Thermal analysis modeling training method for hard connection contact point on wire outlet side of disconnecting link of high-voltage switch

Abstract

The invention discloses a modeling and training method for thermal analysis of a hard connection contact point on the wire outlet side of a disconnecting link of a high-voltage switch, and the method comprises the steps: carrying out the training of a neural network for two times: obtaining simulation data through a thermal shape simulation model for the first time; and training the hard connection contact point at the wire outlet side of the high-voltage switch knife switch to a thermal performance curved surface similar to the thermal analysis structure of the heating curved surface model by using a neural network, and performing thermal analysis identification on the hard connection contact point at the wire outlet side of the high-voltage switch knife switch by using training data obtained on site in the second neural network training. And finally, according to the actual thermal performance curved surface, temperature rise under different currents is predicted, and training is completed. According to the training method, an idealized thermal model is obtained through simulation analysis and serves as an initial training model performance curved surface of AI artificial intelligence training, an initial thermal model is provided for thermal analysis AI training of hard connection contact points, the training speed can be greatly increased, and sample requirements of a training database are reduced.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology for disconnectors, specifically to a method for thermal analysis modeling and training of hard connection contact points on the outgoing side of high-voltage switch disconnectors. Background Technology

[0002] Existing technologies mainly focus on diagnosing the thermal parameters of the disconnect switch body, such as patent CN114819194A (a method for predicting thermal defects in disconnect switches). This method involves acquiring historical data for each disconnect switch, preprocessing the historical data to obtain a training dataset, using a number as an index for the corresponding disconnect switch, establishing various base learners based on different algorithms, training each base learner using the training dataset, obtaining trained base learners, acquiring the data to be predicted for each disconnect switch, preprocessing the data to be predicted to obtain a dataset to be predicted, using each base learner to predict the dataset to be predicted, obtaining classification discriminant values, performing probability transformations on the classification discriminant values ​​obtained by each base learner to obtain probability values, calculating the average of each probability value to obtain a predicted probability value; the data to be predicted includes parameter A of the disconnect switch; the predicted probability value characterizes the probability of a thermal defect occurring in the disconnect switch. This type of thermal analysis training method is generally effective, but: 1. There is no method to consider the heat analysis of the hard connection contact points on the outgoing side.

[0003] 2. The training data requirement is very large. Starting from a learner without a basic model, training data usually requires a large amount of historical data to train a thermal analysis model that is close to reality. Moreover, the training time is long, and model distortion or large errors may occur with sparse and small sample training data. Summary of the Invention

[0004] The present invention aims to provide a thermal analysis modeling and training method for hard-connection contact points on the outgoing side of high-voltage switch disconnectors, specifically for hard-connection contact points connected to the outgoing line of high-voltage switch disconnectors. An idealized thermal model is obtained through simulation analysis, serving as the initial training model performance surface for AI artificial intelligence training. This provides a starting thermal model for AI training of hard-connection contact points, significantly accelerating training speed and reducing the sample requirements of the training database. It is particularly suitable for thermal analysis of hard-connection contact points on the outgoing line of high-voltage switch disconnectors under sparse sample conditions.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for training thermal analysis modeling of hard-connection contact points on the outgoing side of a high-voltage switch disconnector, the training method comprising the following steps: The high-voltage switch disconnectors on site are classified into frames, and then three-dimensional models are built for the classified high-voltage switch disconnectors. The heating surface model of the hard connection contact point on the outgoing side of the high-voltage switch disconnectors under different classifications is simulated by finite element analysis, which serves as the initial thermal shape simulation model provided by the neural network training strategy. The neural network training is performed in two stages: For the first time, simulation data was obtained using a thermal shape simulation model. A neural network was used to train the hard connection contact points on the outgoing side of the high-voltage switch knife switch to a thermal performance surface similar to the thermal analysis structure of the thermal surface model. The second neural network training uses training data obtained on-site to perform thermal analysis and identification of hard connection contact points on the outgoing side of the high-voltage switch knife switch, and obtains the actual thermal performance surface. Finally, the temperature rise under different currents is predicted based on the actual thermal performance curve, thus completing the training.

[0006] In a preferred embodiment, simulation data is obtained for the first time using a thermal shape simulation model. A neural network is then used to train the hard-connection contact points on the outgoing side of the high-voltage switch to a thermal performance surface similar to the thermal analysis structure of the heating surface model. This includes the following steps: After each set of parameters is input into the thermal shape simulation model, the temperature field distribution data of the hard connection contact point and its surrounding area under the corresponding working condition are calculated through the finite element steady-state or transient thermal solution process. Each set of input parameters is associated with its corresponding temperature response to form an input-output data pair set, thus obtaining the simulation dataset. Based on the obtained simulation dataset, a neural network model for thermal performance prediction is constructed. Through a data-driven approach, the nonlinear mapping relationship between the input parameter space and the output thermal response is learned, and the hard connection contact points on the outgoing side of the high-voltage switch disconnector are trained to a thermal performance surface similar to the thermal analysis structure of the heating surface model.

[0007] In a preferred embodiment, the thermal performance surface is a function mapping, with the structural environment and operating conditions of the hard connection contact point as the independent variable and the temperature response of the hard connection contact point as the dependent variable. The mapping relationship from the input parameter space to the output temperature space reflects the coupling law between the structural thermal characteristics and external excitation.

[0008] In a preferred embodiment, the training objective of the neural network model is to continuously adjust the weight parameters within the network using samples from the simulation dataset, so as to minimize the difference between the network's predicted output and the actual simulation output.

[0009] In a preferred embodiment, the temperature field distribution data includes: the surface temperature or core temperature of the hard-connect contact point; the temperature rise of the hard-connect contact point relative to the ambient temperature; and the spatial distribution characteristics of the temperature field on the hard-connect contact point and its adjacent conductors, contacts, and support structures.

[0010] In a preferred embodiment, the input parameters of the thermal shape simulation model include the current intensity of the conductive loop, ambient temperature, contact thermal resistance or contact pressure, material thermal property parameters, and boundary condition parameters.

[0011] In a preferred embodiment, the second neural network training utilizes training data obtained on-site to perform thermal analysis and identification of the hard-connection contact points on the outgoing side of the high-voltage switch disconnector, obtaining the actual thermal performance surface, including the following steps: The neural network architecture's input layer receives actual operating condition parameters, and the output layer predicts the actual temperature response at the hard-connected contact points. Each set of input parameters from the real-world dataset is input into the current neural network model, and the neural network model outputs the predicted thermal response value. The predicted value is compared with the actual temperature response obtained by monitoring, the difference between the two is calculated, and the weight parameters inside the network are updated through the backpropagation algorithm. Through multiple rounds of iterative optimization, the neural network model learns the thermal response characteristics of the hard connection contact points on the outgoing side of the actual high-voltage switch disconnector under real working conditions, forming an actual thermal performance surface.

[0012] In a preferred embodiment, the data in the real-world dataset includes the actual temperature measurement of the hard-connect contact point, the corresponding operating current value, and the ambient temperature.

[0013] In a preferred embodiment, the actual temperature measurement of the hard-connect contact point is recorded in the form of absolute temperature or temperature rise relative to ambient temperature. The corresponding operating current value is the actual load current carried by the outgoing side of the high-voltage switch disconnector at the time of data acquisition, which is obtained in real time through a current transformer or smart meter. The ambient temperature reflects the thermodynamic state of the external environment where the hard connection contact point is located, and is detected by an ambient temperature sensor installed in the switch cabinet.

[0014] In a preferred embodiment, the neural network model learns the thermal response characteristics of the hard-connected contact points on the outgoing side of an actual high-voltage switchgear under real operating conditions, including: Under the same current, if the ambient temperature is higher than the simulation set value, the actual temperature rise will be higher than the predicted value. If the contact surface is oxidized or loose, the temperature at the hard connection contact point will rise abnormally. After long-term operation, due to material aging or surface contamination, the temperature rise under the same operating conditions shows an upward trend.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This application uses a standard high-voltage switch disconnector thermal simulation surface model as the initial condition to approximate the actual thermal performance surface of different high-voltage switch disconnectors in the field, shortening the number of training iterations from the initial surface to the actual model. This method combines simulation analysis with neural network training, which can save a lot of actual data obtained in the field as the amount of data for the second training. It can converge to the true thermal performance surface with very little real field data, and is especially suitable for neural network training with sparse small sample data.

[0016] 2. This application achieves digital twins by combining modeling simulation and AI analysis. Before the neural network training system identifies the circuit breaker, it first classifies and models the high-voltage switchgear according to the current frame level. A family of hard-connection contact point models is established on the outgoing side of the high-voltage switchgear.

[0017] 3. This application uses finite element analysis to classify the hard connection contact point model families on the outgoing side of each high-voltage switch disconnector, and simulates the heating surface model of the hard connection contact point on the outgoing side of the high-voltage switch disconnector under different classifications.

[0018] 4. This application employs a neural network trained in two stages. It selects a family of hard-connection contact point models on the outgoing side of a high-voltage switchgear similar to those used in the field for classification. The first stage uses a large amount of simulation data obtained from the simulated heating surface model for training, approximating the simulated heating surface model. The second stage uses measured heating values ​​from the hard-connection contact points on the outgoing side of actual high-voltage switchgear for training, and the system identifies and adjusts to approximate the real heating surface model.

[0019] 5. The training models of the two neural networks in this application can be the same, or different iteration methods, different time lengths or different hidden layers can be adopted according to the actual needs of the number of nodes and the number of hidden layer neurons.

[0020] 6. By utilizing the results of the second training, this application can predict the temperature value of the hard connection contact point on the outgoing side of the high-voltage switch knife switch under different current, ambient temperature and humidity, and wind speed conditions. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0022] Figure 1 This is a flowchart of the training method of the present invention; Figure 2 This is a flowchart illustrating the classification and modeling process of this invention based on thermally conductive materials and high-voltage switch heat dissipation structures. Figure 3 This is a schematic diagram of the current-temperature rise curve at 0 wind speed in this invention; Figure 4 This is a schematic diagram of the current-temperature rise curve at 0.5 microohms in this invention; Figure 5 This is one of the flowcharts for training the neural network model in this invention; Figure 6 This is the second flowchart of the neural network model training process in this invention; Figure 7 This is a graph showing the initial performance curves of temperature rise prediction at different wind speeds in this invention. Figure 8 This is a graph showing the initial performance of predicted temperature rise under different internal resistance wind speeds in this invention. Figure 9 This is a diagram of the thermal analysis surface model in this invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Traditional methods for training neural network AI to perform thermal analysis on other electrical equipment and components involve designing a neural network, acquiring a large amount of field-based measurement data, starting with an arbitrary initial performance planar model, training the neural network model, and then using the trained system to identify a surface model of the electrical equipment's thermal performance system. Based on this surface model, hard node temperatures are predicted by selecting points. This traditional training method requires a large amount of field-based measurement data because the initial training model is uncertain. If the field-based measurement data is insufficient, it may be impossible to train a surface model of the thermal performance system that matches reality.

[0025] This application mainly focuses on the heat analysis, modeling, and classification design of hard connection contact points connected to the outgoing line side of high-voltage switch disconnectors. These hard connection contact points are key areas for power transmission and distribution operation and maintenance inspection and fault monitoring. With current theoretical research and technology, there are many methods for analyzing the heat of contacts in the hard body of high-voltage switch disconnectors. However, there is no relevant discussion on how to model the heat of the screw connection hard connection points on the outgoing line side that are connected to and affected by the high-voltage switch disconnector, or how to predict the heat generation at the screw connection point.

[0026] The modeling technology for heating at connection contact points has evolved from simplified empirical formulas to high-precision multiphysics simulations. Early methods primarily relied on contact resistance models and steady-state thermal balance equations, neglecting microstructure and dynamic effects, resulting in significant estimation errors. After 2000, finite element analysis was introduced to achieve electro-thermal coupling simulation, but this did not address the microscopic uncertainties of contact resistance. Furthermore, there is currently no strategy for coupling analysis of parameter influences in complex environments; only the structural current ohmic characteristics of the hard-connected contact point itself are considered, leading to significant discrepancies between the results and actual field measurements.

[0027] Therefore, in order to model and evaluate the health status of hard-connection contact points in actual test sites and lay the groundwork for subsequent intelligent analysis, a standardized modeling scheme for the disconnector connection is needed to improve the speed and response of subsequent intelligent analysis of specific hard-connection contact points.

[0028] This application provides a method for training a thermal analysis model of the hard-connection contact points on the outgoing side of a high-voltage switchgear. Please refer to [link / reference needed]. Figure 1 As shown, the training method includes the following steps: High-voltage switchgear on site is classified into frames, and then three-dimensional models are built for each classified high-voltage switchgear. Finite element analysis is used to simulate thermal models, which serve as the initial thermal shape simulation model for neural network training strategies. For hard connection analysis of specific high-voltage switchgear, a suitable typical thermal shape simulation model is selected first. Neural network training is conducted in two stages. The first stage uses the typical thermal shape simulation model to obtain a large amount of simulation data. Neural network training is then used to train the hard connection contact points of specific high-voltage switchgear to a thermal performance surface similar to the thermal analysis structure of the typical three-dimensional high-voltage switchgear thermal shape simulation model. The second stage of neural network training uses the training data obtained on site to identify the thermal analysis system of the hard connection contact points on the outgoing side of the high-voltage switchgear, obtaining the actual thermal performance surface. The temperature rise under different currents is then predicted based on this surface.

[0029] The following is a detailed introduction and explanation of this application: In one disclosed embodiment, the field high-voltage switch disconnectors are classified into frames, and then a three-dimensional model is established for the classified high-voltage switch disconnectors. The simulation thermal model is generated using finite element analysis, which serves as the initial thermal simulation model provided for the neural network training strategy.

[0030] The process of classifying high-voltage disconnectors in the field essentially involves scientifically and rationally categorizing various models, structures, layouts, and electrical parameters of high-voltage disconnectors in actual operating environments based on certain engineering characteristics and structural commonalities. This classification is not arbitrary but based on a series of key structural and functional attributes, such as: the overall mechanical frame structure type of the high-voltage disconnector (e.g., single-column disconnector, double-column disconnector, three-column disconnector), the material and cross-sectional shape of the main conductive components (e.g., copper busbars, aluminum busbars, irregularly shaped connectors), and the typical construction of the outgoing line connection points (e.g., rigid connections). The purpose of this classification is to group high-voltage disconnectors with similar structural features, heat conduction paths, and potential thermal bottleneck locations into one category, enabling the use of a unified modeling strategy and boundary condition settings in subsequent modeling processes, thereby improving the representativeness and generalization ability of the model.

[0031] After completing the framework classification, the next step is to build three-dimensional models of the classified high-voltage switchgear. This step aims to construct a three-dimensional geometric model that accurately reflects the physical structure and material distribution of each defined type of high-voltage switchgear, based on its actual mechanical and electrical structural drawings, field measurement data (such as dimensional parameters, component spacing, connection methods, etc.), and typical design specifications.

[0032] The 3D model is typically built using CAD or CAE pre-processing software, encompassing all key solid components related to heat conduction, convection, and radiation, such as busbars (copper / aluminum), insulating supports, metal frames, connecting bolts, contact areas, busbar clamps, and outgoing conductors. During modeling, close attention must be paid to the contact relationships between components (e.g., tight contact, clearance fit), material properties (e.g., conductivity, specific heat capacity, density, thermal conductivity), and geometric details (e.g., chamfers at connections, transition sections, contact areas), ensuring the model's geometric and physical properties are highly consistent with the actual switchgear. For detailed parameters that cannot be directly obtained (e.g., contact thermal resistance, surface emissivity), reasonable assumptions and values ​​can be made based on experimental data or industry standards of similar switchesgear. Furthermore, after model construction, mesh independence verification of the geometric model is necessary to ensure the convergence and accuracy of subsequent finite element calculations.

[0033] After obtaining a high-quality 3D geometric model, the core step is to use finite element analysis to simulate the thermal model. This step, based on the finite element method, discretizes the 3D model using professional thermal simulation software (such as ANSYS or COMSOL Multiphysics) and applies boundary conditions and thermal loads consistent with actual operating conditions. This allows for the simulation calculation of the temperature field distribution and heat conduction characteristics of the switchgear under a specific current load, ultimately generating a thermal simulation model with thermophysical significance. The processing logic of this process mainly includes: The complete three-dimensional geometric model is divided into a finite number of tiny units (such as tetrahedral and hexahedral units) according to material properties and heat transfer characteristics, forming a finite element mesh. Mesh generation needs to balance computational accuracy and efficiency. High-density meshes are used for critical parts (such as connection nodes, contact areas, and dense wiring areas), while the mesh density is appropriately reduced in non-critical areas to balance computational resources and result accuracy.

[0034] Each finite element is assigned corresponding material thermophysical properties, including thermal conductivity (along anisotropic or isotropic), specific heat capacity, density, etc. These parameters are usually derived from material handbooks, experimental measurement data, or technical parameters provided by suppliers to ensure consistency with the materials used in the actual equipment.

[0035] Based on the actual operating conditions of the high-voltage switchgear, determine the rated current value in the conductive circuit, and base it on Joule heating (I0). 2 The R principle is used to calculate the internal heat generation rate (heat per unit volume per unit time) of each conductive component due to resistive loss, and this heat generation rate is applied as an internal heat source term to the corresponding finite element element. In addition, if there are unsteady-state heat sources such as external electric arcs or partial discharges, appropriate modeling should be performed according to the actual situation.

[0036] Boundary condition settings include convective heat transfer boundaries (such as natural or forced convection between the equipment surface and the surrounding air, usually achieved by defining the convective heat transfer coefficient and the ambient temperature), radiative heat transfer boundaries (such as infrared radiation from a high-temperature surface to the surrounding environment, usually defined based on the Stefan-Boltzmann law, which defines the surface emissivity and the difference in ambient temperature), and adiabatic boundaries (such as certain shielding surfaces inside the equipment that do not exchange heat with the outside). These boundary conditions must be set based on reasonable assumptions and calibrations according to the actual operating environment on site (such as ventilation conditions inside the switchgear, ambient temperature and humidity, equipment installation density, etc.).

[0037] After completing the model construction and parameter settings, the finite element thermal analysis solver is started to calculate the temperature distribution of each hard-connection contact point under steady-state or transient conditions, obtaining complete thermal analysis results including the highest temperature point, temperature gradient distribution, heat flux density distribution, and temperature rise data of key hard-connection contact points. Based on these results, key indicators related to thermal performance (such as temperature rise curves of hard-connection contact points, thermal resistance network parameters, hot spot location distribution, etc.) are further extracted and integrated into a structured thermal simulation model. This model not only includes numerical results of the temperature field but also implicitly contains the physical mapping relationship between equipment structure, materials, heat source distribution, and boundary conditions, which can serve as the benchmark input for the initial thermal simulation model in subsequent neural network training.

[0038] In one disclosed embodiment, simulation data is obtained for the first time using a thermal shape simulation model. The specific high-voltage switch knife switch hard connection contact point is first trained to a thermal performance surface similar to the thermal analysis structure of a typical three-dimensional high-voltage switch knife switch thermal shape simulation model.

[0039] This stage begins with the three-dimensional high-voltage switch thermal shape simulation model (i.e., thermal shape simulation model) built and validated in the preceding steps. Using the controlled variable method or parameter sweep method, key input parameters in the model are systematically changed to drive the simulation program to output a dataset of thermal analysis results covering a typical operating condition range. These input parameters typically include, but are not limited to: Conductive circuit current intensity (I): As the most critical excitation parameter for heat source, its change directly affects the Joule heat generation rate at the hard connection contact point; Ambient temperature (Ta): represents the thermodynamic reference of the external environment where the switch is located, and affects the convection and radiation boundary conditions; Contact thermal resistance or contact pressure (Rc or P): reflects the actual contact state of the conductive contact surface in the hard connection contact point, and significantly affects the local thermal resistance and temperature rise distribution. Material thermophysical parameters (such as thermal conductivity and specific heat capacity): In some sensitivity analyses, they can also be used as variables to assess the impact of material differences on thermal response; Boundary condition parameters (such as convective heat transfer coefficient h and radiative emissivity ε): used to simulate the heat dissipation capacity under different ventilation conditions or surface treatment states.

[0040] In practice, a set of representative input parameter combinations is generated. After each set of parameters is input into the thermal shape simulation model, the temperature field distribution data of the hard connection contact point and its surrounding key areas are calculated through the finite element steady-state or transient thermal solution process. The key simulation outputs extracted include: the surface temperature (Ts) or core temperature (Tn) of the hard connection contact point; the temperature rise of the hard connection contact point relative to the ambient temperature (ΔT=Ts-Ta); and the spatial distribution characteristics of the temperature field on the hard connection contact point and its adjacent conductors, contacts, and support structures (such as temperature gradient and hot spot locations). Finally, each set of input parameters is associated with its corresponding key temperature response (such as ΔT or Ts) to form a high-dimensional set of input-output data pairs. This set is the simulation dataset required for subsequent neural network training. Its essence is a numerical mapping sample of the physical process of structural features + thermal load + boundary conditions → temperature response.

[0041] Based on the obtained simulation dataset, this stage constructs a neural network model specifically for thermal performance prediction. The core objective is to learn the nonlinear mapping relationship between the input parameter space (such as current, ambient temperature, contact state, etc.) and the output thermal response (such as temperature rise or temperature field characteristics of hard-connected contact points) through a data-driven approach, and finally generate a thermal performance surface that is functionally similar to the thermal analysis structure of the hard-connected contact point on the outgoing line side of a typical three-dimensional high-voltage switch knife switch.

[0042] The thermal performance surface is a high-dimensional, implicit function mapping. This function takes the structural environment and operating conditions of the hard-connected contact point as the independent variable and the temperature response (such as temperature rise or absolute temperature) of the hard-connected contact point as the dependent variable. Its mathematical essence can be understood as a mapping relationship f:X→Y from the input parameter space X (such as X=[I,Ta,Rc) to the output temperature space Y (such as Y=ΔT or Ts). This mapping reflects the complex coupling law between the structural thermal characteristics and external excitation. The training objective of the neural network is to continuously adjust the weight parameters inside the network using a large number of samples {xi,yi} in the simulation dataset (where xi is the i-th set of input parameters and yi is the corresponding thermal response output) to minimize the difference between the network's predicted output f(xi) and the actual simulation output yi (usually measured by indicators such as mean square error MSE and mean absolute error MAE), thereby approximating this implicit thermal performance mapping function. This step is existing technology and will not be elaborated here.

[0043] During training, the simulation dataset is first divided into training and validation sets in a specific ratio (e.g., 8:2 or 7:3) to ensure that the model can fit the training data while also generalizing well to unseen data. Then, each set of input parameters xi from the training set is fed into the initialized neural network model. After several hidden layers of nonlinear transformation (e.g., ReLU), a predicted heat response value f(xi) is finally output. This predicted value is compared with the actual simulation value yi, and the loss function (e.g., MSE loss: L=1 / NΣ(f(xi)-yi)) is calculated. 2 (where N is the number of input parameter sets).

[0044] As training iterations continue, the neural network gradually learns the typical temperature response patterns of hard-connection contact points under different combinations of input parameters. For example, when the current increases, the temperature rise of the hard-connection contact point increases rapidly and non-linearly; when the ambient temperature rises, the temperature rise increment under the same current is more significant; when the contact thermal resistance increases (such as due to oxidation or loosening causing an increase in Rc), the temperature of the hard-connection contact point is significantly higher than that under normal contact conditions. These implicit physical laws are internalized by the neural network in a parameterized manner into its weight matrix, ultimately forming a virtual thermal performance surface that can quickly predict the temperature response of the hard-connection contact point under any given input operating conditions.

[0045] In one disclosed embodiment, the second neural network training utilizes training data obtained on-site to perform thermal analysis and identification of the hard connection contact points on the outgoing side of the high-voltage switch knife switch, thereby obtaining the actual thermal performance surface.

[0046] The starting point of this training phase is the hard-connection contact point on the outgoing side of a high-voltage switchgear in actual operation. This is achieved by deploying an online monitoring system (such as an infrared thermal imager, fiber optic temperature sensor, or contact temperature probe) to collect key data reflecting the thermal state of the hard-connection contact point in real time or periodically. This data typically includes: The actual temperature measurement (Tm) of the hard-connect contact point is usually recorded in the form of absolute temperature (such as °C or K) or temperature rise relative to ambient temperature (ΔT=Tm-Ta), and is the direct target output of model training; The corresponding operating current value (Im) is the actual load current carried by the outgoing side of the high-voltage switch disconnector at the time of data acquisition. It is the main heat source excitation that causes the hard connection contact point to heat up. It is usually obtained in real time through a current transformer (CT) or smart meter. Ambient temperature (Ta) reflects the thermodynamic state of the external environment at the hard connection contact point. It is measured by an ambient temperature sensor installed near the switch cabinet and is an important boundary condition affecting convective heat dissipation and final temperature rise.

[0047] Because field-collected data often suffers from noise interference (such as sensor drift, electromagnetic interference, instantaneous load fluctuations, etc.), inconsistent sampling frequencies, or missing parameters, the raw data must be systematically preprocessed before being input into neural network training. This includes data cleaning, data alignment and synchronization, and feature engineering and normalization: standardizing input parameters (such as Im, Ta) (e.g., Min-Max normalization) to eliminate the negative impact of dimensional differences on neural network training and improve model convergence speed and generalization ability.

[0048] After the above preprocessing, a set of high-quality input-output data pairs is finally formed, where the input is the actual operating condition parameters (such as Im, Ta, etc.) and the output is the actual temperature response of the hard-connected contact point at the corresponding time (such as Tm or ΔT). This set is the real-world dataset required for the second neural network training.

[0049] This stage employs the same neural network architecture as the first training stage. Its input layer receives actual operating parameters (such as current and ambient temperature), and its output layer predicts the actual temperature response (such as temperature rise or absolute temperature) of the hard-connection contact points. By introducing real-world data, the parameters of the initially trained model (i.e., the thermal performance prediction model based on simulation data obtained after the first training) are fine-tuned to drive the model to learn and capture detailed features in the actual physical system that are not fully represented by the simulation model, such as the actual non-uniform distribution of contact resistance, the influence of the oxide layer on the conductor surface, local differences in heat dissipation conditions, or thermal characteristic degradation caused by equipment aging. The specific processing logic is as follows: Each set of input parameters xi (such as actual current Im and ambient temperature Ta) from the real-world dataset is input into the current neural network model. The model outputs a predicted thermal response value f(xi). This predicted value is compared with the actual monitored temperature response yi (such as Tm or ΔT), and the difference between the two (mean squared error MSE) is calculated. The weight parameters within the network are then updated using the backpropagation algorithm, allowing the model prediction to gradually approach the actual observation results. Unlike the first training, the training data in this stage is more closely related to the real operating environment and may include more extreme conditions (such as short-term overload, high-temperature seasons, and initial performance of poor contact). Therefore, it can effectively expose and correct prediction biases in the simulation model caused by idealized assumptions (such as uniform contact, constant material parameters, and ideal heat dissipation conditions).

[0050] Through multiple rounds of iterative optimization, the neural network gradually learns the thermal response characteristics of the hard-connection contact points on the outgoing side of actual high-voltage switch disconnectors under real operating conditions. For example, under the same current, if the ambient temperature is higher than the simulation set value, the actual temperature rise is significantly higher than the predicted value; if there is slight oxidation or loosening of the contact surface (manifested as the actual contact thermal resistance being higher than the design value), the temperature of the hard-connection contact point will rise abnormally; or after long-term operation, due to material aging or surface contamination, the temperature rise under the same operating conditions shows a slow upward trend. These unique thermal behavior characteristics in actual operation are internalized by the neural network in a parameterized manner into its weight matrix, ultimately forming the actual thermal performance surface.

[0051] In one disclosed embodiment, training is completed by predicting the temperature rise under different currents based on the actual thermal performance curve.

[0052] To predict temperature rise under different current conditions, it is first necessary to construct a set of input datasets covering the operating range. This dataset is based on the actual operating environment and typically includes the following key parameters: The operating current is the most important thermal excitation input. The selection range should cover the equipment's rated current, common overload conditions (such as 1.1 times or 1.2 times the rated current), and possible extreme conditions (such as the thermal response under short-term overload or fault current, depending on the availability of data). The ambient temperature is selected based on the climate characteristics of the actual installation location of the equipment, with typical values ​​(such as 25℃) chosen, or seasonal changes or extreme environmental conditions can be considered.

[0053] Based on the aforementioned parameter range, a set of multi-dimensional input vectors is generated through structured sampling (such as equal-interval sampling, boundary value sampling, or engineering experience-guided combinations of operating conditions), where each input vector represents a specific combination of operating conditions. The input data, as the independent variables of the actual thermal performance surface, is fed into the trained neural network model. Through the model's forward propagation process—that is, sequentially passing through the input layer, hidden layer, and output layer—the predicted temperature response value corresponding to each input operating condition is finally output.

[0054] The following detailed description of the solution provided in this application, in conjunction with specific implementation methods, illustrates the following: Example The specific application process of the method of this invention in the thermal analysis of hard-connection contact points on the outgoing side of high-voltage switchgear below 4000A is as follows: Step 1: Select the type of disconnecting switch: According to GB / T1985-2023 3.4 Classification of Switchgear, disconnecting switches are divided into single-pole, double-pole, and three-pole types, with single-pole and double-pole being the most common. Refer to GB / T11022-2020 5.5 Rated Continuous Current, GB / 762-2002 Current Rating, and manufacturer model. Confirm that the rated current ratings of the disconnecting switches to be simulated are 2000A, 2500A, 3150A, and 4000A.

[0055] Step Two, as follows Figure 2 As shown: Classification and modeling based on thermally conductive materials and high-voltage switch heat dissipation structures: The research on AI adaptive health status of hard-connection contact points of disconnectors mainly needs to consider three key aspects in modeling: Contact resistance: a major source of heat generation and is affected by a variety of factors.

[0056] Load current: The main factor causing heat generation, which is related to the user's load and is uncontrollable.

[0057] The connection between the base of the disconnector switch and the hard contact, as well as the wiring method, are the main influencing factors and the primary objects of modeling and simulation.

[0058] In hard-contact heating structures, the final temperature value is affected not only by environmental indices and node heating, but also by the heating generated by current flowing through the conductor and other high-temperature nodes. For double-column disconnect switches, since the disconnect switch contacts are relatively close to the outgoing hard-contact connection, their heating effect must be considered, and modeling and simulation should be performed according to the heat source.

[0059] Based on the preliminary analysis of disconnector drawings, it is found that disconnectors from different manufacturers generally need to meet the national standard current carrying capacity requirements. Therefore, for disconnectors of the same type and grade, the cross-sectional area of ​​the conductive outgoing wire material and the shape of the rigid connection of the base are similar. Thus, the most reasonable classification method is to establish a model classification based on the current rating of the disconnector frame.

[0060] Therefore, the projects are categorized according to the requirements of the research subjects as shown in Table 1: Table 1

[0061] Step 3: Establish a simulation structural model, set relevant thermally conductive materials for simulation, and perform finite element analysis of heat generation based on the established model; Current input: The current I is used as the variable, and the unit is A, which facilitates the simulation of multiple current values ​​in the future; Wind speed settings: V_wind is used as the variable, with the unit being m / s, to facilitate subsequent simulations of multiple wind speeds. The wind direction is the Z direction, i.e., the lateral direction. Contact resistance setting: The variable is P, and the unit is W. Calculations facilitate subsequent simulations of multiple contact resistances; Set up contact surface temperature detection to detect the maximum temperature value; Substitute the specific values ​​into the three parameters to form a calculation matrix, calculate each combination, and obtain the heating temperature; This allows us to obtain simulation data on temperature rise under different currents, contact resistances, and wind speeds. For example, the simulation data obtained at a wind speed of 0 is shown in Table 2. Table 2:

[0062] Its constituent thermal structure surfaces are as follows Figure 3 As shown.

[0063] By analogy, the simulated temperature rise surface under different wind speeds can be obtained. For example... Figure 4 As shown, after fitting, the wind speed and current temperature rise performance corresponding surface under a certain resistance can be obtained (taking 0.5 microohms as an example).

[0064] Step 4: Design a neural network training method, selecting an AI-based model to fit the correlation between multiple factors and the temperature of the hard connection contact point of the disconnector, thereby achieving dynamic prediction of the temperature trend of the hard connection contact point. For example, in this case, LSTM (Long Short-Term Memory) network can effectively achieve time series analysis and trend prediction.

[0065] Therefore, based on the historical operating current of the substation equipment, the temperature of the hard connection contact point of the disconnector, the ambient temperature, the wind speed and other detection data, the LSTM model is used to predict the temperature of the hard connection contact point of the disconnector under the influence of multiple factors, so as to realize the dynamic prediction of the development trend of the heating of the hard connection contact point of the disconnector.

[0066] The specific process of the algorithm model is as follows: Figure 5 - Figure 6 As shown, it includes: Selection of model input and output variables: Based on the influencing factors of the contact point temperature of the disconnector, ambient temperature, operating current, and wind speed are selected as the input variables of the prediction model, and the output variable is the contact point temperature; a data sample set is constructed based on historical data and Ansys simulation data; the dataset is normalized and divided into training set and test set according to a certain ratio; LSTM network model construction: Input the training data into the LSTM prediction model, and use LSTM to model the data in the training set to determine the number of input nodes and the number of hidden layer neurons of the LSTM network; LSTM network model training optimization and tuning: Using the mean absolute error as the loss function, the parameters of the LSTM network model are continuously trained and optimized with the goal of minimizing the loss function; the LSTM model architecture is adjusted as necessary until good results are achieved on the test set, and then the model parameters are saved. Prediction of contact point heating in knife switch hard connection: The ambient temperature, operating current and wind speed at time t are input as parameters into the trained LSTM network model to obtain the predicted value of the contact point temperature at time t+1.

[0067] Step 5: In the neural network algorithm described in Step 4, using the data obtained from the finite element simulation thermal analysis in Step 3, several training data points are selected on the surface to perform neural network training in Step 4. In this example, 2000 simulation data points are selected. The first neural network training is completed using PyTorch for the GW4D-40.5ⅡDW / 4000 high-voltage switch, and the following data are obtained: wind speed prediction temperature rise initial performance curve and prediction temperature rise initial performance curves under different internal resistance wind speeds are shown below. Figure 7 - Figure 8 As shown.

[0068] Step 6: On-site measurement data of GW4D-40.5ⅡDW / 4000 high-voltage switch are shown in Table 3: Table 3:

[0069] Using these six data points as historical data, the second neural network training described in step four is performed to obtain a new thermal analysis surface model. This model is then compared with the original six measured data points to observe the measurement error. Figure 9 As shown, with sparse model data, the error of the obtained actual thermal analysis training model is controlled within 1.5 degrees. This indicates that this scheme can also train to approximate the actual measurement points even with limited data.

[0070] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0071] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switch disconnector, characterized in that: The training method includes the following steps: The high-voltage switch disconnectors on site are classified into frames, and then three-dimensional models are built for the classified high-voltage switch disconnectors. The heating surface model of the hard connection contact point on the outgoing side of the high-voltage switch disconnectors under different classifications is simulated by finite element analysis, which serves as the initial thermal shape simulation model provided by the neural network training strategy. The neural network training is performed in two stages: For the first time, simulation data was obtained using a thermal shape simulation model. A neural network was used to train the hard connection contact points on the outgoing side of the high-voltage switch knife switch to a thermal performance surface similar to the thermal analysis structure of the thermal surface model. The second neural network training uses training data obtained on-site to perform thermal analysis and identification of hard connection contact points on the outgoing side of the high-voltage switch knife switch, and obtains the actual thermal performance surface. Finally, the temperature rise under different currents is predicted based on the actual thermal performance curve, thus completing the training.

2. The method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switchgear according to claim 1, characterized in that: For the first time, simulation data was obtained using a thermal shape simulation model. A neural network was used to train the hard-connection contact points on the outgoing side of the high-voltage switch knife switch to a thermal performance surface similar to the thermal analysis structure of the thermal surface model. The steps included: After each set of parameters is input into the thermal shape simulation model, the temperature field distribution data of the hard connection contact point and its surrounding area under the corresponding working condition are calculated through the finite element steady-state or transient thermal solution process. Each set of input parameters is associated with its corresponding temperature response to form an input-output data pair set, thus obtaining the simulation dataset. Based on the obtained simulation dataset, a neural network model for thermal performance prediction is constructed. Through a data-driven approach, the nonlinear mapping relationship between the input parameter space and the output thermal response is learned, and the hard connection contact points on the outgoing side of the high-voltage switch disconnector are trained to a thermal performance surface similar to the thermal analysis structure of the heating surface model.

3. The method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switchgear according to claim 2, characterized in that: The thermal performance surface is a function mapping, with the structural environment and operating conditions of the hard connection contact point as the independent variable and the temperature response of the hard connection contact point as the dependent variable. It reflects the coupling law between the structural thermal characteristics and external excitation by mapping the input parameter space to the output temperature space.

4. The method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switchgear according to claim 3, characterized in that: The training objective of the neural network model is to continuously adjust the weight parameters within the network using samples from the simulation dataset, so as to minimize the difference between the network's predicted output and the actual simulation output.

5. The method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switchgear according to claim 3, characterized in that: The temperature field distribution data includes: the surface temperature or core temperature of the hard connection contact point; the temperature rise of the hard connection contact point relative to the ambient temperature; and the spatial distribution characteristics of the temperature field on the hard connection contact point and its adjacent conductors, contacts, and support structures.

6. The method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switchgear according to claim 2, characterized in that: The input parameters of the thermal shape simulation model include the current intensity of the conductive loop, ambient temperature, contact thermal resistance or contact pressure, material thermal property parameters, and boundary condition parameters.

7. The method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switchgear according to claim 1, characterized in that: The second neural network training utilizes training data obtained on-site to perform thermal analysis and identification of the hard-connection contact points on the outgoing side of the high-voltage switch disconnector, obtaining the actual thermal performance surface, including the following steps: The neural network architecture's input layer receives actual operating condition parameters, and the output layer predicts the actual temperature response at the hard-connected contact points. Each set of input parameters from the real-world dataset is input into the current neural network model, and the neural network model outputs the predicted thermal response value. The predicted value is compared with the actual temperature response obtained by monitoring, the difference between the two is calculated, and the weight parameters inside the network are updated through the backpropagation algorithm. Through multiple rounds of iterative optimization, the neural network model learns the thermal response characteristics of the hard connection contact points on the outgoing side of the actual high-voltage switch disconnector under real working conditions, forming an actual thermal performance surface.

8. The method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switchgear according to claim 7, characterized in that: The data in the real-world dataset includes the actual temperature measurement of the hard-connect contact point, the corresponding operating current value, and the ambient temperature.

9. The method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switchgear according to claim 8, characterized in that: The actual temperature measurement value of the hard connection contact point is recorded in the form of absolute temperature or temperature rise relative to ambient temperature. The corresponding operating current value is the actual load current carried by the outgoing side of the high-voltage switch disconnector at the time of data acquisition, which is obtained in real time through a current transformer or smart meter. The ambient temperature reflects the thermodynamic state of the external environment where the hard connection contact point is located, and is detected by an ambient temperature sensor installed in the switch cabinet.

10. The method for thermal analysis modeling and training of hard-connection contact points on the outgoing side of a high-voltage switchgear according to claim 8, characterized in that: The neural network model learned the thermal response characteristics of the hard-connected contact points on the outgoing side of actual high-voltage switch disconnectors under real operating conditions, including: Under the same current, if the ambient temperature is higher than the simulation set value, the actual temperature rise will be higher than the predicted value. If the contact surface is oxidized or loose, the temperature at the hard connection contact point will rise abnormally. After long-term operation, due to material aging or surface contamination, the temperature rise under the same operating conditions shows an upward trend.

Images