error bounds for biased estimates based on McDiarmid's theorem and the symmetry theorem; Obtain error bounds for unbiased estimates using large deviation bounds based on the U statistic.

7. The variance-guided domain-adaptive power system transient stability adaptive assessment method according to claim 3, characterized in that: Narrowing the difference between the source domain data and the target domain data distribution includes: Based on the variance-guided domain distribution difference measurement method, a loss function of the variance-guided domain distribution alignment mechanism is established to reduce the difference in distribution between the source domain data and the target domain data. The loss function of the variance-guided domain distribution alignment mechanism is expressed as: Among them, τ(x,·) is the basis function defined by the enhanced kernel function, x is the sample data in the source domain data, y is the sample data in the target domain data, E x～p (·) is the mathematical expectation function of data distribution p, E y～q (·) is the mathematical expectation function of the data distribution q, h is the nonlinear continuous function in H, and H' is the reconstructed Hilbert space.

8. A variance-guided domain-adaptive power system transient stability adaptive assessment system, characterized in that: include, A data acquisition module is used to obtain transient stability data of the power system under different operation scenarios; A source domain task model construction module is used to extract voltage trajectory cluster features in the transient stability data to construct source domain data, train the source domain data using a neural network, and construct a source domain task model; A hybrid sample acquisition module, used to acquire transient stability data when the power system operation scenario changes, construct target domain data, merge the source domain data with the target domain data, and construct a hybrid sample set; A target domain task model construction module is used to update the source domain task model using the improved transient stability assessment method and the mixed sample set to construct a target domain task model; The evaluation module is used to input the real-time operation data of the power system into the target domain task model to evaluate the transient stability of the power system.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the variance-guided domain-adaptive power system transient stability adaptive assessment method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the variance-guided domain-adaptive power system transient stability adaptive assessment method as described in any one of claims 1 to 7.

Images