Thermal prediction method, system and device for spacecraft
By constructing a physical information neural network model based on the Transformer architecture, and combining heat transfer formulas and boundary conditions, the problems of computational resources and data dependence in satellite thermal control prediction were solved, and efficient and stable temperature prediction and thermal control design support were achieved.
Patent Information
- Application Number
- CN202511416894.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-30
- Publication Date
- 2026-01-13
AI Technical Summary
Existing technologies for satellite thermal control prediction suffer from problems such as high computational resource requirements, high data dependence, insufficient applicability, and low operating efficiency. In particular, they are difficult to achieve real-time prediction and optimization design in complex and variable heat conduction environments.
The Physical Information Neural Network (PINNsFormer) model based on the Transformer architecture is adopted. By combining the heat transfer formula and initial boundary conditions, a multi-head attention mechanism is used to capture global dependencies, enabling high-precision prediction of spacecraft temperature distribution, simplifying data requirements and improving prediction efficiency.
It achieves efficient and stable temperature prediction in complex environments, reduces computing resource requirements, improves prediction accuracy and operating efficiency, and supports thermal control design.
Smart Images
Figure CN121328296A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of spacecraft thermal prediction technology, and in particular to a method, system and device for predicting the thermal performance of spacecraft. Background Technology
[0002] Accurately predicting the temperature distribution of spacecraft (such as satellites) is crucial for ensuring their proper functioning and extending their lifespan.
[0003] Numerous electronic components and devices on a satellite must operate within specific temperature ranges to maintain their functional stability. When a satellite moves from Earth's shadow into the solar radiation zone, the temperature on the sunlit side rises rapidly. Without effective thermal control measures, internal instruments could be damaged by overheating. Conversely, when a satellite is in Earth's shadow, its surface temperature drops sharply due to continuous heat radiation, accelerating the aging of electronic components and shortening their lifespan.
[0004] Traditionally, heat transfer analysis and prediction for satellites have relied primarily on numerical simulation techniques, such as the finite element method (FEM) and computational fluid dynamics (CFD). Simulation tools like COMSOL Multiphysics, ANSYS, and Thermal Desktop are widely used to simulate heat transfer processes in the complex structures of satellites. These methods utilize accurate physical models to describe heat transfer mechanisms such as conduction, convection, and radiation, and can calculate temperature distributions under various environmental conditions. However, these techniques typically require substantial computational resources and detailed geometric and material data, which may be insufficient, especially in the early stages of satellite design.
[0005] Besides numerical simulation methods, in recent years the industry has begun to explore data-driven spacecraft thermal prediction methods, which utilize machine learning algorithms to train on historical simulation or experimental data. While this can quickly yield prediction results, it relies on a large amount of high-quality historical data for training. For entirely new or complex spacecraft design scenarios, it is often difficult to obtain sufficient experimental or simulation data, thus limiting the applicability of these methods.
[0006] Meanwhile, traditional numerical simulation techniques and data-driven methods lack versatility when dealing with complex and ever-changing application scenarios. Especially when the heat transfer environment changes significantly or the design undergoes substantial alterations, the predictions from data-driven methods may lose reliability, requiring remodeling or model retraining, increasing development costs. Furthermore, numerical simulation techniques typically involve extensive iterative calculations and solving complex physical models, resulting in low operational efficiency and making it difficult to meet the application requirements for real-time prediction and optimized design. Summary of the Invention
[0007] The purpose of this invention is to provide a method, system, and apparatus for predicting the thermal activity of spacecraft, so as to solve at least one of the aforementioned technical problems existing in the prior art.
[0008] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a thermal prediction method for spacecraft, comprising the following steps: Step 1: Obtain the shape model of the spacecraft. Based on the heat transfer formula, use simulation tools to obtain the temperature distribution on the surface of the shape model per unit time. This serves as the reference solution for the Physical Information Neural Network (PINNsFormer) model, thus providing basic data for subsequent model training and verification.
[0009] In one feasible implementation, the heat transfer formula in step 1 includes: ; in, Indicates temperature value; Indicates the density of the material; Indicates specific heat capacity; Indicates thermal conductivity; Indicates time; , and These represent the coordinates of points on the surface of the shape model; The initial conditions for the heat transfer formula include: ; The boundary conditions of the heat transfer formula include boundary heat source conditions and thermal radiation conditions. The specific formulas for the boundary heat source conditions include: ; in, This represents the heat flux applied to the surface of the shape model; This represents the component of the temperature gradient along the normal direction; The normal direction is the direction perpendicular to the outer surface of the spacecraft and points outward from the spacecraft. The specific formulas for the thermal radiation conditions include: ; in, Indicates the emissivity of the surface; This represents the Stefan-Boltzmann constant; Indicates ambient temperature.
[0010] Step 2: Construct and train a physical information neural network model based on the Transformer architecture to replace the traditional MLP structure for modeling physical fields. This scheme uses a multi-head attention mechanism to capture global dependencies, and shows stronger modeling ability and stability when dealing with complex boundary conditions and time dynamic problems.
[0011] In one feasible implementation, the input dimensions of the physical information neural network model include , , and The output is the temperature value (corresponding to the location and time). The physical information neural network model includes an embedding layer and a multi-layer Transformer encoder, used to model the coupling relationship between space and time.
[0012] In one feasible implementation, step 2 specifically includes: Step 21: Set the number of neurons in the Transformer hidden layer, the number of hidden layers in the MLP (Multilayer Perceptron), the number of attention heads used in parallel computation in the multi-head attention mechanism, and the network input dimensions, including... , , and The network output is a temperature value. , as well as the optimizer and initial learning rate used when training the physical information neural network model; Step 22: Construct the loss function calculation formula for the physical information neural network model, specifically including: ; in, Indicates the amount of physical loss; Indicates the boundary loss; Indicates the initial condition loss; Indicates the total loss; Step 23: Based on the heat transfer formula, initial conditions, and boundary conditions, construct... The specific calculation formula includes: ; in, This indicates the number of residual sampling points in the heat transfer formula; Indicates the first The coordinate vector of each residual sampling point; Step 24, Build The specific calculation formula includes: ; in, Indicates the number of initial value sampling points; Indicates the first The coordinate vector of each initial sampling point; Step 25, Build The specific calculation formula includes: ; in, These represent the number of boundary sampling points for the heat source surface and the heat radiation surface (of the physical boundary), respectively. The first two surfaces represent the heat source surface and the heat radiation surface, respectively. The coordinate vector of each boundary sampling point; These represent the physical information neural network model's representation of the first and second layers of the heat source surface and the heat radiation surface, respectively. Predicted solutions for each boundary sampling point; Step 26, based on The physical information neural network model is trained; the optimal solution is obtained by minimizing the loss function value; and the model parameters of the physical information neural network model are iteratively optimized through an optimizer, thereby achieving high-precision prediction of three-dimensional unsteady temperature fields.
[0013] Step 3: Using the trained physical information neural network model, infer the temperature value on the spacecraft surface to obtain the prediction result.
[0014] Step 4: Visualize the prediction results and compare them with the reference solution to evaluate the prediction accuracy and error value of the physical information neural network model.
[0015] In one feasible implementation, the heat prediction method further includes: Step 5: Based on the physical information neural network model, input the coordinates of the point that the spacecraft needs to query and the time to obtain the prediction result of the corresponding temperature value.
[0016] Secondly, based on the same inventive concept, this application also provides a thermal prediction system for spacecraft, including a data receiving module, a data processing module, and a result generation module; The data receiving module is used to acquire the shape model of the spacecraft; The data processing module includes a reference unit, a physical information neural network model unit, a prediction unit, and a comparison unit. The reference unit, based on the heat transfer formula, uses simulation tools to obtain the temperature distribution on the surface of the shape model per unit time, which serves as a reference solution. The physical information neural network model unit constructs and trains a physical information neural network model based on the Transformer architecture. The prediction unit uses a trained physical information neural network model to infer the temperature value on the spacecraft surface and obtain the prediction result. The comparison unit visualizes the prediction results and compares and analyzes them with the reference solution to evaluate the prediction accuracy and error value of the physical information neural network model. The result generation module is used to send the prediction results out.
[0017] Thirdly, based on the same inventive concept, this application also provides a thermal prediction device for a spacecraft, including a processor, a memory, and a bus. The memory stores instructions and data that can be read by the processor. The processor is used to call the instructions and data in the memory to execute the thermal prediction method for a spacecraft as described above. The bus connects the various functional components for transmitting information.
[0018] By adopting the above technical solution, the present invention has the following beneficial effects: This invention provides a method, system, and device for predicting the thermal properties of spacecraft. It uses a physical information neural network model to predict the temperature distribution at various locations on the spacecraft and uses COMSOL simulation data as a validation set to evaluate the accuracy and robustness of the model. Compared with traditional methods, this approach combines the advantages of physical laws and data-driven approaches, eliminating the need for large amounts of experimental and simulation data and computational resources. This simplifies the thermal analysis process for spacecraft, improves the efficiency and effectiveness of thermal prediction, and provides important technical support for thermal control design. Attached Figure Description
[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0020] Figure 1 A flowchart of a spacecraft thermal prediction method provided in an embodiment of the present invention; Figure 2 Example diagram of a unit cube model provided in an embodiment of the present invention; Figure 3 Example graph of COMSOL simulation results for unit cube model temperature provided in the embodiments of the present invention; Figure 4 Example of PINNsFormer prediction results for unit cube model temperature provided in this embodiment of the invention; Figure 5 Example diagram of a complex shape model provided in the embodiments of the present invention; Figure 6 Example of PINNsFormer prediction results for temperature of complex shape models provided in this embodiment of the invention; Figure 7 Example graph of COMSOL simulation results for temperature of complex shape models provided in embodiments of the present invention; Figure 8 A system diagram of a spacecraft thermal prediction method provided in an embodiment of the present invention. Detailed Implementation
[0021] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0023] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0024] The present invention will be further explained below with reference to specific embodiments.
[0025] It should also be noted that the specific embodiments or implementation methods described below are a series of optimized settings listed by the present invention to further explain the specific content of the invention, and these settings can be combined or used in conjunction with each other.
[0026] Example 1: like Figure 1 As shown, taking a satellite as an example, this embodiment provides a spacecraft thermal prediction method, which includes the following steps: Step 1: Obtain the shape model of the satellite and simplify it into a unit cube model. Based on the heat transfer formula, use simulation tools (such as COMSOL) to obtain the temperature distribution on the surface of the shape model within a unit time (e.g., 1 second). This serves as the reference solution for the Physical Information Neural Network (PINNsFormer) model, thus providing basic data for subsequent model training and verification.
[0027] Furthermore, the heat transfer formula (the governing equation of the three-dimensional transient heat transfer process) in step 1 includes: ; in, This indicates a temperature value, expressed in Kelvin (K). This indicates the density of a material, expressed in kilograms per cubic meter (kg / m³). ); Specific heat capacity is expressed in joules per kilogram (Kelvin). ); Thermal conductivity is expressed in watts per meter (kelvin). ); Indicates time, in seconds; , and These represent the coordinates of points on the surface of the shape model, in meters; The initial conditions for the heat transfer formula include: That is, the overall temperature of the shape model is initially set to 273K; The boundary conditions of the heat transfer formula include boundary heat source conditions and thermal radiation conditions. The specific formulas for the boundary heat source conditions include: ; in, This represents the heat flux applied to the surface of the shape model; This represents the component of the temperature gradient along the normal direction; The normal direction is the direction perpendicular to the outer surface of the satellite and points outward from the satellite. The specific formulas for the thermal radiation conditions include: ; in, Indicates the emissivity of the surface; This represents the Stefan-Boltzmann constant; Indicates ambient temperature.
[0028] Step 2: Construct and train a physical information neural network model based on the Transformer architecture to replace the traditional MLP structure for modeling physical fields. This scheme uses a multi-head attention mechanism to capture global dependencies, and shows stronger modeling ability and stability when dealing with complex boundary conditions and time dynamic problems.
[0029] Furthermore, the input dimension of the physical information neural network model is ( , , , The output is the temperature value (corresponding to the location and time). The physical information neural network model includes an embedding layer and a multi-layer Transformer encoder, used to model the coupling relationship between space and time.
[0030] Furthermore, step 2 specifically includes: Step 21: Set the number of hidden layer neurons in the Transformer to 64, the number of hidden layer neurons in the output MLP (Multilayer Perceptron) to 512, the number of attention heads used in parallel computation in the multi-head attention mechanism to 2, and the network input dimension to ( , , , The network output is a temperature value. The optimizer used when training the physical information neural network model is the L-BFGS optimizer with an initial learning rate of 1.0. This can provide sufficient representation capabilities while avoiding model overfitting, and the smaller parameters also help save computational resources. Step 22: Construct the loss function calculation formula for the physical information neural network model, specifically including: ; in, Indicates the amount of physical loss; Indicates the boundary loss; Indicates the initial condition loss; Indicates the total loss; Step 23: Based on the heat transfer formula, initial conditions, and boundary conditions, construct... The specific calculation formula includes: ; in, This represents the number of residual sampling points in the heat transfer formula, for example, for... , , Ten sampling points are selected evenly in each dimension. Six sampling points were selected for each dimension, resulting in a total of 6000 residual sampling points; Indicates the first The coordinate vector of each residual sampling point; Step 24, Build The specific calculation formula includes: ; in, This indicates the number of initial value sampling points, for example, for , , Ten sampling points are selected evenly in each dimension. If the dimension is fixed, there are a total of 1000 sampling points; Indicates the first The coordinate vector of each initial sampling point; Step 25, Build The specific calculation formula includes: ; in, These represent the number of boundary sampling points for the heat source surface and the heat radiation surface (physical boundaries), respectively. For example, for 5 heat radiation surfaces, 10 sampling points are uniformly selected in each of the two spatial dimensions. If 6 sampling points are selected in each dimension, there will be a total of 3000 boundary sampling points; while for the heat source surface, 10 sampling points are evenly selected in each of the two spatial dimensions. If 6 sampling points are selected for the dimension, there will be a total of 600 boundary sampling points; The first two surfaces represent the heat source surface and the heat radiation surface, respectively. The coordinate vector of each boundary sampling point; These represent the physical information neural network model's representation of the first and second layers of the heat source surface and the heat radiation surface, respectively. Predicted solutions for each boundary sampling point; Step 26, based on The physical information neural network model is trained; the optimal solution is obtained by minimizing the loss function value; and the model parameters of the physical information neural network model are iteratively optimized through an optimizer, thereby achieving high-precision prediction of three-dimensional unsteady temperature fields.
[0031] Step 3: Using the trained physical information neural network model, infer the temperature value on the spacecraft surface to obtain the prediction result.
[0032] Step 4: Visualize the prediction results, such as... Figure 2-4 As shown, the prediction accuracy and error value of the physical information neural network model are compared and analyzed with the reference solution to evaluate the prediction accuracy and error value. Specifically, Figure 3-4 A comparison shows that the PINNsFormer prediction result of the satellite in this embodiment at 0.7 seconds ( Figure 4 ) and COMSOL simulation results ( Figure 3 The temperature and color distribution in the image are already very similar.
[0033] Furthermore, the heat prediction method also includes: Step 5: Based on the physical information neural network model, input the coordinates of the point to be queried by the satellite and the time, and obtain the prediction result of the corresponding temperature value.
[0034] Example 2: Taking a satellite with a more complex shape as an example, this embodiment provides a spacecraft thermal prediction method, which includes the following steps: Step 1: Obtain the shape model of the satellite, such as... Figure 5 As shown (m in the figure represents meters), based on the heat transfer formula, the temperature distribution on the surface of the shape model within a unit time (e.g., 1 second) is obtained through simulation tools (e.g., COMSOL), which serves as the reference solution for the Physical Information Neural Network (PINNsFormer) model, thus providing basic data for subsequent model training and validation.
[0035] Furthermore, the heat transfer formula (the governing equation of the three-dimensional transient heat transfer process) in step 1 includes: ; in, This indicates a temperature value, expressed in Kelvin (K). This indicates the density of a material, expressed in kilograms per cubic meter (kg / m³). ); Specific heat capacity is expressed in joules per kilogram (Kelvin). ); Thermal conductivity is expressed in watts per meter (kelvin). ); Indicates time, in seconds; , and These represent the coordinates of points on the surface of the shape model, in meters; The initial conditions for the heat transfer formula include: That is, the overall temperature of the shape model is initially set to 273K; The boundary conditions of the heat transfer formula include boundary heat source conditions and thermal radiation conditions. The specific formulas for the boundary heat source conditions include: ; in, This represents the heat flux applied to the surface of the shape model; This represents the component of the temperature gradient along the normal direction; The normal direction is the direction perpendicular to the outer surface of the satellite and points outward from the satellite. The specific formulas for the thermal radiation conditions include: ; in, Indicates the emissivity of the surface; This represents the Stefan-Boltzmann constant; Indicates ambient temperature.
[0036] Step 2: Construct and train a physical information neural network model based on the Transformer architecture to replace the traditional MLP structure for modeling physical fields. This scheme uses a multi-head attention mechanism to capture global dependencies, and shows stronger modeling ability and stability when dealing with complex boundary conditions and time dynamic problems.
[0037] Furthermore, the input dimension of the physical information neural network model is ( , , , The output is the temperature value (corresponding to the location and time). The physical information neural network model includes an embedding layer and a multi-layer Transformer encoder, used to model the coupling relationship between space and time.
[0038] Furthermore, step 2 specifically includes: Step 21: Set the number of neurons in the Transformer hidden layer to 512, the number of hidden layers in the MLP (Multilayer Perceptron) to 64, the number of attention heads used in parallel computation in the multi-head attention mechanism to 2, and the network input dimension to ( , , , The network output is a temperature value. The optimizer used when training the physical information neural network model is the L-BFGS optimizer, with an initial learning rate of 1.0. Step 22: Construct the loss function calculation formula for the physical information neural network model, specifically including: ; in, Indicates the amount of physical loss; Indicates the boundary loss; Indicates the initial condition loss; Indicates the total loss; Step 23: Based on the heat transfer formula, initial conditions, and boundary conditions, construct... The specific calculation formula includes: ; in, This represents the number of residual sampling points in the heat transfer formula. For example, for each individual training batch, 128 residual sampling points are sampled at any time within the shape model. Indicates the first The coordinate vector of each residual sampling point; Step 24, Build The specific calculation formula includes: ; in, This indicates the number of initial sampling points. For example, for each individual training batch, 128 sampling points are sampled at the initial time within the shape model. Indicates the first The coordinate vector of each initial sampling point; Step 25, Build The specific calculation formula includes: ; in, These represent the number of boundary sampling points for the heat source surface and the heat radiation surface (physical boundary), respectively. For example, 128 boundary sampling points are randomly sampled from the surface of the shape model in each batch. The first two surfaces represent the heat source surface and the heat radiation surface, respectively. The coordinate vector of each boundary sampling point; These represent the physical information neural network model's representation of the first and second layers of the heat source surface and the heat radiation surface, respectively. Predicted solutions for each boundary sampling point; Step 26, based on The physical information neural network model is trained; the optimal solution is obtained by minimizing the loss function value; and the model parameters of the physical information neural network model are iteratively optimized through an optimizer, thereby achieving high-precision prediction of three-dimensional unsteady temperature fields.
[0039] Step 3: Using the trained physical information neural network model, infer the temperature value on the spacecraft surface to obtain the prediction result.
[0040] Step 4: Visualize the prediction results and compare them with the reference solution to evaluate the prediction accuracy and error value of the physical information neural network model. Furthermore, the prediction accuracy and error value can be calculated using the percentage error of conventional simulation results.
[0041] Specifically, Figure 6-7 A comparison shows that the PINNsFormer prediction result of the satellite in this embodiment at 5810 seconds ( Figure 6 ) and COMSOL simulation results ( Figure 7 The temperature and color distribution in the image are very similar, with only slight color differences at the edges (due to...). Figure 6 and Figure 7 There is an angular deflection, so it is related to... Figure 5 (The shapes of the satellites appear inconsistent), with an average percentage extrapolation error of around 10%.
[0042] Furthermore, the heat prediction method also includes: Step 5: Based on the physical information neural network model, input the coordinates of the point to be queried by the satellite and the time, and obtain the prediction result of the corresponding temperature value.
[0043] Example 3: like Figure 8 As shown, this embodiment also provides a thermal prediction system for spacecraft, including a data receiving module, a data processing module, and a result generation module; The data receiving module is used to acquire the shape model of the spacecraft; The data processing module includes a reference unit, a physical information neural network model unit, a prediction unit, and a comparison unit. The reference unit, based on the heat transfer formula, uses simulation tools to obtain the temperature distribution on the surface of the shape model per unit time, which serves as a reference solution. The physical information neural network model unit constructs and trains a physical information neural network model based on the Transformer architecture. The prediction unit uses a trained physical information neural network model to infer the temperature value on the spacecraft surface and obtain the prediction result. The comparison unit visualizes the prediction results and compares and analyzes them with the reference solution to evaluate the prediction accuracy and error value of the physical information neural network model. The result generation module is used to send the prediction results out.
[0044] Example 4: This embodiment provides a thermal prediction device for a spacecraft, including a processor, a memory, and a bus. The memory stores instructions and data that can be read by the processor. The processor is used to call the instructions and data in the memory to execute the thermal prediction method for a spacecraft as described above. The bus connects the various functional components for transmitting information.
[0045] In another embodiment, this solution can also be implemented using an integrated device, which may include corresponding modules that perform one or more steps in the various embodiments described above. A module may be one or more hardware modules specifically configured to perform the corresponding step, or implemented by a processor configured to perform the corresponding step, or stored in a computer-readable medium for implementation by a processor, or implemented through some combination thereof.
[0046] The processor executes the various methods and processes described above. For example, the method implementations in this scheme can be implemented as software programs tangibly contained in a machine-readable medium, such as memory. In some implementations, part or all of the software program can be loaded and / or installed via memory and / or a communication interface. When the software program is loaded into memory and executed by the processor, one or more steps of the methods described above can be performed. Alternatively, in other implementations, the processor can be configured to execute one of the methods described above by any other suitable means (e.g., by means of firmware).
[0047] This device can be implemented using a bus architecture. A bus architecture can include any number of interconnect buses and bridges, depending on the specific application of the hardware and overall design constraints. The bus connects various circuits, including one or more processors, memory, and / or hardware modules. The bus can also connect various other circuits such as peripherals, voltage regulators, power management circuitry, external antennas, etc.
[0048] Buses can be Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, or Extended Industry Standard Component (EISA) buses, etc. Buses can be divided into address buses, data buses, control buses, etc.
[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for predicting the thermal activity of a spacecraft, characterized in that, include: Step 1: Obtain the shape model of the spacecraft. Based on the heat transfer formula, use simulation tools to obtain the temperature distribution on the surface of the shape model per unit time, which serves as the reference solution for the physical information neural network model. Step 2: Construct and train a physical information neural network model based on the Transformer architecture; Step 3: Using the trained physical information neural network model, infer the temperature value on the spacecraft surface to obtain the prediction result; Step 4: Visualize the prediction results and compare them with the reference solution to evaluate the prediction accuracy and error value of the physical information neural network model.
2. The thermal prediction method according to claim 1, characterized in that, The heat transfer formula in step 1 includes: ； in, Indicates the temperature value; Indicates the density of the material; Indicates specific heat capacity; Indicates thermal conductivity; Indicates time; , and These represent the coordinates of points on the surface of the shape model.
3. The thermal prediction method according to claim 2, characterized in that, The initial conditions for the heat transfer formula include: 。 4. The thermal prediction method according to claim 3, characterized in that, The boundary conditions of the heat transfer formula include boundary heat source conditions and thermal radiation conditions. The specific formulas for the boundary heat source conditions include: ； in, This represents the heat flux applied to the surface of the shape model; This represents the component of the temperature gradient along the normal direction; Indicates the normal direction; The specific formulas for the thermal radiation conditions include: ； in, Indicates the emissivity of the surface; This represents the Stefan-Boltzmann constant; Indicates ambient temperature.
5. The thermal prediction method according to claim 2, characterized in that, The input dimensions of the physical information neural network model include , , and The output is a temperature value. The physical information neural network model includes an embedding layer and a multi-layer Transformer encoder, used to model the coupling relationship between space and time.
6. The thermal prediction method according to claim 5, characterized in that, Step 2 specifically includes: Step 21: Set the number of neurons in the Transformer hidden layer, the number of hidden layers in the MLP, the number of attention heads used in parallel computation in the multi-head attention mechanism, and the network input dimensions, including... , , and The network output is a temperature value. , as well as the optimizer and initial learning rate used when training the physical information neural network model; Step 22: Construct the loss function calculation formula for the physical information neural network model, specifically including: ； in, Indicates the amount of physical loss; Indicates the boundary loss; Indicates the initial condition loss; Indicates the total loss; Step 23: Based on the heat transfer formula, initial conditions, and boundary conditions, construct... The specific calculation formula includes: ； in, This indicates the number of residual sampling points in the heat transfer formula; Indicates the first The coordinate vector of each residual sampling point; Step 24, Build The specific calculation formula includes: ； in, Indicates the number of initial value sampling points; Indicates the first The coordinate vector of each initial sampling point; Step 25, Build The specific calculation formula includes: ； in, These represent the number of boundary sampling points on the heat source surface and the heat radiation surface, respectively. The first two surfaces represent the heat source surface and the heat radiation surface, respectively. The coordinate vector of each boundary sampling point; These represent the physical information neural network model's representation of the first and second layers of the heat source surface and the heat radiation surface, respectively. Predicted solutions for each boundary sampling point; Step 26, based on The physical information neural network model is trained; the optimal solution is obtained by minimizing the loss function value; and the model parameters of the physical information neural network model are iteratively optimized using an optimizer.
7. The thermal prediction method according to claim 6, characterized in that, In step 21, the Transformer hidden layer has 512 neurons, the MLP hidden layer has 64 layers, and the multi-head attention mechanism has 2 attention heads computed in parallel.
8. The thermal prediction method according to claim 6, characterized in that, The optimizer used when training the physical information neural network model is the L-BFGS optimizer, with an initial learning rate of 1.
0.
9. A thermal prediction system for a spacecraft, characterized in that, It includes a data receiving module, a data processing module, and a result generation module; The data receiving module is used to acquire the shape model of the spacecraft; The data processing module includes a reference unit, a physical information neural network model unit, a prediction unit, and a comparison unit. The reference unit, based on the heat transfer formula, uses simulation tools to obtain the temperature distribution on the surface of the shape model per unit time, which serves as a reference solution. The physical information neural network model unit constructs and trains a physical information neural network model based on the Transformer architecture. The prediction unit uses a trained physical information neural network model to infer the temperature value on the spacecraft surface and obtain the prediction result. The comparison unit visualizes the prediction results and compares and analyzes them with the reference solution to evaluate the prediction accuracy and error value of the physical information neural network model. The result generation module is used to send the prediction results out.
10. A thermal prediction device for a spacecraft, characterized in that, It includes a processor, a memory, and a bus. The memory stores instructions and data read by the processor. The processor is used to call the instructions and data in the memory to execute the thermal prediction method as described in any one of claims 1-8. The bus connects the functional components for transmitting information.