A temperature field reconstruction method and device of a power equipment

Abstract

This invention discloses a method and apparatus for reconstructing the temperature field of power equipment. The method includes acquiring basic structural data corresponding to the power equipment and constructing a parametric simulation model; generating multiple parameter combination vectors using a sequential sampling method, and training preset surrogate models based on each parameter combination vector and the parametric simulation model to obtain multiple target reconstruction models; when receiving a real-time operating condition vector of the operating condition to be predicted, using each target reconstruction model to predict the prediction mode coefficient matrix corresponding to the real-time operating condition vector; and reconstructing the temperature field based on the prediction mode coefficient matrix to generate a high-dimensional temperature field vector corresponding to the power equipment, thereby better adapting to the real-time state prediction needs under varying operating conditions while taking into account both computational efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of temperature field reconstruction technology, and more particularly to a method and apparatus for temperature field reconstruction of power equipment. Background Technology

[0002] In power transmission and distribution systems, the cable-GIS terminal is the core interface unit for achieving electrical connection and insulation transition between power cables and gas-insulated switchgear (GIS). The cable-GIS terminal integrates various types of solid insulating media, such as cross-linked polyethylene and epoxy resin, current-carrying conductors, and a sealed SF6 gas-insulated chamber. During operation, the current flowing through the current-carrying conductor generates Joule heat, and key properties such as material conductivity change nonlinearly with temperature, creating a nonlinear coupling effect between the electromagnetic field, temperature field, and flow field. Accurately obtaining the temperature distribution of the terminal is of engineering significance for assessing current-carrying capacity, preventing overheating faults, and guiding operation and maintenance decisions.

[0003] Currently, two main methods are used to reconstruct the temperature field of power equipment. One is the thermal circuit model method, which equates the heat transfer path to a thermal resistance and thermal capacity circuit. The other is the traditional finite element method, which uses a three-dimensional electromagnetic-thermal-fluid multiphysics coupled model for numerical solution.

[0004] However, in the above-mentioned temperature field reconstruction methods, the thermal circuit model has significant errors under complex boundary conditions and nonlinear material properties, making it difficult to dynamically reflect the deterioration of the electrical contact surface inside the terminal caused by factors such as vibration and thermal cycling. Furthermore, the finite element method is mostly designed for specific equipment or fixed operating conditions, making it difficult to adapt to the real-time state prediction requirements under varying operating conditions, and thus failing to balance computational efficiency and accuracy. Summary of the Invention

[0005] This invention provides a method and apparatus for reconstructing the temperature field of power equipment, which solves the technical problems that thermal circuit models have significant errors under complex boundary conditions and nonlinear material properties, making it difficult to dynamically reflect the deterioration of electrical contact surfaces inside the terminal caused by factors such as vibration and thermal cycling. Furthermore, the finite element method is mostly designed for specific equipment or fixed operating conditions, making it difficult to adapt to the real-time state prediction requirements under varying operating conditions and failing to balance computational efficiency and accuracy.

[0006] This invention provides a method for reconstructing the temperature field of power equipment, comprising:

[0007] Obtain the basic structural data corresponding to the power equipment and construct a parametric simulation model;

[0008] Multiple parameter combination vectors are generated using a sequential sampling method, and a preset surrogate model is trained based on each parameter combination vector and the parameterized simulation model to obtain multiple target reconstruction models.

[0009] When a real-time operating condition vector to be predicted is received, the prediction mode coefficient matrix corresponding to the real-time operating condition vector is predicted using each of the target reconstruction models.

[0010] The temperature field is reconstructed based on the predicted mode coefficient matrix to generate a high-dimensional temperature field vector corresponding to the power equipment.

[0011] Optionally, it also includes:

[0012] The parameterized simulation model is driven by the combination vectors of the aforementioned parameters to generate multiple measurable data; the measurable data includes load current, ambient temperature, casing temperature, and contact resistance values;

[0013] The initial inversion model is trained using the measurable data described above, and the target inversion model is obtained.

[0014] When real-time input data is collected, the target inversion model performs real-time inversion based on the real-time input data to determine the contact resistance value corresponding to the real-time input data.

[0015] Optionally, the step of acquiring the basic structural data corresponding to the power equipment and constructing a parametric simulation model includes:

[0016] Obtain the basic structural data corresponding to the power equipment;

[0017] Construct a basic geometric model based on the aforementioned basic structural data;

[0018] Create an electromagnetic-thermal-fluid multiphysics coupled model corresponding to the basic geometric model in the finite element simulation platform;

[0019] By setting the load current, ambient temperature, and shell parameters in the electromagnetic-thermal-fluid multiphysics coupling model as variable parameters, a parametric simulation model is obtained.

[0020] Optionally, the step of creating the electromagnetic-thermal-fluid multiphysics coupling model corresponding to the basic geometric model in the finite element simulation platform includes:

[0021] In the finite element simulation platform, create the electromagnetic field model corresponding to the basic geometric model, solve the electromagnetic field control equation of the electromagnetic field model, obtain the Joule heat loss, and input it into the heat flow field model.

[0022] In the finite element simulation platform, a heat flow field model corresponding to the basic geometric model is created, and the temperature distribution under the Joule heat loss is solved by solving the full heat transfer path control equation of the heat flow field model.

[0023] The temperature distribution is fed back to the electromagnetic field model, and the conductivity of the conductor and contact material is updated. The process then jumps to the step of solving the electromagnetic field control equation of the electromagnetic field model, obtaining the Joule heat loss and inputting it into the heat flow field model, until the fluctuation value of the temperature distribution is within the fluctuation range, thus obtaining the electromagnetic-thermal-fluid multiphysics coupling model.

[0024] Optionally, the step of generating multiple parameter combination vectors using a sequential sampling method, and training preset surrogate models based on each parameter combination vector and the parameterized simulation model to obtain multiple target reconstruction models includes:

[0025] A sequential sampling method is used to generate multiple parameter combination vectors within a pre-defined parameter design space;

[0026] The parameterized simulation model is driven by each of the parameter combination vectors to determine the steady-state temperature field corresponding to each parameter combination vector.

[0027] Temperature values ​​are extracted from the finite element mesh nodes associated with each steady-state temperature field, a high-dimensional temperature field snapshot matrix is ​​constructed and its order is reduced to obtain a low-dimensional modal coefficient matrix.

[0028] Using the parameter combination vectors as input and the low-dimensional modal coefficient matrix as output, preset surrogate models are trained to obtain multiple target reconstruction models.

[0029] Optionally, the step of extracting temperature values ​​from the finite element mesh nodes associated with each of the steady-state temperature fields, constructing a high-dimensional temperature field snapshot matrix, and reducing its order to obtain a low-dimensional modal coefficient matrix includes:

[0030] Temperature values ​​are extracted from the finite element mesh nodes associated with each steady-state temperature field to construct a high-dimensional temperature field snapshot vector;

[0031] A high-dimensional temperature field snapshot matrix is ​​created using all the aforementioned high-dimensional temperature field snapshot vectors;

[0032] Calculate the average temperature field vector of all the aforementioned high-dimensional temperature field snapshot vectors;

[0033] Calculate the vector difference between each of the high-dimensional temperature field snapshot vectors and the average temperature field vector, and construct the fluctuation field matrix using all the vector differences;

[0034] Singular value decomposition is performed on the wave field matrix to obtain multiple eigenorthogonal basis modes and their corresponding singular values;

[0035] The cumulative energy contribution rate is calculated according to each singular value, and the target number of eigenorthogonal principal modes are selected from the multiple eigenorthogonal basis modes based on the cumulative energy contribution rate from high to low.

[0036] A reduced-order projection matrix is ​​constructed using all the aforementioned intrinsic orthogonal principal modes;

[0037] The wave field matrix is ​​projected onto the reduced-order subspace according to the transpose of the reduced-order projection matrix to obtain the low-dimensional modal coefficient matrix.

[0038] Optionally, the step of using each of the parameter combination vectors as input and the low-dimensional modal coefficient matrix as output to train preset surrogate models to obtain multiple target reconstruction models includes:

[0039] Multiple training datasets are constructed by taking the parameter combination vectors as inputs and the low-dimensional modal coefficient matrices corresponding to the parameter combination vectors as outputs.

[0040] A single agent model is trained using the training data described above, and target reconstruction models are obtained respectively.

[0041] The surrogate model is a neural network using a radial basis function kernel; the mapping relationship of the surrogate model is as follows:

[0042] ;

[0043] in, Let be the modal coefficients of the j-th mode. It is the parameter combination vector of the i-th group. Let be the connection weight from the i-th hidden layer node to the output layer in the neural network corresponding to the j-th mode. For radial basis kernel functions, To predict the input vector The Euclidean distance between the i-th parameter combination vector and the i-th parameter combination vector, where M is the total number of training data.

[0044] Optionally, the high-dimensional temperature field vector is:

[0045] ;

[0046] in, To predict the modal coefficient matrix, For the reduced-order projection matrix, The average temperature field vector, This is a high-dimensional temperature field vector.

[0047] Optionally, it also includes:

[0048] The target reconstruction model and the target inversion model are modularly encapsulated to create a data interface;

[0049] Connect the data interface to the power equipment;

[0050] When device operation data is received from the data interface, the target reconstruction model is invoked to reconstruct the temperature field according to the device operation data to obtain a real-time temperature field vector.

[0051] The target inversion model is invoked to perform real-time inversion based on the equipment operating data to determine the contact resistance value corresponding to the equipment operating data.

[0052] The real-time temperature field vector and the contact resistance value are visualized in the user interface.

[0053] The present invention also provides a temperature field reconstruction device for power equipment, comprising:

[0054] The data acquisition and modeling module is used to acquire the basic structural data of the power equipment and build a parametric simulation model.

[0055] The surrogate model training module is used to generate multiple parameter combination vectors using a sequential sampling method, and to train preset surrogate models according to each parameter combination vector and the parameterized simulation model to obtain multiple target reconstruction models.

[0056] The modal coefficient prediction module is used to predict the prediction modal coefficient matrix corresponding to the real-time working condition vector when a real-time working condition vector to be predicted is received, using each of the target reconstruction models respectively.

[0057] The temperature field reconstruction module is used to reconstruct the temperature field based on the predicted mode coefficient matrix and generate a high-dimensional temperature field vector corresponding to the power equipment.

[0058] As can be seen from the above technical solutions, the present invention has the following advantages:

[0059] This invention acquires the basic structural data corresponding to power equipment and constructs a parametric simulation model; it uses a sequential sampling method to generate multiple parameter combination vectors, and trains preset surrogate models based on each parameter combination vector and the parametric simulation model to obtain multiple target reconstruction models; when a real-time operating condition vector to be predicted is received, each target reconstruction model is used to predict the prediction mode coefficient matrix corresponding to the real-time operating condition vector; based on the prediction mode coefficient matrix, the temperature field is reconstructed to generate a high-dimensional temperature field vector corresponding to the power equipment, thus better adapting to the real-time state prediction needs under variable operating conditions while taking into account both computational efficiency and accuracy. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 A flowchart illustrating the steps of a method for reconstructing the temperature field of power equipment according to an embodiment of the present invention;

[0062] Figure 2 This is a structural block diagram of a temperature field reconstruction device for power equipment provided in an embodiment of the present invention. Detailed Implementation

[0063] This invention provides a method and apparatus for reconstructing the temperature field of power equipment, which addresses the technical problem that thermal circuit models have significant errors under complex boundary conditions and nonlinear material properties, making it difficult to dynamically reflect the deterioration of electrical contact surfaces inside the terminal caused by factors such as vibration and thermal cycling. Furthermore, the finite element method is mostly designed for specific equipment or fixed operating conditions, making it difficult to adapt to the real-time state prediction requirements under varying operating conditions and failing to balance computational efficiency and accuracy.

[0064] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0065] Please see Figure 1 , Figure 1 A flowchart illustrating the steps of a method for reconstructing the temperature field of power equipment, as provided in an embodiment of the present invention.

[0066] This invention provides a method for reconstructing the temperature field of power equipment, comprising:

[0067] Step 101: Obtain the basic structural data corresponding to the power equipment and construct a parametric simulation model;

[0068] Basic structural data refers to structured fundamental data characterizing the core geometric topology, material constitutive properties, and physical boundary constraints of the target power equipment. It encompasses the dimensional parameters of the core functional components, the physical property parameters of each insulating / conductive medium, and the boundary condition parameters for rated operation. Specifically, it may include, but is not limited to, the relevant structural or property parameters of the center conductor, electrical contact parts, epoxy resin bushings, silicone rubber stress cones, SF6 gas chambers, and aluminum alloy shells. This power equipment may include, but is not limited to, cable-GIS terminals, power cables, GIS equipment, transformers, and other main power transmission and transformation equipment.

[0069] Parametric simulation models refer to numerical simulation models that set key parameters affecting the temperature field distribution of equipment as variable independent variables, and can directly drive multi-physics coupling solutions through independent variable inputs.

[0070] In this embodiment, by collecting the basic structural data corresponding to the design drawings and type test data of the target power equipment, the core paths of electromagnetic conduction and heat transfer within the equipment are identified. A three-dimensional geometric model consistent with the topology of the physical equipment is constructed in a finite element simulation platform, and a multiphysics coupled solution framework of electromagnetic-thermal-fluid is built simultaneously. The equipment operating parameters, environmental parameters, and key contact state parameters are set as independently adjustable variables to complete the construction of a parametric simulation model. This avoids the redundant operation of repetitive single-condition modeling in traditional simulation models, while ensuring the consistency between the simulation results and the operating characteristics of the physical equipment.

[0071] In one example of this application, step 101 may include the following sub-steps S11-S14:

[0072] S11. Obtain the basic structure data corresponding to the power equipment;

[0073] S12. Construct a basic geometric model based on the basic structural data;

[0074] In this embodiment, taking a power equipment cable-GIS terminal as an example, after obtaining its corresponding basic structural data, the shape, position, and assembly relationship of the core components of the equipment are reproduced in the three-dimensional modeling environment of the finite element simulation platform using this basic structural data as input. A basic geometric model without physical properties is constructed, fully preserving the key structural features of the equipment, and providing a stable geometric foundation for subsequent multiphysics coupling modeling. Its core components include a central conductor, electrical contact parts, epoxy resin sleeve, silicone rubber stress cone, SF6 gas cavity, and aluminum alloy shell, etc.

[0075] Among them, components such as conductors, insulation, and shells can be separated into independent geometric units, which makes it easier to configure physical properties for different components individually.

[0076] S13. Create an electromagnetic-thermal-fluid multiphysics coupling model corresponding to the basic geometric model in the finite element simulation platform;

[0077] Furthermore, S13 may include the following sub-steps:

[0078] In the finite element simulation platform, create the electromagnetic field model corresponding to the basic geometric model, solve the electromagnetic field control equation of the electromagnetic field model, obtain the Joule heat loss, and input it into the heat flow field model.

[0079] In the finite element simulation platform, a heat flow field model corresponding to the basic geometric model is created, and the temperature distribution under Joule heat loss is solved by solving the full heat transfer path control equation of the heat flow field model.

[0080] The temperature distribution is fed back to the electromagnetic field model, and the conductivity of the conductor and contact material is updated. The process then jumps to solve the electromagnetic field control equations of the electromagnetic field model, obtains the Joule heat loss, and inputs it into the heat flow field model. This process continues until the temperature distribution fluctuation value is within the fluctuation range, resulting in an electromagnetic-thermal-fluid multiphysics coupling model.

[0081] In the finite element simulation platform, corresponding electromagnetic field and thermal flow field models are built based on the basic geometric model of the power equipment. First, the electromagnetic field model solves the preset electromagnetic field control equations to calculate the Joule heat loss between the current-carrying conductor and the electrical contact area, which is then used as a heat source input to the thermal flow field model. Next, the thermal flow field model solves the full heat transfer path control equations covering heat conduction, natural convection of sealing gases such as SF6, and interfacial thermal radiation to obtain the temperature distribution across the entire equipment under the excitation of this heat source. The temperature distribution results obtained are fed back to the electromagnetic field model, and the conductivity parameters of the conductor and contact materials are updated based on the nonlinear conversion relationship between material conductivity and temperature. The iterative process of solving the electromagnetic field control equations, calculating Joule heat, and solving the thermal flow field temperature distribution is repeated until the temperature distribution fluctuation values ​​of two consecutive iterations fall within the preset convergence fluctuation range, completing the iterative convergence. Finally, an electromagnetic-thermal-fluid multiphysics coupled model that can accurately reproduce the operating characteristics of the equipment is obtained. By using this two-way closed-loop iterative coupling method, and taking into account the nonlinear characteristics of the material conductivity changing with temperature, the result deviation caused by the neglect of multi-field coupling effect in traditional one-way simulation calculation is effectively corrected, so as to accurately restore the entire process of heat source generation and heat transfer in the actual operation of the equipment.

[0082] S14. Set the load current, ambient temperature, and shell parameters in the electromagnetic-thermal-fluid multiphysics coupling model as variable parameters to obtain a parametric simulation model.

[0083] In this embodiment, in the multiphysics coupling model, the load current, ambient temperature, and shell parameters are defined as independent adjustable variables, completing the binding association between the variables and the solution module, and finally obtaining the parameterized simulation model. This enables a single model to adapt to multiple operating conditions, avoids redundant modeling, and improves simulation efficiency.

[0084] Step 102: A sequential sampling method is used to generate multiple parameter combination vectors, and based on each parameter combination vector and the parameterized simulation model, a preset surrogate model is trained to obtain multiple target reconstruction models.

[0085] Sequential sampling methods refer to deterministic sampling methods that can ensure uniform coverage of the entire parameter range by sample points within a multidimensional parameter design space and avoid sample clustering and spatial omissions. Examples include Sobol sequence sampling and Latin hypercube sampling, which can maximize coverage of the entire operating range with a limited sample size.

[0086] A parameter combination vector is a multidimensional vector composed of all variable independent variables in a parameterized simulation model. A single parameter combination vector corresponds to a specific operating condition of the device.

[0087] A surrogate model refers to a data-driven model that constructs a nonlinear mapping relationship between input parameters and output targets. It can replace high-fidelity simulation models to quickly solve for the output targets and is used for online modal coefficient prediction. Examples include radial basis function (RBF) neural networks, Gaussian process regression, and support vector regression.

[0088] In this embodiment, multiple parameter combination vectors covering the entire value range are generated in the multidimensional parameter design space based on the sequential sampling method. Each parameter combination vector is sequentially input into the parameterized simulation model to solve for the equipment temperature field distribution and corresponding low-dimensional modal coefficients under the corresponding operating conditions. With the parameter combination vector as input and the modal coefficients of the corresponding operating conditions as output, the preset multiple surrogate models are supervised and trained respectively. After training is completed, the surrogate model that meets the preset requirements after accuracy verification is determined as the target reconstruction model.

[0089] In one example of this application, step 102 may include the following sub-steps S21-S24:

[0090] S21. Use a sequential sampling method to generate multiple parameter combination vectors within a preset parameter design space;

[0091] In this embodiment, M sets of parameter combinations are generated in a preset parameter design space using a sequential sampling method, and corresponding parameter combination vectors are constructed using each set of parameter combinations.

[0092] S22. Use the parameter combination vectors to drive the parameterized simulation model respectively, and determine the steady-state temperature field corresponding to each parameter combination vector.

[0093] In this embodiment, the generated parameter combination vectors are sequentially input into the parameterized simulation model, automatically triggering electromagnetic-thermal-fluid bidirectional coupling iterative solution until the temperature distribution meets the preset convergence condition, and outputting the steady-state temperature field of the entire equipment under the corresponding operating condition.

[0094] The steady-state temperature field refers to the temperature field in which the global temperature distribution no longer changes with time after the electromagnetic-thermal-fluid multi-field coupling iterative convergence of electronic equipment operating under fixed conditions.

[0095] S23. Extract temperature values ​​from the finite element mesh nodes associated with each steady-state temperature field, construct a high-dimensional temperature field snapshot matrix and reduce its order to obtain a low-dimensional modal coefficient matrix.

[0096] Furthermore, S23 may include the following sub-steps:

[0097] Temperature values ​​are extracted from the finite element mesh nodes associated with each steady-state temperature field to construct a high-dimensional temperature field snapshot vector;

[0098] A high-dimensional temperature field snapshot matrix is ​​created using all high-dimensional temperature field snapshot vectors;

[0099] Calculate the average temperature field vector of all high-dimensional temperature field snapshot vectors;

[0100] Calculate the vector difference between each high-dimensional temperature field snapshot vector and the average temperature field vector, and use all vector differences to construct the fluctuation field matrix;

[0101] Singular value decomposition of the wave field matrix yields multiple eigenorthogonal basis modes and their corresponding singular values;

[0102] The cumulative energy contribution rate is calculated for each singular value, and the target number of eigenorthogonal principal modes are selected from multiple eigenorthogonal basis modes based on the cumulative energy contribution rate from high to low.

[0103] A reduced-order projection matrix is ​​constructed using all intrinsic orthogonal principal modes;

[0104] By projecting the wave field matrix onto the reduced-order subspace using the transpose of the reduced-order projection matrix, a low-dimensional modal coefficient matrix is ​​obtained.

[0105] In this embodiment, temperature values ​​are extracted from the finite element mesh nodes associated with each steady-state temperature field to construct the i-th high-dimensional temperature field snapshot vector. Where N is the total number of grid nodes. A high-dimensional temperature field snapshot matrix is ​​constructed by integrating all high-dimensional temperature field snapshot vectors column-wise. .

[0106] The average temperature field vector is calculated using each high-dimensional temperature field snapshot vector within the aforementioned high-dimensional temperature field snapshot matrix:

[0107] ;

[0108] The wave field matrix is ​​obtained by subtracting the average temperature field vector from each high-dimensional temperature field snapshot vector. .

[0109] Singular value decomposition of the wave field matrix yields multiple sets of one-to-one correspondences of eigenorthogonal basis modes and singular values ​​σ. j :

[0110] ;

[0111] in, column vector ξ i This refers to the intrinsically orthogonal principal modes, specifically the singular values ​​σ on the diagonal of Σ. j It reflects the energy of the corresponding mode.

[0112] Calculate the cumulative energy contribution rate for each singular value. :

[0113] ;

[0114] Based on the cumulative energy contribution rate, the first S dominant modes are selected to meet the target number of orthogonal eigenmodes, from high to low. A reduced-order projection matrix is ​​constructed using all eigenorthogonal principal modes. Selected .

[0115] By projecting the wave field matrix onto the reduced-order subspace using the transpose of the reduced-order projection matrix, we obtain the low-dimensional modal coefficient matrix. :

[0116] ;

[0117] Where, column vector a i This refers to the modal coefficients of the i-th snapshot. By mean centering, the common baseline features of the temperature field are stripped away, and the core orthogonal features of the temperature field are extracted through singular value decomposition. Modal truncation is performed based on the energy contribution rate. Without losing almost all the core distribution features of the temperature field, the ultra-high-dimensional temperature field data is compressed to extremely low-dimensional modal coefficients, which completely solves the problem of high difficulty and low computational efficiency in subsequent model training caused by the dimensionality explosion of high-dimensional temperature field data.

[0118] S24. Using the parameter combination vector as input and the low-dimensional modal coefficient matrix as output, train the preset surrogate models respectively to obtain multiple target reconstruction models.

[0119] Furthermore, S24 may include the following sub-steps:

[0120] Multiple training datasets are constructed by taking the parameter combination vectors as inputs and the low-dimensional modal coefficient matrices corresponding to the parameter combination vectors as outputs.

[0121] A single agent model is trained using each training data to obtain the target reconstruction model;

[0122] The surrogate model is a neural network using radial basis function kernels; the mapping relationship of the surrogate model is as follows:

[0123] ;

[0124] in, Let S be the modal coefficient of the j-th mode, 1≤j≤S; It is the parameter combination vector of the i-th group. Let be the connection weight from the i-th hidden layer node to the output layer in the neural network corresponding to the j-th mode. For radial basis kernel functions, To predict the input vector The Euclidean distance between the i-th parameter combination vector and the i-th parameter combination vector, where M is the total number of training data.

[0125] In this embodiment, for each order of intrinsic orthogonal principal mode, the parameter combination vector in the complete training dataset is used as a unified input, and the modal coefficients corresponding to that order of principal mode are used as the unique fitting target. A separate surrogate model is trained independently. The model parameters are continuously adjusted through an iterative optimization algorithm to minimize the error between the model's predicted value and the true label. After training is completed, the generalization accuracy is verified through a reserved test set. All single-order surrogate models that meet the preset accuracy requirements are the target reconstruction models of the corresponding order. By modeling single-order modes independently, the inter-order fitting interference problem of multi-output coupled models is completely avoided, and the prediction accuracy of each order of modal coefficients is greatly improved. At the same time, the single model structure is simple, the training convergence speed is fast, and it can flexibly adapt to the fitting requirements of different orders of modes.

[0126] Step 103: When the real-time operating condition vector to be predicted is received, the prediction mode coefficient matrix corresponding to the real-time operating condition vector is predicted by each target reconstruction model.

[0127] Real-time operating condition vectors refer to the real-time multidimensional input parameter vectors collected by the sensing and monitoring system during the online operation of the equipment, which correspond to the current operating state and are matched one-to-one with the independent variable dimensions of the parameterized simulation model.

[0128] The predictive modal coefficient matrix refers to a low-dimensional matrix composed of the principal modal coefficients of each order output by the reconstruction model of each objective and corresponding to the real-time operating conditions.

[0129] In this embodiment, the real-time operating condition vector corresponding to the operating condition to be predicted is collected in real time by the data structure deployed on the power equipment. After standardization or preprocessing, the vector is input into the target reconstruction model corresponding to each order of principal mode. Based on the pre-trained nonlinear mapping relationship, each target reconstruction model outputs the modal coefficients of the corresponding order. After all the modal coefficients of the order are integrated according to the preset dimension rules, the prediction modal coefficient matrix corresponding to the real-time operating condition is generated. Thus, the solution time is effectively reduced without having to perform the complex nonlinear iterative solution of traditional finite element simulation.

[0130] Specifically, S target reconstruction models are trained in step 102 above, and the real-time working condition vectors are used to reconstruct these models. Each target reconstruction model is input separately to perform modal coefficient prediction, generating a predicted modal coefficient matrix. .

[0131] Step 104: Reconstruct the temperature field based on the predicted mode coefficient matrix to generate a high-dimensional temperature field vector corresponding to the power equipment.

[0132] A high-dimensional temperature field vector refers to a high-dimensional vector composed of the temperature values ​​of all spatial grid nodes in the equipment simulation model, used to characterize the temperature distribution state of the entire equipment.

[0133] After obtaining the predicted modal coefficient matrix, the pre-generated reduced-order projection matrix and average temperature field vector are acquired. The predicted modal coefficient matrix and the orthogonal basis matrix are linearly mapped and superimposed with the average temperature field vector to restore the temperature values ​​of all spatial grid nodes of the power equipment under the corresponding real-time operating conditions. After integrating the grid nodes according to the spatial topology order, a high-dimensional temperature field vector that can characterize the temperature distribution of the entire equipment is generated, thereby improving the accuracy of the temperature distribution display of the power equipment while ensuring the accuracy of the finite element simulation.

[0134] In one example of this application, the high-dimensional temperature field vector is:

[0135] ;

[0136] in, To predict the modal coefficient matrix, For the reduced-order projection matrix, The average temperature field vector, This is a high-dimensional temperature field vector.

[0137] In another example of this application, the method further includes the following steps:

[0138] A parameterized simulation model is driven by a combination of parameters to generate multiple measurable data points, including load current, ambient temperature, casing temperature, and contact resistance.

[0139] The initial inversion model is trained using all measurable data to obtain the target inversion model;

[0140] When real-time input data is collected, the target inversion model performs real-time inversion based on the real-time input data to determine the contact resistance value corresponding to the real-time input data.

[0141] The initial inversion model refers to a basic algorithm model with nonlinear mapping fitting capability, which can be used to construct the inverse mapping relationship between external measurable parameters and internal contact resistance, such as Huber regression model, Gaussian process regression model or lightweight fully connected neural network.

[0142] Real-time input data can include load current, ambient temperature, and casing temperature, which can be collected in real time by sensing units deployed in the power equipment.

[0143] In this embodiment, similar to step 102, the effective value boundaries of load current, ambient temperature, and contact resistance under all operating conditions in the parameterized simulation model are predefined. Multiple sets of parameter combination vectors covering the entire value range are generated in the multidimensional parameter space through a sequential sampling method. Each set of parameter combination vectors is sequentially input into the parameterized simulation model to complete the electromagnetic-thermal-fluid multiphysics coupled steady-state solution. The load current, ambient temperature, and shell temperature that can be directly obtained through external monitoring under each set of operating conditions are extracted as measurable data. The contact resistance value preset for that set of simulations is matched synchronously to construct multiple sets of training sample sets with one-to-one correspondence between input features and output labels. The initial inversion model is trained with the measurable data in the training sample set as input and the corresponding contact resistance value as output. After the generalization test and error verification meet the preset accuracy requirements, the target inversion model that can be used for online inference is obtained.

[0144] During the actual operation of power equipment, load current, ambient temperature, and casing temperature are collected in real time as real-time input data and fed into the target inversion model. The target inversion model then performs real-time inference according to a nonlinear mapping relationship to determine the contact resistance value corresponding to the real-time input data. This non-invasive external monitoring data enables the quantitative perception of the unmeasurable contact state within the equipment's enclosed cavity.

[0145] It should be noted that after obtaining the contact resistance value, it can be combined with the temperature rise limit or contact resistance limit specified in the standard to determine whether the current contact condition has deteriorated, and to assess the remaining life or trigger an early warning.

[0146] In another example of this application, the method further includes the following steps:

[0147] Modularly encapsulate the target reconstruction model and the target inversion model, and create a data interface;

[0148] Connect the data interface to the power equipment;

[0149] When device operation data is received from the data interface, the target reconstruction model is invoked to reconstruct the temperature field according to the device operation data to obtain the real-time temperature field vector.

[0150] The target inversion model is invoked to perform real-time inversion based on the equipment operation data to determine the contact resistance value corresponding to the equipment operation data.

[0151] The user interface provides a visual representation of the real-time temperature field vector and contact resistance values.

[0152] In this embodiment, after training the target reconstruction model and the target inversion model is completed, they can be modularly encapsulated to shield their underlying algorithm logic, create a standardized data interface that adapts to general communication specifications, and connect the data interface to a specific location of the power equipment, such as an online monitoring system or a sensing unit.

[0153] When equipment operation data is received from the data interface, the encapsulated target reconstruction model is invoked. Using the equipment operation data as input, the temperature field is reconstructed, and the real-time temperature field vector corresponding to the operating condition is output. Simultaneously, the encapsulated target inversion model is invoked. Using the same equipment operation data as input, real-time inversion is performed, and the real-time contact resistance value corresponding to the operating condition is output. Finally, the global temperature distribution, hotspot location information, contact resistance value, contact health level, and degradation warning information corresponding to the real-time temperature field vector are visualized in a graphical user interface (GUI) for maintenance personnel. This GUI can be deployed on the digital twin system of the cable-GIS terminal, and the real-time temperature field can be displayed through 3D temperature field cloud map rendering.

[0154] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0155] The temperature field reconstruction device for power equipment provided in the embodiments of this application is described below. The temperature field reconstruction device for power equipment described below and the temperature field reconstruction method for power equipment described above can be referred to in correspondence.

[0156] Please see Figure 2 The present invention also provides a temperature field reconstruction device for power equipment, comprising:

[0157] The data acquisition and modeling module 201 is used to acquire the basic structural data corresponding to the power equipment and build a parametric simulation model.

[0158] The surrogate model training module 202 is used to generate multiple parameter combination vectors using a sequential sampling method, and to train preset surrogate models based on each parameter combination vector and the parameterized simulation model to obtain multiple target reconstruction models.

[0159] The modal coefficient prediction module 203 is used to predict the prediction modal coefficient matrix corresponding to the real-time working condition vector by using each target reconstruction model when a real-time working condition vector to be predicted is received.

[0160] Temperature field reconstruction module 204 is used to reconstruct the temperature field based on the predicted mode coefficient matrix and generate a high-dimensional temperature field vector corresponding to the power equipment.

[0161] Optionally, the device also includes an inversion module, specifically used for:

[0162] A parameterized simulation model is driven by a combination of parameters to generate multiple measurable data points, including load current, ambient temperature, casing temperature, and contact resistance.

[0163] The initial inversion model is trained using all measurable data to obtain the target inversion model;

[0164] When real-time input data is collected, the target inversion model performs real-time inversion based on the real-time input data to determine the contact resistance value corresponding to the real-time input data.

[0165] Optionally, the data acquisition and modeling module 201 includes:

[0166] The data acquisition submodule is used to acquire the basic structure data corresponding to the power equipment;

[0167] The geometric model construction submodule is used to construct a basic geometric model based on the basic structural data.

[0168] The multiphysics coupling submodule is used to create electromagnetic-thermal-fluid multiphysics coupling models corresponding to the basic geometric models in the finite element simulation platform;

[0169] The parametric simulation model generation submodule is used to set the load current, ambient temperature, and shell parameters in the electromagnetic-thermal-fluid multiphysics coupling model as variable parameters to obtain a parametric simulation model.

[0170] Optionally, the multiphysics coupling submodule is specifically used for:

[0171] In the finite element simulation platform, create the electromagnetic field model corresponding to the basic geometric model, solve the electromagnetic field control equation of the electromagnetic field model, obtain the Joule heat loss, and input it into the heat flow field model.

[0172] In the finite element simulation platform, a heat flow field model corresponding to the basic geometric model is created, and the temperature distribution under Joule heat loss is solved by solving the full heat transfer path control equation of the heat flow field model.

[0173] The temperature distribution is fed back to the electromagnetic field model, and the conductivity of the conductor and contact material is updated. The process then jumps to solve the electromagnetic field control equations of the electromagnetic field model, obtains the Joule heat loss, and inputs it into the heat flow field model. This process continues until the temperature distribution fluctuation value is within the fluctuation range, resulting in an electromagnetic-thermal-fluid multiphysics coupling model.

[0174] Optionally, the surrogate model training module 202 includes:

[0175] The parameter combination generation submodule is used to generate multiple parameter combination vectors within a preset parameter design space using a sequential sampling method.

[0176] The model-driven submodule is used to drive the parameterized simulation model using various parameter combination vectors to determine the steady-state temperature field corresponding to each parameter combination vector.

[0177] The low-dimensional modal coefficient matrix generation submodule is used to extract temperature values ​​from the finite element mesh nodes associated with each steady-state temperature field, construct a high-dimensional temperature field snapshot matrix and reduce its order to obtain a low-dimensional modal coefficient matrix.

[0178] The model training submodule is used to train preset surrogate models by taking the parameter combination vectors as input and the low-dimensional modal coefficient matrix as output, and obtain multiple target reconstruction models.

[0179] Optionally, the low-dimensional modal coefficient matrix generation submodule is specifically used for:

[0180] Temperature values ​​are extracted from the finite element mesh nodes associated with each steady-state temperature field to construct a high-dimensional temperature field snapshot vector;

[0181] A high-dimensional temperature field snapshot matrix is ​​created using all high-dimensional temperature field snapshot vectors;

[0182] Calculate the average temperature field vector of all high-dimensional temperature field snapshot vectors;

[0183] Calculate the vector difference between each high-dimensional temperature field snapshot vector and the average temperature field vector, and use all vector differences to construct the fluctuation field matrix;

[0184] Singular value decomposition of the wave field matrix yields multiple eigenorthogonal basis modes and their corresponding singular values;

[0185] The cumulative energy contribution rate is calculated for each singular value, and the target number of eigenorthogonal principal modes are selected from multiple eigenorthogonal basis modes based on the cumulative energy contribution rate from high to low.

[0186] A reduced-order projection matrix is ​​constructed using all intrinsic orthogonal principal modes;

[0187] By projecting the wave field matrix onto the reduced-order subspace using the transpose of the reduced-order projection matrix, a low-dimensional modal coefficient matrix is ​​obtained.

[0188] Optionally, the model training submodule is specifically used for:

[0189] Multiple training datasets are constructed by taking the parameter combination vectors as inputs and the low-dimensional modal coefficient matrices corresponding to the parameter combination vectors as outputs.

[0190] A single agent model is trained using each training data to obtain the target reconstruction model;

[0191] The surrogate model is a neural network using radial basis function kernels; the mapping relationship of the surrogate model is as follows:

[0192] ;

[0193] in, Let be the modal coefficients of the j-th mode. It is the parameter combination vector of the i-th group. Let be the connection weight from the i-th hidden layer node to the output layer in the neural network corresponding to the j-th mode. For radial basis kernel functions, To predict the input vector The Euclidean distance between the i-th parameter combination vector and the i-th parameter combination vector, where M is the total number of training data.

[0194] Optionally, the high-dimensional temperature field vector is:

[0195] ;

[0196] in, To predict the modal coefficient matrix, For the reduced-order projection matrix, The average temperature field vector, This is a high-dimensional temperature field vector.

[0197] Optionally, the device also includes a visualization module, specifically used for:

[0198] Modularly encapsulate the target reconstruction model and the target inversion model, and create a data interface;

[0199] Connect the data interface to the power equipment;

[0200] When device operation data is received from the data interface, the target reconstruction model is invoked to reconstruct the temperature field according to the device operation data to obtain the real-time temperature field vector.

[0201] The target inversion model is invoked to perform real-time inversion based on the equipment operation data to determine the contact resistance value corresponding to the equipment operation data.

[0202] The user interface provides a visual representation of the real-time temperature field vector and contact resistance values.

[0203] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described device and module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0204] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0205] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0206] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0207] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for reconstructing the temperature field of power equipment, characterized in that, include: Obtain the basic structural data corresponding to the power equipment and construct a parametric simulation model; Multiple parameter combination vectors are generated using a sequential sampling method, and a preset surrogate model is trained based on each parameter combination vector and the parameterized simulation model to obtain multiple target reconstruction models. When a real-time operating condition vector to be predicted is received, the prediction mode coefficient matrix corresponding to the real-time operating condition vector is predicted using each of the target reconstruction models. The temperature field is reconstructed based on the predicted mode coefficient matrix to generate a high-dimensional temperature field vector corresponding to the power equipment.

2. The method for reconstructing the temperature field of power equipment according to claim 1, characterized in that, Also includes: The parameterized simulation model is driven by the combination vectors of the aforementioned parameters to generate multiple measurable data; the measurable data includes load current, ambient temperature, casing temperature, and contact resistance values; The initial inversion model is trained using the measurable data described above, and the target inversion model is obtained. When real-time input data is collected, the target inversion model performs real-time inversion based on the real-time input data to determine the contact resistance value corresponding to the real-time input data.

3. The method for reconstructing the temperature field of power equipment according to claim 1, characterized in that, The steps of acquiring the basic structural data corresponding to the power equipment and constructing a parametric simulation model include: Obtain the basic structural data corresponding to the power equipment; Construct a basic geometric model based on the aforementioned basic structural data; Create an electromagnetic-thermal-fluid multiphysics coupled model corresponding to the basic geometric model in the finite element simulation platform; By setting the load current, ambient temperature, and shell parameters in the electromagnetic-thermal-fluid multiphysics coupling model as variable parameters, a parametric simulation model is obtained.

4. The method for reconstructing the temperature field of power equipment according to claim 3, characterized in that, The steps for creating the electromagnetic-thermal-fluid multiphysics coupled model corresponding to the basic geometric model in the finite element simulation platform include: In the finite element simulation platform, create the electromagnetic field model corresponding to the basic geometric model, solve the electromagnetic field control equation of the electromagnetic field model, obtain the Joule heat loss, and input it into the heat flow field model. In the finite element simulation platform, a heat flow field model corresponding to the basic geometric model is created, and the temperature distribution under the Joule heat loss is solved by solving the full heat transfer path control equation of the heat flow field model. The temperature distribution is fed back to the electromagnetic field model, and the conductivity of the conductor and contact material is updated. The process then jumps to the step of solving the electromagnetic field control equation of the electromagnetic field model, obtaining the Joule heat loss and inputting it into the heat flow field model, until the fluctuation value of the temperature distribution is within the fluctuation range, thus obtaining the electromagnetic-thermal-fluid multiphysics coupling model.

5. The method for reconstructing the temperature field of power equipment according to claim 1, characterized in that, The steps of generating multiple parameter combination vectors using a sequential sampling method, and training preset surrogate models based on each parameter combination vector and the parameterized simulation model to obtain multiple target reconstruction models include: A sequential sampling method is used to generate multiple parameter combination vectors within a pre-defined parameter design space; The parameterized simulation model is driven by each of the parameter combination vectors to determine the steady-state temperature field corresponding to each parameter combination vector. Temperature values ​​are extracted from the finite element mesh nodes associated with each steady-state temperature field, a high-dimensional temperature field snapshot matrix is ​​constructed and its order is reduced to obtain a low-dimensional modal coefficient matrix. Using the parameter combination vectors as input and the low-dimensional modal coefficient matrix as output, preset surrogate models are trained to obtain multiple target reconstruction models.

6. The method for reconstructing the temperature field of power equipment according to claim 5, characterized in that, The steps of extracting temperature values ​​from the finite element mesh nodes associated with each of the steady-state temperature fields, constructing a high-dimensional temperature field snapshot matrix and reducing its order to obtain a low-dimensional modal coefficient matrix include: Temperature values ​​are extracted from the finite element mesh nodes associated with each steady-state temperature field to construct a high-dimensional temperature field snapshot vector; A high-dimensional temperature field snapshot matrix is ​​created using all the aforementioned high-dimensional temperature field snapshot vectors; Calculate the average temperature field vector of all the aforementioned high-dimensional temperature field snapshot vectors; Calculate the vector difference between each of the high-dimensional temperature field snapshot vectors and the average temperature field vector, and construct the fluctuation field matrix using all the vector differences; Singular value decomposition is performed on the wave field matrix to obtain multiple eigenorthogonal basis modes and their corresponding singular values; The cumulative energy contribution rate is calculated according to each singular value, and the target number of eigenorthogonal principal modes are selected from the multiple eigenorthogonal basis modes based on the cumulative energy contribution rate from high to low. A reduced-order projection matrix is ​​constructed using all the aforementioned intrinsic orthogonal principal modes; The wave field matrix is ​​projected onto the reduced-order subspace according to the transpose of the reduced-order projection matrix to obtain the low-dimensional modal coefficient matrix.

7. The method for reconstructing the temperature field of power equipment according to claim 5, characterized in that, The step of using the parameter combination vectors as input and the low-dimensional modal coefficient matrix as output to train preset surrogate models to obtain multiple target reconstruction models includes: Multiple training datasets are constructed by taking the parameter combination vectors as inputs and the low-dimensional modal coefficient matrices corresponding to the parameter combination vectors as outputs. A single agent model is trained using the training data described above, and target reconstruction models are obtained respectively. The surrogate model is a neural network using a radial basis function kernel; the mapping relationship of the surrogate model is as follows: ； in, Let be the modal coefficients of the j-th mode. It is the parameter combination vector of the i-th group. Let be the connection weight from the i-th hidden layer node to the output layer in the neural network corresponding to the j-th mode. For radial basis kernel functions, To predict the input vector The Euclidean distance between the i-th parameter combination vector and the i-th parameter combination vector, where M is the total number of training data.

8. The method for reconstructing the temperature field of power equipment according to claim 1, characterized in that, The high-dimensional temperature field vector is: ； in, To predict the modal coefficient matrix, For the reduced-order projection matrix, The average temperature field vector, This is a high-dimensional temperature field vector.

9. The method for reconstructing the temperature field of power equipment according to claim 2, characterized in that, Also includes: The target reconstruction model and the target inversion model are modularly encapsulated to create a data interface; Connect the data interface to the power equipment; When device operation data is received from the data interface, the target reconstruction model is invoked to reconstruct the temperature field according to the device operation data to obtain a real-time temperature field vector. The target inversion model is invoked to perform real-time inversion based on the equipment operating data to determine the contact resistance value corresponding to the equipment operating data. The real-time temperature field vector and the contact resistance value are visualized in the user interface.

10. A temperature field reconstruction device for power equipment, characterized in that, include: The data acquisition and modeling module is used to acquire the basic structural data of the power equipment and build a parametric simulation model. The surrogate model training module is used to generate multiple parameter combination vectors using a sequential sampling method, and to train preset surrogate models according to each parameter combination vector and the parameterized simulation model to obtain multiple target reconstruction models. The modal coefficient prediction module is used to predict the prediction modal coefficient matrix corresponding to the real-time working condition vector when a real-time working condition vector to be predicted is received, using each of the target reconstruction models respectively. The temperature field reconstruction module is used to reconstruct the temperature field based on the predicted mode coefficient matrix and generate a high-dimensional temperature field vector corresponding to the power equipment.

Images