Method and device for generating high-dimensional security boundary of power system adaptive to online state

Abstract

The invention discloses a power system high-dimensional security boundary generation method and device adaptive to an online state, and the method comprises the steps: carrying out the calculation through a historical data driving strategy and a least square fitting strategy, obtaining a low-dimensional security boundary of a set branch of a whole network, and generating a security boundary set; an AdaBoost.M2 classifier is trained through historical data and the security boundary set, and an out-of-limit risk prediction model is generated; inputting a set online operation state into the out-of-limit risk prediction model for prediction processing, and outputting a possible out-of-limit boundary and a corresponding out-of-limit risk value; and inputting the possible out-of-limit boundary and the corresponding out-of-limit risk value into a generative adversarial network for training processing, and outputting a comprehensive high-dimensional security boundary. According to the method, the high-dimensional security boundary generation efficiency can be greatly improved on the premise of ensuring the accuracy, and efficient and reliable technical support is provided for real-time monitoring, scheduling and decision making of a power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal stability safety domain of power system, and in particular to a method and device for generating a high-dimensional safety boundary of an online power system. Background Art

[0002] With the continuous expansion of the scale of power systems and the widespread access to renewable energy, the operation of power grids has become increasingly complex. Traditional low-dimensional safety boundary methods are difficult to meet the requirements of efficiency and accuracy when facing variable real-time operating conditions. In the planning, design, and dispatching of power systems, if the safety boundary model used has a large deviation, it will directly affect the dispatcher's operation plan, resulting in a waste of resources and even endangering the stable operation of the system. It is of great practical significance to establish a safety boundary model that can accurately reflect the online state characteristics of the power grid in real time.

[0003] At present, in order to meet these challenges, researchers have proposed a method for generating high-dimensional safety boundaries, aiming to more comprehensively describe the safety of power grids under complex working conditions. Existing boundary generation methods often rely on offline analysis of historical data, but due to the dynamic changes of power systems, traditional methods have great limitations when dealing with online conditions.

[0004] Therefore, how to quickly adapt to the real-time status of the power grid and accurately generate high-dimensional safety boundaries has become an important topic of current research. Summary of the invention

[0005] To this end, the present invention provides a method and device for generating a high-dimensional safety boundary for an online power system, which quickly generates a high-dimensional safety boundary for the entire network by combining ensemble learning with a generative adversarial network (GAN). While ensuring accuracy, the efficiency of generating high-dimensional safety boundaries is greatly improved, providing efficient and reliable technical support for real-time monitoring, dispatching and decision-making of power grids.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a method for generating a high-dimensional safety boundary of an online power system, comprising:

[0007] Through the historical data driven strategy and the least squares fitting strategy, the low-dimensional safety boundary of the set branches of the whole network is obtained and the safety boundary set is generated;

[0008] The AdaBoost.M2 classifier is trained using historical data and the safety boundary set to generate a limit crossing risk prediction model;

[0009] Inputting the set online operation status into the over-limit risk prediction model for prediction processing, and outputting the possible over-limit boundary and the corresponding over-limit risk value;

[0010] The possible crossing boundary and the corresponding crossing risk value are input into a generative adversarial network for training processing, and a comprehensive high-dimensional safety boundary is output.

[0011] As a preferred solution for the method of generating high-dimensional safety boundaries of online power systems, in the process of obtaining the low-dimensional safety boundaries of branches set in the entire network, the key nodes of the branches are determined according to the size of the branch power flow sensitivity; based on the key nodes, the low-dimensional safety boundaries of the branches are obtained through the least squares fitting strategy; the low-dimensional safety boundary expression of the branch is:

[0012]

[0013] Where, M is the number of key nodes in the branch; P i is the power injection of node i; j is the branch; c j is the constant term corresponding to the boundary of branch j; is the upper limit of the active power flow of branch j; B is the branch set; β ij is the corresponding coefficient of node i at the safety boundary of branch j.

[0014] As a preferred solution for the method of generating a high-dimensional safety boundary of an online power system, the steps of generating the over-limit risk prediction model by training the AdaBoost.M2 classifier are as follows:

[0015] Initialize the sample weights and error label weights of the training samples;

[0016] Set the maximum number of iterations and calculate the sum of the error label weights, the error label weighting function value and the weight in each iteration;

[0017] The weak hypothesis is calculated for each iteration;

[0018] Calculate the pseudo loss; if the pseudo loss is less than the set threshold, calculate the weight of the base classifier; if the pseudo loss is not less than the set threshold, perform the final step;

[0019] Through iterative calculation, the weights of the wrong labels of the training samples are updated;

[0020] The confidence matrix is ​​superimposed according to the weights of the base classifiers to obtain a final confidence matrix of the integrated classifier; and a final selection result is output according to the final confidence matrix.

[0021] As a preferred scheme for the method of generating high-dimensional safety boundaries of online power systems, in the process of inputting the possible crossing boundaries and the corresponding crossing risk values ​​into the generative adversarial network for processing, the possible crossing boundaries of different branches are expanded according to the confidence ratio to generate real data; the generative adversarial network learns and processes the real data and outputs the comprehensive high-dimensional safety boundary.

[0022] As a preferred solution of a method for generating a high-dimensional safety boundary of an online power system, in the process of training the generative adversarial network and outputting the comprehensive high-dimensional safety boundary, the training processing steps are:

[0023] The generator is fixed; the discriminator is trained and improved through the random boosting gradient strategy to obtain the improved discriminator;

[0024] Fixing the improved judgement device; training and improving the generator through a random descent gradient strategy; obtaining the improved generator;

[0025] Performing iterative game training on the upgraded judger and the upgraded generator;

[0026] When the upgraded judge and the upgraded generator reach "Nash equilibrium", the result generated by the upgraded generator is used as output to obtain the comprehensive high-dimensional security boundary.

[0027] The present invention also provides a device for generating a high-dimensional safety boundary of an online power system, based on the above method for generating a high-dimensional safety boundary of an online power system, comprising:

[0028] The safety boundary set generation module is used to calculate through the historical data driving strategy and the least squares fitting strategy to obtain the low-dimensional safety boundary of the set branches of the entire network and generate a safety boundary set;

[0029] A limit crossing risk prediction model generation module is used to train the AdaBoost.M2 classifier through historical data and the safety boundary set to generate a limit crossing risk prediction model;

[0030] An over-limit risk prediction model processing module is used to input the set online operation status into the over-limit risk prediction model for prediction processing, and output possible over-limit boundaries and corresponding over-limit risk values;

[0031] The comprehensive high-dimensional safety boundary acquisition module is used to input the possible crossing boundary and the corresponding crossing risk value into the generative adversarial network for training processing and output a comprehensive high-dimensional safety boundary.

[0032] As a preferred solution of a high-dimensional safety boundary generation device adapted to an online power system, in the safety boundary set generation module, in the process of obtaining the low-dimensional safety boundary of the set branch of the whole network, the key nodes of the branch are determined according to the size of the branch power flow sensitivity; according to the key nodes, the low-dimensional safety boundary of the branch is obtained through the least squares fitting strategy; the low-dimensional safety boundary expression of the branch is:

[0033]

[0034] Where, M is the number of key nodes in the branch; P i is the power injection of node i; j is the branch; c j is the constant term corresponding to the boundary of branch j; is the upper limit of the active power flow of branch j; B is the branch set; β ij is the corresponding coefficient of node i at the safety boundary of branch j.

[0035] As a preferred solution of a high-dimensional safety boundary generation device adapted to an online power system, in the over-limit risk prediction model generation module, the submodule for generating the over-limit risk prediction model includes:

[0036] The weight initialization submodule is used to initialize the sample weights and error label weights of the training samples;

[0037] The iteration number setting and weight calculation submodule is used to set the maximum number of iterations and calculate the error label weight sum, error label weighting function value and weight in each iteration;

[0038] A weak hypothesis calculation submodule is used to calculate the weak hypothesis of each iteration;

[0039] The pseudo-loss calculation and determination submodule is used to calculate the pseudo-loss; if the pseudo-loss is less than the set threshold, the weight of the base classifier is calculated; if the pseudo-loss is not less than the set threshold, the last step is performed;

[0040] The error label weight update submodule is used to update the error label weights of training samples through iterative calculation;

[0041] The out-of-limit risk prediction model generation submodule is used to superimpose the confidence matrix according to the weights of the base classifiers to obtain the final confidence matrix of the integrated classifier; and output the final selection result according to the final confidence matrix.

[0042] As a preferred solution for a high-dimensional safety boundary generation device that is adapted to an online power system, in the comprehensive high-dimensional safety boundary acquisition module, while inputting the possible out-of-limit boundaries and the corresponding out-of-limit risk values ​​into the generative adversarial network for processing, the possible out-of-limit boundaries of different branches are expanded according to the confidence ratio to generate real data; the generative adversarial network performs learning processing on the real data and outputs the comprehensive high-dimensional safety boundary.

[0043] As a preferred solution for a high-dimensional safety boundary generation device adapted to an online power system, in the comprehensive high-dimensional safety boundary acquisition module, the submodule for training the generative adversarial network includes:

[0044] The generator fixing and judgement improving submodule is used to fix the generator; the judgement is trained and improved through the random improvement gradient strategy to obtain the improved judgement;

[0045] The submodule of fixing the judger and improving the generator is used to fix the improved judger; train and improve the generator through the random descent gradient strategy; and obtain the improved generator;

[0046] An iterative game training submodule, used for performing iterative game training on the upgraded judger and the upgraded generator;

[0047] The comprehensive high-dimensional security boundary output submodule is used to output the result generated by the upgraded generator when the upgraded judge and the upgraded generator reach the "Nash equilibrium" to obtain the comprehensive high-dimensional security boundary.

[0048] The present invention has the following advantages: the present invention calculates through the historical data driven strategy and the least squares fitting strategy, obtains the low-dimensional safety boundary of the set branch of the whole network, and generates a safety boundary set; trains the AdaBoost.M2 classifier through the historical data and the safety boundary set to generate an over-limit risk prediction model; inputs the set online operation state into the over-limit risk prediction model for prediction processing, outputs the possible over-limit boundary and the corresponding over-limit risk value; inputs the possible over-limit boundary and the corresponding over-limit risk value into the generative adversarial network for training processing, and outputs a comprehensive high-dimensional safety boundary. The present invention effectively improves the real-time, accuracy and efficiency of the generation of power grid safety boundaries by combining ensemble learning and generative adversarial network (GAN) technology, and overcomes the limitations of traditional low-dimensional safety boundary methods. Through the over-limit risk prediction model based on AdaBoost.M2, the present invention can accurately identify multiple boundaries that may be over-limited and reduce the risk of misjudgment; at the same time, the innovative concept of "boundary training boundary" is proposed, and the high-dimensional safety boundary that integrates multiple over-limit situations is quickly generated through GAN, which optimizes the real-time monitoring and dispatching decision-making of the power grid, improves the stability and safety of the power grid, and provides more reliable technical support for the operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0050] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0051] Figure 1 A schematic flow chart of a method for generating a high-dimensional safety boundary of an online power system provided in Embodiment 1 of the present invention;

[0052] Figure 2 A schematic diagram of a specific implementation process of a method for generating a high-dimensional safety boundary of an online power system provided in Embodiment 1 of the present invention;

[0053] Figure 3 This is a schematic diagram of comparing the accuracy of different classifiers in a possible embodiment provided in Embodiment 1 of the present invention;

[0054] Figure 4 A schematic diagram of data error comparison between the original low-dimensional boundary and the generated high-dimensional boundary in a possible embodiment provided in Embodiment 1 of the present invention;

[0055] Figure 5 This is a schematic diagram of the architecture of a device for generating a high-dimensional safety boundary for an online power system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0056] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Example 1

[0058] See also Figure 1 and Figure 2 Embodiment 1 of the present invention provides a method for generating a high-dimensional safety boundary of an online power system, comprising the following steps:

[0059] S1. Calculate through historical data driven strategy and least squares fitting strategy to obtain low-dimensional safety boundaries of branches set in the whole network and generate safety boundary sets;

[0060] S2, training the AdaBoost.M2 classifier using historical data and the safety boundary set to generate a limit crossing risk prediction model;

[0061] S3, inputting the set online operation status into the over-limit risk prediction model for prediction processing, and outputting the possible over-limit boundary and the corresponding over-limit risk value;

[0062] S4. Input the possible crossing boundary and the corresponding crossing risk value into a generative adversarial network for training and processing, and output a comprehensive high-dimensional safety boundary.

[0063] In this embodiment, in step S1, the historical data driven strategy and the least squares fitting strategy are used to calculate, obtain the low-dimensional safety boundary of the set branch of the whole network, and generate a safety boundary set;

[0064] Among them, in the process of obtaining the low-dimensional safety boundary of the set branch of the whole network, by sorting the absolute value of the branch flow sensitivity, some nodes are selected according to the absolute value of the sensitivity as the spatial coordinates for dimensionality reduction. Taking branch j as an example, the system has a total of N nodes, of which 1~M are the key nodes of the branch, and M+1~N are non-key nodes. Then the original safety boundary of branch j can be obtained by the least squares fitting method:

[0065]

[0066] Where, M is the number of key nodes in the branch; P i is the power injection of node i; j is the branch; c j is the constant term corresponding to the boundary of branch j; is the upper limit of the active power flow of branch j; B is the branch set; β ij is the corresponding coefficient of node i at the safety boundary of branch j.

[0067] In this embodiment, the safety boundary of the entire power system is fitted to obtain a large number of safety boundary sets. However, the key nodes that are sensitive to different lines are different, making it very difficult for the learner to capture and distinguish each boundary and its own line. In order to facilitate the subsequent work, it is necessary to unify the variables of the entire boundary. It needs to be expanded back to the original boundary dimension. Since most power systems are very large, the non-critical node coefficient α ij It can take a minimum value or even 0, so now the safety margin can be expressed as:

[0068]

[0069] In the formula, α ij is the non-critical node coefficient.

[0070] In this embodiment, in step S2, the AdaBoost.M2 classifier is trained using historical data and the safety boundary set to generate a limit crossing risk prediction model;

[0071] Specifically, the boundary of the entire network is used as the training label Y, and the active power injection and line flow that may cause it to operate in large quantities are used as the training input X. The ensemble learning AdaBoost.M2 algorithm is introduced, and the excellent processing ability of decision trees for large amounts of high-dimensional data is used. The base classifier of this algorithm is used to construct a safety boundary selection model based on AdaBoost.M2. Multiple base classifiers are integrated in this model. According to the training results of the current base classifier, the weight distribution of subsequent training samples is adaptively corrected, so that misclassified samples receive more attention, thereby improving the judgment accuracy of easily confused samples and the stability of the model.

[0072] The steps of generating the limit crossing risk prediction model are as follows:

[0073] S21, initializing the sample weights and error label weights of the training samples;

[0074] Specifically, let the training sample set be S = {x i ,y i}, i = 1, 2, ..., n, where x i is the input of the i-th sample, y i ∈Y is its corresponding label, Y={1,2,…,m} is the label set; initialize the sample weight D1(i) and the error label weight

[0075]

[0076] D1(i)=1 / n

[0077] S22, setting the maximum number of iterations, and calculating the error label weight and the error label weighting function value and the weight in each iteration;

[0078] Specifically, set the maximum number of iterations K, calculate the weight of the wrong label and W in the kth iteration i k , the weighted function value of the wrong label q k (i,y) and weight D k (i)

[0079]

[0080] S23, calculating the weak hypothesis for each iteration;

[0081] Specifically, calculate the weak hypothesis h of the kth iteration k , I is the base learning algorithm selected for this round of iteration;

[0082] h k =I(S,D k )

[0083] S24, calculating the pseudo loss; if the pseudo loss is less than the set threshold, calculating the weight of the base classifier; if the pseudo loss is not less than the set threshold, performing the last step of processing;

[0084] Specifically, calculate the pseudo loss E k , if ε k <0.5, then calculate the weight β of the base classifier k , otherwise go to step S26;

[0085]

[0086] S25, updating the erroneous label weights of the training samples through iterative calculation;

[0087] Specifically, update the wrong label weight of the sample If k+1≤K, return to step S22, otherwise go to step S26;

[0088]

[0089] S26, superimposing the confidence matrix according to the weights of the base classifiers to obtain a final confidence matrix of the integrated classifier; and outputting a final selection result according to the final confidence matrix.

[0090] Specifically, the confidence matrix h k (x,Y) according to their respective weights β k The final confidence matrix of the integrated classifier is obtained by superposition, and the final selection result is output based on the matrix.

[0091]

[0092] Where H(x) is the final confidence matrix of the ensemble classifier.

[0093] In this embodiment, in step S3, the online operation state is set to be input into the over-limit risk prediction model for prediction processing, and the possible over-limit boundary and the corresponding over-limit risk value are output;

[0094] Specifically, the set online operation state is input into the over-limit risk prediction model for prediction processing. In order to fully consider the possibility of over-limit in the current state, the evaluation of the confidence matrix can be fully utilized to enable the classifier to sort the final confidence corresponding to each boundary. For entering a certain online operation state, the largest top k boundaries and their confidences [H1(x),H2(x),…,H k (x)], it can prevent the classifier from capturing imperceptible biases in the training data and causing obvious errors and deviations.

[0095] In this embodiment, in step S4, the possible crossing boundary and the corresponding crossing risk value are input into a generative adversarial network for training processing, and a comprehensive high-dimensional safety boundary is output.

[0096] Specifically, the output of the largest first k boundaries and their confidence [H1(x),H2(x),…,H k(x)], forming a boundary data set. The selected boundaries are likely to come from multiple different lines, and the key nodes of the low-dimensional safety boundaries of these lines are very different. Such differences are difficult to intuitively reflect the distance between the operating state and the actual out-of-limit operating state. In order to facilitate GAN to learn the characteristics of each boundary according to the confidence ratio, these boundaries are expanded according to the confidence ratio to form real data for GAN learning.

[0097] In this embodiment, the thermal stability safety boundaries of multiple branches and multiple key node parameters in the current operating state are analyzed by generating an adversarial network. The characteristics of these boundaries and their confidence are comprehensively considered, and the coefficients of each sensitive node that may cause the limit to be exceeded are integrated, and multiple boundaries are trained to generate a safety boundary that more intuitively expresses the possible limit-exceeding situation.

[0098] The training process steps are:

[0099] S41, fixing the generator; training and improving the discriminator through a random boosting gradient strategy to obtain an improved discriminator;

[0100] Specifically, we hope that the discriminator can separate the real data and the generated data as much as possible. The real data is input1 (derived from the original data), and the generated data is input2 (derived from the generator G). The normal output of input1 is 1, and the normal output of input2 is 0. For training a discriminator D, first fix the generator, and then generate a batch of samples and mix them with the real data for the discriminator to judge. Improve the discriminator by stochastic boosting gradient:

[0101]

[0102] In the formula, x is the real data input; z is the generated data; m is the data dimension;

[0103] The trained discriminator becomes stronger, i.e. the generator is fixed and the discriminator is trained.

[0104] S42, fixing the improved decision device; training and improving the generator through a random descent gradient strategy; obtaining an improved generator;

[0105] Specifically, fix the discriminator and try to confuse it with fake data. The discriminator that has just been trained is very powerful. At this time, you need to adjust the generator to confuse the discriminator. That is, by fixing the discriminator and training the generator, you can improve the generator by stochastic descent gradient:

[0106]

[0107] Finally, the final output of the generated data is also 1.

[0108] S43, performing iterative game training on the upgraded judger and the upgraded generator;

[0109] Specifically, in the process of continuous generation and identification of boundary features, the "zero-sum game" problem formed by G and D is a maximum-minimum problem, which is defined as a function V(D,G) and can be expressed as:

[0110]

[0111] In the formula, x is the real data input; z is the generated data; x~p data (x) indicates that the real data is correctly classified; z~p z (z) indicates that the generated data is correctly classified; Represents the expected probability of correctly distinguishing the real data x; represents the expected probability of misclassification of generated data z.

[0112] In the "game" between G and D, the generator has no effect on the real data, so the first term of the formula is a constant; when the second term of the formula is the largest, D(G(z))→0, indicating that the probability that the generated data is real data is close to 0; when the second term of the formula is the smallest, D(G(z))→1, indicating that the probability that the generated data is true is close to 1.

[0113] S44. When the upgraded judge and the upgraded generator reach "Nash equilibrium", the result generated by the upgraded generator is used as output to obtain the comprehensive high-dimensional security boundary.

[0114] Specifically, after sufficient game and training, G and D reach "Nash equilibrium", at which point the result generated by the generator can be considered as the output. In this way, a new boundary can be generated by integrating the features in the data set.

[0115] In a possible embodiment, a specific example of generating a high-dimensional safety boundary of a power system is provided as follows:

[0116] Taking IEEE 39 as an example, the feasibility and effectiveness of the present invention are demonstrated. In IEEE 39, all safety boundaries are used as training labels Y, which are matched with a large amount of operating status data X as input. In order to evaluate the accuracy of the model in predicting boundary violations, 5,000 known operating data sets that are close to exceeding the boundaries are tested under 5 classification methods. The accuracy is as follows: Figure 3As shown. The boundaries where most power grid operating states may exceed the limit are unknown, and multiple lines may exceed the limit. In the results given by the over-limit risk prediction model, multiple locally similar peaks may appear. The present invention selects an operating state data with two similar over-limit risk peaks. Theoretically, in this case, the features of the two boundaries should be fully learned and mapped to the newly generated boundary by GAN, so that the boundary can effectively assist power grid operation scheduling and decision-making. The comparison diagram of safety boundary errors constructed by different methods is shown in Figure 4 shown.

[0117] In summary, the present invention calculates through the historical data driven strategy and the least squares fitting strategy, obtains the low-dimensional safety boundary of the set branch of the whole network, and generates a safety boundary set; trains the AdaBoost.M2 classifier through historical data and the safety boundary set to generate an over-limit risk prediction model; inputs the set online operation state into the over-limit risk prediction model for prediction processing, outputs the possible over-limit boundary and the corresponding over-limit risk value; inputs the possible over-limit boundary and the corresponding over-limit risk value into the generative adversarial network for training processing, and outputs a comprehensive high-dimensional safety boundary. The present invention effectively improves the real-time, accuracy and efficiency of power grid safety boundary generation by combining ensemble learning and generative adversarial network (GAN) technology, and overcomes the limitations of traditional low-dimensional safety boundary methods. Through the over-limit risk prediction model based on AdaBoost.M2, the present invention can accurately identify multiple boundaries that may be over-limited and reduce the risk of misjudgment; at the same time, the innovative concept of "boundary training boundary" is proposed, and a high-dimensional safety boundary that integrates multiple over-limit situations is quickly generated through GAN, which optimizes the real-time monitoring and dispatching decision-making of the power grid, improves the stability and safety of the power grid, and provides more reliable technical support for the operation of the power grid.

[0118] It should be noted that the method of the embodiment of the present disclosure can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present disclosure, and the multiple devices will interact with each other to complete the described method.

[0119] It should be noted that the above describes some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0120] Example 2

[0121] See also Figure 5 Embodiment 2 of the present invention further provides a device for generating a high-dimensional safety boundary of an online power system, comprising:

[0122] The safety boundary set generation module 001 is used to calculate through the historical data driving strategy and the least squares fitting strategy to obtain the low-dimensional safety boundary of the set branch of the whole network and generate the safety boundary set;

[0123] The over-limit risk prediction model generation module 002 is used to train the AdaBoost.M2 classifier through historical data and the safety boundary set to generate an over-limit risk prediction model;

[0124] The over-limit risk prediction model processing module 003 is used to input the set online operation status into the over-limit risk prediction model for prediction processing, and output the possible over-limit boundary and the corresponding over-limit risk value;

[0125] The comprehensive high-dimensional safety boundary acquisition module 004 is used to input the possible crossing boundary and the corresponding crossing risk value into the generative adversarial network for training processing, and output a comprehensive high-dimensional safety boundary.

[0126] In this embodiment, in the safety boundary set generation module 001, in the process of obtaining the low-dimensional safety boundary of the branch set in the whole network, the key nodes of the branch are determined according to the size of the branch power flow sensitivity; according to the key nodes, the low-dimensional safety boundary of the branch is obtained by the least squares fitting strategy; the low-dimensional safety boundary expression of the branch is:

[0127]

[0128] Where, M is the number of key nodes in the branch; P i is the power injection of node i; j is the branch; c j is the constant term corresponding to the boundary of branch j; is the upper limit of the active power flow of branch j; B is the branch set; β ij is the corresponding coefficient of node i at the safety boundary of branch j.

[0129] In this embodiment, in the limit crossing risk prediction model generation module 002, the submodules for generating the limit crossing risk prediction model include:

[0130] The weight initialization submodule 021 is used to initialize the sample weights and error label weights of the training samples;

[0131] Iteration number setting and weight calculation submodule 022, used to set the maximum number of iterations, and calculate the error label weight sum, error label weighting function value and weight in each iteration;

[0132] A weak hypothesis calculation submodule 023 is used to calculate the weak hypothesis of each iteration;

[0133] The pseudo-loss calculation and determination submodule 024 is used to calculate the pseudo-loss; if the pseudo-loss is less than the set threshold, the weight of the base classifier is calculated; if the pseudo-loss is not less than the set threshold, the last step is performed;

[0134] The error label weight updating submodule 025 is used to update the error label weights of the training samples through iterative calculation;

[0135] The out-of-limit risk prediction model generation submodule 026 is used to superimpose the confidence matrix according to the weights of the base classifiers to obtain the final confidence matrix of the integrated classifier; and output the final selection result according to the final confidence matrix.

[0136] In this embodiment, in the comprehensive high-dimensional safety boundary acquisition module 004, when the possible crossing boundary and the corresponding crossing risk value are input into the generative adversarial network for processing, the possible crossing boundary of different branches is expanded according to the confidence ratio to generate real data; the generative adversarial network learns and processes the real data and outputs the comprehensive high-dimensional safety boundary.

[0137] In this embodiment, in the comprehensive high-dimensional security boundary acquisition module 004, the submodule for training the generative adversarial network includes:

[0138] The generator fixing and judgement improving submodule 041 is used to fix the generator; train and improve the judgement through the random improvement gradient strategy to obtain the improved judgement;

[0139] The judgement device fixing and generator improvement submodule 042 is used to fix the improved judgement device; train and improve the generator through a random descent gradient strategy; and obtain the improved generator;

[0140] An iterative game training submodule 043, used for performing iterative game training on the upgraded judger and the upgraded generator;

[0141] The comprehensive high-dimensional security boundary output submodule 044 is used to output the result generated by the upgraded generator when the upgraded judge and the upgraded generator reach the "Nash equilibrium" to obtain the comprehensive high-dimensional security boundary.

[0142] It should be noted that the information interaction, execution process and other contents between the modules of the above-mentioned system are based on the same concept as the method embodiment in Example 1 of the present application, and the technical effects they bring are the same as those of the method embodiment of the present application. For specific contents, please refer to the description in the method embodiment shown above in the present application, and will not be repeated here.

[0143] Example 3

[0144] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, in which a program code for a method for generating a high-dimensional safety boundary of an online power system is stored. The program code includes instructions for executing embodiment 1 or any possible implementation thereof.

[0145] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0146] Example 4

[0147] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0148] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute a method for generating a high-dimensional safety boundary for an online power system that is adaptable to embodiment 1 or any possible implementation thereof.

[0149] Specifically, the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor implemented by reading software codes stored in a memory. The memory can be integrated in the processor or can be located outside the processor and exist independently.

[0150] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions can be stored in a computer-readable storage medium, or transmitted from a computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.

[0151] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing system, they can be concentrated on a single computing system, or distributed on a network composed of multiple computing systems, and optionally, they can be implemented by a program code executable by a computing system, so that they can be stored in a storage system and executed by the computing system, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0152] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for generating a high-dimensional safety boundary of an online power system, characterized in that: include: Through the historical data driven strategy and the least squares fitting strategy, the low-dimensional safety boundary of the set branches of the whole network is obtained and the safety boundary set is generated; The AdaBoost.M2 classifier is trained using historical data and the safety boundary set to generate a limit crossing risk prediction model; Inputting the set online operation status into the over-limit risk prediction model for prediction processing, and outputting the possible over-limit boundary and the corresponding over-limit risk value; The possible crossing boundary and the corresponding crossing risk value are input into a generative adversarial network for training processing, and a comprehensive high-dimensional safety boundary is output.

2. A method for generating a high-dimensional safety boundary of an online power system according to claim 1, characterized in that: In the process of obtaining the low-dimensional safety boundary of the set branch of the whole network, the key nodes of the branch are determined according to the size of the branch power flow sensitivity; according to the key nodes, the low-dimensional safety boundary of the branch is obtained through the least squares fitting strategy; the low-dimensional safety boundary expression of the branch is: Where, M is the number of key nodes in the branch; P i is the power injection of node i; j is the branch; c j is the constant term corresponding to the boundary of branch j; is the upper limit of active power flow of branch j; B is the branch set; β ij is the corresponding coefficient of node i at the safety boundary of branch j.

3. A method for generating a high-dimensional safety boundary of an online power system according to claim 2, characterized in that: By training the AdaBoost.M2 classifier, the steps of generating the limit crossing risk prediction model are: Initialize the sample weights and error label weights of the training samples; Set the maximum number of iterations and calculate the sum of the error label weights, the error label weighting function value and the weight in each iteration; The weak hypothesis is calculated for each iteration; Calculate the pseudo loss; if the pseudo loss is less than the set threshold, calculate the weight of the base classifier; if the pseudo loss is not less than the set threshold, perform the final step; Through iterative calculation, the weights of the wrong labels of the training samples are updated; The confidence matrix is ​​superimposed according to the weights of the base classifiers to obtain a final confidence matrix of the integrated classifier; and a final selection result is output according to the final confidence matrix.

4. A method for generating a high-dimensional safety boundary of an online power system according to claim 3, characterized in that: In the process of inputting the possible crossing boundaries and the corresponding crossing risk values ​​into the generative adversarial network for processing, the possible crossing boundaries of different branches are expanded according to the confidence ratio to generate real data; The generative adversarial network performs learning processing on the real data and outputs the comprehensive high-dimensional security boundary.

5. A method for generating a high-dimensional safety boundary of an online power system according to claim 4, characterized in that: In the process of training and outputting the comprehensive high-dimensional security boundary through the generative adversarial network, the training processing steps are: The generator is fixed; the discriminator is trained and improved through the random boosting gradient strategy to obtain the improved discriminator; Fixing the improved judgement device; training and improving the generator through a random descent gradient strategy; Get the upgraded generator; Performing iterative game training on the upgraded judger and the upgraded generator; When the upgraded judge and the upgraded generator reach "Nash equilibrium", the result generated by the upgraded generator is used as output to obtain the comprehensive high-dimensional security boundary.

6. A device for generating a high-dimensional safety boundary of an online power system, adopting a method for generating a high-dimensional safety boundary of an online power system according to any one of claims 1 to 5, characterized in that: include: The safety boundary set generation module is used to calculate through the historical data driving strategy and the least squares fitting strategy to obtain the low-dimensional safety boundary of the set branches of the entire network and generate a safety boundary set; A limit crossing risk prediction model generation module is used to train the AdaBoost.M2 classifier through historical data and the safety boundary set to generate a limit crossing risk prediction model; An over-limit risk prediction model processing module is used to input the set online operation status into the over-limit risk prediction model for prediction processing, and output possible over-limit boundaries and corresponding over-limit risk values; The comprehensive high-dimensional safety boundary acquisition module is used to input the possible crossing boundary and the corresponding crossing risk value into the generative adversarial network for training processing and output a comprehensive high-dimensional safety boundary.

7. A device for generating a high-dimensional safety boundary of an online power system according to claim 6, characterized in that: In the safety boundary set generation module, in the process of obtaining the low-dimensional safety boundary of the branch set in the whole network, the key nodes of the branch are determined according to the size of the branch power flow sensitivity; according to the key nodes, the low-dimensional safety boundary of the branch is obtained through the least squares fitting strategy; the low-dimensional safety boundary expression of the branch is: Where, M is the number of key nodes in the branch; P i is the power injection of node i; j is the branch; c j is the constant term corresponding to the boundary of branch j; is the upper limit of active power flow of branch j; B is the branch set; β ij is the corresponding coefficient of node i at the safety boundary of branch j.

8. The device for generating a high-dimensional safety boundary of an online power system according to claim 7, characterized in that: In the over-limit risk prediction model generation module, the submodule for generating the over-limit risk prediction model includes: The weight initialization submodule is used to initialize the sample weights and error label weights of the training samples; The iteration number setting and weight calculation submodule is used to set the maximum number of iterations and calculate the error label weight sum, error label weighting function value and weight in each iteration; A weak hypothesis calculation submodule is used to calculate the weak hypothesis of each iteration; The pseudo-loss calculation and determination submodule is used to calculate the pseudo-loss; if the pseudo-loss is less than the set threshold, the weight of the base classifier is calculated; if the pseudo-loss is not less than the set threshold, the last step is performed; The error label weight update submodule is used to update the error label weights of training samples through iterative calculation; The out-of-limit risk prediction model generation submodule is used to superimpose the confidence matrix according to the weights of the base classifiers to obtain the final confidence matrix of the integrated classifier; and output the final selection result according to the final confidence matrix.

9. The device for generating a high-dimensional safety boundary of an online power system according to claim 8, characterized in that: In the comprehensive high-dimensional safety boundary acquisition module, when the possible crossing boundary and the corresponding crossing risk value are input into the generative adversarial network for processing, the possible crossing boundary of different branches is expanded according to the confidence ratio to generate real data; The generative adversarial network performs learning processing on the real data and outputs the comprehensive high-dimensional security boundary.

10. The device for generating a high-dimensional safety boundary of an online power system according to claim 9, characterized in that: In the comprehensive high-dimensional security boundary acquisition module, the submodule for training the generative adversarial network includes: The generator fixing and judgement improving submodule is used to fix the generator; the judgement is trained and improved through the random improvement gradient strategy to obtain the improved judgement; The submodule of fixing the judger and improving the generator is used to fix the improved judger; train and improve the generator through the random descent gradient strategy; and obtain the improved generator; An iterative game training submodule, used for performing iterative game training on the upgraded judger and the upgraded generator; The comprehensive high-dimensional security boundary output submodule is used to output the result generated by the upgraded generator when the upgraded judge and the upgraded generator reach "Nash equilibrium" to obtain the comprehensive high-dimensional security boundary.

Images