Cable temperature prediction method and device

Abstract

The invention provides a cable temperature prediction method and device. The method comprises the following steps: constructing a data set of each cable layer included in a cable by combining a predefined radius range of each cable layer; constructing a system model constructed by a temperature prediction model, a heat conduction equation of the cable and a loss function; the temperature prediction model predicts and outputs temperature prediction data based on the data set, and the heat conduction equation is configured based on a trainable cable material parameter group; the cable iteratively updates parameters of the system model based on the data set according to a loss function under the constraint of a boundary constraint condition, and performs gradient penalty correction when monitoring that parameters in the cable material parameter group do not conform to corresponding operation thresholds until a convergence condition is reached, so as to obtain a trained system model; and predicting the predicted temperature of the coordinates of the target point on the cable during the operation time of the target cable based on the system model, and if the predicted temperature reaches the evaluation index, determining system model parameters. The precision of the temperature prediction data is ensured, and the credibility of the temperature prediction data is also ensured.

Description

Technical Field

[0001] This application relates to the field of cable technology, and in particular to a method and apparatus for predicting cable temperature. Background Technology

[0002] In power transmission systems, cables serve as the carriers of energy, and their operating temperature directly affects the system's safety and economy. Exceeding the rated temperature of cables not only severely shortens insulation life but also, with long-term overheating accumulation, can lead to cable fires. Therefore, monitoring and predicting cable temperatures is of great importance. Summary of the Invention

[0003] In view of this, this application provides a cable temperature prediction method and apparatus, which aims to improve the accuracy of cable temperature monitoring and prediction.

[0004] In a first aspect, this application provides a method for predicting cable temperature, including:

[0005] Based on the cable's operational data and combined with the predefined radius range of each cable layer, a dataset of each cable layer is constructed. The operational data includes the coordinates of multiple spatial points of the cable under a preset current carrying capacity and the temperature at each point coordinate corresponding to different cable operating times under the current carrying capacity.

[0006] A system model is constructed, including a temperature prediction model, a heat conduction equation for the cable, and a loss function. The temperature prediction model predicts output temperature prediction data based on the dataset, and the heat conduction equation is based on a trainable set of cable material parameters.

[0007] Under boundary constraints, the parameters of the system model are iteratively updated based on the dataset and the loss function. When the parameters in the cable material parameter group do not meet the corresponding running threshold, gradient penalty correction is performed until the convergence condition is reached, and the trained system model is obtained.

[0008] Based on the system model, the predicted temperature of the target point coordinates on the cable during the target cable's operating time is predicted. If the predicted temperature reaches the evaluation index, the system model parameters are determined.

[0009] Optionally, the method further includes

[0010] Collect data during cable operation and process the collected data to obtain actual data. The data processing includes removing outliers from the collected data.

[0011] An electromagnetic thermal coupling simulation model of the cable is established based on its geometric and material parameters.

[0012] Based on the boundary conditions set by the finite element method for cables in actual working conditions and different current carrying capacities, the electromagnetic thermal coupling model is simulated and calculated using the actual data. If the electromagnetic thermal coupling model does not converge, the mesh generation and geometric parameter configuration of the cable in the finite element analysis are readjusted. If the electromagnetic thermal coupling model converges, the simulation data of the cable is output and used as the running data.

[0013] Optionally, the dataset for each cable layer included in the construction of the cable includes:

[0014] For a dataset of cable layers with a thickness less than a threshold, the data in the dataset of the cable layer is increased by rotation or mirroring.

[0015] Optionally, the temperature prediction model sequentially includes an input layer, multiple rounds of feature extraction units, and an output layer; the execution steps of each round of the feature extraction unit include:

[0016] The data received by the input layer is mapped to the first feature tensor of the target dimension through a fully connected layer. The data received by the output layer is a five-dimensional vector composed of the point coordinates, cable running time and current carrying capacity in the dataset.

[0017] The first feature tensor is nonlinearly transformed by the ReLU activation function, and the shape of the first feature tensor activated by the ReLU activation function is adjusted to fit the second feature tensor of the channel attention module.

[0018] The second feature tensor is processed by the channel attention module, and the channel weights generated by the two fully connected layers are applied to the corresponding feature channels in the channel attention module for weighted processing to obtain the third feature tensor. The third feature tensor is then mapped to 1-dimensional temperature prediction data and output through the output layer.

[0019] Optionally, the trainable set of cable material parameters includes the thermal conductivity λ, specific heat capacity c, density ρ, and the ratio K of the shielding layer induced current to the source current for each cable layer included in the cable.

[0020] The heat conduction equation is:

[0021] = ( ( )+ )+ ;

[0022] Among them, heat source item It is mainly generated by the source current in the cable core conductor and the induced current in the metal shielding layer, and is calculated according to the type of cable layer where data point i in the dataset is located, that is, the current q in the cable core conductor.i =I²N1, q in the metal shielding layer i =(KI)²N², where N1 is the resistance of the cable core conductor, N2 is the resistance of the metal shielding layer; I is the current carrying capacity, λ o c is the thermal conductivity of the o-th layer of the cable; o ρ is the specific heat capacity of the o-th layer of the cable. o denoted as the density of the o-th layer of the cable; T is the temperature value of data point i; t is the cable running time corresponding to data point i; z represents the axial length position of data point i in the cable; and r represents the radial position of the data point in the cable.

[0023] Optionally, the boundary constraints include:

[0024] Set the outermost boundary temperature constraint of the cable layer to T=20℃;

[0025] The initial operating time of the cable, and the temperature constraint at all points on the cable: T=20℃;

[0026] The continuity constraint of interlayer heat flow in cables is:

[0027] = ;

[0028] in, This represents the outer boundary of the o-th layer. This represents the inner boundary of the (o+1)th layer.

[0029] Optionally, the loss function is:

[0030] L=V1×L Data +V2×L PDE +V3×L IC +V4×L BC +V5×Lgrad;

[0031] Among them, L Data The data loss term represents the mean square error between the temperature prediction data output by the above temperature prediction model and the temperature data of the corresponding data points in the dataset, L. PDE The loss term in the heat conduction equation is the mean square error of the value on the side of the equation that should be zero after substituting the predicted temperature data into the equation, which has already had one side shifted to zero. L IC The initial loss term constrains the initial temperature of the cable during operation; L BC The boundary loss term constrains the operating environment temperature of the cable; Lgrad is the gradient constraint loss term; V1, V2, V3, V4, and V5 are preset weight values.

[0032] The iterative update of the system model parameters based on the loss function includes:

[0033] When the gradient norm of the loss function exceeds the norm threshold, the gradient is clipped.

[0034] The step of performing gradient penalty correction when the parameters in the cable material parameter group do not meet the corresponding operating threshold includes:

[0035] If one or more parameters in the monitoring cable material parameter group, such as thermal conductivity, specific heat capacity, and density, do not meet the corresponding physical threshold set according to the physical settings of the cable material, or if the induced current ratio K∉[0,1] of the monitoring metal shielding layer is not monitored, gradient penalty correction is performed.

[0036] Optionally, if the predicted temperature reaches the evaluation index, determining the system model parameters includes:

[0037] If the mean square error between the predicted temperature and the corresponding actual temperature at the target point coordinates does not exceed a preset first error threshold, and the mean absolute error between the predicted temperature and the corresponding actual temperature at the target point coordinates does not exceed a preset second error threshold, then the predicted temperature meets the evaluation criteria, and the system model parameters are determined.

[0038] Optionally, the method further includes:

[0039] Using the X, Y, and Z axes as spatial coordinates, color mapping represents temperature, and the boundary contours of each layer of the cable are superimposed to generate a temperature cloud map and radial temperature curve of the cable.

[0040] And / or, deploy the system model to the cable temperature prediction scenario, and when the deviation between the temperature predicted by the system model and the measured temperature exceeds the third error threshold, re-execute the parameters of the system model and subsequent steps based on the temperature data in the dataset and the loss function under the boundary constraint conditions.

[0041] Secondly, this application provides a cable temperature prediction device, the device comprising:

[0042] The processing unit is used to construct a dataset of each cable layer included in the cable based on the cable's operating data and in combination with the predefined radius range of each cable layer; the operating data includes the coordinates of multiple spatial points of the cable under a preset current carrying capacity and the temperature at each of the aforementioned coordinates corresponding to different cable operating times under the current carrying capacity;

[0043] The building unit is used to construct a system model for building a temperature prediction model, a heat conduction equation for the cable, and a loss function; the temperature prediction model predicts output temperature prediction data based on the dataset, and the heat conduction equation is based on a trainable set of cable material parameters.

[0044] The training unit is used to iteratively update the parameters of the system model based on the dataset and the loss function under the boundary constraint conditions of the cable, and to perform gradient penalty correction when the parameters in the cable material parameter group do not meet the corresponding running threshold, until the convergence condition is reached, so as to obtain the trained system model.

[0045] The determining unit is used to predict the temperature of the target point coordinates on the cable during the running time of the target cable based on the system model. If the predicted temperature reaches the evaluation index, the system model parameters are determined.

[0046] Thirdly, this application provides an apparatus comprising a memory and a processor, the memory for storing instructions or code, and the processor for executing the instructions or code to cause the apparatus to perform a cable temperature prediction method as described in any of the first aspects above.

[0047] Fourthly, this application provides a computer storage medium storing code, wherein when the code is executed, a device running the code implements a cable temperature prediction method as described in any of the first aspects above.

[0048] This application provides a method and apparatus for predicting cable temperature. When executing the method, a dataset is constructed based on the cable's operational data and a predefined radius range for each cable layer. The operational data includes the coordinates of multiple spatial points on the cable under a preset current carrying capacity, and the temperature at each coordinate corresponding to different cable operating times under the current carrying capacity. A temperature prediction model, a heat conduction equation for the cable, and a system model constructed using a loss function are constructed. The temperature prediction model predicts output temperature prediction data based on the dataset, and the heat conduction equation is configured based on a trainable set of cable material parameters. Under boundary constraints, the parameters of the system model are iteratively updated based on the dataset and the loss function. Gradient penalty correction is performed when the parameters in the cable material parameter set do not meet the corresponding operating threshold, until convergence is achieved, resulting in a trained system model. The predicted temperature of a target point coordinate on the cable during the target cable's operating time is predicted based on the system model. If the predicted temperature reaches an evaluation index, the system model parameters are determined. Thus, this application configures a trainable set of cable material parameters, automatically learns the thermal properties of cable materials, solves the parameter drift problem of traditional models, and maintains stable prediction accuracy even after the cable has been in service for many years. Therefore, the temperature prediction model is combined with the heat conduction equation constructed based on a trainable set of cable material parameters to build a multivariate loss function that combines data constraints and physical property constraints. After multiple rounds of iterative training, a system model with good generalization ability is formed, ensuring both the accuracy and reliability of the data. This allows for high-precision dynamic tracking of the temperature field even under complex working conditions. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A schematic flowchart illustrating a cable temperature prediction method provided in an embodiment of this application;

[0051] Figure 2 This application provides a schematic diagram of a cable temperature prediction process in an embodiment of the present application.

[0052] Figure 3 A schematic diagram of the temperature field of a cable provided in this application embodiment;

[0053] Figure 4 This is a schematic diagram of a cable temperature prediction device provided in an embodiment of this application. Detailed Implementation

[0054] Currently, cable temperature monitoring technologies mainly present two major approaches: traditional thermal circuit modeling and data-driven methods. Traditional thermal circuit modeling primarily constructs an equivalent thermal circuit network based on Fourier's law of heat conduction, treating each layer of the cable as a series structure of thermal resistance and heat capacity. This method relies on accurate material thermophysical parameters, but in reality, factors such as cable aging can cause parameter drift, making it impossible to obtain the true properties of the material in real time. Furthermore, Fourier's law of heat conduction is insufficient for modeling nonlinear boundary conditions, making it difficult to handle complex laying environments, such as multiple cables laid side-by-side or uneven soil thermal resistance, and failing to encompass all complex external conditions. Data-driven methods rely on various neural networks for cable temperature prediction, but these methods require a large amount of labeled data. Cable operation data is often limited by sensor deployment density, especially for underground cables where dense deployment is difficult, thus often resulting in insufficient effective data in practical engineering. Moreover, data-driven temperature prediction models lack physical constraints, leading to poor extrapolation capabilities. When operating conditions exceed the range of training data, predictions fail, making it difficult to cope with complex and variable operating conditions.

[0055] To address the aforementioned issues, this application provides a method and apparatus for predicting cable temperature. This method configures a trainable set of cable material parameters to learn the thermal properties of the cable material, thus solving the parameter drift problem of traditional models and maintaining stable prediction accuracy even after many years of cable service. Therefore, the temperature prediction model is combined with a heat conduction equation constructed based on the trainable set of cable material parameters to build a multivariate loss function that combines data constraints and physical property constraints. After multiple rounds of iterative training, a system model with good generalization ability is formed, ensuring both data accuracy and reliability. This allows for high-precision dynamic tracking of the temperature field even under complex operating conditions.

[0056] To provide a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this disclosure. In the following technical description, for ease of explanation, several details are used to provide a full understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other cases, well-known structures and devices may be simplified in their depiction to simplify the drawings.

[0057] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be used interchangeably where appropriate. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0058] Unless otherwise stated, the term "multiple" means two or more. In embodiments of this disclosure, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B means: A or B. The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or, A and B.

[0059] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0060] See Figure 1 The diagram shows a flowchart of a cable temperature prediction method. The cable temperature prediction method includes:

[0061] S101. Based on the cable's operating data, construct a dataset of each cable layer included in the cable; the operating data includes the coordinates of multiple spatial points of the cable under a preset current carrying capacity and the temperature value of each point coordinate corresponding to the current carrying capacity under different cable operating times.

[0062] The aforementioned operational data may include data collected directly from the operating cables.

[0063] Of course, the aforementioned operational data can also include data generated from acquired data through radial symmetry transformations (rotation, mirroring) and / or physical extrapolations based on heat conduction equations. This data augmentation is particularly useful for cable layers with thicknesses less than a thickness threshold, such as semiconductor and metal shielding layers.

[0064] The above operating data may include the coordinates (x, y, z) of multiple spatial points collected under the target ambient temperature (e.g., 20 degrees Celsius), based on the preset current carrying capacity I, and the temperature T collected at each point coordinate during different cable operating times t.

[0065] The above-mentioned operational data can be the actual data after removing outliers such as temperature jumps and coordinate errors from the collected data; for example, it can be checked whether it conforms to physical laws (such as conductor temperature > insulation layer temperature, increased current carrying capacity and temperature rise, etc.) and unreasonable data can be removed.

[0066] It should be noted that this application requires the pre-configuration of relevant cable parameters, which may include: cable geometric parameters (radius of each layer), cable material parameter group (thermal conductivity, heat capacity, density, etc. of each layer) and initialization, ambient temperature 20℃, and the ratio K of induced current to source current, the initial value of which can be set to 0.2.

[0067] For example, the aforementioned cable can be a five-layer single-core cylindrical long straight cable, consisting of a cable core conductor, a semiconductor shielding layer, an XLPE insulation layer, a metallic shielding layer, and an outer sheath. In one example, the radius range of each cable layer can be defined, and a Boolean mask function can then be used to generate the dataset for each layer. In another example, the temperatures in the actual data are assigned to the corresponding cable layers based on their corresponding point coordinates. Specifically, the radius corresponding to the coordinate data of the i-th data point (including point coordinates, cable runtime, and current carrying capacity) can be:

[0068] ;

[0069] in, This represents the x-coordinate of the radial section of the i-th data point. This represents the ordinate of the radial section of the i-th data point.

[0070] Then, the calculated radius is compared with the defined radius range of each cable layer to determine which radius range the calculated radius falls into, and thus determine the cable layer where the data point corresponding to the calculated radius is located.

[0071] Furthermore, define and initialize the cable material parameter set. Specific steps include: the trainable cable material parameter set can include 16 material parameters, such as the thermal conductivity λ, specific heat capacity c, density ρ, and the ratio K of the shielding layer induced current to the source current for each cable layer.

[0072] Optionally, the initial values ​​of the material parameters for each cable layer can be determined based on the material properties obtained from the material library.

[0073] It should be noted that the cable material parameters defined above in this application are based on the geometry of the five-layer cable as defined above. These cable material parameters can be configured into each layer of the cable structure.

[0074] In one example, the data points of each cable layer included in the cable are summed to determine if it equals the total number of data points in the running data, ensuring that no data is lost.

[0075] In one possible implementation, see Figure 2The diagram illustrates an application process for cable temperature prediction, and the specific implementation of a cable temperature prediction method is as follows:

[0076] Step A1: First, process the raw data by collecting data during cable operation. Clean the collected data (removing outliers) and extract the effective spatial point coordinates and temperatures at different current-carrying lengths for different cable operating periods to obtain actual data. Then, construct a finite element model based on the cable's geometric parameters (conductor dimensions, radius range of each cable layer, etc.) and material parameters (conductivity, thermal conductivity, specific heat capacity, etc. of each cable layer) to establish an electromagnetic-thermal coupling simulation model. Finally, based on the boundary conditions set by the finite element method in actual operating conditions and different current-carrying capacities, combine the actual data to perform simulation calculations on the electromagnetic-thermal coupling model. If the electromagnetic-thermal coupling model does not converge, readjust the cable mesh and geometric parameter configuration in the finite element analysis. If the electromagnetic-thermal coupling model converges, export the cable finite element simulation data and use this simulation data as the operating data.

[0077] Step A2 involves combining the running data with the predefined radius range of each cable layer to divide the simulation data into the corresponding cable layers. Then, the data is expanded by mirroring or rotating to increase the data, especially for thin-layer structures (such as semiconductor shielding layers and metal shielding layers), forming a dataset for each cable layer.

[0078] Of course, actual data and simulation data can also be combined as running data, and the dataset of each cable layer can be determined by combining the predefined radius range of each cable layer.

[0079] The above-mentioned data expansion is achieved by adding thin-layer cable structures through mirroring or rotation. For example, point coordinates can be rotated ±10° in the xy plane and mirrored along the z-axis to expand the dataset.

[0080] Step A3: Divide the above dataset into a training set and a test set in an 8:2 ratio.

[0081] The training set mentioned above can be used for the model to learn the temperature distribution characteristics of each layer, and the test set can be used to verify the model's generalization ability and prediction accuracy.

[0082] Therefore, when dealing with cable models with significant differences in layer thickness, the temperature distribution data obtained from finite element simulation is prone to imbalance, resulting in insufficient data for thin layers. This application addresses this by performing data augmentation processing to achieve radial layering of the data and supplement the data for thin layers.

[0083] Furthermore, the dataset is converted into PyTorch tensors for easier subsequent computation.

[0084] Optionally, the coordinate data in the dataset is converted into a tensor of shape [N,3], and the temperature data is converted into a tensor of shape [N,1].

[0085] S102, Construct a system model that includes a temperature prediction model, a heat conduction equation for the cable, and a loss function; the temperature prediction model predicts output temperature prediction data based on the dataset, and the heat conduction equation is based on a trainable set of cable material parameters.

[0086] In one example, a temperature prediction model is built.

[0087] The temperature prediction model includes, in sequence, an input layer, a multi-round feature extraction unit, and an output layer;

[0088] The execution structure of the feature extraction unit may include:

[0089] Step B1: Map the data received by the input layer to the first feature tensor of the target dimension through a fully connected layer.

[0090] The data received by the input layer includes the coordinates (x, y, z) of points in the dataset, the cable running time t, and the current carrying capacity I.

[0091] Optionally, the target dimension can be 20 dimensions.

[0092] Step B2: Perform a nonlinear transformation on the first feature tensor using the ReLU activation function, and adjust the shape of the first feature tensor activated by the ReLU activation function to fit the second feature tensor of the channel attention module.

[0093] Optionally, the expression for the ReLU activation function above can be ReLU(x) = max(0,x).

[0094] Optionally, the first feature tensor activated by the ReLU activation function can be adjusted to a shape of (target dimension, 1). For example, if the target dimension is 20, the first feature tensor can be adjusted to a shape of (20, 1) to obtain the second feature tensor after shape adjustment.

[0095] Step B3: Process the second feature tensor through the channel attention module, and apply the channel weights generated by the two fully connected layers to the corresponding feature channels in the channel attention module for weighted processing to obtain the third feature tensor. The third feature tensor is then mapped to 1-dimensional temperature prediction data (x,y,z,t,T) through the output layer and output.

[0096] The channel attention module extracts channel features through global average pooling and max pooling, generates channel weights Mc through a two-layer fully connected network, and performs weighted enhancement of channel features for each feature channel. The specific expression is as follows:

[0097] Att(x) = x × Mc.

[0098] Based on the temperature prediction model described above, the corresponding temperature prediction data is output from the dataset. This can be achieved as follows: First, the input layer can be a 5-dimensional input layer, receiving the coordinates (x, y, z), cable runtime t, and current carrying capacity I from the dataset, i.e., inputting a 5-dimensional vector (x, y, z, t, I) into the input layer. Then, after multiple rounds of feature extraction units (each round of feature extraction units executes steps A1-A3 once), optionally, these multiple rounds can be 3 rounds. Steps A1 (fully connected layer mapping to a feature tensor of the target dimension), A2 (ReLU activation function and shape adjustment), and A3 (channel attention module) are repeated through 3 hidden layers to extract deep features. Finally, the third feature tensor is mapped to 1-dimensional temperature prediction data (x, y, z, t, T) through the output layer and output.

[0099] In one example, constructing the heat conduction equation can include:

[0100] Step C1: Define and initialize the trainable cable material parameter set. Specific steps include: the trainable cable material parameter set can include 16 material parameters, such as the thermal conductivity λ, specific heat capacity c, density ρ, and the ratio K of the shielding layer induced current to the source current for each cable layer.

[0101] Optionally, the initial values ​​of the material parameters for each cable layer can be determined based on the material properties obtained from the material library.

[0102] The data for defining the cable material parameter group in step C1 above is the same as the cable material parameter group mentioned in step S101.

[0103] Step C2: Substitute each material parameter in the above cable material parameter group into the heat conduction equation constructed based on Fourier's heat transfer law, according to the cable layer corresponding to that material parameter. The specific expression is as follows:

[0104] u·∇T+ +∇·q=Q;

[0105] Where ∇ represents the gradient, ∇· represents the divergence, u represents the fluid velocity field, q represents the heat flux density, T represents the cable temperature, and Q represents the cable heat power density.

[0106] Furthermore, the above heat conduction equation is converted into an explicit expression of material parameters, as follows:

[0107] = ( ( )+ )+ ;

[0108] Among them, heat source item It is mainly generated by the source current in the cable core conductor and the induced current in the metal shielding layer, and is calculated according to the type of cable layer where data point i in the dataset is located, that is, the current q in the cable core conductor. i =I²N1, q in the metal shielding layer i =(KI)²N², where N1 is the resistance of the cable core conductor, N2 is the resistance of the metal shielding layer; I is the current-carrying capacity value input to the temperature prediction model input layer, λ o c is the thermal conductivity of the o-th layer of the cable; o ρ is the specific heat capacity of the o-th layer of the cable. o denoted as the density of the o-th layer of the cable; T is the temperature value of data point i; t is the cable running time corresponding to data point i; z represents the axial length position of data point i in the cable; and r represents the radial position of the data point in the cable.

[0109] Step C3: Set the boundary constraints for the cable. For example, the outermost boundary temperature constraint of the cable layer can be set to T=20℃ (the temperature of data points outside the outermost layer of the cable is 20℃). The initial cable running time and the temperature constraint at all points on the cable are also set to T=20℃. The interlayer heat flow continuity constraint is as follows:

[0110] = ;

[0111] in, This represents the outer boundary of the o-th layer. This represents the inner boundary of the (o+1)th layer.

[0112] In one example, step D, constructing the multivariate loss function, can be:

[0113] L=L Data +L PDE +L IC +L BC +Lgrad.

[0114] Among them, L Data The data loss term represents the mean square error between the temperature prediction data output by the above temperature prediction model and the corresponding actual temperature data in the dataset. PDEThe loss term in the heat conduction equation is the mean square error of the value on the side of the equation that should be zero after substituting the predicted temperature data into the equation, which has already had one side shifted to zero. L IC As the initial loss term, the accuracy of the initial temperature is locked in through continuous constraints during the training phase, providing a physically-compliant starting point for predicting the entire temperature evolution process. During model training, the L for the t=0 sample needs to be calculated simultaneously when calculating the total loss L in each iteration. IC Error. During training, L IC This will continuously drive the model to adjust parameters (such as the weights of fully connected layers) to ensure that the predicted temperature at t=0 is within 20°C. It also constrains the initial temperature of the cable during operation. BC This is the boundary loss term, representing the mean square error between the predicted temperature and the ambient temperature at each point on the outermost layer of the cable, used to constrain the cable's operating ambient temperature. L grad A gradient-constrained loss term is used, leveraging the characteristics of the ReLU activation function to prevent radial temperature increase. Thus, the introduction of a multivariate loss function increases the interpretability of the temperature prediction model, providing multiple safeguards for the reliability of the prediction results.

[0115] Optionally, considering the varying importance of different loss terms in the loss function, targeted reinforcement can be achieved by adding weights V to the loss terms, assigning higher weights to more important loss terms. For example, L = V1 × L Data +V2×L PDE +V3×L IC +V4×L BC +V5×Lgrad.

[0116] S103. Under the constraint of boundary conditions, the parameters of the system model are iteratively updated based on the dataset and the loss function. When the parameters in the cable material parameter group do not meet the corresponding running threshold, gradient penalty correction is performed until the convergence condition is reached, and the trained system model is obtained.

[0117] Based on the dataset, the system model is iteratively trained. In each round of iterative training, the parameters of the system model (formed by combining the temperature prediction model, heat conduction equation, and multivariate loss function constructed above) are adjusted. Specifically, forward propagation is performed to calculate the temperature prediction data of each cable layer, calculate the five loss terms in the loss function and sum them according to the weights of each loss term, update the system model (including the temperature prediction model parameters and the material parameters of the cable material parameter group) through backpropagation, and perform gradient clipping (norm ≤ 1.0) to prevent explosion.

[0118] The convergence condition mentioned above can be that the number of iterations reaches the target value, such as 3000 rounds of training iterations.

[0119] Optionally, the above iterative training process can be performed by initializing the Adam optimizer and learning rate scheduler, setting the initial learning rate to 0.01, and enabling automatic differentiation to calculate the temperature gradient and Laplace term.

[0120] The aforementioned gradient clipping (norm ≤ 1.0) refers to scaling the gradient when the calculated gradient norm exceeds 1.0, ultimately keeping the gradient norm below 1.0 to prevent excessively large gradient values ​​from causing training explosion and ensuring stable convergence during model training. When the number of iterations reaches a preset threshold or the loss function value tends to stabilize, the convergence condition is met, and the trained temperature prediction model is obtained for subsequent accurate prediction of cable temperature fields and online monitoring of thermal status.

[0121] Optionally, the average loss on the test set can be calculated and output to the training log after a preset number of iterations (e.g., 100 or 200). In one example, after a preset number of iterations (e.g., 100 or 200), the values ​​of each loss term, the values ​​of the cable material parameter set for the next round, and the ratio K of the induced current to the source current in the metal shield can be printed. On the one hand, the reasonableness of the physical properties of the material parameters in the cable material parameter set is verified in real time during the iterative training process; specifically, the thermal conductivity λ of the cable material parameter set is mainly monitored. i Specific heat capacity c i Density ρ i The optimization results are ensured to conform to the physical properties of the cable material. Simultaneously, the ratio of the induced current to the source current in the metal shielding layer, K∈[0,1], is checked; gradient penalty correction is applied when it exceeds the preset range. The physical property constraints and gradient penalty mechanism work together to ensure that the material parameters are within a reasonable range, thus preventing the iterative update results from deviating from the true physical property range. On the other hand, monitoring the changes in various loss terms allows users to better understand the model's learning progress and also verifies whether the weights V1-V5 of the loss terms are set reasonably. Optionally, considering the different importance of multiple loss terms in the loss function, targeted strengthening can be achieved by adding weights V to the loss terms, setting higher weights for important loss terms in the loss function. For example, L=V1×L Data +V2×L PDE +V3×L IC +V4×L BC +V5×Lgrad. Meanwhile, if L Data With L PDE If a clear discrepancy of "one rising and one falling" occurs, it is necessary to check whether the condition settings of the finite element model match the heat transfer equation (such as checking the geometric structure parameters, material parameters, current carrying capacity settings, etc. of the cable model in the finite element model), check the implementation logic of the heat conduction equation (such as whether the term shifting is correct, whether the heat source term qi is bound to the layer corresponding to the cable), and re-verify the loss function calculation code, etc.

[0122] Thus, during the training process of this temperature prediction model, the heat conduction equation constructed from the trainable cable material parameter set is also introduced for iterative training, taking into account both theoretical accuracy and adaptability to actual working conditions, and enabling high-precision dynamic deduction of the internal temperature field of the multi-layer cable structure.

[0123] S104. Based on the system model, predict the temperature of the target point coordinates on the cable during the running time of the target cable. If the predicted temperature reaches the evaluation index, determine the system model parameters.

[0124] The coordinates of the target point mentioned above can include multiple points.

[0125] Step E1: After the above system model is built and trained, input the point coordinates (x, y, z), cable running time t, and current carrying capacity I of multiple target points to generate a four-dimensional input vector (x, y, z, t, I). Perform forward propagation of the system model according to the radial layer of multiple target point coordinates. Finally, use the Softmax function to weight and integrate the prediction results of each layer (x, y, z, t, T) to obtain the predicted temperature T of each target point coordinate during the target cable running time.

[0126] Optionally, the temperature prediction data can be saved to a text file in the format [x,y,z,t,T].

[0127] Step E2: Determine whether the predicted temperature meets the evaluation criteria.

[0128] If the mean square error between the predicted temperature and the corresponding actual temperature at the target point coordinates does not exceed a preset first error threshold, and the mean absolute error between the predicted temperature and the corresponding actual temperature at the target point coordinates does not exceed a preset second error threshold, then the predicted temperature meets the evaluation criteria. The temperature prediction model parameters are then determined, the system model training is completed and the system model parameters are saved, and subsequent step E3 is executed.

[0129] If the mean square error between the predicted temperature and the corresponding actual temperature at the target point exceeds the preset first error threshold, or the mean absolute error between the predicted temperature and the corresponding actual temperature at the target point exceeds the preset second error threshold, then the above steps S103-S104 are executed again to adjust the training parameters (such as adjusting the values ​​of patience and factor in the learning rate scheduler) and retrain until the system model output meets the evaluation criteria.

[0130] Optionally, the first error threshold can be 5 percent, and the second error threshold can be 3.5 percent.

[0131] Step E3: Deploy the system model to a cable temperature prediction scenario (e.g., a cable online monitoring system), test the real-time prediction performance (e.g., compare the predicted temperature with the sensor measured temperature under different current carrying capacities), and if the deviation between the predicted temperature and the measured temperature of the system model exceeds the third error threshold (e.g., 5%), re-execute steps S103-S104 to trigger system model fine-tuning.

[0132] In this way, the model can be continuously optimized and dynamically updated by regularly introducing the latest monitoring data for incremental training, ensuring that the prediction accuracy in long-term operation is not affected by time-varying factors.

[0133] Optionally, the error between the predicted temperature and the actual temperature after substituting them into the equation should not exceed a preset fourth error threshold. This ensures that the predicted temperature meets the physical laws of cable heat transfer and can be disregarded as a direct evaluation indicator.

[0134] Optional, see Figure 3 This diagram illustrates a visualization of the temperature field of a cable. Matplotlib can be used to construct a 3D temperature field visualization, using the X, Y, and Z axes as spatial coordinates, color mapping to represent temperature, and overlaying the boundary outlines of each cable layer to generate temperature cloud maps and radial temperature curves. These are then saved as PNG images to visually display the internal temperature distribution characteristics of the cable. This multi-dimensional visualization allows maintenance personnel to intuitively grasp the cable's health status, promptly identify potential overheating risks, and provide strong support for the safe and stable operation of the power system.

[0135] This application configures a trainable set of cable material parameters to automatically learn the thermal properties of cable materials, solving the parameter drift problem of traditional models and maintaining stable prediction accuracy even after many years of cable service. Therefore, the temperature prediction model is combined with the heat conduction equation constructed based on the trainable set of cable material parameters to build a multivariate loss function that combines data constraints and physical property constraints. After multiple rounds of iterative training, a system model with good generalization ability is formed, ensuring both data accuracy and reliability. This allows for high-precision dynamic tracking of the temperature field even under complex operating conditions.

[0136] The above describes some specific implementations of the cable temperature prediction method provided in this application. Based on this, this application also provides a corresponding device. The device provided in this application will be described below from the perspective of functional modularity.

[0137] See Figure 4 The diagram shows a structural schematic of a cable temperature prediction device. The cable temperature prediction device includes:

[0138] The processing unit 401 is used to construct a dataset of each cable layer included in the cable based on the cable's operating data and in combination with the predefined radius range of each cable layer; the operating data includes the coordinates of multiple spatial points of the cable under a preset current carrying capacity and the temperature at each of the above coordinates corresponding to different cable operating times under the current carrying capacity.

[0139] The construction unit 402 is used to construct a system model for building a temperature prediction model, a heat conduction equation for the cable, and a loss function; the temperature prediction model predicts output temperature prediction data based on the dataset, and the heat conduction equation is based on a trainable set of cable material parameters.

[0140] Training unit 403 is used to iteratively update the parameters of the system model based on the dataset and the loss function under the boundary constraint conditions of the cable, and to perform gradient penalty correction when the parameters in the cable material parameter group do not meet the corresponding running threshold, until the convergence condition is reached, so as to obtain the trained system model.

[0141] The determining unit 404 is used to predict the temperature of the target point coordinates on the cable during the running time of the target cable based on the system model. If the predicted temperature reaches the evaluation index, the system model parameters are determined.

[0142] Based on the aforementioned device, a trainable set of cable material parameters is configured within the construction unit. Subsequent training units, combined with the heat conduction equation, automatically learn the thermal properties of the cable material, resolving the parameter drift problem of traditional models and maintaining stable prediction accuracy even after many years of cable service. Therefore, the temperature prediction model of the training unit, combined with the heat conduction equation constructed based on the trainable set of cable material parameters, establishes a multivariate loss function that combines data constraints and physical property constraints. Through multiple rounds of iterative training, a system model with excellent generalization ability is formed, ensuring both data accuracy and reliability. This allows for high-precision dynamic tracking of the temperature field even under complex operating conditions.

[0143] In one possible implementation, the device further includes a data generation unit, which is used to collect data during cable operation and process the collected data to obtain actual data. The data processing includes removing outliers from the collected data; establishing an electromagnetic thermal coupling simulation model of the cable based on the cable's geometric and material parameters; performing simulation calculations on the electromagnetic thermal coupling model based on the boundary conditions set by the finite element method and different current carrying capacities of the cable under actual operating conditions, combined with the actual data; readjusting the cable's mesh generation and geometric parameter configuration in the finite element analysis if the electromagnetic thermal coupling model does not converge; and outputting the cable's simulation data if the electromagnetic thermal coupling model converges, using the simulation data as the operating data.

[0144] In one possible implementation, the processing unit is specifically used to increase the data in the dataset of a cable layer with a thickness less than a threshold by rotating or mirroring the dataset.

[0145] In one possible implementation, the temperature prediction model sequentially includes an input layer, a multi-round feature extraction unit, and an output layer. The construction unit is used to map the data received by the input layer into a first feature tensor of the target dimension through a fully connected layer. The data received by the output layer is a five-dimensional vector composed of point coordinates, cable running time, and current carrying capacity in the dataset. The first feature tensor is nonlinearly transformed by the ReLU activation function, and the shape of the first feature tensor activated by the ReLU activation function is adjusted to fit a second feature tensor of the channel attention module. The second feature tensor is processed by the channel attention module, and the channel weights generated by the two fully connected layers are applied to the corresponding feature channels in the channel attention module for weighted processing to obtain a third feature tensor. The third feature tensor is then mapped into 1-dimensional temperature prediction data and output through the output layer.

[0146] In one possible implementation, the trainable set of cable material parameters includes the thermal conductivity λ, specific heat capacity c, density ρ, and the ratio K of the shielding induced current to the source current for each cable layer included in the cable.

[0147] The heat conduction equation is:

[0148] = ( ( )+ )+ ;

[0149] Among them, heat source item It is mainly generated by the source current in the cable core conductor and the induced current in the metal shielding layer, and is calculated according to the type of cable layer where data point i in the dataset is located, that is, the current q in the cable core conductor. i =I²N1, q in the metal shielding layer i =(KI)²N², where N1 is the resistance of the cable core conductor, N2 is the resistance of the metal shielding layer; I is the current carrying capacity, λ o c is the thermal conductivity of the o-th layer of the cable; o ρ is the specific heat capacity of the o-th layer of the cable. o denoted as the density of the o-th layer of the cable; T is the temperature value of data point i; t is the cable running time corresponding to data point i; z represents the axial length position of data point i in the cable; and r represents the radial position of the data point in the cable.

[0150] In one possible implementation, the boundary constraints include:

[0151] Set the outermost boundary temperature constraint of the cable layer to T=20℃;

[0152] The initial operating time of the cable, and the temperature constraint at all points on the cable: T=20℃;

[0153] The continuity constraint of interlayer heat flow in cables is:

[0154] = ;

[0155] in, This represents the outer boundary of the o-th layer. This represents the inner boundary of the (o+1)th layer.

[0156] In one possible implementation, the loss function is:

[0157] L=V1×L Data +V2×L PDE +V3×L IC +V4×L BC +V5×Lgrad;

[0158] Among them, L Data The data loss term represents the mean square error between the temperature prediction data output by the above temperature prediction model and the temperature data of the corresponding data points in the dataset, L. PDE The loss term in the heat conduction equation is the mean square error of the value on the side of the equation that should be zero after substituting the predicted temperature data into the equation, which has already had one side shifted to zero. L IC The initial loss term constrains the initial temperature of the cable during operation; L BC The boundary loss term constrains the operating environment temperature of the cable; Lgrad is the gradient constraint loss term; V1, V2, V3, V4, and V5 are preset weight values.

[0159] In one possible implementation, the training unit is specifically used to clip the gradient when the gradient norm of the loss function exceeds the norm threshold; monitor one or more parameters in the cable material parameter set, such as thermal conductivity, specific heat capacity, and density, that do not meet the corresponding physical threshold set according to the physical settings of the cable material, or monitor the induced current ratio K∉[0,1] of the metal shielding layer and perform gradient penalty correction.

[0160] In one possible implementation, the determining unit is specifically used to determine the system model parameters if the mean square error between the predicted temperature and the corresponding actual temperature of the target point coordinates does not exceed a preset first error threshold, and the average absolute error between the predicted temperature and the corresponding actual temperature of the target point coordinates does not exceed a preset second error threshold, then the predicted temperature meets the evaluation index.

[0161] In one possible implementation, the device further includes a display unit, which is used to generate a temperature cloud map and a radial temperature curve of the cable by using the X-axis, Y-axis and Z-axis as spatial coordinates, color mapping to represent temperature, superimposing the boundary contours of each layer of the cable.

[0162] In one possible implementation, the apparatus further includes an update unit, which is configured to deploy the system model to the cable temperature prediction scenario, and, when the deviation between the predicted temperature and the measured temperature of the system model exceeds a third error threshold, re-execute the parameter update of the system model and subsequent steps based on the dataset and the loss function under the boundary constraint conditions.

[0163] This application also provides corresponding devices and computer storage media for implementing the solutions provided in this application.

[0164] The device includes a memory and a processor. The memory stores instructions or code, and the processor executes the instructions or code to enable the device to perform a cable temperature prediction method according to any embodiment of this application.

[0165] The computer storage medium stores code, and when the code is executed, the device running the code implements a cable temperature prediction method according to any embodiment of this application.

[0166] In the embodiments of this application, the terms "first" and "second" (if they exist) are used only as name identifiers and do not represent the order of first and second.

[0167] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus a general-purpose hardware platform. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a router) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0168] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0169] The above description is merely an exemplary implementation of this application and is not intended to limit the scope of protection of this application.

Claims

1. A method for predicting cable temperature, characterized in that, include: Based on the operational data of the cable, and combined with the predefined radius range of each cable layer, a dataset of each cable layer included in the cable is constructed. The operating data includes the coordinates of multiple spatial points of the cable under a preset current carrying capacity, and the temperature at different current carrying capacities corresponding to different cable operating times at each coordinate. A system model is constructed, including a temperature prediction model, a heat conduction equation for the cable, and a loss function. The temperature prediction model predicts output temperature prediction data based on the dataset, and the heat conduction equation is based on a trainable set of cable material parameters. Under boundary constraints, the parameters of the system model are iteratively updated based on the dataset and the loss function. When the parameters in the cable material parameter group do not meet the corresponding running threshold, gradient penalty correction is performed until the convergence condition is reached, and the trained system model is obtained. Based on the system model, the predicted temperature of the target point coordinates on the cable during the target cable's operating time is predicted. If the predicted temperature reaches the evaluation index, the system model parameters are determined.

2. The method according to claim 1, characterized in that, The method also includes Collect data during cable operation and process the collected data to obtain actual data. The data processing includes removing outliers from the collected data. An electromagnetic thermal coupling simulation model of the cable is established based on its geometric and material parameters. Based on the boundary conditions set by the finite element method for cables in actual working conditions and different current carrying capacities, the electromagnetic thermal coupling model is simulated and calculated using the actual data. If the electromagnetic thermal coupling model does not converge, the mesh generation and geometric parameter configuration of the cable in the finite element analysis are readjusted. If the electromagnetic thermal coupling model converges, the simulation data of the cable is output and used as the running data.

3. The method according to claim 1, characterized in that, The datasets for each cable layer included in the construction of the cable include: For a dataset of cable layers with a thickness less than a threshold, the data in the dataset of the cable layer is increased by rotation or mirroring.

4. The method according to claim 1, characterized in that, The temperature prediction model sequentially includes an input layer, multiple rounds of feature extraction units, and an output layer; the execution steps of each round of the feature extraction unit include: The data received by the input layer is mapped to the first feature tensor of the target dimension through a fully connected layer. The data received by the output layer is a five-dimensional vector composed of the point coordinates, cable running time and current carrying capacity in the dataset. The first feature tensor is nonlinearly transformed by the ReLU activation function, and the shape of the first feature tensor activated by the ReLU activation function is adjusted to fit the second feature tensor of the channel attention module. The second feature tensor is processed by the channel attention module, and the channel weights generated by the two fully connected layers are applied to the corresponding feature channels in the channel attention module for weighted processing to obtain the third feature tensor. The third feature tensor is then mapped to 1-dimensional temperature prediction data and output through the output layer.

5. The method according to claim 1, characterized in that, The trainable set of cable material parameters includes the thermal conductivity λ, specific heat capacity c, density ρ, and the ratio K of the shielding layer induced current to the source current for each cable layer. The heat conduction equation is: = ( ( )+ )+ ； Among them, heat source item It is mainly generated by the source current in the cable core conductor and the induced current in the metal shielding layer, and is calculated according to the type of cable layer where data point i in the dataset is located, that is, the current q in the cable core conductor. i =I²N1, q in the metal shielding layer i =(KI)²N², where N1 is the resistance of the cable core conductor, N2 is the resistance of the metal shielding layer; I is the current carrying capacity, λ o c is the thermal conductivity of the o-th layer of the cable; o ρ is the specific heat capacity of the o-th layer of the cable. o denoted as the density of the o-th layer of the cable; T is the temperature value of data point i; t is the cable running time corresponding to data point i; z represents the axial length position of data point i in the cable; and r represents the radial position of the data point in the cable.

6. The method according to claim 1, characterized in that, The boundary constraints include: Set the outermost boundary temperature constraint of the cable layer to T=20℃; The initial operating time of the cable, and the temperature constraint at all points on the cable: T=20℃; The continuity constraint of interlayer heat flow in cables is: = ； in, This represents the outer boundary of the o-th layer. This represents the inner boundary of the (o+1)th layer.

7. The method according to claim 1, characterized in that, The loss function is: L=V1×L Data +V2×L PDE +V3×L IC +V4×L BC +V5×Lgrad; Among them, L Data The data loss term represents the mean square error between the temperature prediction data output by the above temperature prediction model and the temperature data of the corresponding data points in the dataset, L. PDE The loss term in the heat conduction equation is the mean square error of the value on the side of the equation that should be zero after substituting the predicted temperature data into the equation, which has already had one side shifted to zero. L IC The initial loss term constrains the initial temperature of the cable during operation; L BC The boundary loss term constrains the operating environment temperature of the cable; Lgrad is the gradient constraint loss term; V1, V2, V3, V4, and V5 are preset weight values. The iterative update of the system model parameters based on the loss function includes: When the gradient norm of the loss function exceeds the norm threshold, the gradient is clipped. The step of performing gradient penalty correction when the parameters in the cable material parameter group do not meet the corresponding operating threshold includes: If one or more parameters in the monitoring cable material parameter group, such as thermal conductivity, specific heat capacity, and density, do not meet the corresponding physical threshold set according to the physical settings of the cable material, or if the induced current ratio K∉[0,1] of the monitoring metal shielding layer is not monitored, gradient penalty correction is performed.

8. The method according to claim 1, characterized in that, If the predicted temperature reaches the evaluation index, the system model parameters are determined, including: If the mean square error between the predicted temperature and the corresponding actual temperature at the target point coordinates does not exceed a preset first error threshold, and the mean absolute error between the predicted temperature and the corresponding actual temperature at the target point coordinates does not exceed a preset second error threshold, then the predicted temperature meets the evaluation criteria, and the system model parameters are determined.

9. The method according to claim 8, characterized in that, The method further includes: Using the X, Y, and Z axes as spatial coordinates, color mapping represents temperature, and the boundary contours of each layer of the cable are superimposed to generate a temperature cloud map and radial temperature curve of the cable. And / or, deploy the system model to the cable temperature prediction scenario, and when the deviation between the temperature predicted by the system model and the measured temperature exceeds the third error threshold, re-execute the parameters of the system model and subsequent steps based on the temperature data in the dataset and the loss function under the boundary constraint conditions.

10. A cable temperature prediction device, characterized in that, The device includes: The processing unit is used to construct a dataset of each cable layer included in the cable based on the cable's operating data and in combination with the predefined radius range of each cable layer; the operating data includes the coordinates of multiple spatial points of the cable under a preset current carrying capacity and the temperature at each of the aforementioned coordinates corresponding to different cable operating times under the current carrying capacity; The building unit is used to construct a system model for building a temperature prediction model, a heat conduction equation for the cable, and a loss function; the temperature prediction model predicts output temperature prediction data based on the dataset, and the heat conduction equation is based on a trainable set of cable material parameters. The training unit is used to iteratively update the parameters of the system model based on the dataset and the loss function under the boundary constraint conditions of the cable, and to perform gradient penalty correction when the parameters in the cable material parameter group do not meet the corresponding running threshold, until the convergence condition is reached, so as to obtain the trained system model. The determining unit is used to predict the temperature of the target point coordinates on the cable during the running time of the target cable based on the system model. If the predicted temperature reaches the evaluation index, the system model parameters are determined.

Images