Transient stability assessment method and system based on GRU and LS-CGAN

Abstract

The invention discloses a transient stability evaluation method and system based on a GRU and an LS-CGAN, and the method comprises the steps: collecting power data during the operation of a power system, carrying out the analysis of the power data, obtaining first feature data, and enabling the power data to comprise a voltage amplitude, a phase angle and power; inputting the first feature data into a preset LS-CGAN model to obtain second feature data similar to the first feature data; and inputting the first feature data and the second feature data into a preset GRU model to obtain a transient stability evaluation result of the power system. According to the method, the GRU and the LS-CGAN are fused, so that not only can different power system scenes and evaluation requirements be adapted, but also the transient stability of the power system can be quickly and accurately evaluated and predicted, and thus the safe and stable operation of the power system is ensured.

Description

Technical Field

[0001] The present application relates to the field of stability analysis, and in particular to a transient stability assessment method and system based on GRU and LS-CGAN. Background Art

[0002] The integration of renewable energy and distributed generators in power systems has become an inevitable trend, but these new energy resources are significantly unpredictable and difficult to control, bringing unprecedented challenges to the stable operation of power systems. In power systems, transient stability is a key factor to ensure that the system can still maintain synchronous operation under large disturbances, but the power system will inevitably be affected by various factors during operation. If the transient stability of the power system is insufficient, it may cause power outages or even power system collapse.

[0003] At present, the industry mainly uses the Liapunov direct method and time domain simulation method to analyze the transient stability of power systems, but these methods have their defects in transient stability assessment. Among them, the Liapunov direct method is based on a simplified power system model for analysis, and the analysis results of complex power systems may have errors; while the time domain simulation method requires solving a large number of differential methods, which is difficult to meet the needs of online assessment, and is sensitive to initial conditions, making it difficult to accurately determine the initial conditions, which leads to inaccurate transient stability analysis results. Therefore, it is particularly important to quickly and accurately assess the transient stability of power systems and ensure the safe and stable operation of power systems.

[0004] Application Contents

[0005] This application provides a transient stability assessment method and system based on GRU and LS-CGAN. By integrating GRU and LS-CGAN, it can not only adapt to different power system scenarios and assessment requirements, but also quickly and accurately assess and predict the transient stability of the power system, thereby ensuring the safe and stable operation of the power system.

[0006] In the first aspect, the present application provides a transient stability evaluation method based on GRU and LS-CGAN, including:

[0007] Collecting power data when the power system is running, and analyzing the power data to obtain first characteristic data, wherein the power data includes voltage amplitude, phase angle and power;

[0008] Inputting the first feature data into a preset LS-CGAN model to obtain second feature data similar to the first feature data;

[0009] The first characteristic data and the second characteristic data are input into a preset GRU model to obtain a transient stability assessment result of the power system.

[0010] By analyzing the collected power data, the embodiment of the present application can extract feature data that can reflect the transient stability of the power system, thereby reducing the data dimension and facilitating the subsequent rapid and accurate evaluation and prediction of the transient stability of the power system; by inputting the first feature data into a preset LS-CGAN model, the second feature data similar to the first feature data can be accurately simulated, and then the possible operating state of the power system can be simulated, which is convenient for the subsequent GRU model to evaluate and predict the transient stability of the power system; by inputting the first feature data and the second feature data into a preset GRU model, the GRU model can learn and capture the law of changes in the operating state of the power system over time, and can quickly and accurately evaluate and predict the transient stability of the power system, thereby ensuring the safe and stable operation of the power system.

[0011] Furthermore, the power data is analyzed to obtain first characteristic data, specifically:

[0012] Calculate the stability index value corresponding to each power characteristic of the power system after the disturbance through simulation analysis, and determine the transient stability index of the power system based on the stability index value;

[0013] Calculate the correlation between the power data and the transient stability index by using the maximum information coefficient to obtain a MIC value;

[0014] The first feature data is determined based on the MIC value.

[0015] In this way, by analyzing the collected power data, characteristic data that can reflect the transient stability of the power system can be extracted, thereby reducing the data dimension and facilitating the subsequent rapid and accurate evaluation and prediction of the transient stability of the power system.

[0016] Furthermore, the correlation between the power data and the transient stability index is calculated by the maximum information coefficient to obtain the MIC value, which is specifically:

[0017] Performing grid division on the power data and the transient stability index to obtain a plurality of grid division results;

[0018] Calculating the mutual information corresponding to the plurality of grid division results respectively, and determining the maximum mutual information value among the plurality of grid division results;

[0019] The mutual information value is normalized, and a corresponding MIC value is obtained based on the normalization result.

[0020] In this way, by calculating the correlation between the power data and the transient stability index, characteristic data that can reflect the transient stability of the power system can be extracted, thereby reducing the data dimension, alleviating the complexity of the young model and improving the ability to identify key features, thereby facilitating subsequent rapid and accurate evaluation and prediction of the transient stability of the power system.

[0021] Furthermore, the first feature data is input into a preset LS-CGAN model to obtain second feature data similar to the first feature data, wherein the LS-CGAN model includes a generator and a discriminator, specifically:

[0022] The first feature data is input into a preset LS-CGAN model, and the generator generates a simulated sample based on a randomly generated noise vector and a conditional signal;

[0023] The discriminator obtains a corresponding distribution probability value based on the simulated sample and the first feature data;

[0024] Based on the distribution probability value, second feature data similar to the first feature data is determined.

[0025] In this way, by inputting the first feature data into the preset LS-CGAN model, the second feature data similar to the first feature data can be accurately simulated, and then the possible operating state of the power system can be simulated, which is convenient for the subsequent GRU model to evaluate and predict the transient stability of the power system.

[0026] Furthermore, the objective function of the LS-CGAN model is specifically:

[0027]

[0028] Where V LS-CGAN (D) is the least squares loss function of the discriminator; V LS-CGAN (G) is the least squares loss function of the generator; D is the discriminator; G is the generator; E represents mathematical expectation; x is the first feature data; z is the noise vector; c is the conditional signal; p data (x) and P z (z) are the distribution of the first feature data and the distribution of the noise vector z respectively; G(z) is the simulated sample; D(x) represents the probability that the discriminator D considers the input data x to be true; the simulated sample G(z) generated by D(G(z,y)) is judged to be true by the discriminator D; V is the objective function; a and b are the labels of the simulated sample and the first feature data respectively.

[0029] In the second aspect, the present application provides a transient stability assessment system based on GRU and LS-CGAN, including: an acquisition module, a generation module and an assessment module;

[0030] The acquisition module is used to collect power data when the power system is running, and analyze the power data to obtain first characteristic data, wherein the power data includes voltage amplitude, phase angle and power;

[0031] The generating module is used to input the first feature data into a preset LS-CGAN model to obtain second feature data similar to the first feature data;

[0032] The evaluation module is used to input the first characteristic data and the second characteristic data into a preset GRU model to obtain a transient stability evaluation result of the power system.

[0033] By analyzing the collected power data, the embodiment of the present application can extract feature data that can reflect the transient stability of the power system, thereby reducing the data dimension and facilitating the subsequent rapid and accurate evaluation and prediction of the transient stability of the power system; by inputting the first feature data into a preset LS-CGAN model, the second feature data similar to the first feature data can be accurately simulated, and then the possible operating state of the power system can be simulated, which is convenient for the subsequent GRU model to evaluate and predict the transient stability of the power system; by inputting the first feature data and the second feature data into a preset GRU model, the GRU model can learn and capture the law of changes in the operating state of the power system over time, and can quickly and accurately evaluate and predict the transient stability of the power system, thereby ensuring the safe and stable operation of the power system.

[0034] Further, the acquisition module includes: a first calculation unit, a second calculation unit and a first determination unit;

[0035] The first calculation unit is used to calculate the stability index value corresponding to each power characteristic of the power system after the disturbance through simulation analysis, and determine the transient stability index of the power system based on the stability index value;

[0036] The second calculation unit is used to calculate the correlation between the power data and the transient stability index by using a maximum information coefficient to obtain a MIC value;

[0037] The first determining unit is configured to determine the first characteristic data based on the MIC value.

[0038] Furthermore, the second computing unit is specifically:

[0039] Performing grid division on the power data and the transient stability index to obtain a plurality of grid division results;

[0040] Calculating the mutual information corresponding to the plurality of grid division results respectively, and determining the maximum mutual information value among the plurality of grid division results;

[0041] The mutual information value is normalized, and a corresponding MIC value is obtained based on the normalization result.

[0042] Further, the generating module includes: a third calculating unit, a fourth calculating unit and a second determining unit;

[0043] The third computing unit is used to input the first feature data into a preset LS-CGAN model, and the generator generates a simulated sample based on the randomly generated noise vector and the conditional signal;

[0044] The fourth calculation unit is used for the discriminator to obtain a corresponding distribution probability value based on the simulated sample and the first feature data;

[0045] The second determining unit is configured to determine second feature data similar to the first feature data based on the distribution probability value.

[0046] Furthermore, the objective function of the LS-CGAN model is specifically:

[0047]

[0048] Where V LS-CGAN (D) is the least squares loss function of the discriminator; V LS-CGAN (G) is the least squares loss function of the generator; D is the discriminator; G is the generator; E represents mathematical expectation; x is the first feature data; z is the noise vector; c is the conditional signal; p data (x) and P z (z) are the distribution of the first feature data and the distribution of the noise vector z respectively; G(z) is the simulated sample; D(x) represents the probability that the discriminator D considers the input data x to be true; the simulated sample G(z) generated by D(G(z,y)) is judged to be true by the discriminator D; V is the objective function; a and b are the labels of the simulated sample and the first feature data respectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flowchart of an embodiment of a transient stability evaluation method based on GRU and LS-CGAN provided in this application;

[0050] Figure 2It is a schematic diagram of the data processing flow of the generator and the discriminator provided by this application;

[0051] Figure 3 It is a structural diagram of the GRU model provided in this application;

[0052] Figure 4 It is a structural diagram of an embodiment of a transient stability assessment system based on GRU and LS-CGAN provided in this application. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0054] It should be understood that the step numbers used in this article are only for the convenience of description and are not intended to limit the order in which the steps are executed.

[0055] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0056] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0057] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0058] The integration of renewable energy and distributed generation into the power system has become a trend, but its unpredictability and difficulty in control pose challenges to the stable operation of the power system. Transient stability is the key to maintaining synchronous operation of the power system under large disturbances, but the power system will inevitably be affected by various factors during operation. If the transient stability of the power system is insufficient, it may cause power outages or even power system collapse. At present, the Liapunov direct method and the time domain simulation method are used to analyze transient stability, but each has its own defects: the former Liapunov direct method is based on a simplified power system model for analysis, which may lead to errors in complex system analysis; the latter time domain simulation method requires solving a large number of differential equations, which is difficult to meet the needs of online evaluation, and is sensitive to initial conditions, which affects the accuracy of transient stability analysis results. Therefore, it is crucial to quickly and accurately evaluate the transient stability of the power system to ensure safe and stable operation.

[0059] Next, the nouns involved in this application are analyzed:

[0060] The LS-CGAN model combines the features of LSGAN (Least Squares Generative Adversarial Networks) and CGAN (Conditional Generative Adversarial Nets). Among them, LSGAN is an improvement on the original GAN, and the main improvement lies in the loss function. The original GAN ​​uses the cross entropy loss function, while LSGAN replaces it with the least squares loss function. CGAN is an extension of the original GAN, which controls the features (categories) of the data generated by GAN by adding additional information to both the generator and the discriminator as conditions. In other words, the LS-CGAN model combines the advantages of LSGAN and CGAN, and solves the two defects of traditional CGAN: gradient vanishing and training instability and the limitations of cross entropy loss. The least squares loss function used by LS-CGAN can make the network training more stable, converge faster, and generate higher quality samples.

[0061] GRU (Gated Recurrent Unit) is a recurrent neural network (RNN) used to solve the problem of gradient explosion and gradient vanishing. Traditional RNNs are prone to gradient vanishing or gradient exploding when processing long sequence data, which limits the application of RNNs in long-term dependent tasks. GRU effectively alleviates this problem by introducing a gating mechanism, improving the training stability and generalization ability of the network. Traditional neural networks such as BP neural networks and convolutional neural networks are not good at processing time series information, while GRU, as a variant of RNN, can combine historical information with the current time to predict further information.

[0062] The Maximum Information Coefficient (MIC) is a statistic used to quantify the strength of the dependency relationship between two variables. Compared with other correlation measurement methods such as mutual information (MI), Pearson correlation coefficient, Spearman correlation coefficient, and chi-square test, MIC has higher stability and extensiveness, and can accurately detect various types of correlation relationships such as linear, nonlinear, and periodic, and has significant advantages and application value.

[0063] Based on this, an embodiment of the present application provides a transient stability assessment method and system based on GRU and LS-CGAN. By integrating GRU and LS-CGAN, it can not only adapt to different power system scenarios and assessment requirements, but also quickly and accurately assess and predict the transient stability of the power system, thereby ensuring the safe and stable operation of the power system.

[0064] The transient stability assessment method and system based on GRU and LS-CGAN provided in the embodiments of the present application are specifically illustrated by the following embodiments. First, the transient stability assessment method and system based on GRU and LS-CGAN in the embodiments of the present application are described.

[0065] The transient stability assessment method based on GRU and LS-CGAN provided in the embodiment of the present application relates to the field of stability analysis. The transient stability assessment method and system based on GRU and LS-CGAN provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, and can also be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements a transient stability assessment method based on GRU and LS-CGAN, etc., but is not limited to the above forms.

[0066] The present application can also be used in numerous general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0067] Embodiment 1

[0068] Please refer to Figure 1 , Figure 1 It is a flowchart of an embodiment of a transient stability assessment method based on GRU and LS-CGAN provided by the present application, including steps S101 to S103;

[0069] Step S101, collecting power data when the power system is running, and analyzing the power data to obtain first characteristic data, wherein the power data includes voltage amplitude, phase angle and power;

[0070] It is understandable that the power data during operation can be collected by means of, but is not limited to, sensors or monitoring equipment installed at corresponding locations in the power system, wherein the type of the sensor corresponds to the type of the power data, and the power includes active power and reactive power.

[0071] It is understandable that after obtaining the power data of the power system during operation, the power data needs to be preprocessed, specifically: the format of the power data needs to be converted into a processable format. After obtaining the processable format, the missing values ​​of the power data need to be interpolated, etc., and then the abnormal values ​​in the power data that are obviously inconsistent with other observed values ​​are detected, and the abnormal values ​​are deleted, corrected or marked, etc. Finally, the duplicate data records in the power data are deleted, thereby ensuring the accuracy of the subsequent power data analysis results.

[0072] It can be understood that after the power data is preprocessed, the power data needs to be analyzed, specifically: first, the stability index value corresponding to each power characteristic of the power system after the disturbance is calculated through simulation analysis, and the transient stability index of the power system is determined based on the stability index value, wherein it is necessary to use power system simulation software to simulate the disturbances that the system may encounter under various operating modes, and observe the dynamic response of the power data after the disturbance through the simulation results, and then evaluate the transient stability of the system according to the changes in these data, and determine the corresponding stability index, based on the mathematical model and stability theory of the power system, derive the calculation formula of the transient stability index, substitute the actual parameters of the system into the formula, and calculate to obtain the specific stability index value, and finally, compare the calculated stability index value with the critical stability margin of the system to determine the transient stability index of the power system.

[0073] Secondly, the correlation between the power data and the transient stability index is calculated by the maximum information coefficient to obtain the MIC value; specifically, the power data and the transient stability index are grid-divided to obtain several grid division results G(xy), where x is the number of rows of the grid G, y is the number of columns of the grid G, and x and y are both greater than 1; the mutual information corresponding to several of the grid division results is calculated respectively, and the maximum mutual information value among the several grid division results is determined, where the formula for the mutual information value is MI * ((X,Y),x,y)=maxMI((X,Y)| G ), where (X,Y)| G is the distribution of data points in the x×y grids of the grid G, MI((X,Y)| G ) is (X,Y)| GThe mutual information; perform normalization processing on the mutual information value, and obtain the corresponding MIC value based on the normalization processing result. The calculation formula for normalization processing is as follows: In the formula, M(X,Y) x,y is the normalization processing result, MI * ((X,Y),x,y) is the mutual information value, and the calculation formula for the MIC value is In the formula, xy < C is the constraint during grid division, aiming to limit the scale of the grid.

[0074] In this way, by calculating the correlation between the power data and the transient stability index, characteristic data that can reflect the transient stability of the power system can be extracted, thereby reducing the data dimension to reduce the complexity of the young model and enhancing the recognition ability of key features, and further facilitating the subsequent rapid and accurate evaluation and prediction of the transient stability of the power system.

[0075] Finally, determine the first characteristic data based on the MIC value. Among them, the characteristics are sorted in descending order according to the MIC value of the characteristics, and the characteristics corresponding to the MIC values that meet the preset threshold are used as the first characteristic data. Among them, the first characteristic data In the formula, V i is the voltage amplitude of node i, θ i is the voltage phase angle of node i, G Pi is the active power output of generator i, G Qi is the reactive power output of generator i, P i-j is the active power of line i-j, Q i-j is the reactive power of line i-j, S i-j is the apparent power of line i-j, P loss i-j is the active power loss of line i-j, Q loss i-j is the reactive power loss of line i-j.

[0076] It should be noted that the value range of MIC is in [0,1]. The larger the MIC value, the stronger the correlation between the two variables, that is, the more information they share, and it is more considered to be retained. Generally speaking, the smaller the MIC value, the weaker the correlation between the two variables, that is, the less information they share, and it is more considered to be excluded; when the MIC value is 0, theoretically it indicates that there is no shared information between this feature and the transient stability index.

[0077] In this way, by analyzing the collected power data, characteristic data that can reflect the transient stability of the power system can be extracted, thereby reducing the data dimension and facilitating the subsequent rapid and accurate evaluation and prediction of the transient stability of the power system.

[0078] Step S102: input the first feature data into a preset LS-CGAN model to obtain second feature data similar to the first feature data;

[0079] It is understandable that after the first feature data is obtained, it is necessary to output it to the LS-CGAN model to obtain simulated power data, wherein the LS-CGAN model includes a generator G and a discriminator D. Figure 2 The figure is a schematic diagram of the data processing flow of the generator G and the discriminator D, specifically: first, the first feature data is input into the preset LS-CGAN model, and the generator G generates a simulated sample G(z) based on the randomly generated noise vector z and the conditional signal c, wherein the noise vector z is the random source of the generated data, and the conditional signal c can be any form of additional information, such as category labels, text descriptions or images, etc., which are used to guide the generator G to generate data that meets specific conditions. The noise vector z and the conditional signal c are fused inside the generator G through a series of network layers (such as fully connected layers, convolutional layers, transposed convolutional layers, etc.) (such as embedding and concatenation, etc.), obtain the target data space representation, that is, generate a simulated sample G(z); secondly, the discriminator D obtains the corresponding distribution probability value D(x|c) based on the simulated sample G(z) and the first feature data x, wherein a series of network layers in the discriminator D extract the features of the simulated sample G(z) and the first feature data x, and distinguish them in combination with the conditional signal c, thereby determining whether the input data is real or simulated; finally, based on the distribution probability value D(x|c), determine the second feature data similar to the first feature data, wherein the objective function of the CGAN model is, Where, E represents mathematical expectation, D(x|c) and G(z|c) represent the probability that the input data x of the discriminator D is a true sample and the probability that the generator G generates a simulated sample after mapping the noise vector z under the conditional signal c, respectively, V is the objective function, E represents mathematical expectation, x is the first feature data, z is the noise vector, c is the conditional signal, p data (x) and P z (z) are the distribution of the first feature data and the noise vector z respectively, and the probability that the simulated sample G(z) generated by D(G(z,y)) is judged as true by the discriminator D. After the least squares loss function is introduced, the objective function of the LS-CGAN model is specifically:

[0080]

[0081] Where V LS-CGAN (D) is the least squares loss function of the discriminator; V LS-CGAN(G) is the least squares loss function of the generator; D is the discriminator; G is the generator; E represents mathematical expectation; x is the first feature data; z is the noise vector; c is the conditional signal; p data (x) and P z (z) are the distribution of the first feature data and the distribution of the noise vector z respectively; G(z) is the simulated sample; D(x) represents the probability that the discriminator D considers the input data x to be true; the simulated sample G(z) generated by D(G(z,y)) is judged to be true by the discriminator D; V is the objective function; a and b are the labels of the simulated sample and the first feature data respectively.

[0082] It should be noted that the simulated sample G(z) should be visually or statistically similar to the real data, ie, the first feature data, and conform to the attributes or categories specified by the conditional signal c.

[0083] It should be noted that, during the discriminator's discriminant process, the conditional signal c may be combined with the characteristics of the data sample (that is, the simulated sample G(z) and the first feature data x) in some way (such as weighting, conditional embedding, etc.). The addition of the conditional signal c realizes the guidance of the generated sample, that is, the second feature data category information.

[0084] It should be noted that the distribution probability value D(x|c) represents the probability that the input data x is real data under the conditional signal y. The closer this probability value is to 1, the greater the possibility that the discriminator believes that x is real data; otherwise, the smaller it is.

[0085] It should be noted that the offline training process of the LS-CGAN model is divided into four stages: time domain simulation, sample generation, feature selection based on MIC, and training model. Specifically, PMU is used to collect real-time operating variables of the power system and fault list data of the time interval when the fault causes system instability. After denoising the data signal through time domain simulation, training samples can be generated. Then, the signal feature correlation analysis and judgment are performed based on MIC, and features that are very similar to the system transient stability indicators are selected to construct key data sets, thereby reducing the data dimension and inputting the LS-CGAN model for offline model training.

[0086] In this way, by inputting the first feature data into the preset LS-CGAN model, the second feature data similar to the first feature data can be accurately simulated, and then the possible operating state of the power system can be simulated, which is convenient for the subsequent GRU model to evaluate and predict the transient stability of the power system.

[0087] Step S103: input the first characteristic data and the second characteristic data into a preset GRU model to obtain a transient stability assessment result of the power system.

[0088] It can be understood that the GRU model processes the time series data corresponding to the first feature data and the second feature data, learns and captures the law of system state changes over time, and then extracts the dynamic characteristics of the power system transient process, and then dynamically adjusts the corresponding weights through the update gate and reset gate in the GRU, and then obtains the transient evaluation result of the power system based on the dynamic characteristics. The structural diagram of the GRU model is shown in Figure 3 As shown, the GRU model includes an update gate z t and reset gate r t , by updating the gate z t and reset gate r t To control the direction of the data flow at time t, the calculation method is: Z = σ (W z ·[h t-1 ,x t ],r t =σ(W r ·[h t-1 ,x t ]), where x t is the input data at time t, W z To update the gate z t The weight, W r Reset gate r t The weight, h t-1 is the output of the hidden layer. When calculating the output of the candidate hidden layer at time t When the historical time, i.e. the output information of the hidden layer, needs to be retained, by adjusting r t The value of controls the historical time information, as shown below: Where W is the weight matrix connecting the input, hidden layer and output layer, which is used to linearly transform the input data and hidden state to determine the update of each gate (update gate, reset gate) and hidden state of the GRU unit, · is the matrix multiplication, ⊙ is the matrix element multiplication; at the same time, through z t To control how much information in the hidden layer is forgotten, how much information in the candidate hidden layer is retained, and the final calculated hidden layer output h t As shown in the formula: In the formula, is the output of the candidate hidden layer at time t, h t-1 is the output of the hidden layer at time t-1. In summary, the output of the GRU model is: In the formula, is the output of the last output layer in the time series, W0 is the weight of the output layer, b0 is the output, h last The output of the hidden layer at the last moment.

[0089] It should be noted that if the reset gate is close to 0, the output of the hidden layer will not be retained, and the GRU model will discard some historical information that is irrelevant to the future. The update gate determines how much of the hidden layer information at time t-1 needs to be saved in the output of the hidden layer at time t. If the element is close to 1, the corresponding information of the hidden layer will be copied to time t. The update gate unit with long-distance dependency is active, so long-distance information can be learned: if the element is close to 0, it is equivalent to a standard RNN, which can process short-distance information. The update gate unit with short-distance dependency is active under this mechanism, and historically significant information can flexibly help predict future information.

[0090] It should be noted that in the actual operation of the power system, due to various complex operating conditions and unpredictable changes, the training of the LS-CGAN model and the GRU model may face adaptability challenges, resulting in the inability of the accuracy of the model to meet the actual needs when evaluating the transient stability of the power system. The operation data of the power system is monitored in real time, and the real-time operation data is integrated with the data generated by the LS-CGAN model to enrich the training data set and improve the generalization ability of the model. The generation ability of LS-CGAN is used to simulate the emerging power system operation state, and these simulated data are used for incremental learning to update the existing model. According to the changes in the operation status of the power system, the input conditions and data sources of the LS-CGAN model are adaptively adjusted, and the original samples and the samples generated by LS-CGAN are used together to train the GRU model. By optimizing the parameters of the model, the GRU model can accurately capture the law of system state changes over time and predict the future stable state of the system.

[0091] By analyzing the collected power data, the embodiment of the present application can extract feature data that can reflect the transient stability of the power system, thereby reducing the data dimension and facilitating the subsequent rapid and accurate evaluation and prediction of the transient stability of the power system; by inputting the first feature data into a preset LS-CGAN model, the second feature data similar to the first feature data can be accurately simulated, and then the possible operating state of the power system can be simulated, which is convenient for the subsequent GRU model to evaluate and predict the transient stability of the power system; by inputting the first feature data and the second feature data into a preset GRU model, the GRU model can learn and capture the law of changes in the operating state of the power system over time, and can quickly and accurately evaluate and predict the transient stability of the power system, thereby ensuring the safe and stable operation of the power system.

[0092] Embodiment 2

[0093] Please refer to Figure 4 , Figure 4It is a structural schematic diagram of an embodiment of a transient stability assessment system based on GRU and LS-CGAN provided by the present application, including: an acquisition module 100, a generation module 200 and an assessment module 300;

[0094] The acquisition module 100 is used to collect power data when the power system is running, and analyze the power data to obtain first characteristic data, wherein the power data includes voltage amplitude, phase angle and power;

[0095] The acquisition module 100 includes: a first calculation unit, a second calculation unit and a first determination unit; the first calculation unit is used to calculate the stability index value corresponding to each power characteristic of the power system after the disturbance through simulation analysis, and determine the transient stability index of the power system based on the stability index value; the second calculation unit is used to calculate the correlation between the power data and the transient stability index through the maximum information coefficient to obtain the MIC value; the first determination unit is used to determine the first characteristic data based on the MIC value.

[0096] It is understandable that the power data during operation can be collected by means of, but is not limited to, sensors or monitoring equipment installed at corresponding locations in the power system, wherein the type of the sensor corresponds to the type of the power data, and the power includes active power and reactive power.

[0097] It is understandable that after obtaining the power data of the power system during operation, the power data needs to be preprocessed, specifically: the format of the power data needs to be converted into a processable format. After obtaining the processable format, the missing values ​​of the power data need to be interpolated, etc., and then the abnormal values ​​in the power data that are obviously inconsistent with other observed values ​​are detected, and the abnormal values ​​are deleted, corrected or marked, etc. Finally, the duplicate data records in the power data are deleted, thereby ensuring the accuracy of the subsequent power data analysis results.

[0098] It is understandable that after preprocessing the power data, it is necessary to analyze the power data. Specifically: First, calculate the corresponding stability index values of each power characteristic of the power system after disturbance through simulation analysis, and determine the transient stability index of the power system based on the stability index values. Among them, it is necessary to use power system simulation software to simulate the disturbances that the system may encounter under various operating modes, and through the simulation results, observe the dynamic response of the power data after disturbance, and then evaluate the transient stability of the system according to the changes of these data and determine the corresponding stability index. Based on the mathematical model and stability theory of the power system, derive the calculation formula of the transient stability index, substitute the actual parameters of the system into the formula, and calculate to obtain the specific stability index value. Finally, compare the calculated stability index value with the critical stability margin of the system to determine the transient stability index of the power system.

[0099] Second, calculate the correlation between the power data and the transient stability index through the maximum information coefficient to obtain the MIC value. Specifically, perform grid division on the power data and the transient stability index to obtain several grid division results G(x-y), where x is the number of rows of grid G, y is the number of columns of grid G, and both x and y are greater than 1; calculate the mutual information corresponding to several grid division results respectively, and determine the maximum mutual information value among several grid division results. Among them, the formula for the mutual information value is MI * ((X,Y),x,y) = maxMI((X,Y)| G )), where (X,Y)| G is the distribution generated by the data points in the x×y small grids of grid G, and MI((X,Y)| G ) is the mutual information of (X,Y)| G ; perform normalization processing on the mutual information value, and obtain the corresponding MIC value based on the normalization processing result. Among them, the calculation formula for normalization processing is: where M(X,Y) x,y is the normalization processing result, MI * ((X,Y),x,y) is the mutual information value, and the calculation formula for the MIC value is where xy < C is the constraint during grid division, and the purpose is to limit the scale of the grid.

[0100] In this way, by calculating the correlation between the power data and the transient stability index, characteristic data that can reflect the transient stability of the power system can be extracted, thereby reducing the data dimension, reducing the complexity of the model, and enhancing the recognition ability of key features, and then facilitating the subsequent rapid and accurate evaluation and prediction of the transient stability of the power system.

[0101] Finally, the first feature data is determined based on the MIC value, wherein the features are sorted in descending order according to their MIC values, and the features corresponding to the MIC values ​​that meet the preset threshold are used as the first feature data, wherein the first feature data Where V i is the voltage amplitude at node i, θ i is the voltage phase angle of node i, G Pi is the active power output of generator i, G Qi is the reactive power output of generator i, P i-j is the active power of line ij, Q i-j is the reactive power of line ij, S i-j is the apparent power of line ij, P loss i-j is the active power loss of line ij, Q loss i-j is the reactive power loss of line ij.

[0102] It should be noted that the value range of MIC is [0,1]. The larger the MIC value, the stronger the correlation between the two variables, that is, the more information they share, and they are more likely to be retained. Generally speaking, the smaller the MIC value, the weaker the correlation between the two variables, that is, the less information they share, and they are more likely to be eliminated. When the MIC value is 0, it theoretically indicates that there is no shared information between the feature and the transient stability index.

[0103] In this way, by analyzing the collected power data, characteristic data that can reflect the transient stability of the power system can be extracted, thereby reducing the data dimension and facilitating the subsequent rapid and accurate evaluation and prediction of the transient stability of the power system.

[0104] The generating module 200 is used to input the first feature data into a preset LS-CGAN model to obtain second feature data similar to the first feature data;

[0105] The generation module 200 includes: a third calculation unit, a fourth calculation unit and a second determination unit; the third calculation unit is used to input the first feature data into a preset LS-CGAN model, and the generator generates a simulated sample based on a randomly generated noise vector and a conditional signal; the fourth calculation unit is used for the discriminator to obtain a corresponding distribution probability value based on the simulated sample and the first feature data; the second determination unit is used to determine second feature data similar to the first feature data based on the distribution probability value.

[0106] It is understandable that after the first feature data is obtained, it is necessary to output it to the LS-CGAN model to obtain simulated power data, wherein the LS-CGAN model includes a generator G and a discriminator D. Figure 2The figure is a schematic diagram of the data processing flow of the generator G and the discriminator D, specifically: first, the first feature data is input into the preset LS-CGAN model, and the generator G generates a simulated sample G(z) based on the randomly generated noise vector z and the conditional signal c, wherein the noise vector z is the random source of the generated data, and the conditional signal c can be any form of additional information, such as category labels, text descriptions or images, etc., which are used to guide the generator G to generate data that meets specific conditions. The noise vector z and the conditional signal c are fused inside the generator G through a series of network layers (such as fully connected layers, convolutional layers, transposed convolutional layers, etc.) (such as embedding and concatenation, etc.), obtain the target data space representation, that is, generate a simulated sample G(z); secondly, the discriminator D obtains the corresponding distribution probability value D(x|c) based on the simulated sample G(z) and the first feature data x, wherein a series of network layers in the discriminator D extract the features of the simulated sample G(z) and the first feature data x, and distinguish them in combination with the conditional signal c, thereby determining whether the input data is real or simulated; finally, based on the distribution probability value D(x|c), determine the second feature data similar to the first feature data, wherein the objective function of the CGAN model is, Where, E represents mathematical expectation, D(x|c) and G(z|c) represent the probability that the input data x of the discriminator D is a true sample and the probability that the generator G generates a simulated sample after mapping the noise vector z under the conditional signal c, respectively, V is the objective function, E represents mathematical expectation, x is the first feature data, z is the noise vector, c is the conditional signal, p data (x) and P z (z) are the distribution of the first feature data and the noise vector z respectively, and the probability that the simulated sample G(z) generated by D(G(z,y)) is judged as true by the discriminator D. After the least squares loss function is introduced, the objective function of the LS-CGAN model is specifically:

[0107]

[0108] Where V LS-CGAN (D) is the least squares loss function of the discriminator; V LS-CGAN (G) is the least squares loss function of the generator; D is the discriminator; G is the generator; E represents mathematical expectation; x is the first feature data; z is the noise vector; c is the conditional signal; p data (x) and P z(z) are the distribution of the first feature data and the distribution of the noise vector z respectively; G(z) is the simulated sample; D(x) represents the probability that the discriminator D considers the input data x to be true; the simulated sample G(z) generated by D(G(z,y)) is judged to be true by the discriminator D; V is the objective function; a and b are the labels of the simulated sample and the first feature data respectively.

[0109] It should be noted that the simulated sample G(z) should be visually or statistically similar to the real data, ie, the first feature data, and conform to the attributes or categories specified by the conditional signal c.

[0110] It should be noted that, during the discriminator's discriminant process, the conditional signal c may be combined with the data sample, that is, the simulated sample G(z) and the features of the first feature data x in some way (such as weighting, conditional embedding, etc.). The addition of the conditional signal c realizes the guidance of the generated sample, that is, the second feature data category information.

[0111] It should be noted that the distribution probability value D(x|c) represents the probability that the input data x is real data under the conditional signal y. The closer this probability value is to 1, the greater the possibility that the discriminator believes that x is real data; otherwise, the smaller it is.

[0112] It should be noted that the offline training process of the LS-CGAN model is divided into four stages: time domain simulation, sample generation, feature selection based on MIC, and training model. Specifically, PMU is used to collect real-time operating variables of the power system and fault list data of the time interval when the fault causes system instability. After denoising the data signal through time domain simulation, training samples can be generated. Then, the signal feature correlation analysis and judgment are performed based on MIC, and features that are very similar to the system transient stability indicators are selected to construct key data sets, thereby reducing the data dimension and inputting the LS-CGAN model for offline model training.

[0113] In this way, by inputting the first feature data into the preset LS-CGAN model, the second feature data similar to the first feature data can be accurately simulated, and then the possible operating state of the power system can be simulated, which is convenient for the subsequent GRU model to evaluate and predict the transient stability of the power system.

[0114] The evaluation module 300 is used to input the first characteristic data and the second characteristic data into a preset GRU model to obtain a transient stability evaluation result of the power system.

[0115] It can be understood that the GRU model processes the time series data corresponding to the first feature data and the second feature data, learns and captures the law of system state changes over time, and then extracts the dynamic characteristics of the power system transient process, and then dynamically adjusts the corresponding weights through the update gate and reset gate in the GRU, and then obtains the transient evaluation result of the power system based on the dynamic characteristics. The structural diagram of the GRU model is shown in Figure 3 As shown, the GRU model includes an update gate z t and reset gate r t , by updating the gate z t and reset gate r t To control the direction of the data flow at time t, the calculation method is: Z = σ (W z ·[h t-1 ,x t ],r t =σ(W r ·[h t-1 ,x t ]), where x t is the input data at time t, W z To update the gate z t The weight, W r Reset gate r t The weight, h t-1 is the output of the hidden layer. When calculating the output of the candidate hidden layer at time t When the historical time, i.e. the output information of the hidden layer, needs to be retained, by adjusting r t The value of controls the historical time information, as shown below: Where W is the weight matrix connecting the input, hidden layer and output layer, which is used to linearly transform the input data and hidden state to determine the update of each gate (update gate, reset gate) and hidden state of the GRU unit, · is the matrix multiplication, ⊙ is the matrix element multiplication; at the same time, through z t To control how much information in the hidden layer is forgotten, how much information in the candidate hidden layer is retained, and the final calculated hidden layer output h t As shown in the formula: In the formula, is the output of the candidate hidden layer at time t, h t-1 is the output of the hidden layer at time t-1. In summary, the output of the GRU model is: In the formula, is the output of the last output layer in the time series, W0 is the weight of the output layer, b0 is the output, h last The output of the hidden layer at the last moment.

[0116] It should be noted that if the reset gate is close to 0, the output of the hidden layer will not be retained, and the GRU model will discard some historical information that is irrelevant to the future. The update gate determines how much of the hidden layer information at time t-1 needs to be saved in the output of the hidden layer at time t. If the element is close to 1, the corresponding information of the hidden layer will be copied to time t. The update gate unit with long-distance dependency is active, so long-distance information can be learned: if the element is close to 0, it is equivalent to a standard RNN, which can process short-distance information. The update gate unit with short-distance dependency is active under this mechanism, and historically significant information can flexibly help predict future information.

[0117] It should be noted that in the actual operation of the power system, due to various complex operating conditions and unpredictable changes, the training of the LS-CGAN model and the GRU model may face adaptability challenges, resulting in the inability of the accuracy of the model to meet the actual needs when evaluating the transient stability of the power system. The operation data of the power system is monitored in real time, and the real-time operation data is integrated with the data generated by the LS-CGAN model to enrich the training data set and improve the generalization ability of the model. The generation ability of LS-CGAN is used to simulate the emerging power system operation state, and these simulated data are used for incremental learning to update the existing model. According to the changes in the operation status of the power system, the input conditions and data sources of the LS-CGAN model are adaptively adjusted, and the original samples and the samples generated by LS-CGAN are used together to train the GRU model. By optimizing the parameters of the model, the GRU model can accurately capture the law of system state changes over time and predict the future stable state of the system.

[0118] By analyzing the collected power data, the embodiment of the present application can extract feature data that can reflect the transient stability of the power system, thereby reducing the data dimension and facilitating the subsequent rapid and accurate evaluation and prediction of the transient stability of the power system; by inputting the first feature data into a preset LS-CGAN model, the second feature data similar to the first feature data can be accurately simulated, and then the possible operating state of the power system can be simulated, which is convenient for the subsequent GRU model to evaluate and predict the transient stability of the power system; by inputting the first feature data and the second feature data into a preset GRU model, the GRU model can learn and capture the law of changes in the operating state of the power system over time, and can quickly and accurately evaluate and predict the transient stability of the power system, thereby ensuring the safe and stable operation of the power system.

[0119] The device embodiments described above are merely illustrative, wherein the modules described as separate components may or may not be physically separated, i.e., may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the method of this embodiment.

[0120] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-monitorable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0121] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the scope of protection of the present application.

[0122] It is particularly pointed out that for those skilled in the art, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A transient stability assessment method based on GRU and LS-CGAN, characterized in that: include: Collecting power data when the power system is running, and analyzing the power data to obtain first characteristic data, wherein the power data includes voltage amplitude, phase angle and power; Inputting the first feature data into a preset LS-CGAN model to obtain second feature data similar to the first feature data; The first characteristic data and the second characteristic data are input into a preset GRU model to obtain a transient stability assessment result of the power system.

2. The transient stability assessment method based on GRU and LS-CGAN according to claim 1, characterized in that: The power data is analyzed to obtain the first characteristic data, specifically: Calculate the stability index value corresponding to each power characteristic of the power system after the disturbance through simulation analysis, and determine the transient stability index of the power system based on the stability index value; Calculate the correlation between the power data and the transient stability index by using the maximum information coefficient to obtain a MIC value; The first characteristic data is determined based on the MIC value.

3. The transient stability assessment method based on GRU and LS-CGAN according to claim 2 is characterized in that: The correlation between the power data and the transient stability index is calculated by the maximum information coefficient to obtain the MIC value, which is specifically: Performing grid division on the power data and the transient stability index to obtain a plurality of grid division results; Calculating the mutual information corresponding to the plurality of grid division results respectively, and determining the maximum mutual information value among the plurality of grid division results; The mutual information value is normalized, and a corresponding MIC value is obtained based on the normalization result.

4. The transient stability assessment method based on GRU and LS-CGAN according to claim 1, characterized in that: The first feature data is input into a preset LS-CGAN model to obtain second feature data similar to the first feature data, wherein the LS-CGAN model includes a generator and a discriminator, specifically: The first feature data is input into a preset LS-CGAN model, and the generator generates a simulated sample based on a randomly generated noise vector and a conditional signal; The discriminator obtains a corresponding distribution probability value based on the simulated sample and the first feature data; Based on the distribution probability value, second feature data similar to the first feature data is determined.

5. The transient stability assessment method based on GRU and LS-CGAN according to claim 4, characterized in that: The objective function of the LS-CGAN model is specifically: Where V LS-CGAN (D) is the least squares loss function of the discriminator; V LS-CGAN (G) is the least squares loss function of the generator; D is the discriminator; G is the generator; E represents mathematical expectation; x is the first feature data; z is the noise vector; c is the conditional signal; p data (x) and P z (z) are the distribution of the first feature data and the distribution of the noise vector z respectively; G(z) is the simulated sample; D(x) represents the probability that the discriminator D considers the input data x to be true; the simulated sample G(z) generated by D(G(z,y)) is judged to be true by the discriminator D; V is the objective function; a and b are the labels of the simulated sample and the first feature data respectively.

6. A transient stability assessment system based on GRU and LS-CGAN, characterized in that: include: Acquisition module, generation module and evaluation module; The acquisition module is used to collect power data when the power system is running, and analyze the power data to obtain first characteristic data, wherein the power data includes voltage amplitude, phase angle and power; The generating module is used to input the first feature data into a preset LS-CGAN model to obtain second feature data similar to the first feature data; The evaluation module is used to input the first characteristic data and the second characteristic data into a preset GRU model to obtain a transient stability evaluation result of the power system.

7. The transient stability assessment system based on GRU and LS-CGAN according to claim 6, characterized in that: The acquisition module includes: a first calculation unit, a second calculation unit and a first determination unit; The first calculation unit is used to calculate the stability index value corresponding to each power characteristic of the power system after the disturbance through simulation analysis, and determine the transient stability index of the power system based on the stability index value; The second calculation unit is used to calculate the correlation between the power data and the transient stability index by using a maximum information coefficient to obtain a MIC value; The first determining unit is configured to determine the first characteristic data based on the MIC value.

8. The transient stability assessment system based on GRU and LS-CGAN according to claim 7, characterized in that: The second calculation unit is specifically: Performing grid division on the power data and the transient stability index to obtain a plurality of grid division results; Calculating the mutual information corresponding to the plurality of grid division results respectively, and determining the maximum mutual information value among the plurality of grid division results; The mutual information value is normalized, and a corresponding MIC value is obtained based on the normalization result.

9. The transient stability assessment system based on GRU and LS-CGAN according to claim 6, characterized in that: The generating module comprises: a third calculating unit, a fourth calculating unit and a second determining unit; The third computing unit is used to input the first feature data into a preset LS-CGAN model, and the generator generates a simulated sample based on the randomly generated noise vector and the conditional signal; The fourth calculation unit is used for the discriminator to obtain a corresponding distribution probability value based on the simulated sample and the first feature data; The second determining unit is configured to determine second feature data similar to the first feature data based on the distribution probability value.

10. The transient stability assessment system based on GRU and LS-CGAN according to claim 9, characterized in that: The objective function of the LS-CGAN model is specifically: Where V LS-CGAN (D) is the least squares loss function of the discriminator; V LS-CGAN (G) is the least squares loss function of the generator; D is the discriminator; G is the generator; E represents mathematical expectation; x is the first feature data; z is the noise vector; c is the conditional signal; p data (x) and P z (z) are the distribution of the first feature data and the distribution of the noise vector z respectively; G(z) is the simulated sample; D(x) represents the probability that the discriminator D considers the input data x to be true; the simulated sample G(z) generated by D(G(z,y)) is judged to be true by the discriminator D; V is the objective function; a and b are the labels of the simulated sample and the first feature data respectively.

Images