Fast estimation method of cable temperature field based on neural network

Abstract

A kind of neural network-based cable temperature field fast estimation method, comprising: based on cable structure, establish two-dimensional section model, based on the two-dimensional section model, establish the finite element model coupled with electric field, temperature field, flow field and external non-isothermal flow field;Based on the finite element model, simulation calculation is carried out, and training database is generated;Based on python, neural network model is built, and the neural network model is trained based on the database, to obtain cable temperature neural network prediction model;In online application stage, according to the operating condition parameters acquired in real time, the cable two-dimensional section temperature field distribution of real time is estimated using the cable temperature neural network prediction model.This application combines the way of off-line high-precision simulation and online fast prediction, shortens the calculation time to within minute level, and the prediction accuracy is higher.

Description

Technical Field

[0001] This invention relates to the field of high voltage and insulation technology, and in particular to a method for rapid estimation of cable temperature field based on neural networks. Background Technology

[0002] With the introduction of my country's dual-carbon policy and the increasing energy demand of coastal cities, offshore wind power projects are developing rapidly. High-voltage direct current cross-linked polyethylene (HVDC) cables, as crucial transmission equipment, play a significant role in long-distance power transmission. However, during operation, these cables are subjected to a combination of electrical, thermal, and mechanical forces, which accelerates the aging of the cable insulation layer and negatively impacts its safe and reliable operation.

[0003] Temperature, as a key indicator reflecting the performance of cable insulation, is closely related to the dielectric loss factor of cable insulation materials. Excessive temperature leads to aging of the cable insulation layer, while excessively low temperature results in inefficient use of the cable's current-carrying capacity. Therefore, it is necessary to monitor cable temperature in real time to ensure safe and reliable cable operation. With the advent of power grid digitalization, there is a growing demand not only for accurate calculation of cable temperature but also for visualization and rapid computation.

[0004] In recent years, the calculation of cable temperature fields has mainly employed the equivalent thermal circuit method and the finite element method. The equivalent thermal circuit method treats each layer of the cable structure and the external environment as equivalent thermal resistance or thermal capacity, neglecting the heat exchange between the cable and the external environment, resulting in insufficient accuracy in modeling the cable environment. The finite element method, including the finite difference method and the finite volume method, primarily uses software such as Comsol and Ansys to build the model, apply corresponding boundary conditions and excitations, and obtain the multiphysics distribution of the model. The finite element method provides a more accurate model of the cable's surrounding environment and can reflect the heat exchange between the cable and its environment; however, the simulation time is long, preventing rapid calculations.

[0005] Existing technologies for cable temperature calculation still have the following shortcomings: traditional finite element method calculation is too time-consuming and difficult to achieve real-time calculation; while existing fast calculation methods based on neural networks can either only predict the temperature of a single point, which is insufficient in terms of information dimension, or predict a one-dimensional temperature field along the axis, which cannot provide a two-dimensional cross-sectional temperature field containing all internal structural details that is crucial for assessing the insulation condition.

[0006] Therefore, there is an urgent need to propose a new method that can quickly and accurately calculate the temperature field distribution of the entire two-dimensional cross-section of the cable to meet the needs of comprehensive and real-time monitoring of cable status in engineering sites. Summary of the Invention

[0007] To solve the above-mentioned technical problems, the present invention adopts the following solution.

[0008] A method for rapid estimation of cable temperature field based on neural network includes:

[0009] S1. Establish a two-dimensional cross-sectional model based on the cable structure, and establish a finite element model based on the two-dimensional cross-sectional model that couples the electric field, temperature field, flow field and external non-isothermal flow field;

[0010] S2. Perform simulation calculations based on the finite element model to generate a training database;

[0011] S3. Build a neural network model and train the neural network model based on the database to obtain a neural network prediction model for cable temperature;

[0012] S4. In the online application stage, based on the real-time acquired operating condition parameters, the real-time two-dimensional cross-sectional temperature field distribution of the cable is estimated using the cable temperature neural network prediction model.

[0013] Optionally, step S1 includes:

[0014] A two-dimensional cross-sectional model is established based on the actual physical dimensions of the cable. This model includes the cable's main structure, such as the copper conductor, insulation layer, alloy lead layer, and outer sheath, as well as the external sea area.

[0015] The entire model is physically meshed, and the mesh is refined in key areas such as the cable body and boundary layer.

[0016] Optionally, step S2 includes:

[0017] The electric field, temperature field for solid and fluid heat transfer, and laminar flow physical field are set for the two-dimensional cross-sectional model, and then coupled to form a multiphysics model.

[0018] By setting different combinations of operating parameters, the multiphysics model is solved in steady state to generate the training database.

[0019] Optionally, in step S3, the neural network prediction model includes an input layer, at least one hidden layer, and an output layer;

[0020] The number of nodes in the input layer corresponds to the number of operating condition parameters, and the output layer is used to output the temperature field distribution of the two-dimensional cross-section of the cable.

[0021] Optionally, the cable is a submarine cable, and the external non-isothermal flow field is a laminar flow field in the sea area.

[0022] Optionally, the neural network model uses mean squared error as the loss function and is trained using the Adam optimizer.

[0023] Optionally, a correction step may be included after the online application phase:

[0024] The measured temperature of at least one physical temperature measurement point on the cable is compared with the predicted temperature of the corresponding location output by the cable temperature neural network prediction model.

[0025] If the comparison deviation exceeds a preset threshold, the amount of data in the training database is increased or the structure of the neural network model is adjusted, and the model is retrained.

[0026] Optionally, the cable is a high-voltage DC cross-linked polyethylene submarine cable.

[0027] A fast cable temperature field estimation system based on neural networks includes:

[0028] The model building module is used to build a two-dimensional cross-sectional model based on the cable structure, and to build a finite element model based on the two-dimensional cross-sectional model that couples the electric field, temperature field, flow field and external non-isothermal flow field.

[0029] The simulation calculation module is used to perform simulation calculations based on the finite element model and generate a training database.

[0030] The model training module is used to build a neural network model and train the neural network model based on the database to obtain a neural network prediction model for cable temperature.

[0031] An online application module, used in the online application phase, estimates the real-time two-dimensional cross-sectional temperature field distribution of the cable using the cable temperature neural network prediction model based on real-time acquired operating condition parameters. A computer-readable storage medium is provided for storing a computer program configured to implement the method when invoked by a processor.

[0032] An electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor; wherein the processor implements the method when executing the program.

[0033] Compared with the prior art, the present invention has the following beneficial technical effects:

[0034] 1. Balancing computational efficiency and accuracy: This invention combines offline high-precision simulation with online rapid prediction, reducing the computation time of the traditional finite element method from several hours to within minutes. At the same time, compared with the finite element simulation results, the neural network model has higher prediction accuracy.

[0035] 2. Achieves visualization and comprehensive monitoring of the temperature field: The method of this invention outputs the temperature field distribution of the entire two-dimensional cross-section of the cable, rather than a single or a few discrete temperature points. It can intuitively display the temperature gradient and hot spot distribution at various locations inside the cable in the form of cloud maps, etc., providing more comprehensive data support for assessing the risk of insulation aging. Attached Figure Description

[0036] The accompanying drawings illustrate exemplary embodiments of the invention and, together with the description thereof, serve to explain the principles of the invention. These drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification.

[0037] Figure 1 This is a two-dimensional cross-sectional distribution diagram of a cable in one embodiment of the present invention;

[0038] Figure 2 This is a temperature distribution diagram of a cable cross-section in one embodiment of the present invention;

[0039] Figure 3 This is a neural network model architecture in one embodiment of the present invention;

[0040] Figure 4 This is a graph showing the change in training loss of a neural network model with the number of training iterations in one embodiment of the present invention;

[0041] Figure 5 This is an embodiment of the present invention showing the actual and predicted temperature distribution and absolute temperature difference of three sets of cables under different operating conditions;

[0042] Figure 6 This is a flowchart of a method for rapid estimation of cable temperature field based on neural network in one embodiment of the present invention;

[0043] Figure 7 This is an offline training and modeling subsystem in one embodiment of the present invention;

[0044] Figure 8 This is an online prediction and monitoring subsystem in one embodiment of the present invention;

[0045] Figure 9 This is the software interface of a cable temperature field rapid estimation system based on neural networks in one embodiment of the present invention. Detailed Implementation

[0046] The following is in conjunction with the appendix Figures 1 to 9 The present invention will be further described in detail below with reference to the embodiments. It is to be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present invention are shown in the accompanying drawings.

[0047] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other. The technical solution of this invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0048] Unless otherwise stated, the exemplary embodiments / exemplifications shown are to be understood as providing exemplary features of various details that provide ways in which the technical concept of the invention can be implemented in practice. Therefore, unless otherwise stated, the features of the various embodiments / exemplifications may be additionally combined, separated, interchanged and / or rearranged without departing from the technical concept of the invention.

[0049] The use of crosshairs and / or shading in the accompanying drawings is generally used to clarify the boundaries between adjacent components. Thus, unless otherwise stated, the presence or absence of crosshairs or shading does not convey or indicate any preference or requirement for the specific material, material properties, dimensions, proportions, commonalities between the illustrated components, or any other characteristics, properties, etc., of the components. Furthermore, in the accompanying drawings, the dimensions and relative dimensions of components may be exaggerated for clarity and / or descriptive purposes. When exemplary embodiments can be implemented differently, a specific process sequence may be performed in a different order than that described. For example, two consecutively described processes may be performed substantially simultaneously or in the reverse order of their description. Furthermore, the same reference numerals denote the same components.

[0050] When a component is referred to as being "on" or "above" another component, "connected to," or "joined to" another component, the component may be directly on, directly connected to, or directly joined to the other component, or there may be intermediate components. However, when a component is referred to as being "directly on" another component, "directly connected to," or "directly joined to" another component, there are no intermediate components. Therefore, the term "connection" can refer to a physical connection, an electrical connection, etc., and may or may not have intermediate components.

[0051] For descriptive purposes, the present invention may use spatial relative terms such as “below,” “under,” “below,” “down,” “above,” “above,” “higher,” and “side (e.g., in a “sidewall”)” to describe the relationship between one component and another component as shown in the accompanying drawings. In addition to the orientations depicted in the drawings, the spatial relative terms are also intended to encompass different orientations of the device during use, operation, and / or manufacture. For example, if the device in the drawings is flipped, a component described as “below” or “under” another component or feature would subsequently be positioned “above” said other component or feature. Thus, the exemplary term “below” can encompass both “above” and “below” orientations. Furthermore, the device may be otherwise positioned (e.g., rotated 90 degrees or in other orientations), thus interpreting the spatial relative descriptive terms used herein accordingly.

[0052] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms “a” and “the” are intended to include the plural forms as well. Furthermore, when the terms “comprising” and / or “including” and variations thereof are used in this specification, it indicates the presence of the stated features, integrals, steps, operations, parts, components, and / or groups thereof, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, parts, components, and / or groups thereof. It should also be noted that, as used herein, the terms “substantially,” “about,” and other similar terms are used as approximate terms rather than as terms of degree, thus explaining the inherent biases in measurements, calculated values, and / or provided values ​​that would be recognized by one of ordinary skill in the art.

[0053] In one embodiment, reference is made to Figure 6 This invention discloses a method for fast estimation of cable temperature field based on neural network, comprising:

[0054] S1. Establish a two-dimensional cross-sectional model based on the cable structure, and establish a finite element model based on the two-dimensional cross-sectional model that couples the electric field, temperature field, flow field and external non-isothermal flow field;

[0055] S2. Perform simulation calculations based on the finite element model to generate a training database;

[0056] S3. Build a neural network model and train the neural network model based on the database to obtain a neural network prediction model for cable temperature;

[0057] S4. In the online application stage, based on the real-time acquired operating condition parameters, the real-time two-dimensional cross-sectional temperature field distribution of the cable is estimated using the cable temperature neural network prediction model.

[0058] Optionally, step S1 includes:

[0059] A two-dimensional cross-sectional model is established based on the actual physical dimensions of the cable. This model includes the cable's main structure, such as the copper conductor, insulation layer, alloy lead layer, and outer sheath, as well as the external sea area.

[0060] The entire model is physically meshed, and the mesh is refined in key areas such as the cable body and boundary layer.

[0061] Optionally, step S2 includes:

[0062] The electric field, temperature field for solid and fluid heat transfer, and laminar flow physical field are set for the two-dimensional cross-sectional model, and then coupled to form a multiphysics model.

[0063] By setting different combinations of operating parameters, the multiphysics model is solved in steady state to generate the training database.

[0064] Optionally, in step S3, the neural network prediction model includes an input layer, at least one hidden layer, and an output layer;

[0065] The number of nodes in the input layer corresponds to the number of operating condition parameters, and the output layer is used to output the temperature field distribution of the two-dimensional cross-section of the cable.

[0066] Optionally, the cable is a submarine cable, and the external non-isothermal flow field is a laminar flow field in the sea area.

[0067] Optionally, the neural network model uses mean squared error as the loss function and is trained using the Adam optimizer.

[0068] Optionally, a correction step may be included after the online application phase:

[0069] The measured temperature of at least one physical temperature measurement point on the cable is compared with the predicted temperature of the corresponding location output by the cable temperature neural network prediction model.

[0070] If the comparison deviation exceeds a preset threshold, the amount of data in the training database is increased or the structure of the neural network model is adjusted, and the model is retrained.

[0071] Optionally, the cable is a high-voltage DC cross-linked polyethylene submarine cable.

[0072] In another embodiment, the present invention discloses a method for fast estimation of cable temperature field based on neural network, comprising:

[0073] S1. Establish a finite element model;

[0074] S11. Establish a two-dimensional cross-sectional model;

[0075] In COMSOL Multiphysics software, a two-dimensional cross-sectional model is created based on the actual physical dimensions of the cable. This model includes the cable's main structure, such as the copper conductor, insulation layer, alloy lead layer, and outer sheath, as well as the external sea area.

[0076] S12. Perform mesh generation;

[0077] Appropriate material properties, including thermal conductivity, specific heat capacity, density, and electrical conductivity, are set for each layer of the cable and the sea area. The electrical conductivity of the insulation layer is set as a function of temperature and electric field strength to accurately reflect its physical properties. The entire model is physically meshed, with mesh refinement performed in key areas such as the cable body and boundary layers.

[0078] S2. Perform simulation calculations based on the finite element model to generate a training database; including:

[0079] S21. Set up the electric field, where the field equation for the current is:

[0080]

[0081] Where J is the current density, J e Let σ be the charge density of the flow, σ be the conductivity, E be the electric field strength, V be the electric potential, and Q be the electric potential. j，v It is a volume current source.

[0082] The system uses a copper conductor as the terminal with a voltage of 400kV and an alloy lead layer as the grounding layer.

[0083] S22. Set up the temperature field, where the physical field used is solid-fluid heat transfer, and the field equation is:

[0084]

[0085] Where, d z The thickness of the domain is given by ρ, the density of the fluid material is given by u, the fluid velocity vector is given by T3, the temperature of the material is given by q, the heat flux is given by Q, and the heat source in the material is given by Q. ted For thermoelastic damping, C p Let be the specific heat capacity of the material at normal pressure, k be the thermal conductivity, and q0 be the initial heat flux. The gradient of temperature T3 represents the spatial rate of temperature change. It is the divergence of heat flux q, representing the net inflow or outflow of heat per unit volume per unit time.

[0086] The solid heat transfer region is selected as the submarine cable body, and the fluid heat transfer region is the sea area. The far-field boundary of the sea area is set to a constant temperature, the copper conductor is the heat source (generated by current loss), and the cable surface is set to radiate to the environment.

[0087] S23. A laminar flow physical field is established in a sea area, with the physical model set as weakly compressible flow and a reference pressure level of 1 Pa. The field equations are:

[0088]

[0089] Where ρ is the fluid density, p is the pressure, u is the flow velocity of the seawater, F is the volume force, I is the radiative energy emitted from the surface of the object, and K is the radiative transfer coefficient.

[0090] The left boundary of the sea area is set as the inlet, the right boundary as the outlet, and the upper and lower surfaces are set as no-slip boundary conditions.

[0091] S24. Couple the temperature field, electric field, and flow field, and add a non-isothermal flow multiphysics module with the following field equations:

[0092]

[0093] Among them, Q vd Let τ be the radiative heat transfer rate, τ be the transmittance, and u be the flow velocity.

[0094] S25. Generate the database;

[0095] Add the study as a steady-state condition and add the "Surrogate Model Training" module. Set the input parameters to: cable current I (0-3000A), ambient temperature Tw (0-40℃), and seawater flow velocity v0 (0-2m / s). Set the number of input points to 3000 to generate a dataset containing 3000 sets of different operating conditions. In the export settings, select to export the input parameters (I, Tw, v0) and the corresponding temperature distribution data (T(x,y)) of the entire cross-section of the cable to form the training database.

[0096] S3. Build and train the neural network model;

[0097] In a Python environment, neural network models are built using the TensorFlow or PyTorch frameworks.

[0098] Set the number of training loops and train using the Adam optimizer and mean squared error (MSE) loss function.

[0099] S4. When applying online, input a set of real-time operating condition outputs and predicted cable cross-sectional temperature distributions into the cable temperature neural network prediction model.

[0100] In another embodiment, the present invention discloses a method for rapid estimation of cable temperature field based on neural network, using a 400kV cable model as the simulation object, including:

[0101] S1. Establish a finite element model;

[0102] S11. Establish a two-dimensional cross-sectional model;

[0103] In COMSOL Multiphysics software, according to Figure 1 The actual physical dimensions of the 400kV cable are shown in Table 1. A two-dimensional cross-sectional model is established; this model includes the cable body structure such as the copper conductor, insulation layer, alloy lead layer, and outer sheath, as well as the external sea area. The sea area is set as a 6m × 3m rectangle, and the soil area is set as a 6m × 3m square.

[0104] Table 1 Structural parameters of 400kV submarine cable

[0105]

[0106] S12. Perform mesh generation;

[0107] Appropriate material properties, including thermal conductivity, specific heat capacity, density, and electrical conductivity, are set for each layer of the cable and the sea area. The electrical conductivity of the insulation layer is set as a function of temperature and electric field strength to accurately reflect its physical properties. The entire model is physically meshed, with mesh refinement performed in key areas such as the cable body and boundary layers.

[0108] For example, the number of grid cells is 46405, the minimum cell mass is 0.01387, and the average cell mass is 0.7802.

[0109] S2. Based on the finite element model, perform simulation calculations to generate a training database under different working conditions; including:

[0110] S21. Set up the electric field, where the field equation for the current is:

[0111]

[0112] Where J is the current density, J e Let σ be the charge density of the flow, σ be the conductivity, E be the electric field strength, V be the electric potential, and Q be the electric potential. j，v It is a volume current source.

[0113] The system uses a copper conductor as the terminal with a voltage of 400kV and an alloy lead layer as the grounding layer.

[0114] S22. Set up the temperature field, where the physical field used is solid-fluid heat transfer, and the field equation is:

[0115]

[0116] Where, d z The thickness of the domain is given by ρ, the density of the fluid material is given by u, the fluid velocity vector is given by T3, the temperature of the material is given by q, the heat flux is given by Q, and the heat source in the material is given by Q. ted For thermoelastic damping, C p Let be the specific heat capacity of the material at normal pressure, k be the thermal conductivity, and q0 be the initial heat flux. The gradient of temperature T3 represents the spatial rate of temperature change. It is the divergence of heat flux q, representing the net inflow or outflow of heat per unit volume per unit time.

[0117] The solid heat transfer region is selected as the submarine cable body, and the fluid heat transfer region is the sea area. The far-field boundary of the sea area is set to a constant temperature, the copper conductor is the heat source (generated by current loss), and the cable surface is set to radiate to the environment.

[0118] S23. A laminar flow physical field is established in a sea area, with the physical model set as weakly compressible flow and a reference pressure level of 1 Pa. The field equations are:

[0119]

[0120] Where ρ is the fluid density, p is the pressure, u is the flow velocity of the seawater, F is the volume force, I is the radiative energy emitted from the surface of the object, and K is the radiative transfer coefficient.

[0121] The left boundary of the sea area is set as the inlet, the right boundary as the outlet, and the upper and lower surfaces are set as no-slip boundary conditions.

[0122] S24. Couple the temperature field, electric field, and flow field, and add a non-isothermal flow multiphysics module with the following field equations:

[0123]

[0124] Among them, Q vd Let τ be the radiative heat transfer rate, τ be the transmittance, and u be the flow velocity.

[0125] S25. Generate the database;

[0126] Add the study as a steady-state condition and add the "Surrogate Model Training" module. Set the input parameters to: cable current I (0-3000A), ambient temperature Tw (0-40℃), and seawater flow velocity v0 (0-2m / s). Set the number of input points to 3000 to generate a dataset containing 3000 sets of different operating conditions. In the export settings, select to export the input parameters (I, Tw, v0) and the corresponding temperature distribution data (T(x,y)) of the entire cross-section of the cable to form the training database.

[0127] The range of input parameter values ​​is based on the IEC 60287 standard and data from the National Marine Science Data Center.

[0128] Based on the above-mentioned structural, parameter, and physical field settings for the 400kV high-voltage DC cable, the temperature distribution across the cable cross-section was obtained as follows: Figure 2 As shown.

[0129] S3. Build and train the neural network model;

[0130] In a Python environment, neural network models are built using frameworks such as TensorFlow or PyTorch, with architectures as follows: Figure 3 As shown, the dataset includes: one input layer (3 nodes, corresponding to I, Tw, v0), three hidden layers (64, 128, and 128 nodes respectively), and one output layer (the number of nodes equals the number of nodes in the cable cross-section mesh, outputting the entire temperature field). The generated 3000 sets of data are divided into training and testing sets in a 7:3 ratio. The input range of the testing set should be within the aforementioned range; otherwise, retraining should be performed based on the actual situation.

[0131] The training iterations (epochs) are set to 1000, and the Adam optimizer and mean squared error (MSE) loss function are used for training. During training, the training loss gradually decreases and converges with the increase of the number of training iterations, as shown below. Figure 4 As shown, this indicates that the model fits well.

[0132] To evaluate model performance, mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²) were used. 2 R is used as an evaluation indicator. 2 If the error is below 0.85 or the mean squared error loss function does not converge, it is considered unqualified and needs to be adjusted by changing the number of neurons in the model until the model is qualified.

[0133] In step S2, a dataset D={P, Q} is constructed based on the input parameters under different operating conditions. D is the operating status dataset of the submarine cable under a certain operating condition, where P is a normalized three-dimensional matrix including cable current, ambient temperature, seawater flow velocity, etc., with a total of 3000 sets of data for the input operating conditions. Q is the output ambient temperature data of the cable cross-section, which is a two-dimensional matrix corresponding to the temperature at a specific coordinate (x, y). 70% of the data is used for the training set and 30% for the test set.

[0134] Step S3 includes:

[0135] S31. Constructing a Convolutional Neural Network: The neural network model uses normalized operating condition parameters—cable current I, ambient temperature Tw, and seawater flow velocity v0—as input, with an input dimension of 3. Nonlinear feature extraction of the operating condition information is performed through a multi-layer fully connected network. This network sequentially sets four fully connected layers with 64, 128, 128, and 64 neurons respectively, and the GELU activation function is used between each layer. After passing through this network, each set of operating conditions is mapped to a 64-bit feature vector, used to characterize the overall temperature characteristics under that operating condition. The GELU activation function is shown in the following equation: [Equation omitted for brevity] The cumulative distribution function represents the standard normal distribution. Compared to the traditional ReLU activation function, the GELU activation function has a continuous and smooth output characteristic when the input is close to zero. It does not completely truncate negative inputs, but assigns different retention weights according to the input magnitude, thereby improving the stability of gradient propagation while maintaining nonlinear expressiveness. This characteristic is beneficial to the convergence speed and prediction accuracy of neural networks in complex nonlinear regression tasks.

[0136]

[0137] S32. Constructing the Fully Connected Layer: Normalized spatial coordinates (x, y) are used as input, with an input dimension of 2. The coordinates of all spatial points under each working condition are flattened into a two-dimensional array and then fed into the fully connected layer. This network structure is the same as the working condition network, consisting of four fully connected layers with 64, 128, 128, and 64 neurons respectively, using the GELU activation function. The network output is a 64-dimensional spatial feature vector corresponding to each spatial point, used to describe the temperature distribution characteristics of that location in space.

[0138] S33. Fusion of operating condition features and spatial features: The operating condition feature vector obtained in step S31 is expanded in spatial dimension to match the number of spatial points. Then, it is multiplied element-wise with the spatial feature vector obtained in step S32 to modulate the spatial features by the operating condition information, so that the input under each operating condition corresponds to the temperature output under a specific coordinate.

[0139] S34. Constructing the Output Layer: Using the fused features as input, construct a fully connected regression layer with 1 neuron to output the predicted temperature value for each spatial point. This layer maps the 64-dimensional fused features into a single scalar, realizing the regression prediction from "operating condition + coordinates" to temperature, ultimately obtaining the complete temperature field distribution under each operating condition.

[0140] S35. Constructing the Loss Function: Since the output object is a continuous temperature value, which is a regression problem, Mean Squared Error (MSE) is used as the basic loss function. First, the prediction error of all spatial points under each working condition is averaged to obtain the overall error corresponding to that working condition. Then, weight coefficients are constructed based on the working condition parameters (such as current magnitude and seawater flow velocity), and the errors of different working conditions are weighted and summed. This makes the model pay more attention to key working conditions with poor heat dissipation or stronger heat generation during the training process, thereby improving the prediction accuracy and stability of the model under complex working conditions.

[0141] S36. Model Training and Parameter Update: The Adam optimization algorithm is used to update the model parameters, combined with an adaptive learning rate adjustment strategy. When the training loss no longer decreases within several rounds, the learning rate is automatically reduced to improve training stability and convergence. Simultaneously, the model parameters with the minimum loss are saved during training as the final optimal prediction model.

[0142] S4. When applying online, input a set of real-time operating condition outputs and predicted cable cross-sectional temperature distributions into the cable temperature neural network prediction model.

[0143] If it is necessary to output the cross-sectional temperature distribution of different cable structures, the cross-sectional parameters of the cable structures need to be trained. By using the cross-sectional parameters as input parameters for further training, the cross-sectional temperature distribution of different cable structures can be output.

[0144] In another embodiment, the cable temperature neural network prediction model is used to calculate the cable cross-sectional temperature distribution under different operating conditions. Then, any three sets of operating conditions are selected, and the cable cross-sectional temperature distribution obtained from the COMSOL simulation, the predicted cable cross-sectional temperature distribution, and the magnitude of the absolute temperature difference are compared. Figure 5 As shown. The three working conditions are: "T" w =33.931℃, I=2725.8A, v0=0.891m / s”, “T w =30.826℃, I=514.24A, v0=0.737m / s”, “ T w =15.683℃, I=867.56A, v0=1.955m / s”.

[0145] It can be seen that the absolute temperature difference between the cable cross-section temperature distribution calculated using the COMSOL model and the cable cross-section temperature distribution predicted by the neural network model does not exceed 3.6K, and the simulation accuracy of the neural network model reaches over 94.2%. Furthermore, the calculation time using the neural network model can be less than 1 minute. Compared to the traditional finite element model, using neural network calculations not only ensures calculation accuracy but also significantly reduces calculation time, making it suitable for rapid calculations in engineering applications.

[0146] Correction step: If the absolute difference between the cable cross-section temperature distribution calculated by the COMSOL model and the cable cross-section temperature predicted by the neural network exceeds 3.6K, then additional training data is needed for retraining.

[0147] To increase the training set size, you can either increase the number of COMSOL input parameters or narrow the range of input parameters to allow for retraining within a more precise range. Alternatively, you can increase the number of layers in the neural network or the number of neurons per layer.

[0148] In another embodiment, the present invention provides a method for rapid estimation of cable temperature field based on neural networks, which can monitor and reflect the temperature changes of cables in real time. The simulation method includes the following steps:

[0149] S1. Construct a two-dimensional cross-sectional model based on the cable structure;

[0150] S2. Based on different material properties, set the material and corresponding environmental parameters of the cable, and perform physical model mesh generation;

[0151] S3. Based on the electric field characteristics of the model, set the electric field of the model;

[0152] S4. Set the temperature field of the model according to the temperature field characteristics of the model;

[0153] S5. Based on the flow field characteristics of the model, set the flow field for the model;

[0154] S6. Set up a non-isothermal flow coupling field for the model;

[0155] S7. Set up the research interface and solve multiple cable models;

[0156] S8. Generate databases for different working conditions based on the solution results, which will serve as the database for the neural network model;

[0157] S9. Build a neural network model based on Python, divide the database into training set and test set, and train it for different operating conditions.

[0158] In another embodiment, the present invention provides a fast cable temperature field estimation system based on neural networks, comprising:

[0159] The model building module is used to build a two-dimensional cross-sectional model based on the cable structure, and to build a finite element model based on the two-dimensional cross-sectional model that couples the electric field, temperature field, flow field and external non-isothermal flow field.

[0160] The simulation calculation module is used to perform simulation calculations based on the finite element model and generate a training database.

[0161] The model training module is used to build a neural network model and train the neural network model based on the database to obtain a neural network prediction model for cable temperature.

[0162] An online application module is used during the online application phase to estimate the real-time temperature field distribution of the cable's two-dimensional cross-section using the cable temperature neural network prediction model based on real-time acquired operating condition parameters. In another embodiment, the present invention provides a neural network-based rapid cable temperature field estimation system, comprising: an offline training and modeling subsystem and an online prediction and monitoring subsystem; the offline training and modeling subsystem includes: a cable model construction module, a finite element simulation calculation module, a neural network training module, and a model management module.

[0163] The online prediction and monitoring subsystem includes: a real-time data access module, a temperature field prediction engine module, a visualization and analysis module, and a data storage and output module;

[0164] Based on the offline training and modeling subsystem and the online prediction and monitoring subsystem, a neural network-based rapid estimation system for cable temperature field is formed. The system can predict the temperature distribution of cable cross-section under any working condition in real time and can export the temperature distribution results.

[0165] In another embodiment, the present invention provides a fast cable temperature field estimation system based on neural networks, such as... Figure 7 and Figure 8 As shown, it includes: an offline training and modeling subsystem and an online prediction and monitoring subsystem;

[0166] The offline training and modeling subsystem includes:

[0167] The cable model building module is used to automatically generate a two-dimensional cross-sectional geometric model based on the cable type and structural parameters; and to configure multiphysics coupling parameters.

[0168] The finite element simulation calculation module is used to call finite element analysis software (such as COMSOL and ANSYS) to perform batch parameter simulations, automatically generate training samples covering different operating conditions (current, ambient temperature, wind speed, laying conditions, etc.), and output complete datasets such as temperature, electric field distribution, and flow field distribution at each node of the cable cross section.

[0169] The neural network training module provides neural network architecture design, supports custom layer numbers and activation functions; integrates training database management to achieve data normalization, sample partitioning, and data augmentation; supports distributed training and hyperparameter tuning, and outputs lightweight, high-performance prediction models.

[0170] The model management module is used to implement model version control, performance evaluation, and one-click deployment to the online system.

[0171] The online prediction and monitoring subsystem includes:

[0172] The real-time data access module is used to collect operating parameters such as cable current, ambient temperature, and seawater flow velocity in real time; and to perform data verification, including outlier detection, missing value handling, and data synchronization.

[0173] The temperature field prediction engine module loads a trained neural network model to achieve millisecond-level temperature field prediction and outputs a complete two-dimensional temperature field distribution matrix of the cable cross-section based on real-time operating parameters.

[0174] The visualization and analysis module dynamically displays temperature field cloud maps and supports temperature queries at any location on the cross-section; it also provides early warnings based on preset thresholds.

[0175] The data storage and output module stores prediction results, raw data, and alarm records; it also provides an API for external systems to call.

[0176] The software interface of the neural network-based rapid cable temperature field estimation system formed by the above two subsystems is as follows: Figure 9 As shown, the system can predict the temperature distribution of the cable cross section in real time under any working condition, and the temperature distribution results can be exported.

[0177] In another embodiment, the present invention discloses a computer-readable storage medium for storing a computer program configured to implement the method when invoked by a processor.

[0178] In another embodiment, the present invention discloses an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor; wherein the processor implements the method when executing the program.

[0179] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment / mode or example is included in at least one embodiment / mode or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0180] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0181] Those skilled in the art should understand that the above embodiments are merely for illustrating the present invention and are not intended to limit the scope of the invention. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present invention.

Claims

1. A method for rapid estimation of cable temperature field based on neural network, characterized in that, include: S1. Establish a two-dimensional cross-sectional model based on the cable structure, and establish a finite element model based on the two-dimensional cross-sectional model that couples the electric field, temperature field, flow field and external non-isothermal flow field; S2. Perform simulation calculations based on the finite element model to generate a training database; S3. Build a neural network model and train the neural network model based on the database to obtain a neural network prediction model for cable temperature; S4. In the online application stage, based on the real-time acquired operating condition parameters, the real-time two-dimensional cross-sectional temperature field distribution of the cable is estimated using the cable temperature neural network prediction model.

2. The method according to claim 1, characterized in that, Preferably, step S1 includes: A two-dimensional cross-sectional model is established based on the actual physical dimensions of the cable. This model includes the cable's main structure, such as the copper conductor, insulation layer, alloy lead layer, and outer sheath, as well as the external sea area. The entire model is physically meshed, and the mesh is refined in key areas such as the cable body and boundary layer.

3. The method according to claim 1, characterized in that, Step S2 includes: The electric field, temperature field for solid and fluid heat transfer, and laminar flow physical field are set for the two-dimensional cross-sectional model, and then coupled to form a multiphysics model. By setting different combinations of operating parameters, the multiphysics model is solved in steady state to generate the training database.

4. The method according to claim 3, characterized in that, In step S3, the neural network prediction model includes an input layer, at least one hidden layer, and an output layer; The number of nodes in the input layer corresponds to the number of operating condition parameters, and the output layer is used to output the temperature field distribution of the two-dimensional cross-section of the cable.

5. The method according to claim 1, characterized in that, The cable is a submarine cable, and the external non-isothermal flow field is a laminar flow field of the seawater.

6. The method according to claim 1, characterized in that, The neural network model uses mean squared error as the loss function and is trained using the Adam optimizer.

7. The method according to claim 1, characterized in that, Following the online application phase, a correction step is also included: The measured temperature of at least one physical temperature measurement point on the cable is compared with the predicted temperature of the corresponding location output by the cable temperature neural network prediction model. If the comparison deviation exceeds a preset threshold, the amount of data in the training database is increased or the structure of the neural network model is adjusted, and the model is retrained.

8. A fast cable temperature field estimation system based on neural networks, comprising: The model building module is used to build a two-dimensional cross-sectional model based on the cable structure, and to build a finite element model based on the two-dimensional cross-sectional model that couples the electric field, temperature field, flow field and external non-isothermal flow field. The simulation calculation module is used to perform simulation calculations based on the finite element model and generate a training database. The model training module is used to build a neural network model and train the neural network model based on the database to obtain a neural network prediction model for cable temperature. The online application module is used in the online application phase to estimate the real-time two-dimensional cross-sectional temperature field distribution of the cable based on the real-time acquired operating condition parameters and the cable temperature neural network prediction model.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program configured to implement the method of any one of claims 1-7 when invoked by a processor.

10. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor; wherein, when the processor executes the program, it implements the method of any one of claims 1-7.

Images