MFNN method for Rogowski coil interphase interference compensation

Abstract

The invention discloses an MFNN method for Rogowski coil interphase interference compensation, and the method comprises the steps: building a three-phase alternating current detection experiment table, and measuring the three-phase current through a 0.01% standard CT and a Rogowski coil; the distance between phase lines and A, B and C three-phase current parameters are selected through orthogonal experiment design; collecting current measurement values of the Rogowski coil and the standard CT under different current combinations at a specified phase line distance and a specified current; designing an MFNN based on a full-connection neural network and multi-task learning; and training an MFNN-based Rogowski coil inter-phase interference compensation model by taking the CT measurement value as a standard value to realize Rogowski coil measurement value compensation.

Description

Technical Field

[0001] The present invention relates to the technical field of current measurement, and in particular to an MFNN method for compensating interphase interference of Rogowski coils. Background Art

[0002] Rogowski coil is an AC current sensor, a hollow ring-shaped coil, which is flexible or rigid. It can be directly put on the conductor to measure AC current. Rogowski coil is widely used in industrial production because it is easy to install, does not need to be powered off, and has little physical interference to the equipment. However, the measurement accuracy of Rogowski coil is low, especially when it is used in a multi-phase system, it will produce phase interference due to the electromagnetic coupling effect between the phase lines, resulting in a decrease in measurement accuracy. The phase interference compensation method based on MFNN refers to collecting three-phase current data at different phase distances, training to obtain parameters, so that MFNN compensates for the difference between the original measurement value and the nominal value of the Rogowski coil, and improves the final current measurement accuracy. The multi-branch structure in MFNN improves its generalization, enabling it to effectively cope with the challenges of different complex working conditions. Patent CN117036453A constructs an error prediction model for optical measuring instruments based on computer deep learning methods to achieve error compensation of optical measurement data. Patent CN115856342A compensates for the temperature error of smart sensors by constructing and training a deep learning neural network model.

[0003] The above-mentioned specific patent comparative documents are:

[0004] 1) "A structured light measurement method and system for correcting measurement errors based on deep learning", patent number CN117036453A, this patent discloses a structured light measurement method and system for correcting measurement errors based on deep learning. First, an error prediction model of an optical measuring instrument is constructed based on a computer deep learning method, and then the error compensation coefficient of the optical measuring instrument is determined, and the optical measuring instrument is used to measure the object to be measured to obtain optical measurement data, and then the optical measurement data is input into the error prediction model to obtain the error prediction value, and the error value is used as training data to train the computer deep learning model, and the computer deep learning model is used to predict the errors generated in the measurement process and compensate for them, thereby achieving high-precision measurement. The MFNN method of the present invention is different from the above, adopts a multi-task learning approach, and innovatively uses independent branches to compensate for the three-phase current.

[0005] 2) "A method and device for temperature error compensation of an intelligent sensor based on a deep learning network", patent number CN115856342A, this patent discloses a method and device for temperature error compensation of an intelligent sensor based on a deep learning network. First, based on the deep learning network, a deep learning neural network model is constructed, and the deep learning neural network model is trained using a training data set. Then, the deep learning neural network model is optimized using a network optimization algorithm, and then the optimized deep learning neural network model is used to compensate for the temperature error of the intelligent sensor. This method successfully solves the technical problem that the accuracy of intelligent sensors is not high due to the influence of temperature. The MFNN method of the present invention is different from the above, and uses a deep learning model to solve the technical problem that the measurement accuracy of the Rogowski coil is not high due to the influence of phase-to-phase interference. Summary of the invention

[0006] In order to solve the above technical problems, the object of the present invention is to provide a MFNN method for compensating interphase interference of Rogowski coils.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] An MFNN method for Rogowski coil interphase interference compensation, comprising:

[0009] A. Build a three-phase alternating current detection test bench and use 0.01% standard CT and Rogowski coil to measure the three-phase current;

[0010] B. Select the distance between phase lines and the three-phase current parameters of A, B, and C through orthogonal experimental design;

[0011] C. Collect the current measurement values ​​of Rogowski coil and standard CT at specified phase line distance and specified current under different current combinations;

[0012] D. Design a fully-connected neural network and multi-task learning MFNN;

[0013] E. Taking the CT measurement value as the standard value, the Rogowski coil interphase interference compensation model based on MFNN is trained to realize the Rogowski coil measurement value compensation.

[0014] Furthermore, the three-phase alternating current detection test bench in the above A comprises yellow, green and red three-phase conductors, on which Rogowski coils are respectively loaded, and the three Rogowski coils are all connected to the power quality analyzer; the three 0.01% standard CTs are loaded on the three-phase conductors and connected to the power quality analyzer; the three-phase conductors are fixed on the guide rails, and the distance between the three-phase conductors is adjusted by the guide rails; the power quality analyzer saves the current measurement values ​​of the Rogowski coils and the 0.01% standard CTs in real time through Wi-Fi.

[0015] Furthermore, in the above B, the experimental table is designed according to the orthogonal design principle;

[0016] The variable related to phase interference in the three-phase current system is the distance between phase lines (D αb ,D bc ) and A, B, C three-phase current (I a ,I b ,I c ); Under actual working conditions, phase lines A and C are symmetrical about phase line B, so they are regarded as a distance variable D, that is, D = D ab =D bc ; Four variables (D,I a ,I b ,I c ) are all continuous variables. The interval partitioning method is used to divide the four continuous variables into 6, 9, 9, and 9 discrete levels respectively. The quasi-level method is used to optimize the orthogonal experimental table to obtain 81 variable combinations.

[0017] Furthermore, the above C includes adjusting the distance between phase lines, and traversing all current combinations at the specified distance according to the obtained orthogonal experimental table to adjust the power output; after the power output is stable, the power quality analyzer automatically collects and uploads the Rogowski coil and standard CT current measurement values.

[0018] The uploading of the Rogowski coil and the standard CT current measurement value includes: collecting and uploading four sets of data in each cycle (1s), that is, collecting and uploading one set every 250ms, each set of data includes the distance (D), the standard CT measurement value (I x,CT ,x=a,b,c), Rogowski coil measurement value (I x,R ,x=a,b,c).

[0019] Furthermore, in the above D, MFNN is designed as a structure with four input variables and three output variables, where the input vector is X, including independent variables D, I a,R ,I b,R ,I c,R , expressed as X = [D, I a,R ,I b,R ,I c,R]; Output is the compensation value I of the three-phase current a,AI ,I b,AI ,I c,AI .

[0020] Furthermore, the MFNN adopts an architecture of two shared feature extraction layers and three branch prediction layers, wherein the shared feature extraction layer is used to extract global features from input data, and the branch prediction layer is used to provide predictions for each output variable, i.e., the three-phase compensation currents A, B, and C;

[0021] The two feature extraction layers are both fully connected layers, where the number of hidden neurons in the first layer is 128, so the weight matrix W1 is 128×4, plus the bias term b1, and after rectification by the activation function f, the output Z1 of the first shared feature extraction layer is obtained:

[0022] Z1=f(W1·X+b1)

[0023] f(x)=max(0,x)

[0024] In order to compress the extracted features, the number of neurons in the second layer is 64.

[0025] Each of the three branches is used to predict the corresponding output variable, i.e., I a,AI ,I b,AI ,I c,AI ; Each branch has an independent fully connected layer, and the number of output neurons is 1;

[0026] I a,AI ,I b,AI ,I c,AI The prediction process is:

[0027]

[0028] Among them, Z2 is the output of the second shared matrix extraction layer, W a ,W b ,W c They are the weight matrices of the three branches, and their sizes are all 1×64, b a ,b b ,b c They are the bias items of the three branches respectively. There is no activation function after the fully connected layer. Finally, the compensation current I of the three phases A, B, and C is obtained. a,AI ,I b,AI ,I c,AI .

[0029] Furthermore, the three-phase CT measurement value is used as the standard value in E to calculate the nominal value I a,CT ,I b,CT ,I c,CT With compensation value Ia,AI ,I b,AI ,I c,AI In MFNN, the loss is calculated for each output, that is, the loss of the three phases is calculated separately. a ,ε b ,ε c , the calculation formula is as follows:

[0030]

[0031] Where n is the total number of samples, and the final total loss ε is the weighted sum of the three-phase losses, with the three-phase weights being α, β, and γ respectively:

[0032] ε=αε a +βε b +γε c

[0033] Finally, the stochastic gradient descent method is used to train the parameters θ of the MFNN. The learning rate η of each round adjusts the single-round training step size to obtain the updated parameters θ':

[0034]

[0035] Compared with the prior art, one or more embodiments of the present invention may have the following advantages:

[0036] This method builds a current detection test bench, divides the continuous variables into several discrete levels, optimizes the experimental design and reduces the number of experiments based on the orthogonal experimental design combined with the quasi-level method; trains the Rogowski coil interphase interference compensation model based on MFNN to compensate for the original measurement value of the Rogowski coil and improve the measurement accuracy and stability of the Rogowski coil. The multi-branch structure in the MFNN method improves its generalization, making it perform well under different complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is the workflow diagram of the MFNN method for interphase interference compensation of Rogowski coils;

[0038] Figure 2 This is the MFNN network structure diagram. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below in conjunction with embodiments and drawings.

[0040] like Figure 1 As shown in FIG. 1 , the workflow of the MFNN method for Rogowski coil interphase interference compensation includes the following steps:

[0041] Step 10: Build a three-phase alternating current detection test bench, and use a 0.01% standard CT and a Rogowski coil to measure the three-phase current to obtain the current values ​​(A) of the three phases A, B, and C;

[0042] Step 20: Select the distance (cm) and the three-phase current (A) parameters of A, B, and C by orthogonal experimental design to ensure the scientificity and operability of the method;

[0043] Step 30 collects the current measurement values ​​of the Rogowski coil and the standard CT at the specified phase line distance and the specified current in the three-phase power system;

[0044] Step 40: Innovate MFNN, combine fully connected neural network with multi-task learning, and improve the generalization of compensation method;

[0045] In step 50, the CT measurement value is used as the nominal value to train the Rogowski coil interphase interference compensation model based on MFNN to achieve Rogowski coil measurement value compensation.

[0046] In the above step 10, the three-phase alternating current detection test bench includes yellow, green and red three-phase conductors, on which Rogowski coils are respectively mounted, and the three Rogowski coils are all connected to the power quality analyzer; the three 0.01% standard CTs are mounted on the three-phase conductors and connected to the power quality analyzer; the three-phase conductors are fixed on the guide rails, and the distance between the three-phase conductors is adjusted by the guide rails; on the right are three 0.01% standard CTs, which are respectively mounted on the three-phase conductors, and the three standard CTs are connected to the power quality analyzer. The lower right corner is the connection of the power quality analyzer, which is connected to the Rogowski coil and the standard CT, and can read the current measurement value, and save the current measurement value of the Rogowski coil and the 0.01% standard CT in real time through Wi-Fi.

[0047] In the above step 20, the experimental table is designed according to the orthogonal design principle;

[0048] The variable related to phase interference in the three-phase current system is the distance between phase lines (D αb ,D bc ) and A, B, C three-phase current (I a ,I b ,I c ); Under actual working conditions, phase lines A and C are symmetrical about phase line B, so they are regarded as a distance variable D, that is, D = D ab =D bc ; Four variables (D,I a ,I b ,I c ) are all continuous variables. The interval partitioning method is used to divide the four continuous variables into 6, 9, 9, and 9 discrete levels respectively. The quasi-level method is used to optimize the orthogonal experimental table to obtain 81 variable combinations.

[0049] The above step 30 includes adjusting the distance between phase lines, and traversing all current combinations at the specified distance according to the obtained orthogonal experimental table to adjust the power output; after the power output is stable, the power quality analyzer automatically collects and uploads the Rogowski coil and standard CT current measurement values.

[0050] The uploading of the Rogowski coil and the standard CT current measurement values ​​includes: collecting and uploading four sets of data in each cycle (1s), that is, collecting and uploading one set every 250ms, each set of data includes the distance (D), the standard CT measurement value (I x,CT ,x=a,b,c), Rogowski coil measurement value (I x,R ,x=a,b,c).

[0051] In the above 40, MFNN is designed as a structure with four input variables and three output variables, where the input vector is X, including independent variables D, I a,R ,I b,R ,I c,R , expressed as X = [D, I a,R ,I b,R ,I c,R ]; Output is the compensation value I of the three-phase current a,AI ,I b,AI ,I c,AI .

[0052] like Figure 2 As shown, MFNN adopts an architecture of two shared feature extraction layers and three branch prediction layers, wherein the shared feature extraction layer is used to extract global features from input data, and the branch prediction layer is used to provide predictions for each output variable, i.e., the three-phase compensation currents A, B, and C;

[0053] The two feature extraction layers are both fully connected layers, which are used to deeply extract the potential feature relationship between the four independent variables. The first layer shares the input X=[D,I a,R ,I b,R ,I c,R ], the number of hidden neurons is 128, so the size of its weight matrix W1 is <128×4>, and the size of the bias term b1 is <128> , and use the ReLU activation function f to increase its nonlinear feature extraction capability. After rectification of the activation function f, the output Z1 of the first shared feature extraction layer is obtained:

[0054] Z1=f(W1·X+b1)

[0055] f(x)=max(0,x)

[0056] The number of hidden neurons in the second shared feature extraction layer is 64, so the size of its weight matrix W2 is <64×128>, and the size of the bias term b2 is <64> , and the ReLU activation function is used to increase its nonlinear feature extraction capability. After the shared feature extraction layer inference calculation, the number of output features is 64.

[0057] The three branch prediction layers are all fully connected layers, which are used to map the shared features into compensation current values, that is, to predict the corresponding output variables I a,AI ,I b,AI ,I c,AI ; The three branches predict the three-phase currents A, B, and C respectively, and the hidden neurons are all 1, so the weight matrix W a ,W b ,W c The size of is <64×1>, and the bias term b a ,b b ,b c Size <1> , without using activation function. The three-branch prediction obtains the current compensation value Y = [I a,AI ,I b,AI ,I c,AI ]

[0058] I a,AI ,I b,AI ,I c,AI The prediction process is:

[0059]

[0060] Among them, Z2 is the output of the second shared matrix extraction layer, W a ,W b ,W c are the weight matrices of the three branches, b a ,b b ,b c They are the bias items of the three branches respectively. There is no activation function after the fully connected layer. Finally, the compensation current I of the three phases A, B, and C is obtained. a,AI ,I b,AI ,I c,AI .

[0061] In the above step 50, the three-phase CT measurement value is used as the standard value to calculate the nominal value I a,CT ,I b,CT ,I c,CT With compensation value I a,AI ,I b,AI ,I c,AI In MFNN, the loss is calculated for each output, that is, the loss of the three phases is calculated separately. a ,ε b ,ε c, the calculation formula is as follows:

[0062]

[0063] Where n is the total number of samples, and the final total loss ε is the weighted sum of the three-phase losses, with the three-phase weights being α, β, and γ respectively:

[0064] ε=αε a +βε b +γε c

[0065] Finally, the stochastic gradient descent method is used to train the parameters θ of the MFNN. The learning rate η of each round adjusts the single-round training step size to obtain the updated parameters θ':

[0066]

[0067] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A MFNN method for Rogowski coil interphase interference compensation, characterized in that: The method comprises: A. Build a three-phase alternating current detection test bench and use 0.01% standard CT and Rogowski coil to measure the three-phase current; B. Select the distance between phase lines and the three-phase current parameters of A, B, and C through orthogonal experimental design; C. Collect the current measurement values ​​of Rogowski coils and standard CTs under different current combinations at specified phase line distances and specified current combinations; D. Design MFNN based on fully connected neural network and multi-task learning; E. Taking the CT measurement value as the standard value, the Rogowski coil interphase interference compensation model based on MFNN is trained to realize the Rogowski coil measurement value compensation.

2. The MFNN method for Rogowski coil interphase interference compensation according to claim 1, characterized in that: The three-phase alternating current detection test bench in A comprises yellow, green and red three-phase conductors, on which Rogowski coils are respectively mounted, and all three Rogowski coils are connected to a power quality analyzer; the three 0.01% standard CTs are mounted on the three-phase conductors and connected to the power quality analyzer; the three-phase conductors are fixed on a guide rail, and the distance between the three-phase conductors is adjusted by the guide rail; the power quality analyzer saves the current measurement values ​​of the Rogowski coils and the 0.01% standard CTs in real time via Wi-Fi.

3. The MFNN method for Rogowski coil interphase interference compensation according to claim 1, characterized in that: In said B, the experimental table is designed according to the orthogonal design principle; The variable related to phase interference in the three-phase current system is the distance between phase lines (D αb ,D bc ) and A, B, C three-phase current (I a ,I b ,I c ); Under actual working conditions, phase lines A and C are symmetrical about phase line B, so they are regarded as a distance variable D, that is, D = D ab =D bc ; Four variables (D, I a ,I b ,I c ) are all continuous variables. The interval partitioning method is used to divide the four continuous variables into 6, 9, 9, and 9 discrete levels respectively. The quasi-level method is combined to optimize the orthogonal experimental table to obtain 81 variable combinations.

4. The MFNN method for Rogowski coil interphase interference compensation according to claim 3, characterized in that: The C includes adjusting the distance between phase lines, and traversing all current combinations at the specified distance according to the obtained orthogonal experimental table to adjust the power output; after the power output is stable, the power quality analyzer automatically collects and uploads the Rogowski coil and standard CT current measurement values.

5. The MFNN method for Rogowski coil interphase interference compensation according to claim 4, characterized in that: The uploading of the Rogowski coil and the standard CT current measurement values ​​includes: collecting and uploading four sets of data in each cycle (1s), that is, collecting and uploading one set every 250ms, each set of data includes the distance (D), the standard CT measurement value (I x,CT ,x=a,b,c), Rogowski coil measurement value (I x,R ,x=a,b,c).

6. The MFNN method for Rogowski coil interphase interference compensation according to claim 1, characterized in that: In D, MFNN is designed as a structure with four input variables and three output variables, where the input vector is X, including independent variables D, I a,R ,I b,R ,I c,R , expressed as X = [D, I a,R ,I b,R ,I c,R ]; Output is the compensation value I of the three-phase current a,AI ,I b,AI ,I c,AI .

7. The MFNN method for Rogowski coil interphase interference compensation according to claim 6, characterized in that: The MFNN adopts an architecture of two shared feature extraction layers and three branch prediction layers. The shared feature extraction layer is used to extract global features from input data, and the branch prediction layer is used to provide predictions for each output variable, i.e., the three-phase compensation currents A, B, and C; The two feature extraction layers are both fully connected layers, where the number of hidden neurons in the first layer is 128, so the weight matrix W1 is 128×4, plus the bias term b1, and after rectification by the activation function f, the output Z1 of the first shared feature extraction layer is obtained: Z1=f(W1·X+b1) f(x)=max(0,x) In order to compress the extracted features, the number of neurons in the second layer is 64. Each of the three branches is used to predict the corresponding output variable, i.e., I a,AI ,I b,AI ,I c,AI ; Each branch has an independent fully connected layer, and the number of output neurons is 1; I a,AI ,I b,AI ,I c,AI The prediction process is: Among them, Z2 is the output of the second shared matrix extraction layer, W a ,W b ,W c They are the weight matrices of the three branches, and their sizes are all 1×64, b a ,b b ,b c They are the bias items of the three branches respectively. There is no activation function after the fully connected layer. Finally, the compensation current I of the three phases A, B, and C is obtained. a,AI ,I b,AI ,I c,AI .

8. The MFNN method for Rogowski coil interphase interference compensation according to claim 1, characterized in that: In E, the three-phase CT measurement value is used as the standard value, and the nominal value I is calculated. a,CT ,I b,CT ,I c,CT With compensation value I a,AI ,I b,AI ,I c,AI In MFNN, the loss is calculated for each output, that is, the loss of the three phases is calculated separately. a ,ε b ,ε c , the calculation formula is as follows: Where n is the total number of samples, and the final total loss ε is the weighted sum of the three-phase losses, with the three-phase weights being α, β, and γ respectively: e=ae a +ve b +ge c Finally, the stochastic gradient descent method is used to train the parameters θ of the MFNN. The learning rate η of each round adjusts the single-round training step size to obtain the updated parameters θ':

Images