Power system simulation error parameter correction method and device, equipment and storage medium

Abstract

The invention discloses a power system simulation error parameter correction method, device and equipment and a storage medium, and the method comprises the steps: carrying out the preliminary screening of parameters in a system based on a trajectory sensitivity analysis method, and obtaining a primary parameter set; performing random perturbation on the primary parameter set to obtain a data set of variables measured by each node of the system after each random perturbation; respectively constructing a parameter selection CNN model and a parameter correction CNN model by using the data set; selecting error dominant parameters; and correcting the error dominant parameter. From the global perspective of the system, relevance and difference rules of variable trajectories of all nodes are intelligently extracted and further matched with system error features, and accurate screening of error dominant parameters is achieved. Besides, comprehensive frequency domain features under random perturbation of the error dominant parameters are captured according to the CNN technology and are matched with actual error trajectories, so that intelligent correction of the error dominant parameters is realized.

Description

Technical Field

[0001] The present invention relates to the field of power system simulation error parameter correction, and in particular to a power system simulation error parameter correction method, device, equipment and storage medium. Background Art

[0002] Power system simulation error is an important indicator for evaluating the accuracy and reliability of simulation models. Its source and analysis are crucial to improving simulation quality. Since some parameters are set differently in simulation and in reality, there are errors between power system simulation results and actual results. How to accurately identify the error-dominant parameters and correct them is an urgent problem to be solved in the power system field.

[0003] At present, the methods used to identify error parameters in power system simulation mainly include time domain analysis and frequency domain analysis. The process is to obtain the error dominant parameters by studying the trajectory of the parameter sensitivity to the error variable, and then realize the parameter correction according to the conventional optimization algorithm.

[0004] In the process of identifying the parameters to be corrected, time domain analysis focuses more on the dynamic changes of the signal over time. The effect of the parameters on the variables is determined by studying the response differences of the system at different time points. As a result, time domain analysis is insufficient in capturing complex spatiotemporal characteristics and it is difficult to effectively distinguish the true impact of each parameter on the simulation results. Therefore, the obtained set of parameters to be corrected may contain more unnecessary correction parameters, which can easily lead to multiple solution problems in the parameter correction calculation.

[0005] Frequency domain analysis can reveal the relationship between various parameters and system variables from a frequency perspective by analyzing the frequency characteristics of system signals. However, in high-dimensional power systems, since the error spectrum often contains frequency components of multiple parameters, if the interaction between parameters is not fully considered, the method based on sensitive frequency point matching may cause confusion when extracting error-dominant parameters, resulting in doubts about the accuracy of the constructed error-dominant parameter set.

[0006] Conventional optimization algorithms have limitations in the correction process of the parameters to be corrected. For high-order nonlinear power systems, conventional optimization algorithms may fall into the dilemma of local optimality and fail to reach global optimality. In addition, the applicability and scalability of such algorithms also face challenges in practical applications. For large-scale power systems, the complexity of optimization calculations and the real-time requirements limit the effectiveness of these methods. Therefore, there is still much room for improvement in existing methods in terms of data utilization, global search capabilities, etc. Summary of the invention

[0007] The purpose of the present invention is to provide a method, device, equipment and storage medium for correcting error parameters in power system simulation. Based on parameter perturbation, the change trajectory of variables at each node in the system is obtained, and then the trajectory change characteristics are extracted according to the algorithm to obtain accurate error-dominant parameters. The error-dominant parameters are matched with the frequency domain characteristics of the error trajectory to achieve parameter correction, so as to solve the above-mentioned problems.

[0008] In order to achieve the above objectives, the present invention provides the following problem-solving ideas:

[0009] First, the error-dominant parameters are intelligently identified. The feature extraction based on global variables can make full use of the measurable data in the system. By perturbing the parameters, the convolutional neural network (CNN) is used to intelligently capture the change rules of the variables of each node in the system, so as to accurately obtain the error-dominant parameters. This method first conducts a preliminary screening of the parameters in the system by means of sensitivity analysis based on the error variables in the system, and selects the parameters that are sensitive to the error variables as the primary dominant parameters. After obtaining the primary dominant parameter set, the event playback is performed after the parameter set is randomly perturbed to obtain the trajectory of the variables of each node in the system. Next, an indicator is defined to quantify the difference between the trajectory of these variables and the simulation trajectory, and the change rate before and after the parameter perturbation is calculated. A corresponding CNN training model is constructed based on the difference index between the parameter change rate and the trajectory of each variable. When the model is well trained, the measured trajectory and the simulation trajectory of each variable can be used for error quantification based on the same data processing method, and the error quantification results are input into the trained model, so as to obtain a set of dominant parameter change rates, that is, the degree to which the simulation set values ​​of these parameters deviate from the actual values.

[0010] Secondly, the simulation error dominant parameters are intelligently corrected. After obtaining the error dominant parameters, the variable trajectory of the nodes near the fault point is obtained by further perturbing the parameters, and the spectrum of the trajectory is extracted. CNN is used to obtain the relationship between the parameter perturbation rate and the spectrum data. When the model is well trained, the spectrum data of the measured node data is input to obtain the true value of the error dominant parameter.

[0011] Based on the above ideas, the present invention provides the following technical solutions:

[0012] A method for correcting power system simulation error parameters, comprising:

[0013] Based on the trajectory sensitivity analysis method, the parameters in the system are preliminarily screened to obtain the preliminary parameter set;

[0014] Perform random perturbations on the initially selected parameter set to obtain a data set of variables measured at each node of the system after each random perturbation;

[0015] Using the data set, we build a parameter selection CNN model and a parameter correction CNN model respectively;

[0016] Select CNN model for training based on parameters, select CNN model based on trained parameters, and select error-dominant parameters;

[0017] Training is performed based on a parameter correction CNN model, and the error-dominant parameters are corrected based on the trained parameter correction CNN model and a frequency domain feature extraction method.

[0018] The node refers to an electrical connection point, which can be represented as a power plant, a substation, a load center or a connection point between transmission lines in the system.

[0019] Furthermore, the specific steps of preliminary screening of parameters in the system based on the trajectory sensitivity analysis method include: using the local linearization method in disturbance analysis to calculate the trajectory of sensitivity of different parameters to the variables measured at each node of the system. The local linearization method is a type of trajectory sensitivity analysis method. This method is used here to approximate the nonlinear system as a linear system, thereby simplifying the analysis process.

[0020] Furthermore, the randomly perturbing the primary parameter set to obtain a data set of variables measured by each node of the system after each random perturbation includes:

[0021] The primary parameter set is randomly perturbed to obtain the time series trajectory data of the variables measured at each node in the system, and the similarity of the time series trajectory data of the variables is calculated to obtain the trajectory similarity between the time series trajectory of the variables and the simulation trajectory.

[0022] Further, the selecting a CNN model for training based on parameters, selecting a CNN model based on trained parameters, and selecting error-dominant parameters include:

[0023] Calculate the numerical change rate of the parameters in the preliminary parameter set after random perturbation and before random perturbation, that is, the parameter perturbation rate, and the mapping relationship between the training trajectory similarity and the parameter perturbation rate. Input the trajectory similarity of the time series trajectory of the actually measured variable and the simulation trajectory into the parameter selection CNN model. The parameter selection CNN model outputs a set of parameter perturbation rate values. The absolute value of the value represents the degree of influence of the parameter on the time series waveform. Define the deviation threshold based on the parameter correction scale and calculation cost actually required by the project. The preliminary dominant parameter whose absolute value is higher than the preset deviation threshold is the error dominant parameter to be corrected. Usually, when there is no special requirement, the threshold can be selected as 5% of the maximum deviation value.

[0024] Furthermore, the training is performed based on the parameter correction CNN model, and the error-dominant parameter is corrected based on the trained parameter correction CNN model and the frequency domain feature extraction method, including:

[0025] The error-dominant parameters are randomly perturbed to obtain the time series trajectory of the variables near the fault point. The time series trajectory of the variables is Fourier transformed, and the mapping relationship between the parameter perturbation rate and the frequency domain data is trained using the parameter correction CNN model. The time series trajectory of the actually measured variables is Fourier transformed and brought into the trained parameter correction CNN model. The obtained parameter value is the corrected error-dominant parameter.

[0026] Furthermore, the architecture of the parameter selection CNN model includes three convolutional layers and two fully connected layers, and the functions corresponding to the three convolutional layers are as follows:

[0027] LeakyReLU (Leaky Rectified Linear Unit) activation function:

[0028]

[0029] ELU (Exponential Linear Unit) activation function:

[0030]

[0031] PreLU (Parametric Rectified Linear Unit) activation function:

[0032]

[0033] In the formula, parameters a and b are constants, taking values ​​of 0.1 and 1 respectively, and parameter c is the learnable data;

[0034] x refers to the linear combination of the input data after weighting and bias. Specifically, the result of the convolution operation of the first layer (or the original features directly transmitted by the input layer), that is, the value after the convolution kernel acts on the input data and adds the bias, is x in the LeakyReLU activation function; the output value of the LeakyReLU activation function is x in the ELU activation function; the output value of the ELU activation function is x in the PreLU activation function;

[0035] Furthermore, the architecture of the parameter correction CNN model includes 2 convolutional layers, 1 feature fusion layer and 3 fully connected layers.

[0036] The present invention also provides a power system simulation error parameter correction device, comprising:

[0037] The screening module preliminarily screens the parameters in the system based on the trajectory sensitivity analysis method to obtain the preliminary parameter set;

[0038] The data module randomly perturbs the primary parameter set and obtains a data set of variables measured at each node of the system after each random perturbation;

[0039] The model module uses the data set to build a parameter selection CNN model and a parameter correction CNN model respectively;

[0040] Acquisition module, selects CNN model for training based on parameters, selects CNN model based on trained parameters, and selects error dominant parameters;

[0041] The correction module is trained based on a parameter correction CNN model, and the error-dominant parameters are corrected based on the trained parameter correction CNN model and a frequency domain feature extraction method.

[0042] The present invention also provides a power system simulation error parameter correction device, comprising:

[0043] Memory;

[0044] The processor is used to retrieve the computer execution program of the above-mentioned power system simulation error parameter correction method from the memory, and execute: based on the trajectory sensitivity analysis method, preliminarily screen the parameters in the system to obtain a preliminary parameter set; randomly perturb the preliminary parameter set to obtain a data set of variables measured at each node of the system after each random perturbation; using the data set, respectively construct a parameter selection CNN model and a parameter correction CNN model; based on the parameter selection CNN model, train, and based on the trained parameter selection CNN model, select error-dominant parameters; based on the parameter correction CNN model, train, and based on the trained parameter correction CNN model and the frequency domain feature extraction method, correct the error-dominant parameters.

[0045] The present invention also provides a storage medium storing a computer execution program, wherein the computer execution program is used to be called by a processor to execute the steps of the above-mentioned power system simulation error parameter correction method.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention can intelligently extract the correlation and difference rules of the variable trajectory of each node from the perspective of the system as a whole, and further match them with the system error characteristics to achieve accurate screening of the error-dominant parameters. In addition, the comprehensive frequency domain characteristics of the error-dominant parameters under random perturbation are captured based on CNN technology, and matched with the actual error trajectory, thereby achieving intelligent correction of the error-dominant parameters.

[0048] The present invention makes full use of the sampling data of each node and establishes a mapping relationship based on the difference in the impact of parameter changes on the variables of each node. Compared with the traditional method of constructing a mapping relationship of the degree of impact of parameter changes on a single variable, the data utilization rate is greatly improved, so that more feature information can be discovered.

[0049] The selection of error-dominant parameters in the present invention is highly accurate and representative, the accuracy of parameter correction results is high, error-dominant parameter intelligence and intelligent correction of parameters are realized, the overall structure of the model is clear, and the solution is universal. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic flow chart of a method for correcting power system simulation error parameters provided for the implementation of the present invention.

[0051] Figure 2 It is a schematic diagram of the CNN model architecture for parameter selection provided by the implementation of the present invention;

[0052] Figure 3 It is a schematic diagram of the parameter correction CNN model architecture provided by the implementation of the present invention;

[0053] Figure 4 It is a schematic diagram of an IEEE standard 3-machine 9-node system provided by the implementation of the present invention;

[0054] Figure 5 yes Figure 4 Schematic diagram of the difference between the simulated and measured trajectories at midpoint 2 (B2);

[0055] Figure 6 yes Figure 5 Schematic diagram of the simulation trajectory after parameter correction. DETAILED DESCRIPTION

[0056] The present invention is further described in detail below through the accompanying drawings and specific embodiments.

[0057] like Figure 1 As shown, a method for correcting power system simulation error parameters comprises the following steps:

[0058] Based on the trajectory sensitivity analysis method, the parameters in the system are preliminarily screened. Before extracting the global variable features of the power system, the parameters in the system need to be preliminarily screened to reduce the cost of subsequent calculations. The screening method is to use the local linearization method in the disturbance analysis to calculate the trajectory of the sensitivity of different parameters to various variables:

[0059]

[0060] Among them, x a anda They represent the responses of the state variable x and the algebraic variable y when the specific parameter a changes. The trajectory of the calculated sensitivity is a discrete time series, in which each time sampling point corresponds to a sensitivity data q at that moment. i The energy S contained in the sensitivity trajectory reflects the degree of influence of parameter changes on the variable:

[0061]

[0062] According to the value of S, the sensitive parameters corresponding to each variable are selected and the union of the parameters is taken to construct a preliminary dominant parameter set P, namely the primary dominant parameter set.

[0063] To realize intelligent identification of parameters to be corrected based on deep learning methods, it is necessary to make a large number of perturbations to the preliminary dominant parameter set, and then obtain the data of variables of each node of the system under different conditions.

[0064] Theil inequality coefficient method reflects the relative position and amplitude change characteristics of the signal through trend error and fluctuation error respectively, and defines the similarity of time series data with a value between 0 and 1, which makes it have a unique advantage in evaluating the similarity of two sets of time series signals. Its form is:

[0065]

[0066] Since the time series trajectory after the system disturbance is a decay process, the first few swings of the trajectory contain more energy, so the difference characteristics are often more significant. In order to more clearly obtain the characteristics contained in different node variables, the Gaussian function is combined to amplify the difference in waveforms. Finally, the similarity index is defined

[0067]

[0068] Among them, the time series trajectory Y of the variables obtained by simulation t As a benchmark, is the variable time series data obtained after perturbation, n is the length of the time series, and d is a constant that uses the fluctuation duration as the amplification factor after the system disturbance.

[0069] According to the variable type of high-frequency sampling of WAMS (Wide Area Measurement System), after the variable trajectory of each node is processed by data, an element matrix A with n rows and m columns is finally obtained. The number of rows is the number of perturbations, and the number of columns is the total number of measured node variables, which is in the form of:

[0070]

[0071] Correspondingly, after each perturbation of the primary selected parameter, the relative change rate of the perturbation value of the parameter relative to the original simulation value is taken as another n-row z-column matrix B. The number of rows is the number of perturbations, and the number of columns is the number of primary selected dominant parameters, which is in the form of:

[0072]

[0073] After the above data processing, matrices A and B have been processed into a data set C that conforms to the CNN training format:

[0074] C = [AB];

[0075] Use k% of the standard deviation as noise data added to the feature variable, and use u% of the standard deviation as noise data added to the target variable. The values ​​of k and u depend on the desired generalization ability.

[0076] After dividing the training set and the test set, the neural network training framework is constructed. LeakyReLU and ELU activation functions are used to ensure the gradient flow of negative values, and the PreLU activation function is used to further enhance the expressiveness of the model. The parameters a and b are constants, which are 0.1 and 1 respectively, and the parameter c is learnable.

[0077] Parameter selection CNN model architecture such as Figure 2 As shown, it includes three convolutional layers (Conv.1D) and two fully connected layers (Dense).

[0078]

[0079] The model implements a regression problem. The identification of error-dominant parameters should be considered as a classification problem, that is, the elements in the primary selection of dominant parameters are divided into error-dominant parameters and non-essential correction parameters. Therefore, this goal can be achieved by defining a numerical deviation indicator b, thereby improving the convergence of learning. The definition of the threshold indicator b refers to the numerical range f of the target variable in the training set. max ~f min , the size of e depends on the accuracy of error parameter adjustment required in engineering practice:

[0080] b=max(|f max |,|f min |)×e%.

[0081] After the model is well trained, the error-dominant parameters can be identified. By calculating the similarity of the measured variable trajectory to the simulated trajectory, the similarity dataset of each variable trajectory is obtained. By inputting the data set into the parameter selection CNN model, the offset data of each dominant parameter can be obtained. Based on the threshold index b, the parameter with the largest offset is selected as the selected parameter, that is, the error dominant parameter.

[0082] In order to accurately correct the selected error-dominant parameters, random perturbations are performed based on the selected error-dominant parameters to obtain the trajectory data of the error variables of the nodes near the fault point. The data of different variables are transformed into the frequency domain, and the discrete Fourier transform method (DFT) is used. The feature extraction is based on CNN, and the mapping relationship between the spectrum features and the parameter perturbation rate is obtained after feature fusion, such as Figure 3 As shown in the figure, the parameter correction CNN model architecture includes 2 convolutional layers (Conv.1D), 1 feature fusion layer, and 3 fully connected layers (Dense).

[0083] When the model is well trained, the measured trajectory data is injected into the trained parameter correction CNN model to obtain the correction values ​​of the error-dominant parameters.

[0084] The present invention also provides a power system simulation error parameter correction device, comprising:

[0085] The screening module preliminarily screens the parameters in the system based on the trajectory sensitivity analysis method to obtain the preliminary parameter set;

[0086] The data module randomly perturbs the primary parameter set and obtains a data set of variables measured at each node of the system after each random perturbation;

[0087] The model module uses the data set to build a parameter selection CNN model and a parameter correction CNN model respectively;

[0088] Acquisition module, selects CNN model for training based on parameters, selects CNN model based on trained parameters, and selects error dominant parameters;

[0089] The correction module is trained based on a parameter correction CNN model, and corrects the error-dominant parameters based on the trained parameter correction CNN model and a frequency domain feature extraction method.

[0090] The present invention also provides a power system simulation error parameter correction device, comprising:

[0091] Memory;

[0092] The processor is used to retrieve the computer execution program of the above-mentioned power system simulation error parameter correction method from the memory, and execute: based on the trajectory sensitivity analysis method, preliminarily screen the parameters in the system to obtain a preliminary parameter set; randomly perturb the preliminary parameter set to obtain a data set of variables measured at each node of the system after each random perturbation; using the data set, respectively construct a parameter selection CNN model and a parameter correction CNN model; based on the parameter selection CNN model, train, and based on the trained parameter selection CNN model, select error-dominant parameters; based on the parameter correction CNN model, train, and based on the trained parameter correction CNN model and the frequency domain feature extraction method, correct the error-dominant parameters.

[0093] The present invention also provides a storage medium storing a computer execution program, wherein the computer execution program is used to be called by a processor to execute the steps of the above-mentioned power system simulation error parameter correction method.

[0094] Below Figure 4 The IEEE standard 3-machine 9-node system shown in the figure is used as an example for analysis and verification.

[0095] Assume that in this system, due to the difference between the parameters set in the simulation and the real parameter data of the system, there is an error between the variable trajectory obtained by simulation and the actual trajectory. The data in the standard system are used as real data, and some parameters in the system are randomly selected to perform random perturbations on the values. The system after the modified parameters is used as the simulation scenario.

[0096] The generator is selected as a fourth-order model, and the excitation system is selected as a second-order model. The fault occurs at node 7 (B7) and the fault duration is 0.1s.

[0097] The specific parameter change settings are shown in Table 1, where the digital subscripts are the generator numbers:

[0098] Table 1 Parameter data

[0099]

[0100] Take node 2 as an example to show the difference between the simulation and the actual measurement trajectory. Figure 5 As shown;

[0101] Firstly, the primary dominant parameters are obtained according to the error variables and through the sensitivity method, which are: Kw1, M1, M2, M, 3T0d′, 2X, 2d′, X2d, X2q, X3q′.

[0102] The data are randomly perturbed 10,000 times. The perturbation method is to randomly select 1 to 10 of the 10 parameters. The perturbation range is 70% to 130% of the original simulation value of the parameter.

[0103] The names of the variable trajectories obtained are the phase angle, voltage, active power, reactive power and speed of each node.

[0104] The trajectories of these variables are processed, and the similarity of the trajectories of the perturbed variables relative to the trajectories of the simulated variables is calculated. Finally, an element matrix A with 10,000 rows and 26 columns and an element matrix B with 10,000 rows and 10 columns are obtained.

[0105] Construct a parameter selection CNN model, train the mapping relationship between matrix A and matrix B, divide the data of the validation set to 20% of the total data, and define the numerical deviation indicator b as ±5%.

[0106] The similarity between the actual measured trajectory and the simulated trajectory is calculated. After the parameter selection CNN model is well trained, the trajectory similarity value is brought in. According to the definition range of the deviation index b, the error dominant parameters are obtained as M1, Xd2′, and Xq2.

[0107] The three parameters were randomly perturbed within a range of 70% to 130% of the original simulation values, with a total of 3000 random perturbations, and the variable time series data of the nodes near the fault point were extracted.

[0108] The time series data of the variables are Fourier transformed to obtain the spectrum data of the error-dominant parameter under different perturbations. A parameter correction CNN model is constructed to train the relationship between the spectrum data of different variable types and the parameter perturbation rate, and perform feature superposition.

[0109] After the model training is completed, the trajectory of the actual variable is transformed into the model in the frequency domain to obtain the data correction results of the error-dominant parameters, as shown in Table 2;

[0110] Table 2 Parameter calibration data

[0111]

[0112] The simulation results after parameter correction are shown in Figure 6 shown.

[0113] From the data in Table 1, it can be seen that there are 7 parameters with errors in the system, namely M1, M2, Xd2′, Xq2, Tq01′, Xd1, and Xd3. After screening the parameters, only 3 parameters are retained, namely M1, Xd2′, and Xq2. These parameters are all modified parameters, while the remaining 4 parameters have relatively small errors or do not have a significant effect on the simulation errors. Therefore, these 4 parameters will not have a significant impact on the accuracy of system simulation, so they can be used as non-essential correction parameters, thereby reducing the computational cost of parameter correction. Prevent the occurrence of multiple solution problems and dimensional explosion phenomena.

[0114] In time domain analysis, when the maximum value of parameter sensitivity is used as the leading parameter selection indicator, in addition to the three error-dominant parameters M1, Xd2′, and Xq2, some non-error parameters such as Xq3 and Xq2′ will also be selected, which will greatly increase the correction cost of error parameters. At the same time, the modification of non-error parameters may occur during parameter correction. In frequency domain analysis, the value of the sampling frequency will directly affect the selection of parameters. Otherwise, in frequency matching, if mode aliasing occurs, the parameters that need to be corrected will not be correctly identified, and whether the selection of sensitive frequency points is unique also limits the universality of the method.

[0115] From Table 2 and Figure 6 In the figure, it can be seen intuitively that after the parameters are corrected, the simulation results are well matched to the actual trajectory. Taking the trajectory energy of node 2 before and after phase angle trajectory correction as an example, within 3 seconds, the simulation error of trajectory energy before parameter correction is 20.7%, and the error of trajectory energy after correction is 4.5%, which greatly improves the credibility of the simulation results.

[0116] The above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A method for correcting power system simulation error parameters, characterized in that: The following steps are involved: Based on the trajectory sensitivity analysis method, the parameters in the system are preliminarily screened to obtain the preliminary parameter set; Perform random perturbations on the initially selected parameter set to obtain a data set of variables measured at each node of the system after each random perturbation; Using the data set, we build a parameter selection CNN model and a parameter correction CNN model respectively; Select CNN model for training based on parameters, select CNN model based on trained parameters, and select error-dominant parameters; Training is performed based on a parameter correction CNN model, and the error-dominant parameters are corrected based on the trained parameter correction CNN model and a frequency domain feature extraction method.

2. The method for correcting power system simulation error parameters according to claim 1, characterized in that: The specific steps of performing preliminary screening of parameters in the system based on the trajectory sensitivity analysis method include: using the local linearization method in the disturbance analysis to calculate the trajectory of the sensitivity of different parameters to the variables measured at each node of the system.

3. The method for correcting power system simulation error parameters according to claim 1, characterized in that: The pair of primary parameter sets are randomly perturbed to obtain a data set of variables measured at each node of the system after each random perturbation, including: The primary parameter set is randomly perturbed to obtain the time series trajectory data of the variables measured at each node in the system, and the similarity of the time series trajectory data of the variables is calculated to obtain the trajectory similarity between the time series trajectory of the variables and the simulation trajectory.

4. The method for correcting a power system simulation error parameter according to claim 3, characterized in that: The method of selecting a CNN model for training based on parameters, selecting a CNN model based on trained parameters, and selecting error-dominant parameters includes: The mapping relationship between the training trajectory similarity and the parameter perturbation rate is established. The trajectory similarity between the time series trajectory of the actually measured variable and the simulation trajectory is input into the parameter selection CNN model. The parameter selection CNN model outputs a set of parameter perturbation rate values. The absolute value of the value represents the degree of influence of the parameter on the time series waveform. The preliminary dominant parameter whose absolute value is higher than the preset deviation threshold is the error dominant parameter to be corrected.

5. The method for correcting a power system simulation error parameter according to claim 1, characterized in that: The architecture of the parameter selection CNN model includes three convolutional layers and two fully connected layers. The functions corresponding to the three convolutional layers are as follows: LeakyReLU activation function: ELU activation function: PreLU activation function: In the formula, parameters a and b are constants, taking the values ​​of 0.1 and 1 respectively, and parameter c is the learnable data; x refers to the linear combination of the input data after weighting and bias term.

6. The method for correcting power system simulation error parameters according to claim 1, characterized in that: The architecture of the parameter correction CNN model includes 2 convolutional layers, 1 feature fusion layer and 3 fully connected layers.

7. The method for correcting power system simulation error parameters according to claim 1, characterized in that: The method of training based on the parameter correction CNN model and correcting the error-dominant parameter based on the trained parameter correction CNN model and the frequency domain feature extraction method includes: The error-dominant parameters are randomly perturbed to obtain the time series trajectory of the variables near the fault point. The time series trajectory of the variables is Fourier transformed, and the mapping relationship between the parameter perturbation rate and the frequency domain data is trained using the parameter correction CNN model. The time series trajectory of the actually measured variables is Fourier transformed and brought into the trained parameter correction CNN model. The obtained parameter value is the corrected error-dominant parameter.

8. A power system simulation error parameter correction device, characterized in that: include: The screening module preliminarily screens the parameters in the system based on the trajectory sensitivity analysis method to obtain the preliminary parameter set; The data module randomly perturbs the primary parameter set and obtains a data set of variables measured at each node of the system after each random perturbation; The model module uses the data set to build a parameter selection CNN model and a parameter correction CNN model respectively; Acquisition module, selects CNN model for training based on parameters, selects CNN model based on trained parameters, and selects error dominant parameters; The correction module is trained based on a parameter correction CNN model, and the error-dominant parameters are corrected based on the trained parameter correction CNN model and a frequency domain feature extraction method.

9. A power system simulation error parameter correction device, characterized in that: include: Memory; The processor is used to retrieve the computer execution program of the above-mentioned power system simulation error parameter correction method from the memory, and execute: based on the trajectory sensitivity analysis method, preliminarily screen the parameters in the system to obtain a preliminary parameter set; randomly perturb the preliminary parameter set to obtain a data set of variables measured by each node of the system after each random perturbation; Using the data set, we build a parameter selection CNN model and a parameter correction CNN model respectively; Select CNN model for training based on parameters, select CNN model based on trained parameters, and select error-dominant parameters; Training is performed based on a parameter correction CNN model, and the error-dominant parameters are corrected based on the trained parameter correction CNN model and a frequency domain feature extraction method.

10. A storage medium, characterized in that: A computer execution program is stored, and the computer execution program is used to be called by a processor to execute the steps of a power system simulation error parameter correction method according to any one of claims 1 to 7.

Images