Method for calculating humidity inside medium and high voltage switch cabinet based on simulation analysis

Abstract

The present application belongs to the technical field of simulation analysis, and particularly relates to a method for calculating humidity inversion in a medium-voltage switch cabinet based on simulation analysis, which comprises the following steps: establishing a geometric model of the switch cabinet, importing the geometric model into a finite element simulation software to obtain a simulation analysis model, adding each physical field, its calculation domain and boundary conditions, coupling each physical field, and performing mesh partitioning; using a transient solver to numerically solve the simulation analysis model to obtain the temperature and humidity distribution in the switch cabinet; optimizing the simulation analysis model, constructing a simulation data set, and training a PINN; based on the collected environmental temperature and humidity and operating current in the actual operation of the switch cabinet, the trained PINN is used for prediction to obtain the steady-state relative humidity at the current transformer. The present application realizes efficient and accurate reconstruction of the humidity field in the key area of the switch cabinet based on limited boundary monitoring data, and provides a feasible path for equipment state perception and condensation risk early warning.

Description

Technical Field

[0001] This invention belongs to the field of simulation analysis technology, specifically involving a simulation analysis-based method for calculating the internal humidity of medium and high voltage switchgear, which is particularly applicable to KYN series switchgear. Background Technology

[0002] In modern power systems, medium and high voltage switchgear plays a crucial role in switching, controlling, and protecting circuits, and its operational reliability is vital to the safety of the entire power grid. However, as the operating environment of power grid equipment becomes increasingly complex, the equipment failure rate induced by high humidity factors continues to rise. Taking the KYN-28A switchgear as an example, its compact design saves installation space, but this also results in structural limitations such as smaller insulation distances, narrow internal space, and poor heat dissipation. These characteristics make the insulation performance of key components more susceptible to environmental fluctuations, especially the operational hazards caused by changes in internal humidity. In high humidity environments and when there are drastic temperature differences between the inside and outside of the cabinet, condensation can easily form on the surface or inner walls of electrical equipment, inducing serious consequences such as decreased insulation performance, partial discharge, and even equipment breakdown, directly threatening the stable operation of the power distribution system. In actual engineering, the placement of temperature and humidity sensors can cause electromagnetic interference and damage to the insulation structure, and critical components such as current transformers and insulating bushings are often located in small, enclosed spaces, making it difficult to obtain the actual relative humidity on their surfaces in a timely and accurate manner.

[0003] Currently, due to the complex electromagnetic environment and compact structure inside the cabinet, it is difficult to directly deploy temperature and humidity sensors in the aforementioned critical areas. Adding sensors may also introduce additional insulation risks. Although traditional finite element simulation methods can analyze the temperature and humidity fields inside the cabinet and provide theoretical references for equipment design and operation, they suffer from problems such as complex modeling, high computational costs, and low solution efficiency, making it difficult to achieve rapid diagnosis and online condition assessment in the engineering field.

[0004] In addition, traditional humidity control strategies usually set start and stop thresholds based on the measurement values ​​of single-point humidity sensors inside or outside the cabinet. This cannot reflect the true humidity status of the surface of key insulation parts such as current transformers, and cannot identify the risk of local condensation in advance. This can easily lead to lag or frequent start and stop of dehumidification and heating devices, making it difficult to balance safety and economy. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a simulation analysis-based method for calculating the humidity inside medium and high voltage switchgear. Based on finite boundary monitoring data, it realizes efficient and accurate reconstruction of the humidity field in key areas inside the switchgear, providing a feasible path for equipment status perception and condensation risk early warning.

[0006] To achieve the above objectives, this invention provides a simulation-based method for calculating the internal humidity of medium- and high-voltage switchgear, comprising the following steps:

[0007] S1. For medium and high voltage switchgear, use 3D drawing software to establish a geometric model of the switchgear, including current-carrying busbars, current transformers, circuit breakers, switchgear enclosure, insulators and solid-sealed poles. Import the geometric model into finite element simulation software to obtain a simulation analysis model, and set the material properties of each component.

[0008] S2. Add current physical fields, solid and fluid heat transfer physical fields, turbulence physical fields, and air moisture transport physical fields to the finite element simulation software, and set the computational domain and boundary conditions for each physical field.

[0009] S3. Couple the physical fields and use the free tetrahedral meshing method to mesh the simulation analysis model;

[0010] S4. Add transient study, use transient solver to numerically solve the simulation analysis model, post-process the solution results, and obtain the temperature and humidity distribution inside the switch cabinet.

[0011] S5. Conduct a temperature rise test, measure the actual temperature and humidity of the switchgear during operation, compare the experimental data with the simulation results, and optimize the simulation analysis model.

[0012] S6. In the optimized simulation analysis model, input the operating current and environmental conditions, extract the relative humidity of the current transformer and the inner wall of the cable room, and construct the simulation dataset.

[0013] S7. Use the SMOTE strategy to augment the simulation dataset and construct a loss function that includes data loss and physical loss to train PINN.

[0014] S8. Using the ambient temperature and humidity and operating current collected during the actual operation of the switchgear as input, and using the trained PINN for prediction, the steady-state relative humidity at the current transformer is obtained.

[0015] As a preferred embodiment of the present invention, in S1, a geometric model of the original size of the switch cabinet is constructed using SolidWorks three-dimensional drawing software based on the actual structural parameters of the switch cabinet. The solid-sealed pole is obtained by encapsulating the circuit breaker vacuum interrupter and the moving / static contacts into one unit using an insulating shell as a carrier. The cable compartment is a functional compartment at the bottom of the switch cabinet that accommodates the current transformer.

[0016] Import the geometric model into the COMSOL finite element simulation software, specify copper material for the current-carrying busbar, epoxy resin material for the current transformer, insulator and solid-sealed pole, copper-chromium alloy material for the circuit breaker moving and stationary contacts, aluminum oxide material for the circuit breaker vacuum interrupter bushing, and stainless steel material for the switch cabinet enclosure.

[0017] As a preferred embodiment of the present invention, in S2, a current physical field is added in the COMSOL finite element simulation software, the current field calculation domain is set, and the boundary conditions for electric field analysis are set:

[0018] The current-carrying busbar and the moving and stationary contacts of the circuit breaker are defined as the current field calculation domain. The current input boundary is set by adding surface terminals, and the current outflow boundary is set by adding ground planes. The current magnitude is set to I0, which is the effective current value in actual operation of the switchgear. Contact impedance is added to the contact surface and bolt fastening surface to simulate localized heating caused by contact resistance. The bolt fastening surface is the contact surface between two conductive parts clamped together by bolts. The contact resistance R is calculated using the following formula:

[0019] ;

[0020] In the formula, K is a coefficient related to the contact material; m is a coefficient related to the contact form of the two contact surfaces, which includes surface contact, line contact, and point contact; F is the contact pressure.

[0021] Add solid and fluid heat transfer physics fields, set the temperature field computational domain, and set the boundary conditions for temperature analysis:

[0022] All regions except the vacuum medium inside the vacuum interrupter of the circuit breaker are set as the temperature field calculation domain. The solid domain includes the current-carrying busbar, the circuit breaker body, the current transformer and the switch cabinet shell, and the fluid domain is the air inside the switch cabinet. The heat exchange with the outside air is simulated by applying convective heat flux boundary conditions to the outer surface of the air chamber shell. The heat transfer coefficient is set according to the test conditions. The air chamber is the functional compartment filled with air inside the switch cabinet, and the air chamber shell is the metal wall panel of the functional compartment where each component is located.

[0023] Add based on The turbulent physical field of the model is defined, the computational domain of the flow field is set, and the boundary conditions for fluid flow analysis are set:

[0024] The air domain inside the chamber is set as the flow field computation domain, the fluid physics model is set as compressible flow, including gravity terms and using reduced pressure form, the wall boundary adopts no-slip condition, and pressure point constraints are set.

[0025] Add a physical field for the transport of moisture in the air, set the calculation domain for the humidity field, and set the boundary conditions for the flow of humid air:

[0026] The air domain inside the chamber is set as the calculation domain for moisture transport in the air, i.e., the humidity field calculation domain. A humid air diffusion coefficient is set, and the relative humidity boundary uses a symbol... This indicates that the transport process of moist air is described by humidity boundary conditions that match the environmental conditions.

[0027] As a preferred embodiment of the present invention, in S3, coupling the various physical fields specifically involves:

[0028] The physical field of electric current is coupled with the physical field of heat transfer in solids and fluids to form an electromagnetic thermal field; the physical field of heat transfer in solids and fluids is coupled with the physical field of turbulence to form a non-isothermal flow field; the physical field of turbulence is coupled with the physical field of moisture transport in air to form a moisture flow field; and the physical field of heat transfer in solids and fluids is coupled with the physical field of moisture transport in air to form a thermal-humid field.

[0029] When meshing, a refined mesh is used for the moving and stationary contact areas of the circuit breaker, while a conventional mesh is used for other current-carrying conductors. Other current-carrying conductors include the current-carrying busbar, the conductive parts of the current transformer, the conductive parts of the solid-sealed pole, and the conductive parts of the circuit breaker other than the moving and stationary contacts.

[0030] As a preferred embodiment of the present invention, in step S4, a transient study is added to the COMSOL finite element simulation software, and the output time step is... The transient solver MUMPS was used to numerically solve the simulation analysis model, with a relative tolerance set to 0.0001.

[0031] The solution results are post-processed to obtain the humidity field distribution inside the switch cabinet. Several humidity measurement points on the surface of the current transformer are selected to extract the steady-state relative humidity of each humidity measurement point. At the same time, the inner wall temperature and relative humidity of the cable compartment are extracted.

[0032] As a preferred embodiment of the present invention, in S5, during the temperature rise test, temperature and humidity sensors are arranged according to the actual operating conditions of the switchgear, and the relative humidity is measured at the bottom of the current transformer; at the same time, the air hole at the bottom of the cable compartment is selected as the inner wall temperature and humidity measurement feature point, and temperature and relative humidity data during the operation of the switchgear are collected.

[0033] The temperature and humidity data obtained from the temperature rise test are compared and analyzed with the simulation results. When the error between the experimental data and the simulation results exceeds the preset error threshold, the simulation analysis model is optimized and adjusted by adjusting the material parameters, boundary conditions and mesh density.

[0034] As a preferred embodiment of the present invention, in S6, different load currents, ambient temperatures and relative humidity are set in the optimized simulation analysis model, and the steady-state temperature of the cable room wall, the steady-state relative humidity of the cable room wall and the steady-state relative humidity of the current transformer surface corresponding to each working condition are extracted. A humidity inversion dataset with load current, ambient temperature and relative humidity as input and relative humidity at the cable room wall and current transformer as output is constructed as a simulation dataset.

[0035] The Physical Information Neural Network (PINN) was selected as the inversion model to invert the relative humidity at the current transformer. The number of neurons in the hidden layer of PINN was set to 128, 128, 64 and 64 respectively. Each hidden layer was introduced by layer normalization and the weights were initialized using the Xavier normal distribution. The activation function was the hyperbolic tangent function.

[0036] As a preferred embodiment of the present invention, in S7, the constructed simulation dataset is augmented using the SMOTE strategy, and the augmented samples are initially screened using pre-trained forward PINN.

[0037] Several groups of enhanced samples were randomly selected and substituted into the simulation analysis model for calculation. The consistency between the simulation results and the enhanced sample labels was compared.

[0038] The validated dataset was divided into training, validation, and test sets in a ratio of 60:20:20.

[0039] As a preferred embodiment of the present invention, in S8, the construction includes data loss and physical loss Loss function M:

[0040] ;

[0041] In the formula, , They are respectively , The weights; It is the error between the predicted result and the validated simulation dataset. The error is that the prediction result violates the physical laws inside the switchgear;

[0042] An intermediate physical modeling layer is introduced, dividing the PINN modeling into two stages: The first stage constructs a forward PINN based on ambient temperature and humidity and load current, inverting the steady-state temperature and humidity field of the switchgear inner wall; the second stage embeds the error between the inverted values ​​and measured values ​​of the switchgear inner wall state as a physical consistency constraint term into the loss function, optimizing M to obtain the optimized loss function. , is represented as:

[0043] ;

[0044] In the formula, The optimized loss of the preceding data; The optimized subsequent data loss; , , They are respectively , , Weights; utilizing the enhanced simulation dataset and Train PINN.

[0045] As a preferred embodiment of the present invention, in satisfying Under the constraints, all generated weight combinations are traversed with a step size of 0.1. Ablation experiments are performed on each weight combination, and the prediction accuracy and convergence characteristics on the test set are compared. The weight combination with the best overall performance is selected as the optimal weight combination. The final weight.

[0046] The beneficial effects of this invention are:

[0047] This invention combines multi-physics field coupling simulation with temperature rise test to construct a high-precision temperature and humidity field basic model for medium and high voltage switchgear (especially KYN series switchgear). Compared with methods that rely solely on theoretical calculations or simplified empirical models, this model can more realistically reflect the humidity distribution characteristics under complex environments inside the switchgear.

[0048] This invention introduces the SMOTE data augmentation strategy and combines it with a simulation analysis model to verify the physical consistency of the augmented samples. This effectively alleviates the problem of overfitting inversion models under small sample conditions, enabling stable and reliable prediction performance to be obtained even under limited simulation and experimental data conditions.

[0049] This invention proposes a two-stage physical information neural network modeling approach: the first stage uses environmental parameters to invert the state of the inner wall of the switch cabinet; the second stage uses the inverted value of the inner wall state as both an input feature and a physical consistency constraint, embedding it into the loss function of the final target inversion model. This achieves the dual role of feature point data in both the input and constraint dimensions, significantly improving the accuracy of humidity inversion.

[0050] This invention constructs a complete system solution from multiphysics modeling, experimental calibration, dataset construction and enhancement, PINN training and optimization, online humidity inversion, and condensation risk early warning. Compared with the traditional method of relying solely on humidity controller setpoints for start-stop control, it can predict the relative humidity change trend of key insulation parts in advance, proactively identify condensation risks, and optimize the operation decision of switchgear heating and dehumidification devices, taking into account both safety and economy. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating the principle of this invention;

[0052] Figure 2 This is a schematic diagram of the geometric model of the switch cabinet in Embodiment 1 of the present invention;

[0053] Figure 3 This is a schematic diagram of the simulation analysis model in Embodiment 1 of the present invention;

[0054] Figure 4 This is the predicted data graph in Embodiment 1 of the present invention.

[0055] In the diagram: 1. Current-carrying busbar; 2. Current transformer; 3. Circuit breaker; 4. Switch cabinet enclosure; 5. Insulator. Detailed Implementation

[0056] The embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0057] Example 1: As Figure 1 As shown, the method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis includes the following steps:

[0058] S1. For medium and high voltage switchgear, use 3D drawing software to establish a geometric model of the switchgear, including current-carrying busbar 1, current transformer 2, circuit breaker 3, switchgear shell 4, insulator 5, and solid-sealed pole. Import the geometric model into finite element simulation software to obtain a simulation analysis model, and set the material properties of each component.

[0059] S2. Add current physical fields, solid and fluid heat transfer physical fields, turbulence physical fields, and air moisture transport physical fields to the finite element simulation software, and set the computational domain and boundary conditions for each physical field.

[0060] S3. Couple the physical fields and use the free tetrahedral meshing method to mesh the simulation analysis model;

[0061] S4. Add transient study, use transient solver to numerically solve the simulation analysis model, post-process the solution results, and obtain the temperature and humidity distribution inside the switch cabinet.

[0062] S5. Conduct a temperature rise test, measure the actual temperature and humidity of the switchgear during operation, compare the experimental data with the simulation results, and optimize the simulation analysis model.

[0063] S6. In the optimized simulation analysis model, input the operating current and environmental conditions, extract the relative humidity of current transformer 2 and the inner wall of the cable room, and construct the simulation dataset.

[0064] S7. Use the SMOTE strategy to augment the simulation dataset and construct a loss function that includes data loss and physical loss to train PINN.

[0065] S8. Using the ambient temperature and humidity and operating current collected during the actual operation of the switchgear as input, and using the trained PINN for prediction, the steady-state relative humidity at current transformer 2 is obtained.

[0066] In S1, based on the actual structural parameters of the switchgear, a high-precision three-dimensional geometric model of the switchgear's original dimensions is constructed using SolidWorks 3D drawing software. The solid-sealed pole is formed by encapsulating the vacuum interrupter chamber of circuit breaker 3 and the moving / stationary contacts into a single unit using an insulating shell as a carrier. The cable compartment is a functional compartment at the bottom of the switchgear that houses the current transformer 2. Taking the KYN28A switchgear as an example, the geometric model is as follows: Figure 2 As shown.

[0067] The geometric model includes at least the following components: current-carrying busbar 1, current transformer 2, circuit breaker 3, switch cabinet housing 4, insulator 5, and solid-sealed poles, in order to realistically reflect the internal structural features and heat and moisture transfer path of the cabinet.

[0068] The geometric model was imported into the COMSOL finite element simulation software. The current-carrying busbar 1 was designated as copper, the current transformer 2, insulator 5, and solid-sealed pole as epoxy resin, the moving and stationary contacts of circuit breaker 3 as copper-chromium alloy, the vacuum interrupter bushing of circuit breaker 3 as aluminum oxide, and the switchgear enclosure 4 as stainless steel. Taking the KYN28A switchgear as an example, the simulation analysis model is as follows: Figure 3 As shown.

[0069] In S2, within the COMSOL finite element simulation software, add a current physical field, set the current field calculation domain, and set the boundary conditions for the electric field analysis:

[0070] The current-carrying busbar 1 and the moving and stationary contacts of circuit breaker 3 are defined as the current field calculation domain. The current input boundary is set by adding surface terminals, and the current outflow boundary is set by adding ground planes. The current magnitude is set to I0, which is the effective current value in actual operation of the switchgear. Contact impedance is added to the contact surfaces and bolt fastening surfaces to simulate localized heating caused by contact resistance. The bolt fastening surface is the contact surface between two conductive components clamped together by bolts (e.g., the junction of the main busbar and branch busbar of current-carrying busbar 1, the connection surface between current-carrying busbar 1 and the stationary contact of circuit breaker 3, the connection surface between the primary winding of current transformer 2 and current-carrying busbar 1, etc.). The contact resistance R is calculated using the following formula:

[0071] ;

[0072] In the formula, K is a coefficient related to the contact material; m is a coefficient related to the contact form of the two contact surfaces, which includes surface contact, line contact, and point contact. For surface contact, m is 1; for line contact, m is 1.5~2; and for point contact, m is 2~3. Most bolted conductive connections in switchgear (such as current-carrying busbar 1, current transformer 2, and circuit breaker 3 stationary contacts) are surface contacts; F is the contact pressure, in N.

[0073] Add solid and fluid heat transfer physics fields, set the temperature field computational domain, and set the boundary conditions for temperature analysis:

[0074] All areas except the vacuum medium inside the vacuum interrupter of circuit breaker 3 are set as the temperature field calculation domain. The solid domain includes the current-carrying bus 1, the circuit breaker 3 body, the current transformer 2 and the switch cabinet shell 4, and the fluid domain is the air inside the switch cabinet. The heat exchange with the outside air is simulated by applying convective heat flux boundary conditions to the outer surface of the air chamber shell. The heat transfer coefficient is set according to the test conditions. The air chamber is the functional compartment filled with air inside the switch cabinet, and the air chamber shell is the metal wall panel of the functional compartment where each component is located.

[0075] Add based on The turbulent physical field of the model (two-equation turbulent flow model) is defined, the flow field computational domain is set, and the boundary conditions for fluid flow analysis are set:

[0076] The air domain inside the chamber is set as the flow field computation domain, the fluid physics model is set as compressible flow (Mach number Ma < 0.3), gravity terms are included and reduced pressure is used, the wall boundary adopts no-slip condition, and pressure point constraints are set at locations far from the main flow region to ensure the stability of numerical solution.

[0077] Add a physical field for the transport of moisture in the air, set the calculation domain for the humidity field, and set the boundary conditions for the flow of humid air:

[0078] The air domain inside the chamber is set as the calculation domain for moisture transport in the air, i.e., the humidity field calculation domain, and the humid air diffusion coefficient is set to 2.6 × 10⁻⁶. -5 m 2 / s, relative humidity boundary is indicated by the symbol This indicates that the transport process of moist air is described by humidity boundary conditions that match the environmental conditions. In the COMSOL simulation of the physics field for moisture transport in air, the symbol is used... To standardize the value representing the relative humidity boundary, making it easier to set and calculate the transport process of humid air.

[0079] In S3, the coupling of the various physical fields is specifically as follows:

[0080] The physical field of electric current is coupled with the physical field of heat transfer in solids and fluids to form an electromagnetic thermal field; the physical field of heat transfer in solids and fluids is coupled with the physical field of turbulence to form a non-isothermal flow field; the physical field of turbulence is coupled with the physical field of moisture transport in air to form a moisture flow field; and the physical field of heat transfer in solids and fluids is coupled with the physical field of moisture transport in air to form a thermal-humid field.

[0081] During mesh generation, a fine mesh is used for the moving and stationary contact areas of circuit breaker 3, while a conventional mesh is used for other current-carrying conductors, including the current-carrying busbar 1, the conductive parts of the current transformer 2, the conductive parts of the solid-sealed pole, and the conductive parts of circuit breaker 3 excluding the moving and stationary contacts. A relatively coarse mesh is used for the switch cabinet shell 4 and the air domain inside the air chamber to balance computational accuracy and solution efficiency.

[0082] In S4, a transient study is added to the COMSOL finite element simulation software, and the output time step is... The transient solver MUMPS (Sparse Matrix Direct Solver) was used to numerically solve the simulation analysis model, with a relative tolerance set to 0.0001. This means that starting from 0 seconds into the simulation, the calculation results are recorded every 0.1 hours until the simulation ends after 8 hours, with a total of 81 sets of transient data output. A direct solver is used to improve the stability of the solution when performing transient field calculations.

[0083] The solution results are post-processed to obtain the humidity field distribution inside the switch cabinet. Several humidity measurement points on the surface of current transformer 2 are selected to extract the steady-state relative humidity of each humidity measurement point. At the same time, the inner wall temperature and relative humidity of the cable compartment are extracted.

[0084] In S5, during the temperature rise test, temperature and humidity sensors are arranged according to the actual operating conditions of the switchgear. Under the premise of not affecting the insulation safety, the relative humidity is measured at the bottom of the current transformer 2. At the same time, considering the influence of temperature and humidity sensors on the switchgear insulation and the structural characteristics of the cabinet, the air hole at the bottom of the cable compartment is selected as the characteristic point for measuring the temperature and humidity of the inner wall, and the temperature and relative humidity data during the operation of the switchgear are collected.

[0085] The temperature and humidity data obtained from the temperature rise test are compared and analyzed with the simulation results. When the error between the experimental data and the simulation results exceeds the preset error threshold (5%), the simulation analysis model is optimized and adjusted by adjusting the material parameters, boundary conditions and mesh density, so that the error between the simulation value and the measured value of the key measuring point is controlled within the preset range (meeting the error threshold), thereby obtaining a calibrated high-precision simulation analysis model.

[0086] In S6, different load currents, ambient temperatures and relative humidity are set in the optimized simulation analysis model. The steady-state temperature of the cable room wall, the steady-state relative humidity of the cable room wall and the steady-state relative humidity of the current transformer 2 surface are extracted for each working condition. A humidity inversion dataset is constructed with load current, ambient temperature and relative humidity as inputs and relative humidity of the cable room wall and current transformer 2 as outputs as the simulation dataset.

[0087] The Physical Information Neural Network (PINN) was selected as the inversion model to retrieve the relative humidity at current transformer 2. The number of neurons in the hidden layers of PINN was set to 128, 128, 64, and 64 respectively. Layer normalization was introduced into each hidden layer, and the weights were initialized using a Xavier normal distribution. The hyperbolic tangent function was used as the activation function, and the model's generalization ability was improved by embedding physical constraints. The Xavier normal distribution was used to initialize the weights by generating normally distributed random weights with a mean of 0 and a variance of 2 / (input dimension + output dimension) based on the input and output dimensions of the neuron connections. Essentially, this was done by matching the variance of signal propagation in each layer to avoid activation function saturation (causing gradient vanishing) due to excessively large weights or signal attenuation (causing gradient explosion) due to excessively small weights in the early stages of training.

[0088] In S7, the SMOTE strategy is used to augment the constructed simulation dataset, generating simulation samples that are more uniformly distributed in the physical parameter space (running current, ambient temperature, ambient relative humidity), and the pre-trained forward PINN is used to perform preliminary screening of the augmented samples.

[0089] In the scenario of temperature and humidity simulation in switchgear, due to the high cost of simulation computation, the initial simulation dataset often suffers from a small number of samples and uneven distribution. Therefore, the SMOTE strategy is adopted for sample augmentation. This strategy finds sample pairs with similar features in the existing simulation samples and generates new synthetic samples by interpolation in the sample feature space, thereby expanding the dataset size and optimizing the sample distribution, providing more sufficient material for the subsequent training of PINN. However, SMOTE is a purely data-driven interpolation method, and the generated augmented samples may violate the physical laws of thermal and humidity coupling in the switchgear. Therefore, it is necessary to use a pre-trained forward PINN for preliminary screening. This pre-trained forward PINN has embedded the physical constraints of thermal and humidity transmission and can accurately invert the temperature and humidity of the inner wall of the cable cabinet based on the operating current and environmental conditions. After substituting the input parameters of the augmented sample into this PINN, if the output inner wall temperature and humidity conform to the physical logic (such as humidity in the range of 0~100%, and the temperature and humidity coupling relationship consistent with the simulation law), the augmented sample is retained; otherwise, it is discarded. In the end, this not only solves the problem of insufficient initial samples but also ensures the physical rationality of the augmented samples, avoiding the subsequent PINN learning erroneous data that does not conform to reality.

[0090] Several groups of enhanced samples were randomly selected and substituted into the simulation analysis model for calculation. The consistency between the simulation results and the enhanced sample labels was compared.

[0091] The validated dataset was divided into training, validation, and test sets in a ratio of 60:20:20.

[0092] In S8, a system containing data loss is constructed. and physical loss Loss function M:

[0093] ;

[0094] In the formula, , They are respectively , The weights; It is the error between the predicted result and the validated simulation dataset. The error is that the prediction results violate the physical laws within the switchgear. PINN needs to embed physical constraints such as thermal-humidity coupling and water vapor transmission (e.g., relative humidity cannot exceed 100%, temperature and humidity distribution must conform to the heat and mass transfer equation, mass conservation of humid air, etc.). If the model prediction results do not conform to these laws (e.g., humidity > 100%, temperature and humidity relationship violates heat exchange logic), physical losses will occur. The core is to ensure that the model prediction conforms to the actual physical logic.

[0095] An intermediate physical modeling layer is introduced, dividing the PINN modeling into two stages: The first stage constructs a forward PINN based on ambient temperature and humidity and load current, inverting the steady-state temperature and humidity field of the switchgear inner wall; the second stage embeds the error between the inverted values ​​and measured values ​​of the switchgear inner wall state as a physical consistency constraint term into the loss function, optimizing M to obtain the optimized loss function. , is represented as:

[0096] ;

[0097] In the formula, The optimized loss of the preceding data; The optimized subsequent data loss; , , They are respectively , , Weights; utilizing the enhanced simulation dataset and Train PINN. Corresponding to the task in the first stage, the deviation between the temperature and humidity of the switch cabinet's inner wall retrieved by forward PINN and the temperature and humidity label values ​​of the inner wall in the simulation dataset is measured. Corresponding to the second phase of the task, the deviation between the humidity of current transformer 2 retrieved by PINN and the humidity label value of current transformer 2 in the simulation dataset is measured.

[0098] In satisfying Under the constraints, all generated weight combinations were traversed with a step size of 0.1, resulting in 66 weight combinations. Ablation experiments were performed on each weight combination, and the prediction accuracy and convergence characteristics on the test set were compared. The weight combination with the best overall performance was selected as the final weight combination. The final weight.

[0099] The optimal overall performance is selected by inputting the trained PINN with each weight combination into the ambient temperature, humidity, and load current in the test set to obtain the predicted humidity of current transformer 2. The error (commonly measured by mean square error, mean absolute error, etc.) is calculated between this predicted value and the actual humidity label (calibrated simulation / measured value) in the test set. The smaller the error, the more accurate the PINN's prediction of the target variable (humidity of current transformer 2), which meets the core requirement of switchgear humidity inversion. The weight combination with sufficiently small prediction error in the test set (meeting the accuracy requirements of switchgear humidity monitoring), fast convergence speed, and stable process is selected as the final optimal weight.

[0100] Figure 4 This is an exemplary test set comparison chart of PINN humidity inversion results: the horizontal axis represents the test set number, the vertical axis represents relative humidity, the orange bars represent the actual humidity values, and the green diagonal bars represent the optimized loss function. The blue diagonal bars represent the model's predicted humidity under the baseline loss function (M).

[0101] from Figure 4 The data trends show that across all test set samples, the humidity curve obtained from the optimized loss function consistently approximates the actual humidity value, with a significantly smaller deviation compared to the humidity obtained from the baseline loss function. Particularly in samples with higher humidity, such as test set number 20, the optimized predicted value almost perfectly matches the actual value, while the baseline predicted value still shows a certain gap. This result directly demonstrates the advantages of the optimized loss function (which decomposes and stages data loss and embeds physical constraints). The optimized loss function significantly improves PINN's accuracy in retrieving relative humidity from key parts of the switchgear, making the prediction results more closely reflect the actual conditions and providing more reliable numerical support for subsequent condensation risk warnings and operational strategy optimization.

[0102] Example 2: Based on Example 1, the relative humidity of key parts obtained by inversion is compared with the preset condensation risk threshold (80%~85%RH, where RH represents relative humidity). When the predicted humidity exceeds the condensation risk threshold or the trend is close to the condensation risk threshold, an early warning signal is issued, and the heating and dehumidification devices can be linked to control to optimize the anti-condensation operation strategy of the switch cabinet.

[0103] Furthermore, the simulation results, experimental data, operational data, and prediction results are stored, queried, and visualized, providing operation and maintenance personnel with humidity status trend analysis and decision support.

[0104] Example 3: A simulation-based humidity inversion calculation device for medium and high voltage switchgear, comprising:

[0105] One or more processors;

[0106] Memory, used to store one or more computer programs;

[0107] When one or more programs are executed by one or more processors, the one or more processors perform the method in Embodiment 1 or Embodiment 2.

[0108] Example 4: A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method in Example 1 or Example 2.

Claims

1. A method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis, characterized in that... Includes the following steps: S1. For medium and high voltage switchgear, use 3D drawing software to establish a geometric model of the switchgear, including current-carrying busbars, current transformers, circuit breakers, switchgear enclosure, insulators and solid-sealed poles. Import the geometric model into finite element simulation software to obtain a simulation analysis model, and set the material properties of each component. S2. Add current physical fields, solid and fluid heat transfer physical fields, turbulence physical fields, and air moisture transport physical fields to the finite element simulation software, and set the computational domain and boundary conditions for each physical field. S3. Couple the physical fields and use the free tetrahedral meshing method to mesh the simulation analysis model; S4. Add transient study, use transient solver to numerically solve the simulation analysis model, post-process the solution results, and obtain the temperature and humidity distribution inside the switch cabinet. S5. Conduct a temperature rise test, measure the actual temperature and humidity of the switchgear during operation, compare the experimental data with the simulation results, and optimize the simulation analysis model. S6. In the optimized simulation analysis model, input the operating current and environmental conditions, extract the relative humidity of the current transformer and the inner wall of the cable room, and construct the simulation dataset. S7. Use the SMOTE strategy to augment the simulation dataset and construct a loss function that includes data loss and physical loss to train PINN. S8. Using the ambient temperature and humidity and operating current collected during the actual operation of the switchgear as input, and using the trained PINN for prediction, the steady-state relative humidity at the current transformer is obtained.

2. The method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis according to claim 1, characterized in that, In S1, based on the actual structural parameters of the switchgear, a geometric model of the original dimensions of the switchgear is constructed using SolidWorks 3D drawing software. The solid-sealed pole is obtained by encapsulating the circuit breaker vacuum interrupter and the moving / static contacts into one unit using an insulating shell as a carrier. The cable compartment is a functional compartment at the bottom of the switchgear that accommodates the current transformer. Import the geometric model into the COMSOL finite element simulation software, specify copper material for the current-carrying busbar, epoxy resin material for the current transformer, insulator and solid-sealed pole, copper-chromium alloy material for the circuit breaker moving and stationary contacts, aluminum oxide material for the circuit breaker vacuum interrupter bushing, and stainless steel material for the switch cabinet enclosure.

3. The method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis according to claim 1, characterized in that, In S2, in the COMSOL finite element simulation software, a current physical field is added, the current field calculation domain is set, and the boundary conditions for electric field analysis are set: The current-carrying busbar and the moving and stationary contacts of the circuit breaker are defined as the current field calculation domain. The current input boundary is set by adding surface terminals, and the current outflow boundary is set by adding ground planes. The current magnitude is set to I0, which is the effective current value in actual operation of the switchgear. Contact impedance is added to the contact surface and bolt fastening surface to simulate localized heating caused by contact resistance. The bolt fastening surface is the contact surface between two conductive parts clamped together by bolts. The contact resistance R is calculated using the following formula: ； In the formula, K is a coefficient related to the contact material; m is a coefficient related to the contact form of the two contact surfaces, which includes surface contact, line contact, and point contact; F is the contact pressure. Add solid and fluid heat transfer physics fields, set the temperature field computational domain, and set the boundary conditions for temperature analysis: All regions except the vacuum medium inside the vacuum interrupter of the circuit breaker are set as the temperature field calculation domain. The solid domain includes the current-carrying busbar, the circuit breaker body, the current transformer and the switch cabinet shell, and the fluid domain is the air inside the switch cabinet. The heat exchange with the outside air is simulated by applying convective heat flux boundary conditions to the outer surface of the air chamber shell. The heat transfer coefficient is set according to the test conditions. The air chamber is the functional compartment filled with air inside the switch cabinet, and the air chamber shell is the metal wall panel of the functional compartment where each component is located. Add based on The turbulent physical field of the model is defined, the computational domain of the flow field is set, and the boundary conditions for fluid flow analysis are set: The air domain inside the chamber is set as the flow field computation domain, the fluid physics model is set as compressible flow, including gravity terms and using reduced pressure form, the wall boundary adopts no-slip condition, and pressure point constraints are set. Add a physical field for the transport of moisture in the air, set the calculation domain for the humidity field, and set the boundary conditions for the flow of humid air: The air domain inside the chamber is set as the calculation domain for moisture transport in the air, i.e., the humidity field calculation domain. A humid air diffusion coefficient is set, and the relative humidity boundary uses a symbol... This indicates that the transport process of moist air is described by humidity boundary conditions that match the environmental conditions.

4. The method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis according to claim 1, characterized in that, In S3, the coupling of the various physical fields specifically involves: The physical field of electric current is coupled with the physical field of heat transfer in solids and fluids to form an electromagnetic thermal field; the physical field of heat transfer in solids and fluids is coupled with the physical field of turbulence to form a non-isothermal flow field; the physical field of turbulence is coupled with the physical field of moisture transport in air to form a moisture flow field; and the physical field of heat transfer in solids and fluids is coupled with the physical field of moisture transport in air to form a thermal-humid field. When meshing, a refined mesh is used for the moving and stationary contact areas of the circuit breaker, while a conventional mesh is used for other current-carrying conductors. Other current-carrying conductors include the current-carrying busbar, the conductive parts of the current transformer, the conductive parts of the solid-sealed pole, and the conductive parts of the circuit breaker other than the moving and stationary contacts.

5. The method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis according to claim 1, characterized in that, In S4, a transient study is added to the COMSOL finite element simulation software, and the output time step is... The transient solver MUMPS was used to numerically solve the simulation analysis model, with a relative tolerance set to 0.0001. The solution results are post-processed to obtain the humidity field distribution inside the switch cabinet. Several humidity measurement points on the surface of the current transformer are selected to extract the steady-state relative humidity of each humidity measurement point. At the same time, the inner wall temperature and relative humidity of the cable compartment are extracted.

6. The method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis according to claim 1, characterized in that, In S5, during the temperature rise test, temperature and humidity sensors are arranged according to the actual operating conditions of the switchgear, and the relative humidity is measured at the bottom of the current transformer; at the same time, the air hole at the bottom of the cable compartment is selected as the characteristic point for measuring the temperature and humidity of the inner wall, and the temperature and relative humidity data during the operation of the switchgear are collected. The temperature and humidity data obtained from the temperature rise test are compared and analyzed with the simulation results. When the error between the experimental data and the simulation results exceeds the preset error threshold, the simulation analysis model is optimized and adjusted by adjusting the material parameters, boundary conditions and mesh density.

7. The method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis according to claim 1, characterized in that, In S6, different load currents, ambient temperatures, and ambient relative humidity are set in the optimized simulation analysis model. The steady-state temperature of the cable room wall, the steady-state relative humidity of the cable room wall, and the steady-state relative humidity of the current transformer surface are extracted for each working condition. A humidity inversion dataset is constructed with load current, ambient temperature, and ambient relative humidity as inputs and the relative humidity of the cable room wall and the current transformer as outputs as the simulation dataset. The Physical Information Neural Network (PINN) was selected as the inversion model to invert the relative humidity at the current transformer. The number of neurons in the hidden layer of PINN was set to 128, 128, 64 and 64 respectively. Each hidden layer was introduced by layer normalization and the weights were initialized using the Xavier normal distribution. The activation function was the hyperbolic tangent function.

8. The method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis according to claim 1, characterized in that, In S7, the SMOTE strategy is used to augment the constructed simulation dataset, and the pre-trained forward PINN is used to perform preliminary screening of the augmented samples. Several groups of enhanced samples were randomly selected and substituted into the simulation analysis model for calculation. The consistency between the simulation results and the enhanced sample labels was compared. The validated dataset was divided into training, validation, and test sets in a ratio of 60:20:

20.

9. The method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis according to claim 8, characterized in that, In S8, the construction includes data loss. and physical loss Loss function M: ； In the formula, , They are respectively , The weights; It is the error between the predicted result and the validated simulation dataset. The error is that the prediction result violates the physical laws inside the switchgear; An intermediate physical modeling layer is introduced, dividing the PINN modeling into two stages: The first stage constructs a forward PINN based on ambient temperature and humidity and load current, inverting the steady-state temperature and humidity field of the switchgear inner wall; the second stage embeds the error between the inverted values ​​and measured values ​​of the switchgear inner wall state as a physical consistency constraint term into the loss function, optimizing M to obtain the optimized loss function. , is represented as: ； In the formula, The optimized loss of the preceding data; The optimized subsequent data loss; , , They are respectively , , Weights; utilizing the enhanced simulation dataset and Train PINN.

10. The method for calculating the internal humidity of medium and high voltage switchgear based on simulation analysis according to claim 9, characterized in that, In satisfying Under the constraints, all generated weight combinations are traversed with a step size of 0.

1. Ablation experiments are performed on each weight combination, and the prediction accuracy and convergence characteristics on the test set are compared. The weight combination with the best overall performance is selected as the optimal weight combination. The final weight.

Images