A method, device and storage medium for predicting the frequency security situation of a large power grid

Abstract

The present application provides a method, apparatus, and storage medium for predicting the frequency security situation of a large power grid. The method includes: obtaining generator parameters, where the generator parameters include mechanical power, electromagnetic power, damping coefficient, and initial angular frequency; based on a dynamic prediction model, performing feature extraction on the generator parameters to obtain a first frequency characteristic, and the dynamic prediction model is constructed based on the particle swarm algorithm and support vector regression; based on a steady-state prediction model, performing feature extraction on the generator parameters to obtain a second frequency characteristic, and the steady-state prediction model is constructed based on a lightweight gradient boosting machine and a generative adversarial network; based on a fusion prediction model, fusing the first frequency characteristic and the second frequency characteristic to obtain a target frequency characteristic, and the fusion prediction model is constructed based on a neural network and fuzzy logic; and obtaining a prediction result according to the target frequency characteristic. The present application can improve the prediction accuracy of frequency dynamic response.

Description

Technical Field

[0001] The present application relates to the technical field of power system prediction, and particularly to a large power grid frequency security situation prediction method, device, and storage medium. Background Art

[0002] With the continuous increase in system uncertainty and the increasingly drastic changes in operation modes, the functional requirements for real-time online evaluation have also increased accordingly. From the perspective of evaluation objects, network voltage, frequency, phase, etc. are all important targets for system online evaluation.

[0003] In terms of system frequency situation prediction, since the 1920s, the following several types of mature methods have been proposed: time-domain simulation method based on causal theory, average system frequency model (ASF), system frequency response model (SFR), etc. The time-domain simulation method calculates the frequency dynamics after perturbation through a step-by-step integration strategy, and the calculation results are accurate; both the average system frequency model (ASF) and the system frequency response model (SFR) assume that the system frequency is consistent in each region, but the frequency of a large power grid has complex time and space distribution characteristics, and large errors will occur during dynamic frequency analysis. Therefore, there are many problems when the equivalent model is applied to the post-perturbation dynamic frequency analysis of modern large-scale power grids.

[0004] The existing two major technical systems of power grid online real-time evaluation and frequency security situation prediction are relatively disjointed. The traditional online evaluation method that obtains specific indicators through PMU data processing is difficult to meet the technical requirements for future system frequency evolution prediction, and the traditional frequency situation judgment method based on a deterministic model and moment-by-moment simulation is difficult to be directly applied to the online evaluation of the power grid. Therefore, it is necessary to carry out research on real-time prediction of the frequency security situation of large systems considering multiple uncertain factors. Summary of the Invention

[0005] The present application provides a large power grid frequency security situation prediction method, device, and storage medium to solve the problem of low prediction accuracy of the frequency dynamic response under system perturbation in the prior art.

[0006] In a first aspect, the present application provides a large power grid frequency security situation prediction method, including:

[0007] Obtain generator parameters, where the generator parameters include mechanical power, electromagnetic power, damping coefficient, and initial angular frequency;

[0008] Based on a dynamic prediction model, perform feature extraction on the generator parameters to obtain a first frequency characteristic, where the dynamic prediction model is constructed based on a particle swarm algorithm and support vector regression;

[0009] Based on the steady-state prediction model, feature extraction is performed on the generator parameters to obtain the second frequency characteristic. The steady-state prediction model is constructed based on a lightweight gradient boosting machine and a generative adversarial network;

[0010] Based on the fusion prediction model, the first frequency characteristic and the second frequency characteristic are fused to obtain the target frequency characteristic. The fusion prediction model is constructed based on an adaptive neuro-fuzzy system;

[0011] According to the target frequency characteristic, a prediction result is obtained.

[0012] In a second aspect, the present application provides a large power grid frequency security situation prediction device, including:

[0013] An acquisition module, configured to acquire generator parameters, where the generator parameters include mechanical power, electromagnetic power, damping coefficient, and initial angular frequency;

[0014] A first prediction module, configured to perform feature extraction on the generator parameters based on a dynamic prediction model to obtain a first frequency characteristic. The dynamic prediction model is constructed based on a particle swarm algorithm and support vector regression;

[0015] A second prediction model, configured to perform feature extraction on the generator parameters based on a steady-state prediction model to obtain a second frequency characteristic. The steady-state prediction model is constructed based on a lightweight gradient boosting machine and a generative adversarial network;

[0016] A fusion module, configured to fuse the first frequency characteristic and the second frequency characteristic based on a fusion prediction model to obtain a target frequency characteristic. The fusion prediction model is constructed based on an adaptive neuro-fuzzy system;

[0017] A judgment module, configured to obtain a prediction result according to the target frequency characteristic.

[0018] In a third aspect, the present application provides a terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method in the first aspect or any possible implementation manner of the first aspect are implemented.

[0019] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the method in the first aspect or any possible implementation manner of the first aspect are implemented.

[0020] The present application provides a method, an apparatus, and a storage medium for predicting the frequency security situation of a large power grid. In the present application, a dynamic prediction model and a steady-state prediction model are respectively used to obtain a first frequency characteristic and a second frequency characteristic, and then a fusion prediction model is used to fuse the first frequency characteristic and the second frequency characteristic to obtain a target frequency characteristic. The frequency security situation of the large power grid is judged according to the target frequency characteristic. Since the target frequency characteristic becomes more accurate after fusion, the prediction accuracy of the frequency dynamic response is improved. Description of the Drawings

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 is the implementation flowchart of the method for predicting the frequency security situation of a large power grid provided by an embodiment of the present application;

[0023] Figure 2 is the training flowchart of the dynamic prediction model provided by an embodiment of the present application;

[0024] Figure 3 is the structural schematic diagram of the generative adversarial network provided by an embodiment of the present application;

[0025] Figure 4 is the training flowchart of the steady-state prediction model provided by an embodiment of the present application;

[0026] Figure 5 is the structural schematic diagram of the fusion prediction model provided by an embodiment of the present application;

[0027] Figure 6 is the structural schematic diagram of the apparatus for predicting the frequency security situation of a large power grid provided by an embodiment of the present application;

[0028] Figure 7 is the schematic diagram of the terminal provided by an embodiment of the present application. Detailed Embodiments

[0029] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present application. However, those skilled in the art should understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present application.

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following will illustrate through specific embodiments in conjunction with the accompanying drawings.

[0031] Figure 1 The following is a detailed implementation flowchart of the large power grid frequency security situation prediction method provided by the embodiments of this application:

[0032] In step 101, generator parameters are obtained. The generator parameters include mechanical power, electromagnetic power, damping coefficient, and initial angular frequency.

[0033] When the power system is disturbed, under the action of unbalanced power, it will affect the operating frequencies of each synchronous generator in the system. Therefore, in the embodiments of this application, the obtained generator parameters include mechanical power, electromagnetic power, damping coefficient, and initial angular frequency, which are used to calculate and determine whether the frequency in the power system is safe, thereby ensuring the stability of the large power grid frequency security situation.

[0034] In a possible implementation manner, after step 101, the method may further include:

[0035] Subtract the electromagnetic power from the mechanical power to obtain the change in generator active power, and preprocess the change in generator active power to obtain the preprocessed change in generator active power;

[0036] Based on the identification model, order determination, identification, and conversion are performed on the preprocessed change in generator active power to obtain the frequency change. The identification model is constructed based on the Akaike information criterion and the system identification method;

[0037] Use the swing equation to calculate the inertia time constant for the frequency change, and determine the system inertia center frequency based on the inertia time constant.

[0038] In the embodiments of this application, subtract the electromagnetic power P m from the mechanical power P e obtained in step 101 to get the change in generator active power ΔP, that is, ΔP = P m - P e . Then, perform detrending and low-pass filtering noise reduction preprocessing on the change in generator active power ΔP to obtain the preprocessed change in generator active power ΔP′, and input it into the identification model for order determination, identification, and conversion to obtain the frequency change Input the frequency change into the swing equation to calculate the inertia time constant H, and input the inertia time constant H into the dynamic frequency equation to calculate the system inertia center frequency The calculation formula of the swing equation is as follows:

[0039]

[0040] Among them, D is the damping coefficient of the generator, and w is the initial angular frequency of the generator.

[0041] The dynamic frequency equation is as follows:

[0042]

[0043] Among them, H sys is the sum of the inertia time constants H of each generator set in the system, D i is the damping system of the i-th generator, and w i is the initial angular frequency of the i-th generator.

[0044] Since there is a large amount of noise in the measured power grid data under normal operating conditions of the power system, when the fitting degree of the identification model is very high, overfitting may occur. To avoid the overfitting problem of the identification model caused by a large amount of identification data, in the embodiments of this application, it is necessary to perform detrending and low-pass filtering noise reduction preprocessing on the change in active power ΔP of the generator.

[0045] Among them, the Akaike Information Criterion (AIC) is a standard for measuring the goodness of fit of a statistical model. The Akaike Information Criterion is based on the concept of entropy and can balance the complexity of the estimated model and the goodness of fit of this model to the data.

[0046] System identification is to determine the mathematical model describing the behavior of the system according to the input and output time functions of the system. The purpose of establishing a mathematical model through identification is to estimate the important parameters characterizing the behavior of the system, establish a model that can imitate the behavior of the real system, predict the future evolution of the system output with the current measurable input and output of the system, and design a controller. The main problem in analyzing the system is to determine the output signal according to the input time function and the characteristics of the system.

[0047] For the order of the identification model of the power system, too low an order will make the expression effect of the identification model inaccurate, and too high an order will cause overfitting. To avoid the deviation of the identification result of the power system due to the order of the identification model, in the embodiments of this application, AIC is used to process the change in active power ΔP of the generator after detrending and low-pass filtering noise reduction preprocessing ′ to determine a suitable order of the system identification model, and this identification model focuses on expressing the external characteristics of the measurement data.

[0048] As the most general model selection criterion, AIC is another application of the maximum likelihood principle. According to the maximum likelihood principle, to find a model that approximates the actual process, so that the output probability distribution of this model can approximate the probability distribution of the actual process output as much as possible. The expression formula of the AIC performance index is as follows:

[0049] AIC(n) = Nlnρ N +2N (3)

[0050] where N is the number of parameters in the identification model, and ρ n is the variance of the prediction error of the nth-order model. The meaning of AIC can be understood as: the estimated value of the Kullback-Leibler distance between the probability density function estimated by the AIC model and the actual probability density function. The AIC performance index takes into account both the model complexity and flexibility to adapt to the data, and the smallest AIC value corresponds to the most suitable model.

[0051] After obtaining the identification model, it is necessary to convert the identification model into the form of a first-order transfer function. However, the order reduction process will reduce the fitting degree of the identification model, introduce errors, and cause deviations in the output results. Moreover, the higher the order of the identification model, the greater the error caused by the order reduction of the identification model. To avoid the reduction of the fitting degree of the AIC order determination identification model during the order reduction process, the embodiment of the present application converts the identified discrete-time identification model into a continuous-time model, then gives a step input to the identification model, and takes the frequency change amount of the output value within 0.5 - 2 s after the step response and substitutes it into the swing equation to calculate the inertia time constant H, and calculates the system inertia center frequency according to the inertia time constant

[0052] In step 102, based on the dynamic prediction model, feature extraction is performed on the generator parameters to obtain the first frequency characteristic. The dynamic prediction model is constructed based on the particle swarm algorithm and support vector regression.

[0053] Substitute the mechanical power P m and electromagnetic power P e of the generator obtained in step 101, the damping coefficient D, and the initial angular frequency w into the dynamic prediction model, perform feature extraction on it, and obtain the first frequency characteristic.

[0054] Among them, the particle swarm algorithm (Particle Swarm optimization, PSO), also translated as particle swarm optimization algorithm, particle swarm algorithm, or particle swarm optimization algorithm, is a stochastic search algorithm based on group collaboration developed by simulating the foraging behavior of bird flocks. It is generally considered to be a kind of swarm intelligence (SI).

[0055] Support vector regression (support vector regression, SVR) is a supervised learning algorithm used to predict discrete values. In SVR, the best fitting line is the hyperplane with the largest number of points.

[0056] The essence of SVR is to solve classification problems. For a given training sample set S = {(x i , y i ) | x i ∈ R n , y i ∈ R}, i = 1, 2, …, l, where x i is the n-dimensional input feature vector of the i-th sample, y i is the classification category of the i-th sample, R is the real number space, R n is the n-dimensional real number space, and l is the total number of samples, to solve the optimal classification hyperplane. Compared with the classification problem of the support vector machine, the output of the regression problem of the support vector machine is no longer a discrete value but becomes a continuous value. The regression problem of the support vector machine is to seek the implicit regression function based on limited observation data, that is, to find the mapping relationship f: R n → R from the input space to the output space, that is, to solve the regression function where w is the weight coefficient, b is the bias vector matrix, is the non-linear mapping input to the high-dimensional space, and x is the input feature vector.

[0057] In the embodiment of the present application, v-SVR is used as the dynamic prediction model, and the influence factor v (v ≥ 0) of the insensitive loss coefficient is introduced, and the prediction accuracy is improved by solving the appropriate v. At this time, the optimization problem to be constructed and solved is:

[0058]

[0059] where C is the penalty factor, ξ i and are slack variables, and ε is the insensitive loss coefficient.

[0060] Construct the Lagrange function to obtain the dual problem of formula (4):

[0061]

[0062] where α i and are both Lagrange multipliers, and K(x i , x j ) is the kernel function corresponding to the transformation .

[0063] Solve the convex quadratic optimization problem to obtain the solutions of α i and . Therefore, the regression prediction function can be constructed as:

[0064]

[0065] In a possible implementation, the first frequency characteristic includes a first maximum frequency change rate, a first transient frequency extreme value, and a first quasi-steady state frequency. Step 102 may specifically include:

[0066] Input the mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency into the dynamic prediction model to obtain the first frequency characteristic.

[0067] The transient frequency extreme value f nadir and the rate of change of frequency (RoCoF) are usually used as trigger signals for protection elements and control devices in the power grid. After the power system is disturbed, the maximum frequency change rate is often used The transient frequency extreme value f nadir and the quasi-steady state frequency f ss to determine whether the frequency of the power system can remain stable. Therefore, in the embodiments of the present application, the mechanical power P m of the generator obtained in step 101, the electromagnetic power P e , the damping coefficient D, and the initial angular frequency w are input into the dynamic prediction model, and the obtained first frequency characteristic f SVR , where the first frequency characteristic f SVR includes: the first maximum frequency change rate the first transient frequency extreme value and the first quasi-steady state frequency

[0068] In a possible implementation, the training process of the dynamic prediction model may include:

[0069] Obtain the mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency, and use the feature screening method to obtain a standard sample set according to the mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency. The standard sample set includes training samples and test samples;

[0070] Use the training samples as inputs and the corresponding features of the training samples as outputs to train the support vector regression;

[0071] According to the trained support vector regression, use the particle swarm algorithm to optimize the test samples to obtain the dynamic prediction model.

[0072] Since the dynamic frequency after the power system is disturbed is directly related to the motion equation of the generator rotor, for a multi-unit power system, its dynamic frequency equation is shown in formula (2). Considering that the electromagnetic power P e generated by the generator in the system at steady state is balanced with the power consumed in the power grid, then there is:

[0073]

[0074] Among them, P lossij is the line loss of the power grid from node i to node j in the power system, P lj is the active power consumed by the jth load, m is the number of generator nodes in the power system, and n is the number of load nodes in the power system.

[0075] The obtained mechanical power P m , electromagnetic power P e , damping coefficient D, initial angular frequency w, and the system inertia center frequency are used to form a sample set. After obtaining the standard sample set through feature screening, the training samples in the standard sample set are used to train the SVR model, and the performance of the trained SVR model is tested using another part of the test samples. Among them, during the training process, the parameters of the SVR model need to be optimized. In this application, the particle swarm optimization algorithm is used as the algorithm for finding the optimal parameters of the SVR model, so as to obtain a dynamic prediction model. The trained dynamic prediction model is used to receive the disturbed power grid data measured by the power system, and the frequency characteristics after the disturbance of the power system can be quickly predicted, that is, the first frequency characteristic f SVR , including the first maximum frequency change rate transient frequency extreme value and quasi-steady state frequency Specific training process can be referred to Figure 2 .

[0076] In step 103, based on the steady-state prediction model, feature extraction is performed on the generator parameters to obtain the second frequency characteristic. The steady-state prediction model is constructed based on the light gradient boosting machine and the generative adversarial network.

[0077] The mechanical power P m , electromagnetic power P e , damping coefficient D, and initial angular frequency w of the generator obtained in step 101 are input into the dynamic prediction model, and feature extraction is performed on it to obtain the second frequency characteristic.

[0078] Among them, the light gradient boosting machine (lightGBM) is a boosting ensemble learning model based on the gradient boosting decision tree (GBDT), and uses the classification and regression trees (CART) as the base learner. The CART algorithm recursively divides all samples to construct a binary tree, divides the feature space into a finite number of sub-regions, and takes the average value of all samples in the sub-region as the output of the sub-region. The construction process of the CART algorithm is as follows:

[0079] (1) Construct a root node that contains all samples.

[0080] (2) Select the optimal splitting feature j and splitting point s, with the minimum mean squared error as the optimal splitting criterion. The calculation formula is as follows:

[0081]

[0082] where y i is the actual value of the variable, c 1 is the predicted value of the left node after splitting, and c 2 is the predicted value of the right node after splitting. Traverse the variable j, scan the splitting point s for feature j, and select (j, s) that minimizes formula (8). R 1 and R 2 refer to the left and right sub-regions.

[0083] (3) Use the selected (j, s) to divide the sub-region to determine the output value.

[0084]

[0085] (4) Continue to apply steps (2) and (3) to the two sub-regions until the stopping condition is met.

[0086] (5) Divide the input space into N sub-regions R 1 , R 2 , …, R n to generate a regression tree.

[0087]

[0088] Gradient Boosting Decision Tree (GBDT) adopts the boosting idea. In each round of iteration, a new CART tree is generated to fit the residuals of the previous round of results, and the gap between the fitted value and the target value becomes smaller and smaller. The output values of the CART trees generated in each round of iteration are accumulated to obtain the final learning result. LightGBM has made improvements on the basis of GBDT, integrating histogram algorithms, exclusive feature bundling, and unilateral gradient sampling, reducing memory usage and improving the training speed. The leaf-wise growth strategy with depth limit is used to avoid generating deep trees and prevent overfitting.

[0089] A generative adversarial network (GAN) contains a generative model and a discriminative model. For the specific structure, see Figure 3, for the purpose of generating similar sample data, the real samples are learned in an unsupervised manner to simulate their data distribution. The training process is as follows: First, fix the generative model, input the real samples into the discriminative model for training until a certain accuracy is achieved, then fix the discriminative model, input a set of random variables into the generative model, and then input the generated samples output by the generative model into the discriminative model for discrimination. During the training process, the backpropagation gradient is used for both the generative model and the discriminative model, and the model parameters are continuously updated until the discriminative model can no longer distinguish between real samples and generated samples. The training process can be expressed by formula (11), and formula (11) is as follows:

[0090]

[0091] where, xP data (x) is that x is taken from the real sample distribution, zP z (z) is that z is taken from the simulated sample distribution, G(z) is the generative model, D(x) is the discriminative model, and G and D are differentiable functions to ensure that the error of the model can be backpropagated.

[0092] After the system topology changes, the data dimension becomes m-dimensional. Input the n-dimensional random variables into the generator trained with a small number of samples by the new system, and output the m-dimensional features, and enter the discriminator to check whether they are close enough to the real samples.

[0093] In a possible implementation manner, the second frequency characteristic includes the second maximum frequency change rate, the second transient frequency extreme value, and the second quasi-steady-state frequency. Step 103 may specifically include:

[0094] Input the mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency into the steady-state prediction model to obtain the second frequency characteristic.

[0095] This application inputs the mechanical power P m , electromagnetic power P e , damping coefficient D, and initial angular frequency w of the generator obtained in step 101 into the steady-state prediction model to obtain the second frequency characteristic f GBM , where the second frequency characteristic f GBM includes: the second maximum frequency change rate the second transient frequency extreme value and the second quasi-steady-state frequency

[0096] In a possible implementation manner, the training process of the steady-state prediction model may include:

[0097] Divide the mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency into a training set and a test set;

[0098] Input the training set into the maximum frequency change rate prediction model, use the lightweight gradient boosting machine to calculate the first feature importance, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the maximum frequency change rate prediction model after feature processing;

[0099] Input the training set into the transient frequency extreme value prediction model, use the lightweight gradient boosting machine to calculate the second feature importance, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the transient frequency extreme value prediction model after feature processing;

[0100] Input the training set into the quasi-steady state frequency prediction model, use the lightweight gradient boosting machine to calculate the third feature importance, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the quasi-steady state frequency prediction model after feature processing;

[0101] According to the maximum frequency change rate prediction model, transient frequency extreme value prediction model and quasi-steady state frequency prediction model after feature processing, use the production adversarial network to verify the test set to obtain the steady state prediction model.

[0102] The specific training process can refer to Figure 4 , in the embodiment of the present application, the obtained mechanical power P m , electromagnetic power P e , damping coefficient D, initial angular frequency w and system inertia center frequency are divided into a training set and a test set: input the training set into the maximum frequency change rate prediction model, transient frequency extreme value prediction model and quasi-steady state frequency prediction model respectively, use lightGBM to calculate the first feature importance, second feature importance and third feature importance respectively, and then remove the corresponding lowest N 1 , N 2 , N 3 features according to the attention mechanism, and judge whether the accuracy of the current prediction model is reduced to the first preset accuracy, second preset accuracy and third preset accuracy respectively. If so, determine the trained maximum frequency change rate prediction model, transient frequency extreme value prediction model and quasi-steady state frequency prediction model. If not, N i = i +1 (i = 1, 2, 3) returns to the step of deleting the frequency feature with the lowest score according to the attention mechanism and continues to execute until the current prediction model meets the preset requirements; then input the test set into the trained maximum frequency change rate prediction model, transient frequency extreme value prediction model and quasi-steady state frequency prediction model for verification respectively. If the performance requirements of the model are met, determine the steady state prediction model according to the trained maximum frequency change rate prediction model, transient frequency extreme value prediction model and quasi-steady state frequency prediction model.

[0103] In step 104, based on the fusion prediction model, the first frequency characteristic and the second frequency characteristic are fused to obtain the target frequency characteristic, and the fusion prediction model is constructed based on the adaptive neuro-fuzzy system.

[0104] Since the first frequency characteristic f obtained according to the dynamic prediction model SVR and the second frequency characteristic f obtained from the steady-state prediction model GBM Considering the influence of all generators, loads, and network structures in the power system on the frequency dynamic response, its characteristic is that the accuracy of the prediction result increases with the increase of the sample accuracy and data. The disadvantage is that the prediction model does not involve the physical process of the power system frequency response. Therefore, there are deficiencies in the reliability of the prediction result. The Adaptive Network-based Fuzzy Inference System (ANFIS) is a comprehensive algorithm that combines the respective advantages of neural networks and fuzzy logic. It can take into account the strong self-learning and adaptive capabilities of neural networks while fully utilizing the ability of fuzzy systems to accurately express objective physical laws. Therefore, in the embodiments of the present application, ANFIS is used to construct a fusion prediction model, and the first frequency characteristic f obtained in step 102 SVR and the second frequency characteristic f obtained in step 103 GBM are input into the fusion prediction model for fusion to obtain the target frequency characteristic f ANFIS , so as to obtain a more accurate and reliable frequency prediction result. Among them, ANFIS is a new type of neural network formed by organically combining fuzzy logic and neural networks. Its basic idea is based on the Sugeno fuzzy model, and the parameters of the fuzzy inference system are adjusted by using the backpropagation algorithm or a hybrid algorithm combining the backpropagation algorithm and the least squares method, so that the designed fuzzy inference system can best simulate the relationship between the actual input and output.

[0105] In a possible implementation manner, the target frequency characteristic includes the target maximum frequency change rate, the target transient frequency extreme value, and the target quasi-steady-state frequency. Step 104 may specifically include:

[0106] Input the first frequency characteristic and the second frequency characteristic into the fusion prediction model to obtain the target frequency characteristic.

[0107] In the embodiments of the present application, the first frequency characteristic f obtained in step 102 SVR and the second frequency characteristic f obtained in step 103 GBM are input into the fusion prediction model for fusion to obtain the target frequency characteristic f ANFIS , where the target frequency characteristic f ANFIS includes: the target maximum frequency change rate the target transient frequency extreme value And the target steady-state frequency

[0108] In a possible implementation, the training process of the fusion prediction model may include:

[0109] Adopt the fuzzy logic method to perform fuzzy extraction on the first frequency characteristic and the second frequency characteristic, and obtain the first fuzzy rule corresponding to the first frequency characteristic and the second fuzzy rule corresponding to the second frequency characteristic;

[0110] Based on the neural network, fuse the first fuzzy rule and the second fuzzy rule to obtain the fusion prediction model.

[0111] For the structure of the fusion prediction model constructed in the embodiments of the present application, see Figure 5 , where ANFIS has five layers, and the inputs of the fusion prediction model are the first frequency characteristic f obtained by the dynamic prediction model SVR and the second frequency characteristic f obtained by the steady-state prediction model GBM . The fusion process of the two prediction results is divided into: a fuzzification layer, a fuzzy rule layer, a normalization layer, an input connection layer, and an output layer. Figure 5 In, A 1 , A 2 , A 3 , B 1 , B 2 , B 3 represent fuzzy sets, P ord represents the vector product, N orm represents the norm, f 1 -f 6 represents the fuzzy rule.

[0112] The main function of the fuzzification layer is to perform fuzzy processing on each input using multiple membership functions. In the embodiments of the present application, the bell-shaped membership function and the Gaussian membership function are mainly selected. The output of the fuzzification layer after fuzzy processing is the fuzzy feature of each input variable, that is:

[0113]

[0114] Among them, the first digits "1", "2", "3", "4" of the subscript of O respectively represent the node number, and O 1,A,a , O 1,B,b are the membership degrees of the fuzzy sets, and are respectively represented by the subscripts A and B for the input quantities composed of the prediction results of the two sub-model methods, is the membership function, and a and b are the membership function serial numbers.

[0115] The main function of the fuzzy rule layer is to perform the product operation on all input fuzzy values. The output of each node in this layer represents the excitation strength of a fuzzy rule. The establishment of fuzzy rules can fully combine the fuzzy features of the prediction results of the two sub-model methods. The excitation strength of each fuzzy rule obtained through the fuzzy rule layer is:

[0116] O 2,1 = O 1,A,a ·O 1,B,b , a, b = 1, 2, 3 (13)

[0117] The main function of normalization is to obtain the normalized excitation strength by dividing the fuzzy rule strength of each output of the previous layer by the total strength of all fuzzy rules, that is:

[0118] O 3,1 = O 2,1 / ∑ 1 O 2,1 (14)

[0119] The main function of the input connection layer is to connect the output of the normalization layer to the input of the ANFIS. Its output represents the final contribution value of each rule to the total output, that is:

[0120] O 4,1 = O 3,1 (a 1 f SVR + b 1 f GBM + c 1 ) (15)

[0121] Among them, a 1 , b 1 , c 1 are the parameter sets of this node and are called consequent parameters.

[0122] The main function of the output layer is to add all the output signals of the previous layer to finally obtain the target frequency characteristic of the fusion prediction model, that is:

[0123] f ANFIS = ∑ 1 O 4,1 (16)

[0124] Based on the above, the input data of the fusion prediction model is mainly composed of the frequency characteristics of the dynamic prediction model and the steady-state prediction model, and the output data is the target frequency characteristic generated by algorithm simulation. Among them, the backpropagation gradient descent method is used to learn its premise parameters, and the least squares method is used to determine its conclusion parameters. By continuously looping this learning process until the training sample accuracy condition is met, a trained fusion prediction model is finally obtained.

[0125] In step 105, a prediction result is obtained according to the target frequency characteristics.

[0126] According to the target frequency characteristic f obtained in step 104 ANFIS , that is, the target maximum frequency change rate The target transient frequency extreme value and the target quasi-steady-state frequency Determine whether the active frequency safety situation of the current large power grid is stable.

[0127] In a possible implementation manner, step 105 may specifically include:

[0128] Judge whether the target maximum frequency change rate is not greater than a first preset value, and judge whether the target transient frequency extreme value is not greater than a second preset value, and judge whether the target quasi-steady-state frequency is not greater than a third preset value;

[0129] If the target maximum frequency change rate is not greater than the first preset value, and the target transient frequency extreme value is not greater than the second preset value, and the target quasi-steady-state frequency is not greater than the third preset value, then determine that the active frequency safety situation of the large power grid is stable according to the target maximum frequency change rate, the target transient frequency extreme value, and the target quasi-steady-state frequency;

[0130] If at least one of the target maximum frequency change rate not being greater than the first preset value, the target transient frequency extreme value not being greater than the second preset value, and the target quasi-steady-state frequency not being greater than the third preset value does not meet the requirements, then determine that the active frequency safety situation of the large power grid is unstable.

[0131] In the embodiment of the present application, according to the target frequency characteristic f obtained in step 104 ANFIS , that is, the target maximum frequency change rate The target transient frequency extreme value and the target quasi-steady-state frequency Respectively set a first preset value ε for the target maximum frequency change rate the target transient frequency extreme value and the target quasi-steady-state frequency 1 a second preset value ε 2 and a third preset value ε 3 , and then judge whether the target maximum frequency change rate is not greater than the first preset value ε 1 , and judge whether the target transient frequency extreme value is not greater than the second preset value ε 2 , and judge whether the target quasi-steady-state frequency is not greater than the third preset value ε 3 .

[0132] If the target maximum frequency change rate Not greater than the first preset value ε 1 , and the extreme value of the target transient frequency Not greater than the second preset value ε 2 , and the target steady-state frequency Not greater than the third preset value ε 3 , that is Then it is determined that the active frequency security situation of the current large power grid is stable.

[0133] If the target maximum frequency change rate Not greater than the first preset value ε 1 , the extreme value of the target transient frequency Not greater than the second preset value ε 2 , the target steady-state frequency Not greater than the third preset value ε 3 , and at least one of them does not meet the requirements, that is If at least one of them is satisfied, it is determined that the active frequency security situation of the current large power grid is unstable.

[0134] The present application provides a method for predicting the frequency security situation of a large power grid. The present application respectively obtains the first frequency characteristic and the second frequency characteristic through a dynamic prediction model and a steady-state prediction model, and then fuses the first frequency characteristic and the second frequency characteristic through a fusion prediction model to obtain the target frequency characteristic, and judges the frequency security situation of the large power grid according to the target frequency characteristic. Since the target frequency characteristic becomes more accurate after fusion, the prediction accuracy of the frequency dynamic response is improved.

[0135] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0136] The following is an apparatus embodiment of the present application. For the details not described in detail, reference may be made to the corresponding method embodiment above.

[0137] Figure 6 The structural schematic diagram of the large power grid frequency security situation prediction apparatus provided by the embodiment of the present application is shown. For the convenience of description, only the parts related to the embodiment of the present application are shown and are described in detail as follows:

[0138] As Figure 6 shown, the large power grid frequency security situation prediction apparatus 6 includes:

[0139] An acquisition module 61, configured to acquire generator parameters, where the generator parameters include mechanical power, electromagnetic power, damping coefficient, and initial angular frequency;

[0140] The first prediction module 62 is configured to extract features of the generator parameters based on a dynamic prediction model to obtain a first frequency characteristic. The dynamic prediction model is constructed based on a particle swarm algorithm and support vector regression.

[0141] The second prediction model 63 is configured to extract features of the generator parameters based on a steady-state prediction model to obtain a second frequency characteristic. The steady-state prediction model is constructed based on a lightweight gradient boosting machine and a generative adversarial network.

[0142] The fusion module 64 is configured to fuse the first frequency characteristic and the second frequency characteristic based on a fusion prediction model to obtain a target frequency characteristic. The fusion prediction model is constructed based on an adaptive neuro-fuzzy system.

[0143] The judgment module 65 is configured to obtain a prediction result according to the target frequency characteristic.

[0144] This application provides a large power grid frequency security situation prediction device. In this application, the first frequency characteristic and the second frequency characteristic are respectively obtained through the dynamic prediction model and the steady-state prediction model, and then the first frequency characteristic and the second frequency characteristic are fused through the fusion prediction model to obtain the target frequency characteristic. The frequency security situation of the large power grid is judged according to the target frequency characteristic. Since the target frequency characteristic becomes more accurate after fusion, the prediction accuracy of the frequency dynamic response is improved.

[0145] In a possible implementation manner, after the acquisition module, the device may further include:

[0146] The preprocessing module is configured to obtain the change amount of the generator active power by taking the difference between the mechanical power and the electromagnetic power, and preprocess the change amount of the generator active power to obtain the preprocessed change amount of the generator active power.

[0147] The identification module is configured to perform order determination, identification, and conversion on the preprocessed change amount of the generator active power based on an identification model to obtain the frequency change amount. The identification model is constructed based on the Akaike information criterion and a system identification method.

[0148] The calculation module is configured to calculate the inertia time constant for the frequency change amount by using a swing equation, and determine the system inertia center frequency according to the inertia time constant.

[0149] In a possible implementation manner, the first frequency characteristic includes a first maximum frequency change rate, a first transient frequency extreme value, and a first quasi-steady state frequency. Specifically, the first prediction module may be configured to:

[0150] Input the mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency into the dynamic prediction model to obtain the first frequency characteristic.

[0151] In a possible implementation, the training process of the dynamic prediction model may include:

[0152] Obtain mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency, and use a feature screening method to obtain a standard sample set based on the mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency. The standard sample set includes training samples and test samples;

[0153] Use the training samples as input and the corresponding features of the training samples as output to train the support vector regression;

[0154] According to the trained support vector regression, use the particle swarm optimization algorithm to optimize the test samples to obtain the dynamic prediction model.

[0155] In a possible implementation, the second frequency characteristic includes the second maximum frequency change rate, the second transient frequency extreme value, and the second quasi-steady state frequency. The second prediction module can specifically be used for:

[0156] Input the mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency into the steady-state prediction model to obtain the second frequency characteristic;

[0157] The training process of the steady-state prediction model includes:

[0158] Divide the mechanical power, electromagnetic power, damping coefficient, initial angular frequency, and system inertia center frequency into a training set and a test set;

[0159] Input the training set into the maximum frequency change rate prediction model, use the lightweight gradient boosting machine to calculate the first feature importance, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the maximum frequency change rate prediction model after feature processing;

[0160] Input the training set into the transient frequency extreme value prediction model, use the lightweight gradient boosting machine to calculate the second feature importance, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the transient frequency extreme value prediction model after feature processing;

[0161] Input the training set into the quasi-steady state frequency prediction model, use the lightweight gradient boosting machine to calculate the third feature importance, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the quasi-steady state frequency prediction model after feature processing;

[0162] According to the maximum frequency change rate prediction model, transient frequency extreme value prediction model, and quasi-steady state frequency prediction model after feature processing, use the generative adversarial network to verify the test set to obtain the steady-state prediction model.

[0163] In a possible implementation, the target frequency characteristic includes a target maximum frequency change rate, a target transient frequency extreme value, and a target steady-state frequency. Specifically, the fusion module can be used to:

[0164] Input the first frequency characteristic and the second frequency characteristic into a fusion prediction model to obtain the target frequency characteristic;

[0165] The training process of the fusion prediction model can include:

[0166] Adopt a fuzzy logic method to perform fuzzy extraction on the first frequency characteristic and the second frequency characteristic, and obtain a first fuzzy rule corresponding to the first frequency characteristic and a second fuzzy rule corresponding to the second frequency characteristic;

[0167] Based on a neural network, fuse the first fuzzy rule and the second fuzzy rule to obtain the fusion prediction model.

[0168] In a possible implementation, the judgment module can be specifically used to:

[0169] Judge whether the target maximum frequency change rate is not greater than a first preset value, and judge whether the target transient frequency extreme value is not greater than a second preset value, and judge whether the target steady-state frequency is not greater than a third preset value;

[0170] If the target maximum frequency change rate is not greater than the first preset value, and the target transient frequency extreme value is not greater than the second preset value, and the target steady-state frequency is not greater than the third preset value, then determine that the large power grid active frequency security situation is stable according to the target maximum frequency change rate, the target transient frequency extreme value, and the target steady-state frequency;

[0171] If at least one of the conditions that the target maximum frequency change rate is not greater than the first preset value, the target transient frequency extreme value is not greater than the second preset value, and the target steady-state frequency is not greater than the third preset value is not satisfied, then determine that the large power grid active frequency security situation is unstable.

[0172] Figure 7 is a schematic diagram of the terminal provided by the embodiment of the present application. As Figure 7 shown, the terminal 7 of this embodiment includes: a processor 70, a memory 71, and a computer program 72 stored in the memory 71 and executable on the processor 70. When the processor 70 executes the computer program 72, it implements the steps in the embodiments of the above various large power grid frequency security situation prediction methods, such as Figure 1 the steps 101 to 105 shown. Or, when the processor 70 executes the computer program 72, it implements the functions of each module in the above device embodiments, such as Figure 6 the functions of the modules 61 to 65 shown.

[0173] Exemplarily, the computer program 72 may be divided into one or more modules, which are stored in the memory 71 and executed by the processor 70 to complete this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program 72 in the terminal 7. For example, the computer program 72 may be divided into Figure 6 the modules 61 to 65 shown.

[0174] The terminal 7 may be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal 7 may include, but is not limited to, a processor 70 and a memory 71. Those skilled in the art can understand that Figure 7 merely examples of the terminal 7, which do not constitute a limitation on the terminal 7, and may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the terminal may further include input / output devices, network access devices, a bus, etc.

[0175] The so-called processor 70 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0176] The memory 71 may be an internal storage unit of the terminal 7, such as the hard disk or memory of the terminal 7. The memory 71 may also be an external storage device of the terminal 7, such as a plug-in hard disk equipped on the terminal 7, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory 71 may also include both the internal storage unit and the external storage device of the terminal 7. The memory 71 is used to store the computer program and other programs and data required by the terminal 7. The memory 71 may also be used to temporarily store data that has been output or will be output.

[0177] Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the above division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of this application. The specific working processes of the units and modules in the above system can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated here.

[0178] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0179] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0180] In the embodiments provided in this application, it should be understood that the disclosed device / terminal and method can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For example, the division of the module or unit is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be in electrical, mechanical or other forms.

[0181] The unit described as a separated component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0182] In addition, the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware or in the form of software functional units.

[0183] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. Based on such an understanding, all or part of the processes in the above-mentioned method embodiments of the present application can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned embodiments of the method for predicting the frequency security situation of a large power grid can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0184] The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for predicting the frequency security situation of a large power grid, characterized in that, it includes: Obtain generator parameters, where the generator parameters include mechanical power, electromagnetic power, damping coefficient, and initial angular frequency; Based on a dynamic prediction model, perform feature extraction on the generator parameters to obtain a first frequency characteristic, and the dynamic prediction model is constructed based on the particle swarm algorithm and support vector regression; Based on a steady-state prediction model, perform feature extraction on the generator parameters to obtain a second frequency characteristic, and the steady-state prediction model is constructed based on a lightweight gradient boosting machine and a generative adversarial network; Based on a fusion prediction model, fuse the first frequency characteristic and the second frequency characteristic to obtain a target frequency characteristic, and the fusion prediction model is constructed based on an adaptive neuro-fuzzy system; Obtain a prediction result according to the target frequency characteristic; Among them, after obtaining the generator parameters, the method further includes: Subtract the electromagnetic power from the mechanical power to obtain the change in generator active power, and preprocess the change in generator active power to obtain the preprocessed change in generator active power; Based on an identification model, perform order determination, identification, and conversion on the preprocessed change in generator active power to obtain the frequency change, and the identification model is constructed based on the Akaike information criterion and system identification methods; Use the swing equation to calculate the inertia time constant for the frequency change, and determine the system inertia center frequency according to the inertia time constant; The second frequency characteristic includes the second maximum frequency change rate, the second transient frequency extreme value, and the second quasi-steady state frequency. Based on the steady-state prediction model, performing extraction on the generator parameters to obtain the second frequency characteristic includes: Input the mechanical power, the electromagnetic power, the damping coefficient, the initial angular frequency, and the system inertia center frequency into the steady-state prediction model to obtain the second frequency characteristic; The training process of the steady-state prediction model includes: Divide the mechanical power, the electromagnetic power, the damping coefficient, the initial angular frequency, and the system inertia center frequency into a training set and a test set; Input the training set into the maximum frequency change rate prediction model, use a lightweight gradient boosting machine to calculate the first feature importance, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the maximum frequency change rate prediction model after feature processing; Input the training set into the transient frequency extreme value prediction model, use a lightweight gradient boosting machine to calculate the second feature importance, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the transient frequency extreme value prediction model after feature processing; Input the training set into the quasi-steady state frequency prediction model, use a lightweight gradient boosting machine to calculate the third feature importance, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the quasi-steady state frequency prediction model after feature processing; According to the maximum frequency change rate prediction model, transient frequency extreme value prediction model, and quasi-steady state frequency prediction model after feature processing, use a generative adversarial network to verify the test set to obtain the steady-state prediction model.

2. The large power grid frequency security situation prediction method according to claim 1, characterized in that the first frequency characteristics include the first maximum frequency change rate, the first transient frequency extreme value, and the first quasi-steady state frequency. Based on the dynamic prediction model, feature extraction is performed on the generator parameters to obtain the first frequency characteristics, including: Inputting the mechanical power, the electromagnetic power, the damping coefficient, the initial angular frequency, and the system inertia center frequency into the dynamic prediction model to obtain the first frequency characteristics.

3. The large power grid frequency security situation prediction method according to claim 2, characterized in that the training process of the dynamic prediction model includes: Obtaining the mechanical power, the electromagnetic power, the damping coefficient, the initial angular frequency, and the system inertia center frequency, and using a feature screening method to obtain a standard sample set according to the mechanical power, the electromagnetic power, the damping coefficient, the initial angular frequency, and the system inertia center frequency. The standard sample set includes training samples and test samples; Using the training samples as inputs and the features corresponding to the training samples as outputs to train the support vector regression; According to the trained support vector regression, using the particle swarm algorithm to optimize the test samples to obtain the dynamic prediction model.

4. The large power grid frequency security situation prediction method according to claim 1, characterized in that the target frequency characteristics include the target maximum frequency change rate, the target transient frequency extreme value, and the target quasi-steady state frequency. Based on the fusion prediction model, the first frequency characteristics and the second frequency characteristics are fused to obtain the target frequency characteristics, including: Inputting the first frequency characteristics and the second frequency characteristics into the fusion prediction model to obtain the target frequency characteristics; the training process of the fusion prediction model includes: Using the fuzzy logic method to perform fuzzy extraction on the first frequency characteristics and the second frequency characteristics to obtain the first fuzzy rule corresponding to the first frequency characteristics and the second fuzzy rule corresponding to the second frequency characteristics; Based on the neural network, fusing the first fuzzy rule and the second fuzzy rule to obtain the fusion prediction model.

5. The large power grid frequency security situation prediction method according to claim 4, characterized in that obtaining the prediction result according to the target frequency characteristics includes: Judging whether the target maximum frequency change rate is not greater than the first preset value, and judging whether the target transient frequency extreme value is not greater than the second preset value, and judging whether the target quasi-steady state frequency is not greater than the third preset value; If the target maximum frequency change rate is not greater than the first preset value, and the target transient frequency extreme value is not greater than the second preset value, and the target quasi-steady state frequency is not greater than the third preset value, then determine that the active frequency security situation of the large power grid is stable according to the target maximum frequency change rate, the target transient frequency extreme value, and the target quasi-steady state frequency; If at least one of the following conditions is not met: the target maximum frequency change rate is not greater than the first preset value, the target transient frequency extreme value is not greater than the second preset value, and the target quasi-steady state frequency is not greater than the third preset value, it is determined that the active frequency security situation of the large power grid is unstable.

6. A large power grid frequency security situation prediction device, characterized in that, it includes: An acquisition module for acquiring generator parameters, where the generator parameters include mechanical power, electromagnetic power, damping coefficient, and initial angular frequency; A first prediction module for extracting features from the generator parameters based on a dynamic prediction model to obtain a first frequency characteristic, where the dynamic prediction model is constructed based on a particle swarm algorithm and support vector regression; A second prediction model for extracting features from the generator parameters based on a steady-state prediction model to obtain a second frequency characteristic, where the steady-state prediction model is constructed based on a lightweight gradient boosting machine and a generative adversarial network; A fusion module for fusing the first frequency characteristic and the second frequency characteristic based on a fusion prediction model to obtain a target frequency characteristic, where the fusion prediction model is constructed based on an adaptive neuro-fuzzy system; A judgment module for obtaining a prediction result according to the target frequency characteristic; Wherein, after acquiring the generator parameters, the device further includes: A preprocessing module for subtracting the electromagnetic power from the mechanical power to obtain the change in generator active power, and preprocessing the change in generator active power to obtain the preprocessed change in generator active power; An identification module for performing order determination, identification, and conversion on the preprocessed change in generator active power based on an identification model to obtain a frequency change amount, where the identification model is constructed based on the Akaike information criterion and a system identification method; A calculation module for calculating the inertia time constant using the swing equation for the frequency change amount, and determining the system inertia center frequency according to the inertia time constant; The second frequency characteristic includes a second maximum frequency change rate, a second transient frequency extreme value, and a second quasi-steady state frequency, and the second prediction model is used for: Inputting the mechanical power, the electromagnetic power, the damping coefficient, the initial angular frequency, and the system inertia center frequency into the steady-state prediction model to obtain a second frequency characteristic; The training process of the steady-state prediction model includes: Dividing the mechanical power, the electromagnetic power, the damping coefficient, the initial angular frequency, and the system inertia center frequency into a training set and a test set; Inputting the training set into the maximum frequency change rate prediction model, calculating the first feature importance using a lightweight gradient boosting machine, and deleting the frequency feature with the lowest score according to the attention mechanism to obtain the maximum frequency change rate prediction model after feature processing; Inputting the training set into the transient frequency extreme value prediction model, calculating the second feature importance using a lightweight gradient boosting machine, and deleting the frequency feature with the lowest score according to the attention mechanism to obtain the transient frequency extreme value prediction model after feature processing; Input the training set into the quasi-steady-state frequency prediction model, calculate the third feature importance using a lightweight gradient boosting machine, and delete the frequency feature with the lowest score according to the attention mechanism to obtain the quasi-steady-state frequency prediction model after feature processing; According to the maximum frequency change rate prediction model, transient frequency extreme value prediction model, and quasi-steady-state frequency prediction model after feature processing, use a generative adversarial network to verify the test set to obtain the steady-state prediction model.

7. A terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the steps of the large power grid frequency security situation prediction method according to any one of claims 1 to 5 above.

8. A computer-readable storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the steps of the large power grid frequency security situation prediction method according to any one of claims 1 to 5 above.

Images