High-voltage switchgear temperature field monitoring method and system based on digital twinning

Abstract

The invention belongs to the technical field of high-voltage switch equipment state monitoring, and particularly relates to a high-voltage switch equipment temperature field monitoring method and system based on digital twinning, and the method comprises the steps: obtaining a three-dimensional model and real-time operation data of high-voltage switch equipment based on digital twinning; obtaining a temperature field prediction result of the high-voltage switchgear according to the obtained real-time operation data and the deep learning reduced-order model; wherein the deep learning order reduction model comprises a deep feed-forward neural network used for mapping the acquired real-time operation data from a working condition parameter space to a low-dimensional feature space and a convolutional auto-encoder used for performing temperature field reconstruction on low-dimensional feature vector data output by the deep feed-forward neural network; and associating the obtained temperature field prediction result with the three-dimensional model, and performing visual interaction of the temperature field of the high-voltage switchgear to complete monitoring of the temperature field of the high-voltage switchgear based on digital twinning.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of high-voltage switchgear state monitoring, and particularly relates to a high-voltage switchgear temperature field monitoring method and system based on digital twinning. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] High-voltage switchgear is the core device of the power grid, and strong coupling effects of multiple physical fields such as electromagnetism, heat, and fluid mechanics occur during the breaking of large currents and the long-term operation; for example, intense Joule heat is generated at the conductor and the contact finger during short-circuit current, causing the overheat and deformation of the conductive components; the arc burning and arc extinguishing process is accompanied by huge thermal shock and gas flow; the thermal expansion of the device structure affects parameters such as contact resistance. In order to ensure the safe and reliable operation of high-voltage switchgear, the temperature field distribution of high-voltage switchgear under various operating conditions needs to be accurately obtained, and the temperature field distribution is combined to perform fault warning and life assessment of high-voltage switchgear.

[0004] At present, the temperature field of high-voltage switchgear is mainly obtained by full-size numerical simulation calculation, and the finite element / finite volume method is used to solve the partial differential equations of multi-field coupling; however, for a 550kV GIS device with complex structure and huge scale, millions or even tens of millions of grids need to be divided in the process of constructing a full-order simulation model, and highly nonlinear control equations need to be solved, and a single operating condition simulation calculation often takes tens of hours, which is extremely low in efficiency.

[0005] Based on the existing research, the temperature field of high-voltage switchgear can be obtained through digital twinning technology; although the existing digital twinning technology can realize high-fidelity rendering, assembly, and operation and maintenance animation simulation, VR interaction, and other functions of the three-dimensional model of high-voltage switchgear, it still uses traditional numerical calculation methods or classic reduced-order models (such as POD reduction) to predict the temperature field, which has the disadvantages of low calculation efficiency and inability to update in real time; especially in the case of strong nonlinear boundary conditions or material properties, the traditional POD-Galerkin reduced-order model needs to rely on a linear trial function space and is difficult to accurately represent nonlinear effects, and the calculation accuracy is seriously reduced. Taking the nonlinear change of the contact resistance of the contact finger in the GIS switch with temperature as an example, the error of the traditional POD method in this case will increase significantly.

[0006] The existing deep learning reduced order model (DL-ROM) is mostly oriented to general equipment such as aviation blades and turbine blades, lacks targeted design for the electromagnetic-thermal-structural strong coupling and short-time large current impact of high-voltage switch equipment, and does not consider the integration requirement of real-time data interaction with digital twin technology. Therefore, in view of the obvious deficiency of the existing technology in the real-time prediction of the temperature field of GIS equipment, a customized rapid monitoring method is needed to meet the special needs of the field. SUMMARY

[0007] To solve the above problems, the present application provides a high-voltage switch equipment temperature field monitoring method and system based on digital twin, which adopts a temperature field calculation method combining deep learning reduced order model and digital twin technology, realizes real-time and rapid prediction of high-voltage switch equipment temperature field at any time and under any working condition, realizes real-time data interaction of high-voltage switch equipment by combining digital twin, realizes real-time state monitoring of high-voltage switch equipment, and improves the safety and reliability of high-voltage switch equipment operation.

[0008] According to some embodiments, the first aspect of the present application provides a high-voltage switch equipment temperature field monitoring method based on digital twin, which adopts the following technical scheme: A high-voltage switch equipment temperature field monitoring method based on digital twin, comprising: Obtaining a three-dimensional model and real-time operation data of the high-voltage switch equipment based on digital twin; Obtaining a temperature field prediction result of the high-voltage switch equipment according to the obtained real-time operation data and a deep learning reduced order model; wherein the deep learning reduced order model comprises a deep feedforward neural network for mapping the obtained real-time operation data from a working condition parameter space to a low-dimensional feature space, and a convolutional autoencoder for reconstructing a temperature field from a low-dimensional feature vector data output by the deep feedforward neural network; Associating the obtained temperature field prediction result with the three-dimensional model, performing visual interaction of the high-voltage switch equipment temperature field, and completing monitoring of the high-voltage switch equipment temperature field based on digital twin.

[0009] As a further technical limitation, the deep feedforward neural network is used to learn the nonlinear mapping relationship from the wide working condition parameter space of the high-voltage switch equipment to the low-dimensional feature space, and the working condition parameter vector in the obtained real-time operation data is input. The input information is mapped to the low-dimensional feature space by the network hidden layer adopting multiple fully connected layers, to obtain the low-dimensional feature representation of the real-time operation data; the convolutional autoencoder comprises an encoder convolutional neural network and a decoder convolutional neural network, and the obtained low-dimensional feature representation vector of the real-time operation data is converted into a reduced dimension feature under the action of the encoder convolutional neural network and the decoder convolutional neural network, to complete the prediction from the temperature field parameter to the temperature field and realize the temperature field reconstruction.

[0010] As a further technical limitation, in the process of acquiring the three-dimensional model of the high-voltage switching device, a three-dimensional geometric model of the high-voltage switching device is constructed according to the structure of the high-voltage switching device, mesh division of the three-dimensional geometric model is performed according to the structure type of the high-voltage switching device, coupling control equations of different physical fields are acquired based on the geometric model after mesh division, and a multi-physical field coupling model of the high-voltage switching device is constructed; full-order simulation calculation of the high-voltage switching device is performed according to the constructed multi-physical field coupling model, a multi-working-condition full-order simulation data set of the high-voltage switching device is constructed, and a three-dimensional model of the high-voltage switching device is obtained.

[0011] Further, the loss function of the deep learning reduced-order model is a weighted sum of a global error function of temperature field reconstruction and a key area error function; wherein the global error function adopts a mean square error of a reconstructed temperature field and an original simulation field, the reconstructed temperature field is a temperature field prediction result, the original simulation field is a multi-working-condition full-order simulation data, and the key area error function is an error term of a key part in the high-voltage switching device; the network parameters of the deep feedforward neural network and the convolutional autoencoder are adjusted through a back propagation algorithm to reduce the loss function of the deep learning reduced-order model; when the loss function of the deep learning reduced-order model reaches a preset loss threshold, the training of the deep learning reduced-order model is completed, and the mapping from the real-time running data working condition parameter to the temperature field is completed.

[0012] As a further technical limitation, after the temperature field prediction result is associated with the three-dimensional model, a three-dimensional geometric model displaying the temperature distribution of each part of the high-voltage switching device in real-time running in a cloud chart is obtained, the temperature data and the change curve of the hotspot temperature of the key part of the high-voltage switching device are displayed in combination with digital twinning, the temperature field evolution of the high-voltage switching device is viewed, and the visualization interaction of the high-voltage switching device is completed.

[0013] As a further technical limitation, when the temperature of the high-voltage switching device abnormally rises, a temperature abnormality warning is issued according to the visualization interaction result of the high-voltage switching device in combination with digital twinning, the reason for the abnormal rise of the temperature is analyzed according to the temperature field prediction result, and the monitoring of the temperature field of the high-voltage switching device is completed.

[0014] According to some embodiments, the second aspect of the present application provides a high-voltage switching device temperature field monitoring system based on digital twinning, which adopts the following technical scheme: A high-voltage switching device temperature field monitoring system based on digital twinning, comprising: An acquisition module configured to acquire a three-dimensional model and real-time running data of a high-voltage switching device based on digital twinning; a prediction module configured to obtain a temperature field prediction result of the high-voltage switching device according to the obtained real-time operation data and a deep learning reduced-order model, wherein the deep learning reduced-order model comprises a deep feedforward neural network for mapping the obtained real-time operation data from a working condition parameter space to a low-dimensional feature space and a convolutional autoencoder for reconstructing a temperature field from low-dimensional feature vector data output by the deep feedforward neural network; a monitoring module configured to associate the obtained temperature field prediction result with the three-dimensional model, perform visual interactive monitoring of the temperature field of the high-voltage switching device, and complete monitoring of the temperature field of the high-voltage switching device based on digital twinning.

[0015] According to some embodiments, a third aspect of the present application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium, which stores a program, the program being executed by a processor to implement the steps in the method for monitoring a temperature field of a high-voltage switching device based on digital twinning according to the first aspect of the present application.

[0016] According to some embodiments, a fourth aspect of the present application provides an electronic device, which adopts the following technical solution: An electronic device, comprising a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor implements the steps in the method for monitoring a temperature field of a high-voltage switching device based on digital twinning according to the first aspect of the present application when executing the program.

[0017] According to some embodiments, a fifth aspect of the present application provides a computer program product, which adopts the following technical solution: A computer program product, comprising software code, wherein a program in the software code implements the steps in the method for monitoring a temperature field of a high-voltage switching device based on digital twinning according to the first aspect of the present application.

[0018] Compared with the prior art, the present application has the following beneficial effects: The present application adopts a temperature field calculation method combining a deep learning reduced-order model and digital twinning technology, extracts low-dimensional features of high-dimensional working condition parameters of a high-voltage switching device through a deep feedforward neural network, and maps the extracted low-dimensional features to a temperature field through a convolutional autoencoder; according to a high-voltage switching device temperature field distribution result obtained by the deep learning reduced-order model and a high-voltage switching device temperature field prediction result obtained based on digital twinning, real-time simulation prediction of the high-voltage switching device temperature field is completed, real-time calculation of a multi-physical field temperature field of the high-voltage switching device based on twinning digital technology is realized, and the "calculation-simulation-diagnosis-decision" of the temperature field is integrated, real-time state monitoring of the high-voltage switching device is realized, and the safety and reliability of the operation of the high-voltage switching device are improved. BRIEF DESCRIPTION OF DRAWINGS

[0019] The drawings constituting a part of this embodiment are used to provide further understanding of this embodiment, the illustrative embodiments of this embodiment and the description thereof are used to explain this embodiment, and do not constitute improper limitation on this embodiment.

[0020] Figure 1 Flow chart of the high-voltage switchgear temperature field monitoring method based on digital twinning in the first embodiment of the present application; Figure 2 Overall block diagram of the high-voltage switchgear temperature field monitoring method based on digital twinning in the first embodiment of the present application; Figure 3 Structure block diagram of the high-voltage switchgear temperature field monitoring system based on digital twinning in the second embodiment of the present application. DETAILED DESCRIPTION

[0021] The present application will be further described below in conjunction with the drawings and embodiments.

[0022] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0023] It should be noted that the terms used herein are only for the purpose of describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and furthermore, it should be understood that when the terms "comprise" and / or "include" are used in the specification, there is a feature, step, operation, device, component and / or combination thereof.

[0024] In the present application, the terms such as "upper", "lower", "left", "right", "front", "back", "vertical", "horizontal", "side", "bottom", etc. indicate the orientation or positional relationship shown in the drawings, which is only a relationship word determined for the convenience of describing the structural relationship of the components or elements of the present application, and cannot be understood as a limitation on the present application.

[0025] In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0026] Embodiment one The first embodiment of the present application introduces a high-voltage switchgear temperature field monitoring method based on digital twinning.

[0027] As Figure 1A temperature field monitoring method for high-voltage switchgear based on digital twin is shown, comprising: Obtain 3D models and real-time operating data of high-voltage switchgear based on digital twins; A temperature field prediction result of the high-voltage switchgear is obtained based on the acquired real-time operating data and a deep learning reduced-order model; wherein the deep learning reduced-order model includes a deep feedforward neural network for mapping the acquired real-time operating data from an operating parameter space to a low-dimensional feature space and a convolutional autoencoder for reconstructing the temperature field using the low-dimensional feature vector data output by the deep feedforward neural network; The obtained temperature field prediction results are associated with the three-dimensional model to perform visual interaction of the high-voltage switchgear temperature field and complete the monitoring of the high-voltage switchgear temperature field based on digital twins.

[0028] This embodiment combines the deep learning reduced-order model (DL-ROM) with the 550kV GIS high-voltage switch digital twin platform to quickly calculate the temperature field of high-voltage switchgear. Specifically, by constructing a deep neural network model to replace traditional numerical simulation, the multi-physics temperature field of GIS equipment can be predicted within seconds. Based on the high-fidelity 3D model rendering, equipment assembly and operation animation, and VR interactive functions provided by the existing digital twin platform, the DL-ROM algorithm module is integrated to achieve real-time and rapid prediction of the temperature field of GIS equipment at any time and under any operating conditions, thereby enhancing the integrity and intelligence of the platform functions.

[0029] like Figure 2 As shown, the temperature field monitoring method of high-voltage switchgear based on digital twin in this embodiment is divided into an offline training phase and an online prediction phase, and mainly includes the following steps: Step 1: Construction of multi-physics simulation model of 550kV GIS high-voltage switch; In the offline stage, a high-precision full-scale simulation model of the GIS equipment is established to generate a large amount of temperature field data required for neural network training. Specifically: Based on the design drawings and physical structure of the 550kV GIS equipment, a 3D geometric model of the main components of the GIS equipment was constructed and meshed. In order to balance accuracy and computational cost, the complex structure was appropriately simplified and modeled. For example, the conductive circuit model includes conductive moving contacts, contact holders, spring contacts, petal-shaped static contacts, conductive rods and other conductive components; the heat dissipation structure includes pot-type insulators, insulating rods and aluminum alloy shells; the fluid area includes the area filled with During the meshing process, a locally encrypted tetrahedral mesh is used in key hot spots such as conductive contacts to obtain a higher-resolution temperature gradient; a boundary layer mesh is created at the junction of the gas chamber and the metal wall to more accurately capture the fluid heat transfer boundary layer effect.

[0030] After the three-dimensional geometric model mesh is divided, the coupling control equation groups of the internal electromagnetic field, temperature field and flow field of the GIS device are defined; specifically: The electromagnetic field adopts the Maxwell equation under quasi-static approximation, uses the magnetic vector potential A to represent the electromagnetic field state, satisfies the electromagnetic field control equation, and under quasi-static approximation, the magnetic vector potential A satisfies: ; wherein, is the magnetic permeability, is the electrical conductivity, is the current density.

[0031] The electric field strength is derived from the magnetic vector potential, that is, .

[0032] The heat conduction and convection control equation is ; wherein, is the density, is the specific heat capacity, is the thermal conductivity, is the fluid velocity, is the Joule heat source term.

[0033] For the fluid mechanics control equation (incompressible Navier-Stokes), the fluid velocity and pressure satisfy the continuity equation and the momentum equation can be realized as: ; ; wherein, is the dynamic viscosity, is the body force including electromagnetic force.

[0034] Through the above electromagnetic field, thermal field and flow field equations, the control equation group of the internal multi-physical field coupling of the high-voltage switchgear is obtained, which is used for high-fidelity simulation calculation of the temperature field distribution of the device under different working conditions. Through the above multi-physical field coupling model, the physical response of the GIS device in the process of current passing, heating, gas flowing and structure stress can be comprehensively simulated.

[0035] According to the obtained multi-physical field coupling model, full-order simulation calculation is performed on the typical working condition group, and rich temperature field distribution data are obtained. The design of the typical working condition should cover the main operating range and extreme cases of the GIS device, for example: the current level covers light load 500A, rated 5000A and overload 6000A, breaking short-circuit current 63kA, etc.; the time nodes include short-circuit instant 0.01s, breaking process 1s, temperature rise process 100s and long-time steady state 10000s, etc.; the environmental temperature considers winter low temperature-20℃, normal temperature 25℃ and summer high temperature 40℃, etc.

[0036] It should be noted that the full-order simulation calculation in the embodiment is to solve the simulation model of the boundary conditions under multiple working conditions by using the existing finite element numerical calculation technology, so as to create a training data set.

[0037] Through the combination of different current amplitudes, durations and environmental conditions, a simulation scheme such as 48 typical working conditions (4 currents x 4 times x 3 environments) can be formed. For each working condition, electromagnetic-thermal-fluid transient coupling simulation is performed on the full-size model (i.e. the pre-processed simplified geometric model of the simulated real GIS device size) to obtain the temperature field distribution from the initial time to the target time. Since each group of working condition simulation requires solving a large number of nonlinear equations, it may still take tens of hours on a high-performance computing platform. With 48 groups of working conditions, a total of about 48x1.5 million node scale temperature field snapshot data is generated; these high-precision, high-resolution temperature field data will be used as the basic data set for training the deep learning reduced-order model.

[0038] Step two: DL-ROM model construction (DFNN+CAE architecture) After obtaining sufficient simulation data, the temperature field is reduced and modeled by a deep learning method. The DL-ROM model in the embodiment is composed of a deep feedforward neural network (DFNN) and a convolutional autoencoder (CAE) decoder two sub-networks, which respectively undertake the functions of mapping from input parameters to low-dimensional features and reconstructing the full-field temperature from low-dimensional features; specifically including: (1) Deep feedforward neural network (DFNN) DFNN is used to learn the nonlinear mapping relationship from the wide working condition parameter space of the GIS device to the low-dimensional feature space. The input of DFNN is a vector containing the main working condition parameters, for example, three dimensions can correspond to the running current I (such as 0-63 kA), environmental temperature (such as -20-+40℃) and running time t (simulation evolution time).

[0039] The embodiment adopts 4 layers of hidden layers (the hidden layers use multi-layer fully connected layers), and the number of neurons is 256, 128, 64 and 32 in turn, and ReLU activation function (parameter a=0.01) is used in each layer to enhance the nonlinear expression ability and avoid gradient disappearance. The output layer of DFNN is a dimension of low-dimensional feature vector, which is used to represent the main characteristic mode of the temperature field under the corresponding working condition. For example, different dimensions of the feature vector can represent the temperature rise characteristic quantities of different parts / components in the GIS device (such as the average temperature of the contact, the temperature gradient of the insulator, the average temperature of the conductive rod, the shell temperature rise speed, etc.). Through the pre-training of the DFNN, the information of the temperature field affected by a wide range of working conditions (hundreds of thousands of possible combinations) can be effectively compressed, providing a concise feature input for subsequent CAE decoding.

[0040] The training process of DFNN (deep feedforward neural network supervised learning) is to establish the mapping relationship between operating parameters such as current I , ambient temperature T amb , running time t and temperature field dimensionality reduction features , to realize the fast calculation of "parameter input temperature field prediction".

[0041] Let DFNN be a multi-layer fully connected network, and the mapping function be , then:

[0042] wherein, is the DFNN layer weight and bias, is the activation function.

[0043] With the CAE-generated dimensionality reduction features as the target, the mean square error is calculated: , and the historical temperature field data is encoded by the CAE to obtain , and the corresponding parameters form a training pair (x, y). (2) Convolutional Autoencoder (CAE) CAE is responsible for realizing the reconstruction of GIS temperature field from low-dimensional feature representation to full three-dimensional temperature field distribution; CAE is composed of an encoder and a decoder two convolutional neural networks, and the bottleneck layer in the middle is connected with the feature vector output by DFNN.

[0044] CAE unsupervised pre-training (temperature field data), the high-dimensional temperature field data matrix , wherein N is the number of samples (temperature field snapshots under different parameters), D is the spatial dimension (number of grid nodes), through self-learning, the encoder reconstructs the data to approximate the original data , so as to learn the low-dimensional feature representation of the data.

[0045] The bottleneck layer is arranged at the output end of the CAE encoder and the input end of the decoder, and is an intermediate layer with the lowest feature dimension in the network; the bottleneck layer is a low-dimensional feature interaction node, which on one hand receives the device parameter feature vector output by the DFNN, and on the other hand drives the CAE decoder to reconstruct the temperature field, so as to realize the mapping closed loop of 'parameter->feature physical field'.

[0046] As the core hub connecting the parameter space and the physical field space, the bottleneck layer breaks through the traditional CAE limitation through multi-source feature fusion and dynamic mapping mechanism; the core function of the bottleneck layer is to integrate the temperature field space mode features extracted by the CAE encoder And The generated parameter associated features Dynamically balance the contributions of physical prior and parameter driven.

[0047]

[0048] Wherein, is an activation function, is a weight, represents element-wise multiplication, represents the spatial distribution mode of the temperature field captured by the encoder (such as the hot spot position of the contact and the temperature gradient of the heat dissipation channel), represents the current load I , the ambient temperature T env , the running time t and the like are mapped into a low-dimensional vector.

[0049] The encoder of the CAE takes high-fidelity temperature field data as input, and compresses and extracts the low-dimensional features; for example, for each group of temperature field snapshots (about 1.5 million grid nodes), interpolation or down-sampling can be used to regularize it into a three-dimensional tensor with a size of about 200x200x3 (equivalent to 120,000 effective nodes), and then input into several layers of three-dimensional convolution for encoding.

[0050] In this embodiment, the encoder adopts 3 layers of convolution, each with a kernel size of 3x3 and a step size of 2, and a batch normalization layer is added after convolution to speed up convergence; the encoder finally outputs an 8-dimensional feature vector to represent the main mode features of the input temperature field.

[0051] The encoder compresses the high-dimensional temperature field into a low-dimensional feature vector (the dimension reduction space); specifically, assuming that the encoder is composed of multiple convolution layers, and the mapping function is then:

[0052] Wherein, is the weight and bias of the i th layer, is the activation function, To reduce the dimension.

[0053] The decoder of CAE performs the opposite operation of the encoder. Specifically, it inputs the above-mentioned 8-dimensional feature vector, and through three layers of transposed convolution, it gradually "enlarges" the features and reconstructs them back to the original temperature field distribution. The decoder convolution kernel adopts 4×4 with a step size of 2. Skip connections are introduced in the network to fuse the intermediate features of the encoding stage with the corresponding decoding layer to restore the temperature distribution details of complex structures such as contact details and narrow channels. At the same time, the encoder part of CAE adds the fusion of the geometric topological features of GIS equipment. For example, the geometric identification or coordinates of the key components of the equipment are used as additional inputs, so that the network retains the topological information when compressing features. Through structural customization and optimization, CAE can more accurately reconstruct the temperature field in areas with complex geometric structures and significantly reduce the error of reduced-order reconstruction.

[0054] The decoder converts the low-dimensional feature vector Restore to high-dimensional temperature field data ; The decoder is the inverse process of the encoder, and the mapping function is ,but:

[0055] in, For decoder layer weights and biases, is the activation function, For the prediction target, .

[0056] It's important to note that in the CAE decoder, skip connections utilize a cross-layer feature reuse mechanism to directly transmit detailed temperature field features captured at each encoder level (such as contact edge gradients and narrow channel geometry) to the corresponding decoder level. These features are then concatenated or weightedly fused with the upsampled feature maps on a channel-by-channel basis. This process preserves high-frequency spatial information from the encoding stage, avoiding detail loss caused by downsampling. This allows accurate temperature field reconstruction of complex structures within GIS equipment (such as localized temperature spikes caused by contact resistance and strong gradients within narrow SF6 gas channels). Furthermore, the embedding of geometric topological features further enhances physical realism, significantly reducing reconstruction errors.

[0057] (3) The DFNN and CAE sub-networks are trained together to obtain an end-to-end reduced-order mapping model. During the combined training process, the parameters of the DFNN and CAE are jointly optimized using the large amount of multi-condition temperature field snapshot data generated in the previous step.

[0058] Update the encoder and decoder parameters through backpropagation to minimize reconstructing low-dimensional feature vectors , containing the key features of the temperature field.

[0059] The reconstruction error adopts the mean square error: .

[0060] This embodiment adopts a joint training strategy, and defines a loss function as a weighted sum of the global error of temperature field reconstruction and the error of key areas; wherein the global error function adopts the mean square error (MSE) of the reconstructed temperature field (temperature field prediction result) and the original simulation field (multi-condition full-order simulation data), and the error function of the key area is an error term with a higher weight defined for the key parts such as the heat-prone points and insulator interfaces in the GIS device, so as to ensure the prediction accuracy of these areas. The network parameters of the DFNN and the CAE are adjusted through the back propagation algorithm, so that the training error is continuously reduced. When a predetermined accuracy threshold is reached on the validation set, the training process is ended, and the trained DL-ROM reduced-order model is obtained. The obtained DL-ROM reduced-order model essentially contains a fast mapping from the working condition parameters to the temperature field, and can instantaneously output the corresponding three-dimensional temperature field distribution when the input condition is given.

[0061] It should be noted that the back propagation algorithm in this embodiment is a mature existing technology in the field of deep learning. The back propagation algorithm adjusts the parameters of the DFNN and the CAE to reduce the training error through the following mechanism: first, the input data is forward propagated through the DFNN and the CAE to generate a feature vector, which is dynamically fused through the bottleneck layer and then reconstructed by the CAE decoder to generate a temperature field, and the loss (such as MSE) between the reconstructed value and the true value is calculated; then, the gradient of the loss with respect to the parameters of each layer is derived in reverse using the chain rule, wherein the gradient directly optimizes the local feature extraction capability of the CAE encoder through the jump connection, and is transmitted to the output features of the DFNN through the fusion layer weight, and then the internal weights of the DFNN are updated in reverse; at the same time, the transpose convolution parameters of the CAE decoder and the convolution kernel parameters of the encoder are also adjusted synchronously through the reconstruction loss gradient; the optimizer (such as Adam) updates all parameters according to the gradient direction, so that the feature mapping of the DFNN is more consistent with the physical law, and the encoding and decoding of the CAE are more accurate in capturing spatial patterns, and through iterative training, a closed loop of “loss driven gradient→gradient updates parameters→parameters optimize features→features reduce loss” is formed, and finally the training error is continuously reduced under the collaborative optimization of the two networks.

[0062] Step three: digital twin platform integration and real-time temperature field calculation The DL-ROM model trained in step two is deployed and integrated into the digital twin platform of the GIS device to realize real-time temperature field prediction and visualization; specifically: In the background server or cloud of the digital twin platform, load the trained DL-ROM model, and connect it with real-time data stream through the API interface provided by the platform; During the platform operation, periodically or on demand, obtain the running parameters of GIS devices (such as real-time current, switch opening and closing state, gas pressure, ambient temperature, etc.), call the deployed DL-ROM model interface, and real-time transfer these input parameters to the model for calculation.

[0063] When the digital twin platform receives new working condition parameters, the DL-ROM model performs prediction calculation immediately, the DL-ROM reduced order model calculation result is mapped into a new color json file by the backend server, the old file is updated, and after the update is completed, it is transmitted to the digital twin model through post communication, the twin model reads the updated json file, and simultaneously updates the variable color for rendering; Specifically: the platform sends the input multi-dimensional working condition parameters into the DFNN sub-network, and quickly maps to obtain a low-dimensional feature vector; input the feature into the CAE decoder network to reconstruct a complete three-dimensional temperature field. Due to the small amount of calculation of neural network inference (compared to solving complex PDE), the entire prediction process only takes milliseconds, and the temperature field result can be output; This second-level response capability enables the platform to realize real-time refreshing of the GIS device temperature field; Whether it is daily rated operation, temporary overload, or fault short-circuit transient temperature rise, it can achieve fast prediction and timely reflect the current thermal distribution of the device.

[0064] The digital twin platform associates the temperature field result output by the DL-ROM model with the three-dimensional model of the device, and performs visual presentation. For example, display the temperature distribution of each part on the three-dimensional geometric model of the GIS device in the form of a cloud chart, and mark the high temperature area as a red hotspot. At the same time, through the UI interface of the platform, the temperature values of key measuring points and the temperature curve of hot spots over time can be displayed. The operation and maintenance personnel wearing VR equipment or on the monitoring interface can directly view the temperature field evolution inside the device and obtain comprehensive thermal information complementary to the actual sensor temperature measuring point. When the temperature abnormally rises, the platform can issue a warning and analyze the possible fault reasons combined with the simulation results; On-line monitoring of current load or environmental temperature information is transmitted into the reduced order model, and after fast calculation, the temperature field chromatogram cloud chart rendered according to the digital twin model can determine whether the GIS internal temperature is locally too high or the temperature in the gas chamber is generally too high to determine the simulated local overheating or poor heat dissipation fault in the gas chamber. For example, local overheating of a contact may indicate poor contact or excessive resistance; General high temperature in the gas chamber may be related to poor heat dissipation or overload.

[0065] The embodiment integrates the multi-field coupling effects of electromagnetism, heat, fluid, etc. applied to GIS equipment at the data level, synchronously collects temperature field snapshots under the action of multiple factors such as current, time, and environmental temperature, and directly trains a neural network to learn the nonlinear mapping relationship of “multi-field input → temperature field output”. This end-to-end data-driven modeling breaks through the limitations of relying only on physical simplification assumptions in the past, and greatly improves the ability of the model to cope with complex coupled working conditions.

[0066] For the wide working condition range and complex geometric topology of 550 kV GIS switchgear, the embodiment adopts a double-branch deep neural network structure; through a deep feedforward neural network (DFNN), a low-dimensional feature representation of high-dimensional working condition parameters (such as current from 500 A to 5000 A, time from millisecond level to hour level, and environmental temperature from -20℃ to 40℃) is extracted, and efficient learning of the relationship between wide working condition parameters and low-dimensional features is realized; through a convolutional autoencoder (CAE), the above low-dimensional features are mapped back to the complete temperature field, and the reconstruction of low-dimensional features to full-field temperature field is realized; through joint optimization of the composite loss function of reconstruction error and key area error, the prediction accuracy of the fault-prone parts is highlighted, and the accuracy of the reduced-order model in the focus area is ensured.

[0067] The DL-ROM reduced-order model in the embodiment can be seamlessly embedded in the existing digital twin platform, and directly use the real-time data stream of the platform for online calculation; with the help of the interactive interface of the digital twin platform, the user can input or obtain the running condition data of the GIS equipment at any time (such as the current value at the moment of a short-circuit fault or the steady-state load after long-term operation), and the DL-ROM model will output the temperature field distribution result at the corresponding time immediately, and bind it on the three-dimensional model of the equipment in the form of a three-dimensional cloud chart for display; the operation and maintenance personnel can intuitively observe the temperature field evolution, hot spot position, and thermal stress change trend inside the equipment through the platform UI or VR equipment, realize the “simulation-diagnosis” closed loop, and take maintenance decisions in time when abnormalities are found. This field-customized deep learning rapid simulation technology integration greatly enhances the practicality and intelligent decision support capability of the digital twin system, and should belong to the application level innovation and integration, rather than the simple application of the basic deep learning algorithm itself.

[0068] The embodiment adopts a temperature field calculation mode combining deep learning reduced order model and digital twin technology, extracts low-dimensional features of high-dimensional working condition parameters of the high-voltage switchgear through a deep feedforward neural network, and maps the extracted low-dimensional features to a temperature field in combination with a convolutional autoencoder; according to the high-voltage switchgear temperature field distribution result obtained by the deep learning reduced order model and the high-voltage switchgear temperature field prediction result obtained based on the digital twin, real-time simulation prediction of the high-voltage switchgear temperature field is completed, real-time calculation of the high-voltage switchgear multi-physical field temperature field based on the digital twin technology is realized, the 'calculation-simulation-diagnosis-decision' of the temperature field is integrated, real-time state monitoring of the high-voltage switchgear is realized, and the safety and reliability of the high-voltage switchgear operation are improved.

[0069] Embodiment two The embodiment two of the present application introduces a high-voltage switchgear temperature field monitoring system based on digital twin.

[0070] As shown in Figure 3 A high-voltage switchgear temperature field monitoring system based on digital twin, comprising: An acquisition module configured to acquire a three-dimensional model and real-time operation data of the high-voltage switchgear based on digital twin; A prediction module configured to obtain a temperature field prediction result of the high-voltage switchgear according to the acquired real-time operation data and a deep learning reduced order model; wherein the deep learning reduced order model comprises a deep feedforward neural network for mapping the acquired real-time operation data from a working condition parameter space to a low-dimensional feature space and a convolutional autoencoder for reconstructing a temperature field from the low-dimensional feature vector data output by the deep feedforward neural network; A monitoring module configured to associate the obtained temperature field prediction result with the three-dimensional model, perform visual interactive monitoring of the high-voltage switchgear temperature field, and complete monitoring of the high-voltage switchgear temperature field based on digital twin.

[0071] The detailed steps are the same as those of the high-voltage switchgear temperature field monitoring method based on digital twin provided in embodiment one, and will not be repeated here.

[0072] Embodiment three The embodiment three of the present application provides a computer readable storage medium.

[0073] A computer readable storage medium having a program stored thereon, the program being executed by a processor to implement the steps in the high-voltage switchgear temperature field monitoring method based on digital twin provided in embodiment one of the present application.

[0074] The detailed steps are the same as those of the high-voltage switchgear temperature field monitoring method based on digital twin provided in embodiment one, and will not be repeated here.

[0075] Embodiment four Embodiment four of the present application provides an electronic device.

[0076] An electronic device, comprising a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor implements the steps in the high-voltage switch device temperature field monitoring method based on digital twinning as described in embodiment one of the present application.

[0077] The detailed steps are the same as the high-voltage switch device temperature field monitoring method based on digital twinning provided in embodiment one, which will not be repeated here.

[0078] Embodiment five Embodiment five of the present application provides a computer program product.

[0079] A computer program product, comprising software code, wherein the program in the software code implements the steps in the high-voltage switch device temperature field monitoring method based on digital twinning as described in embodiment one of the present application.

[0080] The detailed steps are the same as the high-voltage switch device temperature field monitoring method based on digital twinning provided in embodiment one, which will not be repeated here.

[0081] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0082] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks.

[0083] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow Figure 1 The flow or flows and / or blocks Figure 1 The flow or flows and / or blocks

[0084] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 The flow or flows and / or blocks Figure 1 The flow or flows and / or blocks

[0085] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments described and illustrated herein, without departing from the spirit and scope of the application. Therefore, it is intended that such additional variations and modifications be included within the scope of the application. Accordingly, the appended claims are intended to cover all such modifications and variations as falling within the scope of the application.

[0086] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

[0087] The above description is only preferred embodiments of the present application, and is not intended to limit the present application. The present application can be variously changed and modified by those skilled in the art without departing from the spirit and scope of the present application. Therefore, any modification, equivalent replacement, improvement, and the like made by those skilled in the art should be included in the scope of the present application.

Claims

1. A temperature field monitoring method for high-voltage switchgear based on digital twin, characterized in that: include: Obtain 3D models and real-time operating data of high-voltage switchgear based on digital twins; A temperature field prediction result of the high-voltage switchgear is obtained based on the acquired real-time operating data and a deep learning reduced-order model; wherein the deep learning reduced-order model includes a deep feedforward neural network for mapping the acquired real-time operating data from an operating parameter space to a low-dimensional feature space and a convolutional autoencoder for reconstructing the temperature field using the low-dimensional feature vector data output by the deep feedforward neural network; The obtained temperature field prediction results are associated with the three-dimensional model to perform visual interaction of the high-voltage switchgear temperature field and complete the monitoring of the high-voltage switchgear temperature field based on digital twins.

2. A method for monitoring the temperature field of high-voltage switchgear based on digital twins as claimed in claim 1, characterized in that: The deep feedforward neural network is used to learn the nonlinear mapping relationship from the wide operating parameter space of the high-voltage switchgear to the low-dimensional feature space, and inputs the operating parameter vector in the acquired real-time operation data. The input information is mapped to the low-dimensional feature space by the hidden layer of the network using multiple layers of fully connected layers to obtain the low-dimensional feature representation of the real-time operation data; the convolutional autoencoder includes an encoder convolutional neural network and a decoder convolutional neural network, and the obtained low-dimensional feature representation vector of the real-time operation data is converted into a dimensionality reduction feature under the action of the encoder convolutional neural network and the decoder convolutional neural network, completing the prediction from temperature field parameters to temperature field and realizing temperature field reconstruction.

3. A method for monitoring the temperature field of high-voltage switchgear based on digital twins as claimed in claim 1, characterized in that: In the process of obtaining the three-dimensional model of the high-voltage switchgear, a three-dimensional geometric model of the high-voltage switchgear is constructed according to the structure of the high-voltage switchgear, and the three-dimensional geometric model is meshed according to the structural type of the high-voltage switchgear. Based on the meshed geometric model, the coupling control equations of different physical fields are obtained, and a multi-physical field coupling model of the high-voltage switchgear is constructed; full-order simulation calculations of the high-voltage switchgear are performed based on the constructed multi-physical field coupling model, and a multi-operating condition full-order simulation data set of the high-voltage switchgear is constructed to obtain the three-dimensional model of the high-voltage switchgear.

4. A method for monitoring the temperature field of high-voltage switchgear based on digital twins as described in claim 3, characterized in that: The loss function of the deep learning reduced-order model is the weighted sum of the global error function of temperature field reconstruction and the key area error function; wherein, the global error function adopts the mean square error between the reconstructed temperature field and the original simulation field, and the reconstructed temperature field is the temperature field prediction result; the original simulation field is multi-operating condition full-order simulation data; the key area error function is the error term of the key parts in the high-voltage switchgear; the network parameters of the deep feedforward neural network and the convolutional autoencoder are adjusted by the back-propagation algorithm to reduce the loss function of the deep learning reduced-order model; when the loss function of the deep learning reduced-order model obtains the preset loss threshold, the training of the deep learning reduced-order model is completed, and the mapping from the real-time operating data operating condition parameters to the temperature field is completed.

5. A method for monitoring the temperature field of high-voltage switchgear based on digital twins as claimed in claim 1, characterized in that: After the temperature field prediction results are associated with the three-dimensional model, a three-dimensional geometric model is obtained that uses a cloud map to display the temperature distribution of various parts of the high-voltage switchgear in real time. Combined with the digital twin, the temperature data of key parts of the high-voltage switchgear and the change curve of the hot spot temperature are displayed, the temperature field evolution of the high-voltage switchgear is viewed, and the visual interaction of the high-voltage switchgear is completed.

6. A method for monitoring the temperature field of high-voltage switchgear based on digital twins as claimed in claim 1, characterized in that: When the temperature of the high-voltage switchgear rises abnormally, a temperature abnormality warning is issued based on the visual interaction results of the high-voltage switchgear combined with the digital twin. The cause of the abnormal temperature increase is analyzed based on the temperature field prediction results to complete the monitoring of the temperature field of the high-voltage switchgear.

7. A high-voltage switchgear temperature field monitoring system based on digital twin, characterized in that: include: an acquisition module configured to acquire a three-dimensional model and real-time operation data of the high-voltage switchgear based on the digital twin; A prediction module configured to obtain a temperature field prediction result of the high-voltage switchgear based on the acquired real-time operating data and a deep learning reduced-order model; wherein the deep learning reduced-order model includes a deep feedforward neural network for mapping the acquired real-time operating data from an operating parameter space to a low-dimensional feature space and a convolutional autoencoder for reconstructing the temperature field using the low-dimensional feature vector data output by the deep feedforward neural network; The monitoring module is configured to associate the obtained temperature field prediction results with the three-dimensional model, perform visual interaction of the high-voltage switchgear temperature field, and complete the monitoring of the high-voltage switchgear temperature field based on digital twins.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the high-voltage switchgear temperature field monitoring method based on digital twin are implemented as described in any one of claims 1 to 6.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the steps of the high-voltage switchgear temperature field monitoring method based on digital twin are implemented as described in any one of claims 1 to 6.

10. A computer program product comprising software code, characterized in that The program in the software code executes the steps of the high-voltage switchgear temperature field monitoring method based on digital twin according to any one of claims 1 to 6.

Images