Fast digital twinning modeling method for switch cabinet
By combining virtual material method and deep operator network, the accuracy problem of temperature field prediction of switchgear due to equipment aging is solved, realizing fast and accurate temperature field prediction and improving the prediction capability of digital twin system.
Patent Information
- Application Number
- CN202510967826.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-14
- Publication Date
- 2025-10-17
AI Technical Summary
Existing technologies cannot fully consider the impact of equipment aging on key electrical parameters, and cannot accurately warn of the risk of local overheating caused by the synergistic effect of contact degradation and insulation aging, resulting in a decrease in the prediction accuracy of the digital twin system after long-term operation.
A virtual material method model is used to handle contact resistance. Combined with a deep operator network, a dynamic coupling mechanism between the time-varying material parameter model and boundary conditions is established through a fully connected neural network structure and Fourier transform. The neural network is then trained to quickly predict the temperature field of the switchgear.
It significantly improves the accuracy of temperature field prediction for equipment with different service years, with a fast calculation time of 4.53s, which shortens the calculation time compared to finite element simulation, and realizes accurate prediction of the temperature field evolution of power equipment.
Smart Images
Figure CN120805598A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of rapid digital twinning of switch cabinets, in particular to a rapid digital twinning modeling method of a switch cabinet. BACKGROUND
[0002] The switch cabinet is responsible for distributing electric energy and isolating faults, and the safety and reliability of the entire power distribution network depend on the performance of the switch cabinet, which plays an important role in the power distribution network. Therefore, it has strong engineering practical significance to study the switch cabinet and master its temperature rise characteristics. At present, the calculation method of the switch cabinet temperature field is mainly solved by finite element simulation, but the forward solution by finite element simulation alone will have a large amount of work and take a long time, which is difficult to apply to switch cabinet digital twinning. Therefore, the prior art considers combining finite elements and neural networks to realize rapid calculation of switch cabinet temperature field.
[0003] However, the prior art generally adopts a fully connected neural network architecture, which has a relatively simple structure and usually uses fixed parameters for calculation, and cannot fully consider the influence of device aging on key electrical parameters. In particular, it fails to consider the growth characteristics of conductor contact resistance with operating time and does not reflect the decay phenomenon of insulation material resistivity under the action of electrical and thermal aging. This idealized modeling method leads to a significant decrease in prediction accuracy of the digital twinning system after long-term operation, and cannot accurately predict the risk of local overheating caused by the synergistic effect of contact degradation and insulation aging.
[0004] Therefore, it is necessary to further solve the above problems, and the present applicant proposes a rapid digital twinning modeling method of a switch cabinet. SUMMARY
[0005] In view of the deficiencies of the prior art, the present application provides a rapid digital twinning modeling method of a switch cabinet, which solves the problem that the prior art cannot fully consider the influence of device aging on key electrical parameters and cannot accurately predict the risk of local overheating caused by the synergistic effect of contact degradation and insulation aging.
[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions: A rapid digital twinning modeling method of a switch cabinet, the method comprising the following steps: S1: considering the influence of device aging of the switch cabinet due to operating time on the material properties, mainly considering the contact resistance between conductors and the resistivity of insulation materials; S2: establishing a finite element model of the object, determining the boundary conditions and structure parameters, and solving the temperature field under different working conditions by using the finite element method after mesh division; S3: Derive the grid point coordinates on each finite element grid point, the temperature of the switch cabinet, the ambient temperature and current when the switch cabinet is running, and the contact resistance and the conductivity of the insulating material, and divide the above derived data into a training set and a test set; S4: The deep operator network is divided into two parts, the branch network input is the ambient temperature and current when the switch cabinet is running, and the contact resistance and the conductivity of the insulating material, and the backbone network input is the result of the grid point coordinates after Fourier transform; S5: Neural network training is performed; S6: The accuracy of the trained model is evaluated by the normalized mean absolute error NMAE of the test set; S7: The hot spot temperature of the switch cabinet with different running years is monitored, the field collected data is taken as the input of the algorithm, and the output of the algorithm is compared with the experimental results to verify the accuracy of the algorithm for predicting the temperature field of the switch cabinet; Step S2 is specifically: The equations are as shown in the following formulas 2-4, and the boundary conditions are as shown in the following formula 5: is the current density vector, with units of A / m 2 ; is the electric field intensity vector, with units of V / m; is the electric displacement vector, C / m 2 ; is the charge density change, A / m 3 ; subscript j,v represents the change of the body current density under the three-dimensional model, is the conductivity, =1Ω·m; is the electric potential, with units of V; is the thermal conductivity of the material; is the density of the material; is the constant pressure heat capacity of the material; is the fluid velocity field; is the heat; is the energy generated per unit volume, with units of W / m 3 ; is the temperature, with units of K; is the time, with units of s; is the fluid dynamic viscosity; is the acceleration of gravity; is the inertial force; is the pressure gradient; is the viscous force; is the unit tensor; is the fluid pressure; is the heat flux vector, with the unit of W / m²; is the unit normal vector; is the convective heat transfer coefficient, with the unit of W / (m²·K); is the external fluid temperature, with the unit of K; is the boundary surface temperature, with the unit of K.
[0007] As a further scheme of the present application: step S1 is specifically: For the contact resistance, it cannot be directly introduced in the switch cabinet digital twin research, therefore a contact resistance model of virtual material method is adopted, the resistivity of the virtual material of the contact and the bolted place equivalent to the equivalent resistance cylinder or square piece structure is adjusted to control the overall resistance value, as shown in the following formula 1, the resistivity of the insulating material can be directly introduced in the digital twin research, is the contact resistance value, is the virtual material resistivity, is the equivalent cross-sectional area, is the current path length.
[0008] As a further scheme of the present application: step S4 is specifically: Both networks adopt a fully connected neural network structure, and the Fourier transform formula is shown in the following formula 6: wherein is the input grid point coordinate, satisfies a three-dimensional normal distribution with a mean of 0, a covariance matrix as shown in the following formula 7, and a correlation coefficient of 0, the size of is 3 x m / 2, and m is the number of neurons in the first hidden layer of the neural network; wherein represents a variable, here referring to the grid point x, y, z coordinates, , , are the variances of x, y and z respectively.
[0009] As a further scheme of the present application: step S5 is specifically: The neural network is trained, wherein a loss function is shown in the following formula 8, an optimizer is selected as Adam, and a linear layer activation function is selected as LeakyRelu, wherein, is composed of an ambient temperature and a current when the switch cabinet is running, and a contact resistance and an electrical conductivity of an insulating material , under the condition that the neural network with parameters , a sampling point , and a value of is a temperature calculated by a finite element, is a number of sampling points.
[0010] As a further scheme of the present application, the calculation formula of step S6 is specifically shown in the following formula 9: .
[0011] Compared with the prior art, the present application has the following beneficial effects: The present method uses two fully connected neural networks to extract boundary conditions and grid coordinate information on the basis of a deep operator network, uses Fourier transform to increase the weight of a high-frequency part, so that the neural network can learn the high-frequency part, i.e., a high gradient area, in the data more quickly, uses varying boundary conditions, grid vertex coordinates and corresponding finite element solutions to construct a training set, and the network after training can quickly predict the temperature field distribution under untrained boundary conditions. The calculation time of the fast algorithm is 4.53s, and for the same model, the calculation time of finite element simulation is about 8h. Meanwhile, in a digital twin system, in order to realize accurate prediction of the temperature field evolution of power equipment, the influence of material aging effect on thermodynamic characteristics needs to be considered. The present method significantly improves the temperature field prediction accuracy of equipment with different service life by establishing a dynamic coupling mechanism between a time-varying material parameter model and boundary conditions. At the same time, the hotspot temperature of a switch cabinet with different service life is monitored, the data collected on site is used as the input of the algorithm, and the output of the algorithm is compared with the experimental results, and the effect is good. BRIEF DESCRIPTION OF DRAWINGS
[0012] Figure 1 is a switch cabinet model schematic diagram of the present application; Figure 2 is a deep operator network schematic diagram of the present application; Figure 3 is a finite element solution schematic diagram of the present application; Figure 4 is a neural network solution schematic diagram of the present application; Figure 5 A point-by-point error estimation diagram for a rapid digital twin modeling method of a switch cabinet of the present application; Figure 6 A point-by-point error estimation diagram for a rapid digital twin modeling method of a switch cabinet of the present application without embedded Fourier features; Figure 7 A point-by-point error estimation diagram for a rapid digital twin modeling method of a switch cabinet of the present application with embedded Fourier features. DETAILED DESCRIPTION
[0013] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0014] Embodiment 1 Please refer to Figures 1-7 A rapid digital twin modeling method of a switch cabinet, the method comprising the following steps: Step 1: Consider the influence of device aging on the material properties of the switch cabinet due to the running time, mainly considering the contact resistance between conductors and the resistivity of insulating materials For the contact resistance, it cannot be directly introduced in the switch cabinet digital twin research, so the contact resistance model of the virtual material method is adopted, and the overall resistance value is controlled by adjusting the resistivity of the virtual material of the contact and the bolted place which is equivalent to a resistive cylinder or a square sheet structure, as shown in the following formula 1, and the resistivity of the insulating material can be directly introduced in the digital twin research, For the contact resistance value, For the virtual material resistivity, For the equivalent cross-sectional area, For the current path length.
[0015] Step 2: Establish a finite element model of the object, determine the boundary conditions and structural parameters, and after meshing, solve the temperature field under different working conditions by using the finite element method The solution equation is shown in the following formulas 2-4, and the boundary condition is shown in the following formula 5: For the current density vector, the unit is A / m 2 ; For the electric field intensity vector, the unit is V / m; For the electric displacement vector, C / m2 ; is the change of charge density, A / m 3 ; subscript j,v represents the change of current density of the lower body in the three-dimensional model, is the electrical conductivity, =1Ω·m; is the electric potential, with the unit of V; is the thermal conductivity of the material; is the density of the material; is the constant pressure heat capacity of the material; is the fluid velocity field; is the heat; is the energy generated per unit volume, with the unit of W / m 3 ; is the temperature, with the unit of K; is the time, with the unit of s; is the fluid dynamic viscosity; is the acceleration of gravity; is the inertial force; is the pressure gradient; is the viscous force; is the unit tensor; is the fluid pressure; is the heat flux vector, with the unit of W / m²; is the unit normal vector; is the convective heat transfer coefficient, with the unit of W / (m²·K); is the temperature of the external fluid, with the unit of K; is the boundary surface temperature, with the unit of K.
[0016] Step 3: Derive the lattice point coordinates on each finite element grid, the temperature of the switch cabinet, the ambient temperature and current when the switch cabinet is running, and the contact resistance and conductivity of the insulation material. Divide the above derived data into training set and test set.
[0017] Step 4: The deep operator network is divided into two parts. The input of the branch network is the ambient temperature and current when the switch cabinet is running, and the contact resistance and conductivity of the insulation material. The input of the trunk network is the result of the Fourier transform of the lattice point coordinates Both networks use a fully connected neural network structure. The Fourier transform formula is shown in the following formula 6: in is the input grid coordinate, A three-dimensional normal distribution with a mean of 0, a covariance matrix as shown in Equation 7, and a correlation coefficient of 0 is satisfied. The size of is 3×m / 2, where m is the number of neurons in the first hidden layer of the neural network; in represents the variables, here the x, y, and z coordinates of the grid points, 、 、 are the variances of x, y, and z, respectively.
[0018] Step 5: Train the neural network, where the loss function is shown in Equation 8 below. The optimizer is Adam, and the linear layer activation function is LeakyRelu. in, It is composed of the ambient temperature and current when the switch cabinet is running, as well as its contact resistance and the conductivity of the insulation material. Under the condition of Neural network, sampling points The value of is the temperature calculated by finite element, is the number of sampling points.
[0019] Step 6: The trained model is evaluated for accuracy using the normalized mean absolute error (NMAE) of the test set, as shown in Equation 9 below: Step 7: Monitor the hotspot temperatures of switchgear with different operating years, use the data collected on-site as the input of the algorithm, compare the output of the algorithm with the experimental results, and verify the accuracy of the algorithm in predicting the switchgear temperature field.
[0020] Example 2 Specifically, the KYN28-12 medium-voltage switchgear was used as the research object. The current and ambient temperature during operation were changed. The effects of aging on its material properties were fully considered. The contact resistance and conductivity of the insulating material were changed. The contact resistance and conductivity of the insulating material were changed by multiplying them by coefficients. Resistivity and conductivity are inversely proportional to each other. The specific parameters are shown in Table 1 below: 1. Isometrically select 11 groups of ambient temperature, 6 groups of current and 5 groups of contact resistance and 4 groups of conductivity, a total of 320 groups of cases, use the finite element method for parameterized calculation to obtain the data set; 2. Export Figure 1 the grid point coordinates on each finite element grid point in the switch cabinet shown in the figure, the temperature of the switch cabinet, the ambient temperature and current of the switch cabinet during operation, and the contact resistance and conductivity of the insulating material, select 300 groups of the entire data set as the training set, and the remaining as the test set; 3. As Figure 2 shown, the ambient temperature and current of the switch cabinet during operation, and the contact resistance and conductivity of the insulating material are taken as the input of the branch network in the Figure 2 , the grid point coordinates are taken as the input of the main network, and Fourier transform is performed, the branch network adopts a fully connected neural network structure, the main network adopts a fully connected neural network structure, Figure 2 where n is the number of grid point coordinates, the optimizer selects Adam, the loss function is shown in equation (8), and the neural network is trained; 4. The normalized mean absolute error is used to evaluate the accuracy, the average value of the test set is 0.51%, the minimum relative error is 0.26%, and the maximum relative error is 0.77%, the accuracy meets the requirements, a group of data is randomly taken in the test set, the finite element solution, the neural network solution and the point-by-point error are shown in Figure 3 , Figure 4 and Figure 5 , and it can be seen from the two figures of Figure 6 and Figure 7 that after the embedding balance of the Fourier features, the error of the switch cabinet internal conductor is obviously small, in [-0.5, 0.5]; 5. Monitor the hot spot temperature of the switch cabinet with different running years, the hot spot temperature of the switch cabinet changes with the running years as shown in the following table 1: As can be seen from table 1, the difference between the predicted temperature and the actual temperature is not more than 1℃, the algorithm has good effect for predicting the temperature of the switch cabinet, the field environment temperature is 20.5℃, and the current passing through the switch cabinet is 1350A.
[0021] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, which should be covered in the scope of the claims of the present application.
Claims
1. A fast digital twin modeling method for switchgear, characterized by: The method comprises the following steps: S1: Consider the impact of aging of switchgear components on their material properties due to years of operation, mainly considering the contact resistance between conductors and the resistivity of the insulation material; S2: Establish a finite element model of the object, determine the boundary conditions and structural parameters, perform meshing, and use the finite element method to solve the temperature field under different working conditions; S3: Export the grid coordinates of each finite element grid point, the temperature of the switchgear, the ambient temperature and current during the operation of the switchgear, as well as its contact resistance and the conductivity of the insulating material, and divide the above-exported data into a training set and a test set; S4: The deep operator network is divided into two parts. The input of the branch network is the ambient temperature and current of the switch cabinet during operation, as well as its contact resistance and the conductivity of the insulation material. The input of the main network is the result of Fourier transform of the grid coordinates. S5: Perform neural network training; S6: The trained model is evaluated for accuracy by the normalized mean absolute error (NMAE) of the test set. S7: Monitor the hot spot temperatures of switchgear with different operating years, use the data collected on site as the input of the algorithm, compare the output of the algorithm with the experimental results, and verify the accuracy of the algorithm in predicting the switchgear temperature field; Step S2 is specifically as follows: The solution equations are shown in Equations 2-4 below, and the boundary conditions are shown in Equation 5 below: is the current density vector, in A / m 2 ; is the electric field strength vector, unit is V / m; is the electric displacement vector, C / m 2 ; is the charge density change, A / m 3 ; Subscript j,v represents the change of current density in the lower body of the three-dimensional model, is the conductivity, =1Ω·m; is the electric potential, in V; is the thermal conductivity of the material; is the density of the material; is the constant pressure heat capacity of the material; is the fluid velocity field; for heat; The energy generated per unit volume, in W / m 3 ; is the temperature in K; is the time, the unit is s; is the fluid dynamic viscosity; is the acceleration due to gravity; is the inertial force; is the pressure gradient; is the viscous force; is the unit tensor; is the fluid pressure; is the heat flux vector, in W / m²; is the unit normal vector; is the convective heat transfer coefficient, in W / (m²·K); is the external fluid temperature in K; is the boundary surface temperature in K.
2. The rapid digital twin modeling method for a switch cabinet according to claim 1, characterized in that: Step S1 is specifically as follows: Contact resistance cannot be directly introduced in the switchgear digital twin research. Therefore, a contact resistance model based on the virtual material method is used. The overall resistance value is controlled by adjusting the resistivity of the virtual materials at the contacts and bolt joints, which are equivalent to equal-resistance cylinders or square sheets. As shown in the following formula 1, the resistivity of the insulating material can be directly introduced in the digital twin research. is the contact resistance value, is the resistivity of the fictitious material, is the equivalent cross-sectional area, is the current path length.
3. The rapid digital twin modeling method for a switch cabinet according to claim 1, characterized in that: Step S4 is specifically as follows: Both networks use a fully connected neural network structure, and the Fourier transform formula is shown in Equation 6 below: in is the input grid coordinate, A three-dimensional normal distribution with a mean of 0, a covariance matrix as shown in Equation 7, and a correlation coefficient of 0 is satisfied. The size of is 3×m / 2, where m is the number of neurons in the first hidden layer of the neural network; in represents the variables, here the x, y, and z coordinates of the grid points, 、 、 are the variances of x, y, and z, respectively.
4. The rapid digital twin modeling method for a switch cabinet according to claim 1, characterized in that: Step S5 is specifically as follows: Train the neural network, where the loss function is shown in Equation 8 below, Adam is selected as the optimizer, and LeakyRelu is selected as the linear layer activation function. in, It is composed of the ambient temperature and current when the switch cabinet is running, as well as its contact resistance and the conductivity of the insulation material. Under the condition of Neural network, sampling points The value of is the temperature calculated by finite element, is the number of sampling points.
5. The rapid digital twin modeling method for a switch cabinet according to claim 4, characterized in that: The calculation formula of step S6 is specifically shown in the following formula 9: 。