Satellite Component Layout Temperature Field Prediction Method Based on Neural Network with Physical Priors

Abstract

The present invention discloses a method for predicting the temperature field of a satellite component layout based on a neural network with physical prior, including: establishing a structural model of the satellite component layout according to the layout characteristics of the satellite components; obtaining a plurality of training data and preprocessing the training data; constructing a loss function embedded with physical prior according to the steady-state heat conduction equation followed by the temperature field of the satellite component layout; determining the weight of each prediction point of the temperature field by using online data mining, updating the loss function according to the prediction point weight, constructing a regularization term of the loss function by using regional heat flux conservation, and determining the final loss function; constructing a deep neural network model, and training the deep neural network model by using the preprocessed training data and the final loss function; predicting the temperature field of the satellite component layout by using the trained deep neural network model. The present invention can realize the stable and rapid training of the deep neural network model by using unlabeled training data and ensure the prediction accuracy of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of component layout optimization design, and particularly relates to a method for predicting the temperature field of a satellite component layout based on a neural network with physical prior knowledge. Background Art

[0002] To achieve different functions and perform various tasks, a large number of components are usually integrated inside a satellite. During normal operation, the components generate a large amount of heat. If effective heat dissipation of the components cannot be ensured, the excessive ambient temperature will lead to a reduction in the reliability of the components or even their failure. Since the satellite structure and the heat dissipation method of the satellite components are fixed, currently, the temperature field corresponding to the satellite component layout is mainly reduced by adjusting the satellite component layout, so as to achieve effective heat dissipation.

[0003] When performing the optimization design of the satellite component layout, to determine the optimal satellite component layout, it is necessary to calculate the temperature fields corresponding to different satellite component layouts. Currently, two methods are used to calculate the temperature fields corresponding to different satellite component layouts. Among them, the first method is to use numerical methods such as the finite element method, the finite difference method, and the finite volume method to perform simulation analysis on the satellite component layout to calculate the temperature field corresponding to the satellite component layout. The second method is to use a deep learning surrogate model to predict the temperature field of the satellite component layout. This method constructs and trains a neural network in deep learning by giving a certain number of training data including the satellite component layout and its temperature field to obtain a neural network surrogate model, and uses this surrogate model to predict the temperature field corresponding to the satellite component layout.

[0004] However, since the optimization design of the satellite component layout is a process of repeated iteration, using numerical methods to perform simulation analysis on the satellite component layout to determine the corresponding temperature field requires a large amount of computing resources and computing time, and the computing time increases exponentially with the computing accuracy. When using a deep learning surrogate model to predict the temperature field of the satellite component layout, a large amount of training data including the satellite component layout and its temperature field is required to train the neural network model. Since it is difficult to obtain the real data of the temperature field corresponding to the satellite component layout, the acquisition of each training data needs to perform simulation analysis on the satellite component layout by using numerical methods, which will also consume a lot of computing resources and computing time. Summary of the Invention

[0005] To solve some or all of the above-mentioned technical problems existing in the prior art, the present invention provides a method for predicting the temperature field of a satellite component layout based on a neural network with physical prior knowledge.

[0006] The technical solution of the present invention is as follows:

[0007] A method for predicting the temperature field of satellite component layout based on a neural network with physical prior is provided. The method includes:

[0008] According to the layout characteristics of satellite components, establish a structural model of the satellite component layout;

[0009] Based on the structural model of the satellite component layout, obtain multiple training data and preprocess the training data. Among them, the training data includes the satellite component layout;

[0010] According to the heat conduction steady-state equation that the satellite component layout temperature field obeys, construct a loss function embedded with physical prior;

[0011] Use online data mining to determine the weight of each prediction point of the temperature field, update the loss function according to the prediction point weight, and use the regional heat flux conservation to construct a regularization term of the loss function to determine the final loss function;

[0012] Construct a deep neural network model, and use the preprocessed training data and the final loss function to train the deep neural network model to fit the mapping relationship between the satellite component layout and the temperature field;

[0013] Use the trained deep neural network model to predict the temperature field of the satellite component layout.

[0014] In some possible implementation manners, according to the layout characteristics of satellite components, the following method is used to establish a structural model of the satellite component layout:

[0015] Set the satellite component layout area as a square layout area, set a small hole with a set length on one of the four sides of the square layout area as a heat dissipation hole, the temperature of the heat dissipation hole area is fixed, and the remaining boundaries except the heat dissipation hole area are adiabatic. The satellite components are distributed at different positions in the square layout area, and one satellite component is regarded as a heat source.

[0016] In some possible implementation manners, the obtaining of multiple training data includes:

[0017] According to the number of satellite components, randomly select corresponding positions in the square layout area to place the satellite components to obtain a training data including the satellite component layout. Repeat the random selection process multiple times until the preset number of training data is obtained.

[0018] In some possible implementation manners, the preprocessing of the training data includes:

[0019] Divide the square layout area of the satellite component layout into M1×M2 grids, represent the satellite component layout with an M1×M2 matrix, and the matrix element corresponding to the grid position with a component is the component power, and the matrix element corresponding to the grid position without a component is 0.

[0020] In some possible implementation manners, set M1 = M2, and construct a loss function incorporating physical prior as follows:

[0021]

[0022] where T i,j represents the temperature at the grid point of the i-th row and j-th column in the temperature field of the satellite component layout, Δh represents the distance between two adjacent grid points in the same row or the same column, Δh = l / m, l represents the side length of the satellite component layout area, m represents the number of grids into which the satellite component layout area is divided, m = M1×M2, and φ i,j represents the component power at the grid point of the i-th row and j-th column in the satellite component layout area, T i-1,j represents the temperature at the grid point of the (i - 1)-th row and j-th column in the temperature field of the satellite component layout, T i+1,j represents the temperature at the grid point of the (i + 1)-th row and j-th column in the temperature field of the satellite component layout, T i,j-1 represents the temperature at the grid point of the i-th row and (j - 1)-th column in the temperature field of the satellite component layout, T i,j+1 represents the temperature at the grid point of the i-th row and (j + 1)-th column in the temperature field of the satellite component layout.

[0023] In some possible implementation manners, the weight of the prediction point is set as:

[0024]

[0025] where w i,j represents the weight corresponding to the grid point of the i-th row and j-th column in the temperature field of the satellite component layout, min(T) represents the lowest temperature of the temperature field of the satellite component layout, and max(T) represents the highest temperature of the temperature field of the satellite component layout.

[0026] In some possible implementation manners, according to the prediction point weight, the loss function is updated as:

[0027]

[0028] In some possible implementation manners, the regularization term of the loss function is set as:

[0029]

[0030] where Ω q represents the region satisfying heat flux conservation.

[0031] In some possible implementation manners, the deep neural network model adopts a U-net neural network.

[0032] In some possible implementation manners, the method further includes:

[0033] During the training process of the deep neural network model, according to the boundary conditions satisfied by the temperature field of the satellite component layout, a hard constraint method is used to limit the predicted temperature field output by the deep neural network model;

[0034] Calculate the final loss function value based on the processed predicted temperature field.

[0035] The main advantages of the technical solution of the present invention are as follows:

[0036] The method for predicting the temperature field of the satellite component layout based on the physics prior neural network of the present invention constructs a loss function with physical prior by introducing the steady-state equation of heat conduction of the temperature field, determines the weight of each prediction point of the temperature field by using online data mining, and constructs a regularization term of the loss function by using the conservation of regional heat flux. It can realize the stable and rapid training of the deep neural network model by using unlabeled training data, accelerate the training convergence speed, improve the prediction accuracy of the model, reduce the calculation time and calculation resource consumption required during the training process, reduce the component layout optimization design cost, and improve the component layout optimization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0038] Figure 1 It is a flowchart of the method for predicting the temperature field of the satellite component layout based on the physics prior neural network according to an embodiment of the present invention;

[0039] Figure 2 It is a schematic structural diagram of a component layout according to an embodiment of the present invention;

[0040] Figure 3 It is a schematic diagram of the discretization of a component layout area according to an embodiment of the present invention;

[0041] Figure 4a 、 Figure 4b and Figure 4c They are schematic diagrams of three different regions divided by the conservation of heat flux;

[0042] Figure 5 It is a schematic diagram of the training process of a deep neural network model according to an embodiment of the present invention;

[0043] Figure 6 It is a schematic diagram of the training process of another deep neural network model according to an embodiment of the present invention. Detailed implementation manners

[0044] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0045] The technical solutions provided by the embodiments of the present invention will be described in detail below in conjunction with the drawings.

[0046] See Figure 1 , an embodiment of the present invention provides a method for predicting the temperature field of the satellite component layout based on a neural network with physical prior. The method includes the following steps:

[0047] S1. According to the layout characteristics of the satellite components, establish a structural model of the satellite component layout;

[0048] S2. Based on the structural model of the satellite component layout, obtain a plurality of training data and preprocess the training data. Among them, the training data includes the satellite component layout;

[0049] S3. According to the heat conduction steady-state equation obeyed by the temperature field of the satellite component layout, construct a loss function embedded with physical prior;

[0050] S4. Use online data mining to determine the weight of each prediction point of the temperature field, update the loss function according to the prediction point weight, and construct a regularization term of the loss function by using regional heat flux conservation to determine the final loss function;

[0051] S5. Construct a deep neural network model, and use the preprocessed training data and the final loss function to train the deep neural network model to fit the mapping relationship between the satellite component layout and the temperature field;

[0052] S6. Use the trained deep neural network model to predict the temperature field of the satellite component layout.

[0053] The satellite component layout temperature field prediction method based on the neural network with physical prior provided by an embodiment of the present invention constructs a loss function with physical prior by introducing the steady-state equation of temperature field heat conduction, determines the weight of each prediction point of the temperature field by using online data mining, constructs a regularization term of the loss function by using regional heat flux conservation, can realize the stable and rapid training of the deep neural network model by using unlabeled training data, can accelerate the training convergence speed, improve the prediction accuracy of the model, reduce the computing time and computing resource consumption required in the training process, reduce the component layout optimization design cost, and improve the component layout optimization efficiency.

[0054] The steps and principles of the satellite component layout temperature field prediction method based on the neural network with physical prior provided by an embodiment of the present invention are specifically described below.

[0055] Step S1, according to the layout characteristics of the satellite components, establish a structural model of the satellite component layout.

[0056] Specifically, the satellite components usually have the following layout characteristics:

[0057] The satellite components are installed on the satellite cabin board, and a certain number of satellite components with different sizes, heating powers and functions are distributed on the satellite cabin board. The satellite components will continuously generate heat during operation, and the heat generated by the satellite components is conducted to the cabin board through heat conduction, and then the heat is exported to the outside of the satellite through the heat pipe.

[0058] See Figure 2 , for the layout characteristics of the above satellite components, the following method is used to establish a structural model of the satellite component layout:

[0059] Set the satellite component layout area as a square layout area, set a small hole with a set length on one of the four sides of the square layout area as a heat dissipation hole, the temperature of the heat dissipation hole area is fixed, the remaining boundaries except the heat dissipation hole area are adiabatic, and the satellite components are distributed at different positions in the square layout area, and one satellite component is regarded as a heat source.

[0060] Since the space environment where the satellite operates is a vacuum environment, there is no convective heat transfer, and the radiative heat transfer can also be ignored. In an embodiment of the present invention, by opening a heat dissipation hole with a certain size on one side of the square layout area, the temperature at the heat dissipation hole is fixed, and the four boundaries of the square layout area except the heat dissipation hole are adiabatic to simulate the heat pipe heat dissipation method of the satellite cabin board.

[0061] Step S2, based on the structural model of the satellite component layout, obtain a plurality of training data and preprocess the training data, where the training data includes the satellite component layout.

[0062] Specifically, based on the above structural model of the satellite component layout, multiple training data are obtained, including:

[0063] According to the number of satellite components, randomly select corresponding positions from the square layout area to place the satellite components, obtaining a training data including the satellite component layout. Repeat the random selection process multiple times until the preset number of training data is obtained.

[0064] Among them, the specific number of training data can be determined according to the actual required model accuracy and component layout optimization time. For example, the number of training data can be 8000, 10000, etc. Generally, the more training data, the better the training effect of the model and the higher the prediction accuracy of the model.

[0065] Furthermore, based on the above-obtained training data, preprocess the training data, including:

[0066] Divide the square layout area of the satellite component layout into M1×M2 grids, represent the satellite component layout with an M1×M2 matrix. The matrix element corresponding to the grid position with a component is the component power, and the matrix element corresponding to the grid position without a component is 0.

[0067] Step S3, construct a loss function embedding physical prior according to the heat conduction steady-state equation that the satellite component layout temperature field obeys.

[0068] According to the above layout characteristics of the satellite components and the structural model of the satellite component layout, assume that the satellite component layout temperature field does not change with time. Then the satellite component layout temperature field can be solved by the heat conduction steady-state equation shown in the following formula;

[0069]

[0070] Among them, (x,y) represents the coordinates of a point in the two-dimensional plane of the satellite component layout area, T represents the temperature at the point (x,y), k represents the heat conduction coefficient, and φ(x,y) represents the component power at the point (x,y), that is, the heat source intensity.

[0071] According to different boundary heat dissipation conditions, the temperature field satisfies different boundary conditions. The boundary conditions are divided into three types, including the first type of boundary condition (Dirichlet boundary condition), the second type of boundary condition (Neumann boundary condition), and the third type of boundary condition (Robin boundary condition).

[0072] The first type of boundary condition is a constant temperature boundary condition; the second type of boundary condition stipulates the heat flux density value of the boundary, and the heat flux density value of the boundary is set to a fixed value q, that is n is the normal direction on the boundary Γ. If q = 0, then This is the adiabatic boundary condition; the third boundary condition stipulates the surface heat transfer coefficient between the object on the boundary and the surrounding fluid and the temperature of the surrounding fluid.

[0073] In an embodiment of the present invention, the heat dissipation hole region satisfies the first type of boundary condition, that is, the temperature of the heat dissipation hole region is fixed, and the remaining boundaries except the heat dissipation hole region satisfy the second type of boundary condition, which is the adiabatic boundary condition, that is, the boundary change condition of the remaining boundaries except the heat dissipation hole region is the adiabatic boundary condition.

[0074] Since the temperature field of the satellite component layout obeys the steady-state heat conduction equation, in an embodiment of the present invention, the steady-state heat conduction equation is used as physical prior information to construct a loss function embedding physical prior.

[0075] Specifically, referring to Figure 3 , since the satellite component layout area is divided into M1×M2 grids, for the steady-state heat conduction equation, it can be expressed in the difference form of the second-order differential as:

[0076]

[0077] where, T i+1,j represents the temperature at the grid point of the (i + 1)-th row and the j-th column in the temperature field of the satellite component layout, T i,j represents the temperature at the grid point of the i-th row and the j-th column in the temperature field of the satellite component layout, T i-1,j represents the temperature at the grid point of the (i - 1)-th row and the j-th column in the temperature field of the satellite component layout, T i,j+1 represents the temperature at the grid point of the i-th row and the (j + 1)-th column in the temperature field of the satellite component layout, T i,j-1 represents the temperature at the grid point of the i-th row and the (j - 1)-th column in the temperature field of the satellite component layout, Δx represents the distance between two adjacent grid points in the same row, and Δy represents the distance between two adjacent grid points in the same column.

[0078] Assume that the side length of the satellite component layout area is l, and each row and each column of the satellite component layout area have the same number of grids, that is, M1 = M2, then the steady-state heat conduction equation can be expressed as:

[0079] Δh 2 ·φ i,j +T i-1,j +T i+1,j +T i,j-1 +T i,j+1 -4T i,j = 0 Equation Three

[0080] where, φ i,j represents the component power at the grid point of the i-th row and the j-th column in the satellite component layout area, Δh = Δx = Δy = l / m, and m represents the number of grids divided in the satellite component layout area, m = M1×M2.

[0081] In discrete form, for the temperature at each grid point, the temperature at the grid points nearby satisfies Equation (III) above. Therefore, a system of linear equations can be formed for all grid points. Jacobi iteration is a commonly used iterative method for solving linear equations. Based on Equation (III) above, the following iterative form can be constructed:

[0082]

[0083] where T′ i,j represents the temperature at the grid point in the i-th row and j-th column of the temperature field of the satellite component layout after iteration.

[0084] Based on the above derivation process, the loss function incorporating physical prior knowledge is constructed as follows:

[0085]

[0086] Step S4: Use online data mining to determine the weight of each prediction point in the temperature field, update the loss function according to the prediction point weights, and construct a regularization term for the loss function using regional heat flux conservation to determine the final loss function.

[0087] For the predicted temperature field, it is expected that each prediction point in the predicted temperature field satisfies Equation (IV). However, during the training process, the training difficulty of each prediction point is not the same. Therefore, when calculating the gradient with the same weight assigned to all prediction points in the temperature field for training, problems such as slow network convergence speed and low final prediction accuracy of the model exist. In an embodiment of the present invention, to solve the situation where the training difficulty of different prediction points is inconsistent, improve the convergence speed during training, and improve the prediction accuracy of the model, an online data mining method is used to determine the weight of each prediction point in the temperature field.

[0088] Specifically, based on the satellite component layout after the above grid division, the weight of the prediction point is set as follows:

[0089]

[0090] where w i,j represents the weight corresponding to the grid point in the i-th row and j-th column of the temperature field of the satellite component layout, T i,j represents the temperature at the grid point in the i-th row and j-th column of the temperature field of the satellite component layout, max(T) represents the highest temperature of the temperature field of the satellite component layout, and min(T) represents the lowest temperature of the temperature field of the satellite component layout.

[0091] According to the weights of the prediction points set above, the loss function incorporating physical prior knowledge shown in Equation (V) is updated as follows:

[0092]

[0093] Furthermore, since the space environment where the satellite operates is a vacuum environment, there is no convective heat transfer, and radiative heat transfer can also be ignored. The heat generated by the satellite components is propagated through heat conduction. Therefore, in an embodiment of the present invention, for any region in the satellite component layout that includes a number of prediction points, the following formula is satisfied among the heat inflow, heat outflow, and heat generated by the satellite components in this region, that is, the heat flowing into the region plus the heat generated by the satellite components in the region is the same as the heat flowing out of the region;

[0094] q out =q in +q g Formula VIII

[0095] wherein, q out represents the heat flowing out of the region, q in represents the heat flowing into the region, and q g represents the heat generated by the satellite components in the region.

[0096] See Figure 4a - Figure 4c , in an embodiment of the present invention, in order to further accelerate network convergence and improve the training efficiency of the network model, the conservation of heat flux in the horizontal direction region, vertical direction region, and square region is considered during the training process of the deep neural network model, and the regional heat flux conservation is used as a regularization term of the loss function during the training process to accelerate the training of the deep neural network model.

[0097] Specifically, the regularization term of the loss function is set as:

[0098]

[0099] wherein, Ω q represents the region that satisfies the conservation of heat flux.

[0100] According to the updated loss function and the constructed regularization term of the loss function above, the final loss function is obtained as:

[0101]

[0102] wherein, λ represents the weight of the regularization term of the loss function, and λ is a hyperparameter of a real constant.

[0103] Step S5, construct a deep neural network model, and use the preprocessed training data and the final loss function to train the deep neural network model to fit the mapping relationship between the satellite component layout and the temperature field.

[0104] In one embodiment of the present invention, the deep neural network model adopts a U-net neural network, which is a fully convolutional neural network structure of encoder-decoder. Among them, the encoder is used to capture the context information in the image, and the symmetric decoder increases the resolution of the feature map to finally achieve image-to-image prediction. In addition, through the skip connection architecture, the features on different encoder paths and decoder paths are fused.

[0105] Further, referring to Figure 5 , after determining the initial deep neural network model, the preprocessed training data is used as the input of the deep neural network model to obtain the corresponding predicted temperature field. Based on the predicted temperature field, the loss function value is calculated using the determined final loss function, and the parameters of the deep neural network model are iteratively updated in a backpropagation manner according to the loss function value until the set number of iterations is reached and then stopped, and the trained deep neural network model is saved. Among them, the number of iterations during training can be set according to the actual required prediction accuracy and training time. Generally, the more the number of iterations, the higher the prediction accuracy of the trained deep neural network model, and the longer the required training time. For example, the number of iterations can be set to 50 times, 60 times, etc.

[0106] In one embodiment of the present invention, since the training data input to the deep neural network model is represented by an M1×M2 matrix, the corresponding predicted temperature field output by the deep neural network model is also represented by an M1×M2 matrix.

[0107] Step S6, using the trained deep neural network model to predict the temperature field of the satellite component layout.

[0108] Specifically, after completing the training of the deep neural network model, the deep neural network model is loaded, and the satellite component layout whose temperature field is to be calculated is input into the deep neural network model, and the temperature field corresponding to the satellite component layout can be obtained, thereby assisting in the optimal design of the satellite component layout.

[0109] Further, referring to Figure 6 , in one embodiment of the present invention, the method may further include:

[0110] During the training process of the deep neural network model, according to the boundary conditions satisfied by the temperature field of the satellite component layout, the predicted temperature field output by the deep neural network model is restricted by a hard constraint method;

[0111] Calculate the final loss function value based on the processed predicted temperature field.

[0112] In one embodiment of the present invention, the heat dissipation hole area of the satellite component layout satisfies the first type of boundary condition, and the remaining boundaries except the heat dissipation hole area satisfy the second type of boundary condition, which is an adiabatic boundary condition.

[0113] Specifically, for the heat dissipation hole area under the action of the first type of boundary condition, a constant temperature value is used for filling processing. For the area under the action of the second type of boundary condition, this area satisfies the following difference form:

[0114]

[0115] Based on the above difference form, nodes are filled in the area under the action of the second type of boundary condition for predicting the temperature field, so that the corresponding action area satisfies the second type of boundary condition.

[0116] By performing filling processing on the boundary area of the temperature field and calculating the final loss function value based on the predicted temperature field after filling processing for iterative update of the model, the prediction accuracy of the trained model can be further improved.

[0117] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. In addition, in this article, "front", "rear", "left", "right", "upper", and "lower" are all referenced with respect to the placement state shown in the drawings.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the temperature field of satellite component layout based on a neural network with physical prior, characterized in that Including: Establish a structural model of the satellite component layout according to the layout characteristics of the satellite components; Based on the structural model of the satellite component layout, obtain multiple training data and preprocess the training data, where the training data includes the satellite component layout; Construct a loss function embedding physical prior according to the steady-state heat conduction equation obeyed by the temperature field of the satellite component layout; Use online data mining to determine the weights of each prediction point of the temperature field, update the loss function according to the prediction point weights, and construct a regularization term of the loss function using regional heat flux conservation to determine the final loss function; Construct a deep neural network model, and use the preprocessed training data and the final loss function to train the deep neural network model to fit the mapping relationship between the satellite component layout and the temperature field; Use the trained deep neural network model to predict the temperature field of the satellite component layout; According to the layout characteristics of the satellite components, establish a structural model of the satellite component layout in the following way: Set the satellite component layout area as a square layout area, set a small hole with a set length on one of the four sides of the square layout area as a heat dissipation hole, the temperature of the heat dissipation hole area is fixed, and the rest of the boundaries except the heat dissipation hole area are adiabatic. The satellite components are distributed at different positions in the square layout area, and one satellite component is regarded as a heat source; The obtaining of multiple training data includes: According to the number of satellite components, randomly select corresponding positions in the square layout area to place the satellite components to obtain a training data including the satellite component layout. Repeat the random selection process multiple times until the preset number of training data is obtained; The preprocessing of the training data includes: Divide the square layout area of the satellite component layout into M1×M2 grids, represent the satellite component layout with an M1×M2 matrix, and the matrix element corresponding to the grid position with a component is the component power, and the matrix element corresponding to the grid position without a component is 0; Set M1 = M2, and the loss function embedding physical prior is constructed as: Among them, T i,j represents the temperature at the grid point in the i-th row and j-th column of the satellite component layout temperature field. Δh represents the distance between two adjacent grid points in the same row or the same column. Δh = l / m, where l represents the side length of the satellite component layout area, m represents the number of grids divided in the satellite component layout area, and m = M1×M2. φ i,j represents the component power at the grid point in the i-th row and j-th column of the satellite component layout area. T i-1,j represents the temperature at the grid point in the (i - 1)-th row and j-th column of the satellite component layout temperature field. T i+1,j represents the temperature at the grid point in the (i + 1)-th row and j-th column of the satellite component layout temperature field. T i,j-1 represents the temperature at the grid point in the i-th row and (j - 1)-th column of the satellite component layout temperature field. T i,j+1 represents the temperature at the grid point in the i-th row and (j + 1)-th column of the satellite component layout temperature field.

2. The method for predicting the temperature field of the satellite component layout based on the physically prior neural network according to claim 1, wherein The weight of the prediction point is set as: where w i,j represents the weight corresponding to the grid point at the i-th row and j-th column in the temperature field of the satellite component layout, min(T) represents the lowest temperature of the temperature field of the satellite component layout, and max(T) represents the highest temperature of the temperature field of the satellite component layout.

3. The method for predicting the temperature field of the satellite component layout based on the physically prior neural network according to claim 2, wherein According to the prediction point weight, the loss function is updated as:

4. The method for predicting the temperature field of the satellite component layout based on the neural network with physical prior according to claim 3, wherein The regularization term of the loss function is set as: where, Ω q represents the region satisfying the heat flux conservation.

5. The method for predicting the temperature field of the satellite component layout based on the neural network with physical prior according to claim 1, characterized in that The deep neural network model adopts a U-net neural network.

6. The method for predicting the temperature field of the satellite component layout based on the physics prior neural network according to any one of claims 1 to 5, characterized in that, The method further includes: During the training process of the deep neural network model, according to the boundary conditions satisfied by the temperature field of the satellite component layout, perform a limiting process on the predicted temperature field output by the deep neural network model in a hard constraint manner; Calculate the value of the final loss function based on the processed predicted temperature field.

Images