A low-voltage current transformer quality detection method based on digital twinning

Abstract

The application discloses a kind of low-voltage current transformer quality detection methods based on digital twinning, comprising the following steps: S1, create the digital twin model of the initial state of low-voltage current transformer to be detected, according to the simulation calculation dynamic response of the low-voltage current transformer obtained from digital twin model, and the simulation calculation dynamic response is converted into code and input into the digital twin model in initial state, obtain simulation dynamic response;S2, according to the frequency, magnetic field intensity, temperature and humidity, phase, noise signal under the working state of low-voltage current transformer to establish correlation function, the correlation function value is calculated using regular attention cycle network, to obtain the stability evaluation coefficient of current low-voltage current transformer.The application can accurately determine whether the current low-voltage current transformer exists fault point by constructing the digital twin of low-voltage current transformer, by constructing the way of virtual-real data fusion to comprehensively analyze the parameter change of current low-voltage current transformer.

Description

Technical Field

[0001] This invention relates to the field of low-voltage current transformer testing technology, specifically a low-voltage current transformer quality testing method based on digital twins. Background Technology

[0002] With the rapid development of cities and the continuous increase in electricity demand, the requirements for the stability of the power system are also constantly increasing. As a device in the power system that provides measurement and protection signals to protect equipment, the accuracy of current measurement of current transformers is an important indicator to ensure the normal and reliable operation of high and low voltage electrical equipment. This highlights the importance of numerical calibration management of current transformers.

[0003] During operation, existing low-voltage current transformers require parameter testing and evaluation. However, conventional testing and evaluation methods may span multiple physical dimensions and exceed the scope of standard testing methods for low-voltage current transformers. This not only easily leads to deviations in test results but also makes the entire process cumbersome, thus affecting overall testing efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a method for quality testing of low-voltage current transformers based on digital twins, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for quality detection of low-voltage current transformers based on digital twins, comprising the following steps:

[0006] S1. Create a digital twin model of the low-voltage current transformer under the initial state, obtain the simulation calculation dynamic response of the low-voltage current transformer based on the digital twin model, and convert the simulation calculation dynamic response into code and input it into the digital twin model under the initial state to obtain the simulation dynamic response.

[0007] S2. Establish a correlation function based on the frequency, magnetic field strength, temperature and humidity, phase and noise signals of the low-voltage current transformer under the working state. Then, plot the correlation parameters obtained under the simulation dynamic response on the correlation function and generate dynamic error. At the same time, determine the true function value of the low-voltage current transformer under the current dynamic error.

[0008] S3. Establish a dynamic threshold under the true function value, obtain the functional coupling relationship and directed edge between the components of the low-voltage current transformer based on the dynamic threshold, create the fault coupling network of the low-voltage current transformer, obtain the possible fault coupling connection relationship in the fault coupling network using the method of proper subset and multi-color set, and generate multiple sets of fault geometry groups.

[0009] S4. Monitor multiple fault geometric groups and collect the activity trajectories of different fault geometric groups in the same time period. Repeat the test on the collected activity trajectories of different fault geometric groups and compare them with the true function value under dynamic error to obtain the mean error.

[0010] S5. Trace back to obtain the correlation function value corresponding to the mean error, and use a recurrent network with regular attention to generate the stable evaluation coefficient Val of the correlation function value at the current time. t Among them, if Val t If Val is ≥1, it indicates that the current low-voltage current transformer meets the quality standards. t If the value is less than 1, it indicates that the current low-voltage current transformer is of substandard quality.

[0011] Furthermore, the digital twin model mentioned in step S1 includes a geometric model, a dynamic simulation model, a physical model, a three-dimensional image model, and a rule model.

[0012] Furthermore, after the dynamic error mentioned in step S2 is generated, the dynamic parameters existing in the historical data of the current correlation function are corrected in a timely manner, the missing values ​​are traced and nodes are found, and they are marked and extracted to obtain the fault propagation path.

[0013] Furthermore, in step S5, the correlation function value corresponding to the mean error is obtained by tracing back, and a stable evaluation coefficient Val of the correlation function value is generated using a recurrent network of regular attention. t The processing procedure of the recurrent network for regular attention is as follows:

[0014] R t =τ(W r1 Out t-1 +W r2 T t +r1)

[0015] Where R t This is a stock unit, representing the intermediate unit (Out) that the network needs to save at the current time step from the previous time step. t-1 The weights, τ is the Sigmoid function, W r1 and W r2 T is a learnable weight matrix. t It is a signal matrix generated by using an embedding function to obtain the correlation function values ​​of frequency, magnetic field strength, temperature and humidity, phase, and noise signals, where r1 is a correction parameter;

[0016] S t =tanh(W s1 ·RegAtt(R t Out t-1 )+W s2 T t +r)

[0017] Where S t Representing the network intermediate values ​​at the current moment, tanh() is a hyperbolic tangent activation function, W s1 and W s2 Here, is a learnable weight matrix, r is a correction parameter, and RegAtt() is a regularized attention function that adjusts Val... t-1 Multiplying by three randomly initialized weight matrices of the same size yields three matrices of the same size, Q, K, and V, representing different mappings. For matrix Q, arbitrarily select a row q by slicing. i Similarly, we can obtain k j v j ;

[0018]

[0019] In the formula ||q i ||and||k j || represent the values ​​of q respectively i and k j Perform L2 regularization, where n is the number of rows in matrices Q, K, and V;

[0020] U t =τ(W u1 Out t-1 +W u2 T t +r2)

[0021] U t For updating units, representing units used to update the intermediate unit Out at the current time. t The weight, W u1 and W u2 R1 is a learnable weight matrix, and R2 is a correction parameter.

[0022] Out t =γU t ⊙Out t-1 +(1-γU t )⊙S t

[0023] Where ⊙ represents pointwise multiplication, and γ is a hyperparameter used to optimize intermediate units;

[0024] Val t =W val Out t +r val

[0025] Finally, through the learnable weight matrix W val and correction parameter r val Obtain the current stability evaluation coefficient Val t .

[0026] Furthermore, after the true function values ​​are determined, the rated operating current and rated operating voltage are extracted from the standard operation and maintenance mode corresponding to the low-voltage current transformer. The extracted rated operating current and rated operating voltage are used as a reference group and compared with multiple generated fault geometry groups to obtain the reference parameters. Then, a conventional rated operating current and rated operating voltage experimental group is established. The two rated operating currents and rated operating voltages are compared according to the reference standard, and the reference current inductance deviation, reference current amplitude, reference current frequency, and reference phase difference corresponding to the experimental group are obtained to reduce the error.

[0027] Furthermore, when the stability evaluation coefficient value Val t When the error is less than 1, the geometric model, dynamic simulation model, physical model, three-dimensional image model, and rule model of the digital twin are promptly corrected until the error stabilizes and is below the dynamic threshold.

[0028] Furthermore, T t The acquisition method is as follows: First, the correlation function values ​​of frequency, magnetic field strength, temperature and humidity, phase, and noise signals are converted into one-dimensional sequences of different lengths through one-dimensional embedding. By truncating empty bits and redundant bits, the sequence length is unified to 512. Then, the splicing operation is used to splice multiple one-dimensional sequences into a matrix.

[0029] Furthermore, after obtaining the fault propagation path, the model simulation is performed on its next operating trajectory and activity range, and a specified operating channel under the current fault propagation path is created. Then, the current channel is traced and the root cause of the fault is found, thus completing the fault tracing.

[0030] Furthermore, after obtaining the simulated dynamic response, image information of the low-voltage current transformer under working conditions is acquired. After smoothing, filtering and denoising the acquired image information, the image part to be detected in the image information is separated from the overall image part. Then, edge segmentation is performed on the image part to be detected, and the image information of this part is clipped and extracted.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention constructs a digital twin of the low-voltage current transformer and comprehensively analyzes the parameter changes of the current low-voltage current transformer by constructing a virtual-real data fusion method, thereby accurately determining whether there is a fault point in the current low-voltage current transformer. Moreover, it obtains the possible fault coupling connection relationship in the fault coupling network by using proper subsets and multi-color sets, and generates multiple sets of fault geometry groups. This allows for simulation and exercise of possible fault points in the subsequent operation of the low-voltage current transformer, facilitating timely maintenance of the low-voltage current transformer and greatly reducing the probability of faults occurring in subsequent operation. Attached Figure Description

[0032] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Please see Figure 1 This invention provides a method for quality testing of low-voltage current transformers based on digital twins, comprising the following steps:

[0035] S1. Create a digital twin model of the low-voltage current transformer under the initial state. Obtain the simulation dynamic response of the low-voltage current transformer based on the digital twin model. Then, convert the simulation dynamic response into code and input it into the digital twin model under the initial state to obtain the simulation dynamic response. Specifically, creating the digital twin model includes: using data attributes of multiple physical quantities, multiple dimensions, multiple reference objects, and multiple probabilities, and building the model information under the digital twin in the model software through digital modeling.

[0036] S2. Establish a correlation function based on the frequency, magnetic field strength, temperature and humidity, phase, and noise signals of the low-voltage current transformer under its operating state (i.e., based on the normal operating state of the low-voltage current transformer (parameter information generated during normal operation), which can be achieved by any of the tabular method, graphical method, or analytical method). Then, plot the correlation parameters obtained under the simulated dynamic response (i.e., the parameters generated by the low-voltage current transformer under the simulated state) on the correlation function and generate dynamic error to eliminate the error value existing in the correlation function, thereby determining the true function value of the low-voltage current transformer under the current dynamic error.

[0037] S3. Establish a dynamic threshold under the true function value. Based on the dynamic threshold, obtain the functional coupling relationship and directed edges between the components of the low-voltage current transformer. Create a fault coupling network for the low-voltage current transformer. The fault coupling network is based on the functional coupling relationship and directed edges. In the virtual network, build fault nodes between the components of the low-voltage current transformer. When the dynamic threshold reaches the fault node, issue a fault coupling signal to remind. Use proper subsets and polychromatic sets to obtain the possible fault coupling in the fault coupling network (the part where the fault coupling occurs is determined by the overlapping part of the proper subset and polychromatic set), and generate multiple sets of fault geometry groups.

[0038] S4. Monitor multiple fault geometric groups and collect the activity trajectories of different fault geometric groups in the same time period. Repeat the test on the collected activity trajectories of different fault geometric groups, compare them with the true function value under dynamic error to obtain the mean error, and generate three different time intervals, where the first time interval to the third time interval are represented as [0, t1], (t1, t2], and (t2, t3] respectively, and t1 > t2 > t3.

[0039] S5. Trace back to obtain the correlation function value corresponding to the mean error, and use a recurrent network with regular attention to generate a stable evaluation coefficient Val for the correlation function value. t The processing procedure of the recurrent network for regular attention is as follows:

[0040] R t =τ(W r1 Out t-1 +W r2 T t +r1)

[0041] Where R t This is a stock unit, representing the intermediate unit (Out) that the network needs to save at the current time step from the previous time step. t-1 The weights, τ is the Sigmoid function, W r1 and W r2 T is a learnable weight matrix. t It is a signal matrix generated by using an embedding function to obtain the correlation function values ​​of frequency, magnetic field strength, temperature and humidity, phase, and noise signals, with r1 being a correction parameter.

[0042] S t =tanh(W s1 ·RegAtt(R t Out t-1 )+W s2 T t +r)

[0043] Where S t Representing the network intermediate values ​​at the current moment, tanh() is a hyperbolic tangent activation function, W s1 and W s2 Here, r is a learnable weight matrix, r is a correction parameter, and RegAtt() is a regularized attention function that assigns the stable evaluation coefficient Val at time t-1 to the matrix. t-1 Multiplying by three randomly initialized weight matrices of the same size yields three matrices of the same size, Q, K, and V, representing different mappings. For matrix Q, arbitrarily select a row q by slicing. i Similarly, we can obtain k j v j .

[0044]

[0045] In the formula ||q i ||and||k j || represent the values ​​of q respectively i and k j L2 regularization is performed, where n is the number of rows in matrices Q, K, and V. By using RegAtt(), the correlation between different dimensions of the data can be further obtained. Since the softmax operation in traditional attention is omitted, a lot of computation is saved while improving accuracy.

[0046] U t =τ(W u1 Out t-1 +W u2 T t +r2)

[0047] U t For updating units, representing units used to update the intermediate unit Out at the current time. t The weight, W u1 and W u2 R1 is a learnable weight matrix, and R2 is a correction parameter.

[0048] Out t =γU t ⊙Out t-1 +(1-γU t )⊙S t

[0049] Where ⊙ represents pointwise multiplication, and γ is a hyperparameter used to optimize intermediate units.

[0050] Val t =W val Out t +r val

[0051] Finally, through the learnable weight matrix W val and correction parameter r val Obtain the current stability evaluation coefficient Val t Among them, if Val t If Val is ≥1, it indicates that the current low-voltage current transformer meets the quality standards. t If the value is less than 1, it indicates that the current low-voltage current transformer is of substandard quality.

[0052] Furthermore, digital twin models include geometric models, dynamic simulation models, physical models, 3D image models, and rule models.

[0053] Furthermore, after dynamic errors occur, they should be corrected in a timely manner from the historical data of the relevant functions, and missing values ​​should be traced and nodes should be found, marked and extracted to obtain the fault propagation path.

[0054] Furthermore, after the true function values ​​are determined, the rated operating current and rated operating voltage should be extracted from the standard operation and maintenance mode corresponding to the low-voltage current transformer. The extracted rated operating current and rated operating voltage should be used as a reference group and compared with multiple generated fault geometry groups to obtain the reference parameters. Then, a conventional rated operating current and rated operating voltage experimental group should be established. The two rated operating currents and rated operating voltages should be compared according to the reference standard to obtain the reference current inductance deviation, reference current amplitude, reference current frequency, and reference phase difference under the experimental group, so as to reduce errors and maintain the accuracy of the data during model operation.

[0055] Furthermore, when Val t When the error is less than 1, the geometric model, dynamic simulation model, physical model, three-dimensional image model, and rule model of the digital twin should be corrected in a timely manner, that is, the range of the initial set parameters should be adjusted until the error is stable and lower than the dynamic threshold.

[0056] Furthermore, T t The acquisition method is as follows: First, the correlation function values ​​of frequency, magnetic field strength, temperature and humidity, phase, and noise signals can be converted into one-dimensional sequences of different lengths through one-dimensional embedding. By truncating empty bits and redundant bits, the sequence length is unified to 512. Then, the splicing operation is used to splice multiple one-dimensional sequences into a matrix.

[0057] Furthermore, after obtaining the fault propagation path, it is necessary to perform model simulation of its next operating trajectory and activity range, create a specified operating channel under the current fault propagation path, trace the current channel, find the root cause of the fault, and complete the fault tracing.

[0058] Furthermore, after obtaining the simulated dynamic response, image information of the low-voltage current transformer under operating conditions should be acquired. After smoothing, filtering and denoising the acquired image information, the image part to be detected in the image information is separated from the overall image part. Then, edge segmentation is performed on the image part to be detected, and the image information of this part is edited and extracted to ensure the integrity and authenticity of the information acquired.

[0059] This invention constructs a digital twin of a low-voltage current transformer and comprehensively analyzes the parameter changes of the current low-voltage current transformer by fusing virtual and real data. It calculates stability evaluation coefficients through a cyclic network of regular attention, thereby accurately determining whether there are fault points in the current low-voltage current transformer. Furthermore, it obtains the possible fault coupling connections in the fault coupling network by using proper subsets and multi-color sets, and generates multiple sets of fault geometry groups. This allows for simulation and exercise of potential fault points in the subsequent operation of the low-voltage current transformer, facilitating timely maintenance and greatly reducing the probability of faults occurring during subsequent operation.

[0060] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for quality inspection of low-voltage current transformers based on digital twins, characterized in that, Includes the following steps: S1. Create a digital twin model of the low-voltage current transformer under the initial state, obtain the simulation calculation dynamic response of the low-voltage current transformer based on the digital twin model, and convert the simulation calculation dynamic response into code and input it into the digital twin model under the initial state to obtain the simulation dynamic response. S2. Establish a correlation function based on the frequency, magnetic field strength, temperature and humidity, phase and noise signals of the low-voltage current transformer under the working state. Then, plot the correlation parameters obtained under the simulation dynamic response on the correlation function and generate dynamic error. At the same time, determine the true function value of the low-voltage current transformer under the current dynamic error. S3. Establish a dynamic threshold under the true function value, obtain the functional coupling relationship and directed edge between the components of the low-voltage current transformer based on the dynamic threshold, create the fault coupling network of the low-voltage current transformer, obtain the possible fault coupling connection relationship in the fault coupling network using the method of proper subset and multi-color set, and generate multiple sets of fault geometry groups. S4. Monitor multiple fault geometric groups and collect the activity trajectories of different fault geometric groups in the same time period. Repeat the test on the collected activity trajectories of different fault geometric groups and compare them with the true function value under dynamic error to obtain the mean error. S5. Trace back to obtain the correlation function value corresponding to the mean error, and use a recurrent network with regular attention to generate the stable evaluation coefficient Val of the correlation function value at the current time. t Among them, if Val t If Val is ≥1, it indicates that the current low-voltage current transformer meets the quality standards. t If the value is less than 1, it indicates that the current low-voltage current transformer is of substandard quality. In step S5, the correlation function value corresponding to the mean error is obtained by tracing back, and a stable evaluation coefficient Val of the correlation function value is generated using a recurrent network with regular attention. t The processing procedure of the recurrent network for regular attention is as follows: R t =τ(W r1 Out t-1 +W r2 T t +r1) Where R t This is a stock unit, representing the intermediate unit (Out) that the network needs to save at the current time step from the previous time step. t-1 The weights, τ is the Sigmoid function, W r1 and W r2 T is a learnable weight matrix. t It is a signal matrix generated by using an embedding function to obtain the correlation function values ​​of frequency, magnetic field strength, temperature and humidity, phase, and noise signals, where r1 is a correction parameter; S t =tanh(W s1 ·RegAtt(R t Out t-1 )+W s2 T t +r) Where S t Representing the network intermediate values ​​at the current moment, tanh() is a hyperbolic tangent activation function, W s1 and W s2 Here, r is a learnable weight matrix, r is a correction parameter, and RegAtt() is a regularized attention function that assigns the stable evaluation coefficient Val at time t-1 to the matrix. t-1 Multiplying by three randomly initialized weight matrices of the same size yields three matrices of the same size, Q, K, and V, representing different mappings. For matrix Q, arbitrarily select a row q by slicing. i Similarly, we can obtain k j v j ; In the formula ||q i ||and||k j || represent the values ​​of q respectively i and k j Perform L2 regularization, where n is the number of rows in matrices Q, K, and V; U t =τ(W u1 Out t-1 +W u2 T t +r2) U t For updating units, representing units used to update the intermediate unit Out at the current time. t The weight, W u1 and W u2 R1 is a learnable weight matrix, and R2 is a correction parameter. Out t =γU t ⊙Out t-1 +(1-γU t )⊙S t Where ⊙ represents pointwise multiplication, and γ is a hyperparameter used to optimize intermediate units; Val t =W val Out t +r val Finally, through the learnable weight matrix W val and correction parameter r val Obtain the current stability evaluation coefficient Val t .

2. The method for quality detection of low-voltage current transformers based on digital twins according to claim 1, characterized in that: The digital twin model mentioned in step S1 includes a geometric model, a dynamic simulation model, a physical model, a three-dimensional image model, and a rule model.

3. The method for quality detection of low-voltage current transformers based on digital twins according to claim 1, characterized in that: After the dynamic error mentioned in step S2 is generated, the dynamic parameters existing in the historical data of the current correlation function are corrected in a timely manner, the missing values ​​are traced and nodes are found, and they are marked and extracted to obtain the fault propagation path.

4. The method for quality detection of low-voltage current transformers based on digital twins according to claim 1, characterized in that: After the true function values ​​are determined, the rated operating current and rated operating voltage are extracted from the standard operation and maintenance mode corresponding to the low-voltage current transformer. The extracted rated operating current and rated operating voltage are used as a reference group and compared with multiple generated fault geometry groups to obtain the reference parameters. Then, a conventional rated operating current and rated operating voltage experimental group is established. The two rated operating currents and rated operating voltages are compared according to the reference standard, and the reference current inductance deviation, reference current amplitude, reference current frequency, and reference phase difference under the experimental group are obtained to reduce the error.

5. The method for quality detection of low-voltage current transformers based on digital twins according to claim 1, characterized in that: When the stability evaluation coefficient Val t When the error is less than 1, the geometric model, dynamic simulation model, physical model, 3D image model, and rule model of the digital twin model are promptly corrected until the error stabilizes and is below the dynamic threshold.

6. The method for quality detection of low-voltage current transformers based on digital twins according to claim 1, characterized in that: T t The acquisition method is as follows: First, the correlation function values ​​of frequency, magnetic field strength, temperature and humidity, phase, and noise signals are converted into one-dimensional sequences of different lengths through one-dimensional embedding. By truncating empty bits and redundant bits, the sequence length is unified to 512. Then, the splicing operation is used to splice multiple one-dimensional sequences into a matrix.

7. The method for quality detection of low-voltage current transformers based on digital twins according to claim 1, characterized in that: After obtaining the fault propagation path, the model simulation is performed on its next running trajectory and activity range, and a specified running channel under the current fault propagation path is created. Then, the current channel is traced and the root cause of the fault is found, thus completing the fault tracing.

8. The method for quality detection of low-voltage current transformers based on digital twins according to claim 1, characterized in that: After obtaining the simulated dynamic response, image information of the low-voltage current transformer under working conditions is acquired. The acquired image information is smoothed, filtered and denoised. The image part to be detected in the image information is separated from the overall image. Then, the image part to be detected is segmented at the edges, and the image information of this part is clipped and extracted.

Images