High-voltage power grid current transformer transmission signal distortion reconstruction method and system

Abstract

The invention discloses a high-voltage power grid current transformer transmission signal distortion reconstruction method and system, and the method comprises the steps: building a transfer function model of a main circuit and an electromagnetic current transformer through electromagnetic transient simulation software, and collecting the current data of a primary side and a secondary side of the electromagnetic current transformer, fault working conditions and simulation parameters are set; carrying out normalization processing on the collected primary side current data and the collected secondary side current data; building a mapping model framework from the secondary side current to the primary side current based on the stacked long-short-term memory neural network; obtaining a mapping model from the secondary side current to the primary side current based on the mapping model framework; and taking the secondary side current data as model input, and taking model output as prediction data of the primary side current, thereby realizing reconstruction of the primary side current data of the electromagnetic current transformer. According to the method, the problem of distortion of the secondary transient quantity current signal of the electromagnetic current transformer of the high-voltage power grid can be solved, and the reliability of transient quantity protection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission line protection, and in particular to a method and system for reconstructing distortion of a current transformer transmission signal of a high-voltage power grid. Background Art

[0002] After a high-voltage transmission line fails, the current wave will be reflected back and forth between the fault point and the substation busbar, and the transient voltage and current generated contain rich fault information. Unlike traditional relay protection that uses power frequency information, transient quantity protection that uses transient quantity information has faster performance and is not affected by factors such as current transformer saturation, power system oscillation, and transition resistance.

[0003] In traditional substations, line monitoring devices cannot directly obtain transient information of the line, but must transmit transient information through a transmission circuit composed of a capacitor voltage transformer or an electromagnetic current transformer, a secondary cable, and a secondary load. Although electromagnetic current transformers have good high-frequency transmission characteristics and have been used in engineering, they still cause waveform attenuation and ringing phenomena that cause severe waveform distortion when faced with high-frequency transient information, thereby affecting the reliability of transient protection action. Therefore, it is crucial to deduce the correct primary current signal from the distorted secondary current signal. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a method and system for reconstructing the distortion of the transmission signal of the high-voltage power grid current transformer, which can solve the problem of distortion of the secondary transient current signal of the electromagnetic current transformer of the high-voltage power grid and improve the reliability of transient protection.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A method for reconstructing distortion of a current transformer transmission signal of a high-voltage power grid, comprising:

[0007] Use electromagnetic transient simulation software to build the transfer function model of the main circuit and electromagnetic current transformer, collect the primary and secondary current data of the electromagnetic current transformer, and set the fault conditions and simulation parameters;

[0008] The collected primary and secondary current data are normalized, and the normalized data are divided into a training set, a validation set, and a test set;

[0009] Based on the stacked long short-term memory neural network, a mapping model framework from secondary current to primary current is built;

[0010] Setting the learning rate of the mapping model framework, using the data of the training set for training, and obtaining the weight parameters of the mapping model framework; using the data of the validation set to evaluate the mapping model framework, and using the data of the test set to evaluate the generalization ability of the mapping model framework, adjusting the learning rate, and performing inverse normalization to obtain a mapping model from the secondary side current to the primary side current;

[0011] The mapping model is deployed to an actual power system, the secondary side current data of the electromagnetic current transformer is used as the model input, and the model output is used as the prediction data of the primary side current, thereby realizing the reconstruction of the primary side current data of the electromagnetic current transformer.

[0012] Furthermore, the transfer function model of the main circuit and the electromagnetic current transformer is built by electromagnetic transient simulation software, the primary and secondary current data of the electromagnetic current transformer are collected, and the fault condition and simulation parameters are set, including:

[0013] The two-end power supply system is built through electromagnetic transient simulation software. The two-end power supplies are connected through transformers, busbars and transmission lines to build the main circuit model.

[0014] A transfer function is obtained according to the circuit model of the electromagnetic current transformer, and the coefficient of the transfer function is written into the N-order transfer function element of the electromagnetic transient simulation software as the transfer function model of the electromagnetic current transformer in the electromagnetic transient simulation software; the primary side current signal measured by the current measuring element is input into the transfer function model to obtain the secondary side current of the electromagnetic current transformer;

[0015] Fault conditions and simulation parameters are set, wherein the fault conditions include fault type, fault initial phase angle, fault transition resistance and fault location, and the simulation parameters include secondary side circuit transmission parameters of the electromagnetic current transformer.

[0016] Furthermore, the collected primary and secondary current data are normalized, and the normalized data are divided into a training set, a validation set, and a test set, including:

[0017] The primary and secondary current data of the electromagnetic current transformer are normalized by using the deviation normalization method, the primary and secondary current dimensions are eliminated, and the currents are mapped to the interval [0, 1].

[0018] A sample data set is constructed according to the normalized primary and secondary current data, wherein the sample data set includes multiple data matrices; the data of the sample data set are randomly shuffled with rows as basic units, the internal order of each row is kept unchanged during the shuffling process, and the multiple data matrices are shuffled in the same way, so that each row of the multiple shuffled data matrices still has a corresponding relationship, and then divided into a training set, a validation set and a test set according to a preset ratio.

[0019] Furthermore, the stacked long short-term memory neural network is used to build a mapping model framework from the secondary current to the primary current, including:

[0020] Constructing a stacked long short-term memory neural network, the stacked long short-term memory neural network includes an input layer, a timing layer and a fully connected layer; the neurons of the timing layer can receive the output vector of the neurons of the previous layer, and transform it through the weight matrix and the bias vector to obtain the output value; the fully connected layer can receive the output vector of the neurons of the previous layer, and perform linear transformation through the weight matrix and the bias vector to obtain new state information;

[0021] A mapping model framework of secondary-side current to primary-side current is built based on a stacked long short-term memory neural network, and model quality evaluation criteria are set, which include mean square error and mean absolute error.

[0022] Further, the setting of the learning rate of the mapping model framework, using the data of the training set for training, obtaining the weight parameters of the mapping model framework; using the data of the validation set to evaluate the mapping model framework, and using the data of the test set to evaluate the generalization ability of the mapping model framework, adjusting the learning rate, performing inverse normalization, and obtaining a mapping model from the secondary side current to the primary side current, includes:

[0023] Setting the learning rate of the mapping model framework, combining the data of the training set, adjusting the parameters according to the gradient descent method, and updating the weight parameters of the mapping model framework;

[0024] Based on the updated mapping model framework, the mean square error and the mean absolute error are calculated and judged using the validation set data until the iteration end condition is met. The weight parameter after the iteration end is the value at the end of the mapping model framework training.

[0025] Bring the iteratively obtained weight parameters into the mapping model framework, predict the data of the test set, calculate the corresponding mean square error and mean absolute error, so as to evaluate the generalization effect of the mapping model framework and determine whether to continue updating the weight parameters;

[0026] The predicted value of the actual current is obtained by the inverse normalization method as the reconstructed primary current value, so as to realize the reconstruction of the primary current and obtain the mapping model of the secondary current to the primary current.

[0027] A high-voltage power grid current transformer transmission signal distortion reconstruction system, comprising:

[0028] The transfer function model building module is configured to build a transfer function model of the main circuit and the electromagnetic current transformer through electromagnetic transient simulation software, collect primary and secondary current data of the electromagnetic current transformer, and set fault conditions and simulation parameters;

[0029] The data normalization and division module is configured to normalize the collected primary and secondary current data, and divide the normalized data into a training set, a validation set, and a test set;

[0030] A mapping model framework building module is configured to build a mapping model framework from secondary current to primary current based on a stacked long short-term memory neural network;

[0031] A mapping model framework training module is configured to set a learning rate of the mapping model framework, use the data of the training set for training, and obtain weight parameters of the mapping model framework; use the data of the validation set to evaluate the mapping model framework, and use the data of the test set to evaluate the generalization ability of the mapping model framework, adjust the learning rate, perform inverse normalization, and obtain a mapping model from secondary current to primary current;

[0032] The transmission signal distortion reconstruction module is configured to deploy the mapping model to the actual power system, use the secondary side current data of the electromagnetic current transformer as the model input, and use the model output as the predicted data of the primary side current, thereby realizing the reconstruction of the primary side current data of the electromagnetic current transformer.

[0033] Furthermore, in the transfer function model construction module, the transfer function model of the main circuit and the electromagnetic current transformer is constructed by electromagnetic transient simulation software, the primary and secondary current data of the electromagnetic current transformer are collected, and the fault conditions and simulation parameters are set, including:

[0034] The two-end power supply system is built through electromagnetic transient simulation software. The two-end power supplies are connected through transformers, busbars and transmission lines to build the main circuit model.

[0035] A transfer function is obtained according to the circuit model of the electromagnetic current transformer, and the coefficient of the transfer function is written into the N-order transfer function element of the electromagnetic transient simulation software as the transfer function model of the electromagnetic current transformer in the electromagnetic transient simulation software; the primary side current signal measured by the current measuring element is input into the transfer function model to obtain the secondary side current of the electromagnetic current transformer;

[0036] Fault conditions and simulation parameters are set, wherein the fault conditions include fault type, fault initial phase angle, fault transition resistance and fault location, and the simulation parameters include secondary side circuit transmission parameters of the electromagnetic current transformer.

[0037] Furthermore, in the data normalization and division module, the collected primary and secondary current data are normalized, and the normalized data are divided into a training set, a validation set and a test set, including:

[0038] The primary and secondary current data of the electromagnetic current transformer are normalized by using the deviation normalization method, the primary and secondary current dimensions are eliminated, and the currents are mapped to the interval [0, 1].

[0039] A sample data set is constructed according to the normalized primary and secondary current data, wherein the sample data set includes multiple data matrices; the data of the sample data set are randomly shuffled with rows as basic units, the internal order of each row is kept unchanged during the shuffling process, and the multiple data matrices are shuffled in the same way, so that each row of the multiple shuffled data matrices still has a corresponding relationship, and then divided into a training set, a validation set and a test set according to a preset ratio.

[0040] Furthermore, in the mapping model framework construction module, a mapping model framework from secondary current to primary current is constructed based on a stacked long short-term memory neural network, including:

[0041] Constructing a stacked long short-term memory neural network, the stacked long short-term memory neural network includes an input layer, a timing layer and a fully connected layer; the neurons of the timing layer can receive the output vector of the neurons of the previous layer, and transform it through the weight matrix and the bias vector to obtain the output value; the fully connected layer can receive the output vector of the neurons of the previous layer, and perform linear transformation through the weight matrix and the bias vector to obtain new state information;

[0042] A mapping model framework of secondary-side current to primary-side current is built based on a stacked long short-term memory neural network, and model quality evaluation criteria are set, which include mean square error and mean absolute error.

[0043] Furthermore, in the mapping model framework training module, the learning rate of the mapping model framework is set, the data of the training set is used for training, and the weight parameters of the mapping model framework are obtained; the mapping model framework is evaluated using the data of the validation set, and the generalization ability of the mapping model framework is evaluated using the data of the test set, the learning rate is adjusted, and denormalization is performed to obtain a mapping model from the secondary side current to the primary side current, including:

[0044] Setting the learning rate of the mapping model framework, combining the data of the training set, adjusting the parameters according to the gradient descent method, and updating the weight parameters of the mapping model framework;

[0045] Based on the updated mapping model framework, the mean square error and the mean absolute error are calculated and judged using the validation set data until the iteration end condition is met. The weight parameter after the iteration end is the value at the end of the mapping model framework training.

[0046] Bring the iteratively obtained weight parameters into the mapping model framework, predict the data of the test set, calculate the corresponding mean square error and mean absolute error, so as to evaluate the generalization effect of the mapping model framework and determine whether to continue updating the weight parameters;

[0047] The predicted value of the actual current is obtained by the inverse normalization method as the reconstructed primary current value, so as to realize the reconstruction of the primary current and obtain the mapping model of the secondary current to the primary current.

[0048] The beneficial effects of the present invention are:

[0049] Aiming at the problem of transient current signal distortion caused by the transmission circuit composed of electromagnetic current transformer, secondary cable and secondary load, the present invention obtains the secondary current to primary current mapping model through fault condition simulation, data preprocessing, model training, etc., which has good adaptability to different fault conditions, different secondary transmission circuit parameters, etc., and can improve the reliability of transient protection action; at the same time, the amount of calculation is mainly concentrated in the offline part, which is easy to deploy to the actual power system, and ensures the timeliness of the online monitoring system to reconstruct the current, thereby ensuring the speed of the transient protection action. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A current reconstruction flow chart of an embodiment of the present invention.

[0051] Figure 2 Schematic diagram of the main circuit model of an embodiment of the present invention.

[0052] Figure 3 Schematic diagram of a circuit model of an embodiment of the present invention.

[0053] Figure 4 Schematic diagram of an equivalent model of an electromagnetic current transformer circuit according to an embodiment of the present invention.

[0054] Figure 5 It is a simplified equivalent circuit schematic diagram of the secondary circuit of the electromagnetic current transformer according to an embodiment of the present invention.

[0055] Figure 6 It is a graph of the artificial intelligence model training loss function of an embodiment of the present invention.

[0056] Figure 7 It is a comparison chart of the primary value, secondary value and predicted value of the current according to the embodiment of the present invention.

[0057] Figure 8 It is a comparison chart of the primary value, secondary value and predicted value of the current according to the embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific implementation methods of the present invention are now described. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the embodiments described are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0059] Example 1

[0060] In traditional substations, line monitoring devices cannot directly obtain transient information of the line, but must transmit transient information through a transmission circuit composed of a capacitor voltage transformer or an electromagnetic current transformer, a secondary cable, and a secondary load. Although electromagnetic current transformers have good high-frequency transmission characteristics and have been used in engineering, they still cause waveform attenuation and ringing phenomena that cause severe waveform distortion when faced with high-frequency transient information, thereby affecting the reliability of transient protection action. Therefore, it is crucial to deduce the correct primary current signal from the distorted secondary current signal.

[0061] With the continuous development of the field of artificial intelligence, artificial intelligence models suitable for various scenarios have emerged. Among them, the recurrent neural network (RNN) introduces the concept of time series and has a good effect on the prediction of time series information. The long short-term memory neural network (LSTM) model is an improvement on the RNN model. It inherits the good performance of the RNN model in predicting time series information, and solves the problems of gradient disappearance, gradient explosion, long-term dependence, etc. in the RNN model, so that the model can better utilize long-distance information. By taking the distorted secondary signal as the input of the LSTM model and the primary signal as the output, it is expected that the distorted secondary current signal can be accurately reconstructed into the primary current signal using the LSTM model.

[0062] Based on this, this embodiment provides a method for reconstructing distortion of a current transformer transmission signal of a high-voltage power grid, including:

[0063] Use electromagnetic transient simulation software to build the transfer function model of the main circuit and electromagnetic current transformer, collect the primary and secondary current data of the electromagnetic current transformer, and set the fault conditions and simulation parameters;

[0064] The collected primary and secondary current data are normalized, and the normalized data are divided into a training set, a validation set, and a test set;

[0065] Based on the stacked long short-term memory neural network, a mapping model framework from secondary current to primary current is built;

[0066] Set the learning rate of the mapping model framework, use the data of the training set for training, and obtain the weight parameters of the mapping model framework; use the data of the validation set to evaluate the mapping model framework, and use the data of the test set to evaluate the generalization ability of the mapping model framework, adjust the learning rate, perform inverse normalization, and obtain the mapping model from the secondary current to the primary current;

[0067] The mapping model is deployed to the actual power system, the secondary side current data of the electromagnetic current transformer is used as the model input, and the model output is used as the predicted data of the primary side current, thereby realizing the reconstruction of the primary side current data of the electromagnetic current transformer.

[0068] Specifically, Figure 1 As shown, the method for reconstructing the distortion of the current transformer transmission signal of the high-voltage power grid of this embodiment can be implemented by the following steps:

[0069] Step 1: By building the main circuit and electromagnetic current transformer transfer function model in the electromagnetic transient simulation software, and setting different secondary circuit transmission parameters of the electromagnetic current transformer, different fault types, fault locations, fault phase angles, and transition resistances, a large amount of primary and secondary side current simulation data can be obtained.

[0070] Step 2: Normalize the primary and secondary side current data collected in Step 1, and randomly divide the normalized data into training set, validation set and test set.

[0071] Step 3: Based on the stacked long short-term memory neural network (LSTM), a secondary current to primary current mapping model framework is built by setting the quality evaluation criteria.

[0072] Step 4: Set the learning rate of the model framework in Step 3, use the training set data in Step 2 for training, obtain the weight parameters of the secondary current to primary current mapping model framework, evaluate the model using the validation set data in Step 2, evaluate the model generalization ability using the test set data in Step 2, adjust the model learning rate, and then perform denormalization to obtain the secondary current to primary current mapping model.

[0073] Step 5: Deploy the secondary current to primary current mapping model in Step 4 to the actual power system, use the secondary current data of the electromagnetic current transformer as the input of the model, and use the output of the model as the predicted data of the primary current for protection logic judgment, thereby realizing the reconstruction of the primary current data of the electromagnetic current transformer.

[0074] Preferably, in this embodiment, Step 1 can be implemented by the following steps:

[0075] Step 1: Build the main circuit model. Use electromagnetic transient simulation software to build a two-terminal power supply system with a voltage level of 220kV. The two power sources are connected through transformers, busbars, and transmission lines. The main circuit model is as follows: Figure 2 The transmission line model selects the 3-conductor flat tower model in the frequency dependent (phase) model, and the length is set to 100km. The transmission line model is as follows Figure 3As shown in the figure, the total stray capacitance of the busbar and various devices connected to it to the ground is 0.1μF; the transformer is a three-phase two-winding transformer (3Phase2Winding Transformer); the current measuring element (Current Meter) is connected to the left busbar outlet to measure the primary current. Figure 2 In the figure, S represents the system power supply, T is the transformer, M is the local bus, N is the opposite bus, L1 is the transmission line, and k1 and k2 are the short-circuit fault locations that occur at 5 km and 80 km away from the local bus M. Figure 3 In the figure, G1 and G2 represent lightning protection wires, and C1, C2, and C3 represent three-phase transmission wires.

[0076] Step 2: According to the electromagnetic current transformer circuit model, obtain its transfer function, and build the transfer function model of the electromagnetic current transformer in the electromagnetic transient simulation software.

[0077] The equivalent model of electromagnetic current transformer circuit is as follows Figure 4 As shown, R1 is the equivalent resistance of the primary coil of the current transformer, R2 is the equivalent resistance of the secondary coil of the current transformer, k is the current transformer ratio, C 10 , C 20 are the self-capacitance of the primary and secondary sides respectively; C 120 is the mutual capacitance between the primary and secondary sides, and N1 and N2 are the number of coil turns on both sides respectively. Figure 4 The corresponding electromagnetic current transformer secondary simplified equivalent circuit is as follows: Figure 5 As shown, R C and L C They represent the total resistance and inductance of the cable lumped parameter model, R2 is the equivalent resistance of the secondary coil of the current transformer, and R L is the secondary load resistance of the current transformer, R m is the core loss resistance, C 2s is the stray capacitance of the current transformer to ground.

[0078] The transfer function H(s) of the electromagnetic current transformer corresponding to the simplified equivalent circuit is:

[0079]

[0080] R s =R2+R C +R L (2)

[0081] In the formula, R C and L C They represent the total resistance and inductance of the cable lumped parameter model, R2 is the equivalent resistance of the secondary coil of the current transformer, and R Lis the secondary load resistance of the current transformer, R m is the core loss resistance, C 2s is the stray capacitance of the current transformer to ground.

[0082] The coefficients of the transfer function are written into the Nth-order transfer function element of the electromagnetic transient simulation software. This element is the transfer function model of the electromagnetic current transformer in the electromagnetic transient simulation software. The primary side current signal measured by the current measuring element is input into the above transfer function model to obtain the secondary side current of the electromagnetic current transformer.

[0083] Step 3: Fault conditions and simulation parameter settings. Considering the influence of secondary circuit transmission parameters such as the secondary cable length and secondary load of the electromagnetic current transformer, as well as the fault type, fault initial phase angle, fault transition resistance and fault location on the secondary current waveform, simulation is carried out by setting different fault conditions and simulation parameters. The simulation platform sets the sampling rate f s The frequency is 1MHz, the fault occurrence time is 1s, and the fault duration is 0.5s.

[0084] The specific settings can be configured as follows:

[0085] Fault type: single-phase grounding (Ag, Bg, Cg), two-phase grounding fault (ABg, ACg, BCg), three-phase grounding fault (ABCg), three-phase short circuit (ABC), two-phase short circuit (AB, AC, BC), a total of 7 types of grounding faults and 4 types of interphase faults;

[0086] Fault initial phase angle: 0~90°, setting interval step 15°, a total of 7 fault phase angles;

[0087] Ground fault transition resistance: 0~300Ω, with 15Ω as the interval step, a total of 21 transition resistances;

[0088] Phase-to-phase fault transition resistance: 0~50Ω, with 10Ω as the interval step, a total of 6 transition resistances;

[0089] Fault location: 0~100km, with a step length of 5km, a total of 21 fault locations;

[0090] ·Cable length of secondary side of current transformer: 200~1000m, with 200m interval step, a total of 5 cable lengths;

[0091] Current transformer secondary load: 0.02~0.06Ω, with 0.02Ω as the interval step, a total of 3 types of secondary loads.

[0092] Step 4: Under the above model and simulation conditions, there are a total of 7×(7×21+4×6)×21×5×3=377055 fault scenarios. The electromagnetic transient simulation software is used to simulate the primary and secondary side current data of 1ms before and after the fault. At a sampling rate of 1MHz, 2000 data of each type of current can be obtained, thus forming a sample data set D:I apri ,I bpri ,I cpri ,I asec ,I bsec ,I csec , the specific structure can be:

[0093]

[0094]

[0095]

[0096]

[0097]

[0098]

[0099] Among them, the value range of m is 1, 2...377055, the value range of n is 1, 2, 3...2000, and I a、b、cpri[m,n] It represents the value of the nth sampling point of the primary current of phase a, phase b, and phase c in the mth simulation, I a、b、csec[m,n] Represents the value of the nth sampling point of the secondary side current of phase a, phase b, and phase c in the mth simulation.

[0100] Preferably, in this embodiment, Step 2 can be implemented by the following steps:

[0101] Step 1: Data normalization. Use the Min-Max normalization method to normalize the primary and secondary current data, eliminate the primary and secondary current dimensions, and map them to the [0,1] interval. The normalization calculation formula is as follows:

[0102]

[0103]

[0104] In the formula, I a、b、cpri[m,n] It represents the value of the nth sampling point of the primary current of phase a, phase b, and phase c in the mth simulation, I a、b、csec[m,n] I' represents the value of the nth sampling point of the secondary current of phase a, phase b, and phase c in the mth simulation. a、b、cpri[m,n] and I' a、b、csec[m,n] Ia、b、cpri[m,n] and I a、b、csec[m,n] The normalized result of I pri_min and I pri_max are the minimum and maximum values ​​of the primary current data, I sec_min and I sec_max are the minimum and maximum values ​​in the secondary current data respectively.

[0105] Using the normalized I' a、b、cpri[m,n] and I' a、b、csec[m,n] Construct a new sample dataset D':I' apri , I' bpri , I' cpri , I' asec , I' bsec , I' csec .

[0106] Step 2: Randomly shuffle the sample set data D' with rows as the basic unit. Keep the internal order of each row unchanged during the shuffling process, and ensure that the six matrices are shuffled in the same way to ensure that each row of the six matrices still has a corresponding relationship after shuffling. Then, divide it into training set, validation set and test set according to a preset ratio (for example, 7:2:1).

[0107] Preferably, in this embodiment, Step 3 can be implemented by the following steps:

[0108] Step 1: The stacked LSTM neural network is divided into an input layer, a timing layer, and a fully connected layer. First, the input layer of the stacked LSTM neural network is set to 2000 neurons, which can accept 2000 data inputs at the same time. During training, the training set data obtained in Step 2 is input row by row, and the input vector x = [I' a、b、csec[p,1] :I' a、b、csec[p,2000] ] T , p represents the pth row of the training set data, and its value is 1, 2, 3…, 263938.

[0109] Step 2: The time layer based on the stacked long short-term memory neural network is set to two layers, each layer has 128 neurons, and each time layer neuron receives the output vector of the previous layer neuron. The time layer neuron transforms the output vector of the previous layer neuron through the weight matrix and bias vector to obtain the output value. The output vector h of the first time layer [1] As shown in formula (5). The output vector h of the second time series layer [2] As shown in formula (6).

[0110] h [1] =T [1] ·x 2 +W [1] ·x+b [1] (5)

[0111] h [2] =T [2] ·h [1] 2 +W [2] ·h [1] +b [2] (6)

[0112] In the formula, h [1] is a 128-dimensional vector, which is the output vector of the first temporal layer; T [1] , W [1] is a matrix of size 128×2000, which is used to calculate the secondary weight matrix and the primary weight matrix from the input vector to the output vector; b [1] It is a 128-dimensional vector, which is the bias vector of the first temporal layer. [2] is a 128-dimensional vector, which is the output vector of the second temporal layer; T [2] , W [2] is a matrix of size 128×128, which is used to calculate the secondary weight matrix and the primary weight matrix from the input vector to the output vector of the second sequential layer; b [2] It is a 128-dimensional vector and is the bias vector of the second temporal layer.

[0113] Step 3: The fully connected layer based on the stacked long short-term memory neural network is set to 2000 neurons. Each neuron in the fully connected layer is connected to all neurons in the previous layer and can receive the output vector of the neurons in the previous layer. The neurons in the fully connected layer perform linear transformation on the output vector of the LSTM neurons in the previous layer through the weight matrix and the bias vector to obtain the new state information, as shown in formula (7):

[0114] y=W [3] ·h [2] +b [3] (7)

[0115] Where y is a 2000-dimensional vector, which is the output of the neurons in this layer, that is, the predicted normalized current value:

[0116]

[0117] W [3] is a matrix of size 2000×128, which is the weight matrix of the fully connected layer; b [3] It is a 2000-dimensional vector, which is the bias vector for calculating the output information.

[0118] Step 4: The secondary current to primary current mapping model framework can be constructed by the above formulas (5) to (7). In order to evaluate the accuracy of the mapping model framework, the model quality evaluation criteria are set. The mean square error (MSE) and mean absolute error (MAE) are used as the evaluation criteria for the accuracy of the mapping model framework. The calculation formulas for MSE and MAE are:

[0119]

[0120]

[0121] In the formula, I' a、b、cpri[p,n] is the actual value of the normalized current, is the predicted normalized current value, n=1,2…,2000.

[0122] Preferably, in this embodiment, Step 4 can be implemented by the following steps:

[0123] Step 1: Set the learning rate λ = 0.1 and adjust the parameters according to the gradient descent method. First, set the weight parameter T [1] ,W [1] ,b [1] ,T [2] ,W [2] ,b [2] ,W [3] ,b [3] A total of 803024 parameters are used in sequence θ: θ [1] ,θ [2] ,θ [3] ,…,θ [j] ,…,θ [803024] Represents, where j = 1, 2, 3, ..., 803024. The training set data obtained in Step 2 is used as the input of the model framework row by row, and the sum of MSE and MAE will be θ [j] Then set all the values ​​in the sequence θ to 1, and use the sequence θ 0 :θ 0 [1] ,θ 0 [2] ,θ 0 [3] ,…,θ 0 [j] ,…θ 0 [803024] According to formula (10), for each θ 0 [j] Find the partial derivative and get the sequence θ after iterative calculation 1 .

[0124]

[0125] In the formula, θ k-1 represents the sequence θ obtained after the k-1th iteration, θ k-1 [j] ,θ k [j] Respectively represent the parameters θ calculated after the k-1th and kth iterations [j] , k = 1, 2, 3... Then, the obtained sequence θ 1 The parameters are used to update the model framework weight parameters T [1] , W [1] 、b [1] , T [2] , W [2] 、b [2] , W [3] 、b [3] , and obtain a new model framework.

[0126] Step 2: Use the above optimized and updated model framework to calculate MSE and MAE on the validation set data. If MSE>K MSE or MAE>K MAE Then continue to perform the next round of iterative calculation according to step 1 to obtain a new model framework, and use the validation set data again to perform MSE and MAE calculation and discrimination until the iteration end condition is met, that is, the MSE and MAE on the validation set are both less than the threshold K MSE , K MAE。 Among them, the threshold K MSE =3×10 -5 , K MAE =3×10 -3 The weight parameter T after the iteration is completed [1] , W [1] 、b [1] , T [2] , W [2] 、b [2] , W [3] 、b [3] This is the final value at the end of the model framework training.

[0127] Step 3: Bring the above weight parameters into the secondary current to primary current mapping model framework, predict the current on the test set, calculate the MSE and MAE on the test set, and evaluate the generalization effect of the framework model. MSE or MAE>K MSE , then return to step 1 and set the learning rate to 0.9λ, and re-obtain the weight parameters until MSE <K MSE And MAE <K MSE , which indicates that the model has good generalization performance, and finally determines T[1] , W [1] 、b [1] , T [2] , W [2] 、b [2] , W [3] 、b [3] matrix.

[0128] Step 4: Perform inverse normalization to obtain the actual predicted value. Use Min-Max inverse normalization to obtain the predicted value of the actual current. The inverse normalization formula is as follows:

[0129]

[0130] In the formula, is the predicted normalized current value, is the predicted current value after denormalization, n=1,2…,2000.

[0131] In this way, the reconstruction of the primary side current is achieved, and a secondary side current to primary side current mapping model is obtained. The predicted current value after denormalization is the reconstructed primary side current value.

[0132] Preferably, in this embodiment, Step 5 can be implemented by the following steps:

[0133] Step 1: Deploy the final secondary current to primary current mapping model trained in Step 4 to the protection system at the transmission line protection installation, and use the electromagnetic current transformer secondary current data as the input data of the model.

[0134] Step 2: Current reconstruction is achieved. The model will calculate the output data based on the data input in step 1, that is, the reconstructed current data, which will be used for protection logic judgment to achieve reconstruction of the primary side current data of the electromagnetic current transformer.

[0135] In this embodiment, a data set is obtained through circuit simulation, and on this basis, a stacked long short-term memory neural network model is built, the model structure is determined, the data set is used for model training, and a mapping model from the secondary side current to the primary side current after the fault is established, and finally the trained mapping model is deployed in the current monitoring system. The method for reconstructing the distortion of the transmission signal of the high-voltage power grid current transformer of this embodiment can accurately reconstruct the primary side current based on the secondary side current of the electromagnetic current transformer, and has good adaptability to different fault conditions and different electromagnetic current transformer secondary circuit parameters; the amount of calculation is concentrated in the offline stage, and the amount of online calculation is small, which ensures the accuracy and timeliness of the electromagnetic current transformer transient current prediction, and is easy to deploy in the existing current monitoring system, thereby effectively improving the reliability of transient protection.

[0136] Example 2

[0137] This embodiment is based on embodiment 1:

[0138] This embodiment selects Figure 2 The two-terminal power supply system at the 220kV voltage level shown in the figure is taken as the research object. The two power sources are connected through transformers, busbars, and transmission lines. The transmission line model selects the 3-conductor flat tower model in the frequency dependent (phase) model. The length is set to 100km. The transmission line model is as follows Figure 3 As shown in the figure, the total stray capacitance of the busbar and various devices connected to it to the ground is 0.1μF; the transformer is a three-phase two-winding transformer (3Phase 2Winding Transformer); the current measuring element (Current Meter) is connected to the left busbar outlet to measure the primary side current. Figure 2 In the figure, S represents the system power supply, T is the transformer, M is the local bus, N is the opposite bus, L1 is the transmission line, and k 1、 k2 represents different fault points, k1 is a three-phase short circuit fault 5 km away from busbar M, and k2 is a phase A short circuit fault 80 km away from busbar M. Figure 3 In the figure, G1 and G2 represent lightning protection wires, and C1, C2, and C3 represent three-phase transmission wires.

[0139] In this embodiment, the steps of constructing the secondary current to primary current mapping model of the electromagnetic current transformer are as follows:

[0140] Step 1: Build the main circuit model in the electromagnetic transient simulation software, build the current transformer transfer function model, and write the transfer function coefficients into the N-order transfer function elements of the electromagnetic transient simulation software. Set the sampling rate to 1MHz, set multiple operating parameters, run the simulation, and obtain the fault data set D:I apri ,I bpri ,I cpri ,I asec ,I bsec ,I csec .

[0141] Step 2: Perform Min-Max normalization on the primary and secondary current data collected in Step 1 to obtain the sample data set D':I' apri , I' bpri , I' cpri , I' asec , I' bsec , I' csecThe six matrices are randomly shuffled with rows as the basic unit. The internal order of each row is kept unchanged during the shuffling process, and the six matrices are shuffled in the same way to ensure that each row of the six matrices still has a corresponding relationship after shuffling. Then they are divided into training set, validation set and test set in a ratio of 7:2:1.

[0142] Step 3: Based on the stacked long short-term memory neural network (LSTM), determine the input layer, timing layer and fully connected layer of the secondary current to primary current mapping model framework. The input layer is set to 2000 neurons, the timing layer is set to two layers, each with 128 neurons, and the fully connected layer is set to 2000 neurons, as shown below:

[0143]

[0144] Set the following evaluation criteria:

[0145]

[0146]

[0147] Step 4: Set the learning rate λ = 0.01, T [1] , W [1] 、b [1] , T [2] , W [2] 、b [2] , W [3] 、b [3] The initial value of the sequence θ consisting of a total of 803024 parameters is set to 1. The threshold value is set to K MSE =3×10 -5 , K MAE =3×10 -3 . Perform iterative calculation according to the following formula.

[0148]

[0149] After the first round of calculation, MSE on the validation set is 2.20×10 -3 , MAE = 1.43 × 10 -1 After 184 rounds of parameter iteration, MSE on the validation set is 2.94×10 -5 、MAE=1.97×10 -3 All are less than the threshold K MSE , K MAE , MSE on the test set = 2.98×10 -5 and MAE=2.67×10 -3 Also smaller than K MSE , K MAE, indicating that the model has good generalization performance. The sum of MSE and MAE during training is as follows Figure 6 As shown, the final T [1] , W [1] , b [1] , T [2] , W [2] , b [2] , W [3] , b [3] Matrix, where:

[0150] T [1] : 128×2000

[0151] 0.00600 0.04135 -0.01195 ... -0.00600 0.02117 0.07185 -0.00600 0.03110 0.03331 ... 0.02621 0.02754 0.02269 -0.00600 -0.00390 0.03393 ... 0.05470 -0.00598 -0.00600 ... ... ... ... ... ... ... 0.01178 -0.00599 0.04058 ... 0.02586 -0.00385 -0.00875 0.00536 0.06120 -0.00599 ... 0.01767 0.05470 0.01178 -0.00599 -0.00540 0.02170 ... -0.00600 -0.00599 0.04314

[0152] W [1] : 128×2000

[0153] -0.02445 -0.00599 -0.00599 ... -0.00453 0.01767 0.03499 0.03393 0.06120 -0.00599 ... 0.02797 0.05470 -0.00598 0.02536 0.02244 -0.00600 ... -0.00597 0.01616 -0.00241 ... ... ... ... ... ... ... -0.00690 0.04852 -0.00600 ... 0.02329 -0.00599 0.04396 0.01178 -0.00599 0.02186 ... -0.00572 0.00536 0.02671 0.06120 -0.00599 0.02797 ... 0.05470 -0.00600 0.06120

[0154] b [1] T : 128×1

[0155] 0.02211 0.04085 0.05462 ... -0.01648 0.02209 -0.04412

[0156] T [2] : 128×2000

[0157] 0.08075 -0.07249 -0.09069 ... -0.03315 0.05373 -0.06454 -0.05557 0.01743 -0.08969 ... 0.08158 -0.05338 0.03486 -0.11939 0.03100 0.08042 ... -0.05098 -0.02584 -0.04045 ... ... ... ... ... ... ... -0.01608 0.11305 -0.01398 ... -0.03396 -0.00832 -0.00833 0.02744 -0.01396 0.05094 ... 0.04713 0.08501 -0.00833 0.14260 -0.01397 0.06517 ... 0.03523 0.03117 -0.00833

[0158] W [2] : 128×2000

[0159] 0.07106 -0.06379 -0.07981 ... -0.02917 0.04728 -0.05679 -0.04891 0.01534 -0.07893 ... 0.07179 -0.04697 0.03068 -0.10506 0.02728 0.07077 ... -0.04487 -0.02274 -0.03560 ... ... ... ... ... ... ... -0.01415 0.09949 -0.01230 ... -0.02988 -0.00732 -0.00733 0.02414 -0.01228 0.04483 ... 0.04147 0.07481 -0.00733 0.12549 -0.01229 0.05735 ... 0.03100 0.02743 -0.00733

[0160] b [2] T : 128×1

[0161] 0.05566 0.00529 0.05755 ... 0.05334 0.07613 0.07587

[0162] W [3] : 128×2000

[0163]

[0164] b [3] T : 128×1

[0165] -0.20537 -0.18345 -0.16318 ... -0.14626 0.01498 0.03311

[0166] This enables the reconstruction of the primary current and determines T [1] , W [1] , b [1] , T[2] , W [2] 、b [2] , W [3] 、b [3] Matrix. Get the mapping model from secondary current to primary current.

[0167]

[0168] The output result of the mapping model is denormalized according to the above formula to obtain the actual prediction value.

[0169] Example 3

[0170] This embodiment is based on Embodiment 2:

[0171] In this embodiment, a three-phase short circuit grounding fault occurs, the fault phase angle is 90°, the short circuit position is 5km, the transition resistance is 0.01Ω, and the current transformer transfer function is:

[0172]

[0173] The fault is simulated to obtain the primary and secondary current data after the fault. Using the secondary current to primary current mapping model obtained above, the secondary current of 1ms before and after the fault is input into the model. The reconstructed primary current waveform is as follows: Figure 7 shown.

[0174] Compared with the actual primary side data of the simulation, the mean square error (MSE) is 2.35×10 -5 , the mean absolute error MAE is 2.59×10 -3 , the prediction effect is good.

[0175] Example 4

[0176] This embodiment is based on Embodiment 2:

[0177] In this embodiment, the A phase short circuit grounding fault occurs, the fault phase angle is 90°, the short circuit position is 80km, the transition resistance is 200Ω, and the current transformer transfer function is as follows:

[0178]

[0179] The fault is simulated to obtain the primary and secondary current data after the fault. Using the secondary current to primary current mapping model obtained above, the secondary current of 1ms before and after the fault is input into the model. The reconstructed primary current waveform is as follows: Figure 8 shown.

[0180] Compared with the actual primary side data of the simulation, the mean square error (MSE) is 3.1×10 -5 , the mean absolute error MAE is 1.4×10-3 , the prediction effect is good.

[0181] In summary, the present invention adopts a stacked long short-term memory neural network, which can realize the reconstruction of distorted secondary current to primary current under different fault conditions (fault type, transition resistance, fault location) and different electromagnetic current transformer transfer functions. The effectiveness of the present invention is verified by combining the above analysis and simulation.

[0182] Example 5

[0183] The present embodiment provides a high-voltage power grid current transformer transmission signal distortion reconstruction system, including: a transfer function model construction module, which is configured to build a transfer function model of the main circuit and the electromagnetic current transformer through electromagnetic transient simulation software, collect the primary and secondary current data of the electromagnetic current transformer, and set the fault condition and simulation parameters; a data normalization and division module, which is configured to normalize the collected primary and secondary current data, and divide the normalized data into a training set, a verification set and a test set; a mapping model framework construction module, which is configured to build a mapping model from the secondary current to the primary current based on a stacked long short-term memory neural network Framework; a mapping model framework training module, configured to set the learning rate of the mapping model framework, use the data of the training set for training, and obtain the weight parameters of the mapping model framework; use the data of the validation set to evaluate the mapping model framework, and use the data of the test set to evaluate the generalization ability of the mapping model framework, adjust the learning rate, perform inverse normalization, and obtain a mapping model from the secondary side current to the primary side current; a transmission signal distortion reconstruction module, configured to deploy the mapping model to the actual power system, use the secondary side current data of the electromagnetic current transformer as the model input, and use the model output as the predicted data of the primary side current, thereby realizing the reconstruction of the primary side current data of the electromagnetic current transformer.

[0184] Preferably, in the transfer function model construction module, the transfer function model of the main circuit and the electromagnetic current transformer is constructed by electromagnetic transient simulation software, the primary and secondary current data of the electromagnetic current transformer are collected, and the fault conditions and simulation parameters are set, including:

[0185] The two-end power supply system is built through electromagnetic transient simulation software. The two-end power supplies are connected through transformers, busbars and transmission lines to build the main circuit model.

[0186] A transfer function is obtained according to the circuit model of the electromagnetic current transformer, and the coefficient of the transfer function is written into the N-order transfer function element of the electromagnetic transient simulation software as the transfer function model of the electromagnetic current transformer in the electromagnetic transient simulation software; the primary side current signal measured by the current measuring element is input into the transfer function model to obtain the secondary side current of the electromagnetic current transformer;

[0187] The fault conditions and simulation parameters are set. The fault conditions include the fault type, the initial phase angle of the fault, the fault transition resistance and the fault location. The simulation parameters include the secondary side circuit transmission parameters of the electromagnetic current transformer.

[0188] Preferably, in the data normalization and division module, the collected primary and secondary current data are normalized, and the normalized data are divided into a training set, a validation set and a test set, including:

[0189] The primary and secondary current data of the electromagnetic current transformer are normalized by using the deviation normalization method, the primary and secondary current dimensions are eliminated, and the currents are mapped to the interval [0, 1].

[0190] A sample data set is constructed based on the normalized primary and secondary current data, and the sample data set includes multiple data matrices. The data of the sample data set are randomly shuffled with rows as the basic unit. During the shuffling process, the internal order of each row is kept unchanged, and multiple data matrices are shuffled in the same way, so that each row of the shuffled data matrices still has a corresponding relationship, and then divided into training set, validation set and test set according to a preset ratio.

[0191] Preferably, in the mapping model framework building module, a mapping model framework from secondary current to primary current is built based on a stacked long short-term memory neural network, including:

[0192] Construct a stacked long short-term memory neural network, which includes an input layer, a timing layer, and a fully connected layer; the neurons in the timing layer can receive the output vector of the neurons in the previous layer, and transform it through the weight matrix and the bias vector to obtain the output value; the fully connected layer can receive the output vector of the neurons in the previous layer, and perform linear transformation through the weight matrix and the bias vector to obtain the new state information;

[0193] A mapping model framework from secondary current to primary current is built based on a stacked long short-term memory neural network, and model quality evaluation criteria are set. The model quality evaluation criteria include mean square error and mean absolute error.

[0194] Preferably, the learning rate of the mapping model framework is set, and the training set data is used for training to obtain the weight parameters of the mapping model framework; the mapping model framework is evaluated using the validation set data, and the generalization ability of the mapping model framework is evaluated using the test set data, the learning rate is adjusted, and the inverse normalization is performed to obtain the mapping model from the secondary side current to the primary side current, including:

[0195] Set the learning rate of the mapping model framework, adjust the parameters according to the gradient descent method based on the data of the training set, and update the weight parameters of the mapping model framework;

[0196] Based on the updated mapping model framework, the mean square error and mean absolute error are calculated and judged using the validation set data until the iteration end condition is met. The weight parameter after the iteration is the value at the end of the mapping model framework training.

[0197] The iteratively obtained weight parameters are brought into the mapping model framework, the test set data is predicted, and the corresponding mean square error and mean absolute error are calculated to evaluate the generalization effect of the mapping model framework and determine whether to continue updating the weight parameters;

[0198] The predicted value of the actual current is obtained by the inverse normalization method as the reconstructed primary current value, so as to realize the reconstruction of the primary current and obtain the mapping model of the secondary current to the primary current.

[0199] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

Claims

1. A method for reconstructing distortion of a current transformer transmission signal of a high-voltage power grid, characterized in that: include: Use electromagnetic transient simulation software to build the transfer function model of the main circuit and electromagnetic current transformer, collect the primary and secondary current data of the electromagnetic current transformer, and set the fault conditions and simulation parameters; The collected primary and secondary current data are normalized, and the normalized data are divided into a training set, a validation set, and a test set; Based on the stacked long short-term memory neural network, a mapping model framework from secondary current to primary current is built; Setting the learning rate of the mapping model framework, using the data of the training set for training, and obtaining the weight parameters of the mapping model framework; using the data of the validation set to evaluate the mapping model framework, and using the data of the test set to evaluate the generalization ability of the mapping model framework, adjusting the learning rate, and performing inverse normalization to obtain a mapping model from the secondary side current to the primary side current; The mapping model is deployed to an actual power system, the secondary side current data of the electromagnetic current transformer is used as the model input, and the model output is used as the prediction data of the primary side current, thereby realizing the reconstruction of the primary side current data of the electromagnetic current transformer.

2. A method for reconstructing distortion of a current transformer transmission signal of a high-voltage power grid according to claim 1, characterized in that: The transfer function model of the main circuit and the electromagnetic current transformer is built by electromagnetic transient simulation software, the primary and secondary current data of the electromagnetic current transformer are collected, and the fault conditions and simulation parameters are set, including: The two-end power supply system is built through electromagnetic transient simulation software. The two-end power supplies are connected through transformers, busbars and transmission lines to build the main circuit model. A transfer function is obtained according to the circuit model of the electromagnetic current transformer, and the coefficient of the transfer function is written into the N-order transfer function element of the electromagnetic transient simulation software as the transfer function model of the electromagnetic current transformer in the electromagnetic transient simulation software; the primary side current signal measured by the current measuring element is input into the transfer function model to obtain the secondary side current of the electromagnetic current transformer; Fault conditions and simulation parameters are set, wherein the fault conditions include fault type, fault initial phase angle, fault transition resistance and fault location, and the simulation parameters include secondary side circuit transmission parameters of the electromagnetic current transformer.

3. The method for reconstructing the distortion of the current transformer transmission signal of a high-voltage power grid according to claim 1 is characterized in that: The collected primary and secondary current data are normalized, and the normalized data are divided into a training set, a validation set, and a test set, including: The primary and secondary current data of the electromagnetic current transformer are normalized by using the deviation normalization method, the primary and secondary current dimensions are eliminated, and the currents are mapped to the interval [0, 1]. A sample data set is constructed according to the normalized primary and secondary current data, wherein the sample data set includes multiple data matrices; the data of the sample data set are randomly shuffled with rows as basic units, the internal order of each row is kept unchanged during the shuffling process, and the multiple data matrices are shuffled in the same way, so that each row of the multiple shuffled data matrices still has a corresponding relationship, and then divided into a training set, a validation set and a test set according to a preset ratio.

4. A method for reconstructing distortion of a current transformer transmission signal of a high-voltage power grid according to claim 1, characterized in that: The stacked long short-term memory neural network is used to build a mapping model framework from secondary current to primary current, including: Constructing a stacked long short-term memory neural network, the stacked long short-term memory neural network includes an input layer, a timing layer and a fully connected layer; the neurons of the timing layer can receive the output vector of the neurons of the previous layer, and transform it through the weight matrix and the bias vector to obtain the output value; the fully connected layer can receive the output vector of the neurons of the previous layer, and perform linear transformation through the weight matrix and the bias vector to obtain new state information; A mapping model framework of secondary-side current to primary-side current is built based on a stacked long short-term memory neural network, and model quality evaluation criteria are set, which include mean square error and mean absolute error.

5. The method for reconstructing the distortion of the current transformer transmission signal of a high-voltage power grid according to claim 1 is characterized in that: The step of setting the learning rate of the mapping model framework, using the data of the training set for training, and obtaining the weight parameters of the mapping model framework; using the data of the validation set to evaluate the mapping model framework, and using the data of the test set to evaluate the generalization ability of the mapping model framework, adjusting the learning rate, and performing inverse normalization to obtain a mapping model from the secondary side current to the primary side current includes: Setting the learning rate of the mapping model framework, combining the data of the training set, adjusting the parameters according to the gradient descent method, and updating the weight parameters of the mapping model framework; Based on the updated mapping model framework, the mean square error and the mean absolute error are calculated and judged using the validation set data until the iteration end condition is met. The weight parameter after the iteration end is the value at the end of the mapping model framework training. Bring the iteratively obtained weight parameters into the mapping model framework, predict the data of the test set, calculate the corresponding mean square error and mean absolute error, so as to evaluate the generalization effect of the mapping model framework and determine whether to continue updating the weight parameters; The predicted value of the actual current is obtained by the inverse normalization method as the reconstructed primary current value, so as to realize the reconstruction of the primary current and obtain the mapping model of the secondary current to the primary current.

6. A high-voltage power grid current transformer transmission signal distortion reconstruction system, characterized in that: include: The transfer function model building module is configured to build a transfer function model of the main circuit and the electromagnetic current transformer through electromagnetic transient simulation software, collect primary and secondary current data of the electromagnetic current transformer, and set fault conditions and simulation parameters; The data normalization and division module is configured to normalize the collected primary and secondary current data, and divide the normalized data into a training set, a validation set, and a test set; A mapping model framework building module is configured to build a mapping model framework from secondary current to primary current based on a stacked long short-term memory neural network; A mapping model framework training module is configured to set a learning rate of the mapping model framework, use the data of the training set for training, and obtain weight parameters of the mapping model framework; use the data of the validation set to evaluate the mapping model framework, and use the data of the test set to evaluate the generalization ability of the mapping model framework, adjust the learning rate, perform inverse normalization, and obtain a mapping model from secondary current to primary current; The transmission signal distortion reconstruction module is configured to deploy the mapping model to the actual power system, use the secondary side current data of the electromagnetic current transformer as the model input, and use the model output as the predicted data of the primary side current, thereby realizing the reconstruction of the primary side current data of the electromagnetic current transformer.

7. A high-voltage power grid current transformer transmission signal distortion reconstruction system according to claim 6, characterized in that: In the transfer function model building module, the transfer function model of the main circuit and the electromagnetic current transformer is built by electromagnetic transient simulation software, the primary and secondary current data of the electromagnetic current transformer are collected, and the fault conditions and simulation parameters are set, including: The two-end power supply system is built through electromagnetic transient simulation software. The two-end power supplies are connected through transformers, busbars and transmission lines to build the main circuit model. A transfer function is obtained according to the circuit model of the electromagnetic current transformer, and the coefficient of the transfer function is written into the N-order transfer function element of the electromagnetic transient simulation software as the transfer function model of the electromagnetic current transformer in the electromagnetic transient simulation software; the primary side current signal measured by the current measuring element is input into the transfer function model to obtain the secondary side current of the electromagnetic current transformer; Fault conditions and simulation parameters are set, wherein the fault conditions include fault type, fault initial phase angle, fault transition resistance and fault location, and the simulation parameters include secondary side circuit transmission parameters of the electromagnetic current transformer.

8. A high-voltage power grid current transformer transmission signal distortion reconstruction system according to claim 6, characterized in that: In the data normalization and division module, the collected primary and secondary current data are normalized, and the normalized data are divided into a training set, a validation set, and a test set, including: The primary and secondary current data of the electromagnetic current transformer are normalized by using the deviation normalization method, the primary and secondary current dimensions are eliminated, and the currents are mapped to the interval [0, 1]. A sample data set is constructed according to the normalized primary and secondary current data, wherein the sample data set includes multiple data matrices; the data of the sample data set are randomly shuffled with rows as basic units, the internal order of each row is kept unchanged during the shuffling process, and the multiple data matrices are shuffled in the same way, so that each row of the multiple shuffled data matrices still has a corresponding relationship, and then divided into a training set, a validation set and a test set according to a preset ratio.

9. A high-voltage power grid current transformer transmission signal distortion reconstruction system according to claim 6, characterized in that: In the mapping model framework construction module, a mapping model framework from secondary current to primary current is constructed based on a stacked long short-term memory neural network, including: Constructing a stacked long short-term memory neural network, the stacked long short-term memory neural network includes an input layer, a timing layer and a fully connected layer; the neurons of the timing layer can receive the output vector of the neurons of the previous layer, and transform it through the weight matrix and the bias vector to obtain the output value; the fully connected layer can receive the output vector of the neurons of the previous layer, and perform linear transformation through the weight matrix and the bias vector to obtain new state information; A mapping model framework of secondary-side current to primary-side current is built based on a stacked long short-term memory neural network, and model quality evaluation criteria are set, which include mean square error and mean absolute error.

10. A high-voltage power grid current transformer transmission signal distortion reconstruction system according to claim 6, characterized in that: In the mapping model framework training module, the learning rate of the mapping model framework is set, the data of the training set is used for training, and the weight parameters of the mapping model framework are obtained; the mapping model framework is evaluated using the data of the validation set, and the generalization ability of the mapping model framework is evaluated using the data of the test set, the learning rate is adjusted, and denormalization is performed to obtain a mapping model from the secondary side current to the primary side current, including: Setting the learning rate of the mapping model framework, combining the data of the training set, adjusting the parameters according to the gradient descent method, and updating the weight parameters of the mapping model framework; Based on the updated mapping model framework, the mean square error and the mean absolute error are calculated and judged using the validation set data until the iteration end condition is met. The weight parameter after the iteration end is the value at the end of the mapping model framework training. Bring the iteratively obtained weight parameters into the mapping model framework, predict the data of the test set, calculate the corresponding mean square error and mean absolute error, so as to evaluate the generalization effect of the mapping model framework and determine whether to continue updating the weight parameters; The predicted value of the actual current is obtained by the inverse normalization method as the reconstructed primary current value, so as to realize the reconstruction of the primary current and obtain the mapping model of the secondary current to the primary current.

Images