CN112733275A - Satellite assembly thermal layout temperature field prediction method based on semi-supervised learning - Google Patents

Satellite assembly thermal layout temperature field prediction method based on semi-supervised learning Download PDF

Info

Publication number
CN112733275A
CN112733275A CN202110068649.7A CN202110068649A CN112733275A CN 112733275 A CN112733275 A CN 112733275A CN 202110068649 A CN202110068649 A CN 202110068649A CN 112733275 A CN112733275 A CN 112733275A
Authority
CN
China
Prior art keywords
satellite component
satellite
layout
temperature field
loss function
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110068649.7A
Other languages
Chinese (zh)
Other versions
CN112733275B (en
Inventor
龚智强
张俊
彭伟
姚雯
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
National Defense Technology Innovation Institute PLA Academy of Military Science
Original Assignee
National Defense Technology Innovation Institute PLA Academy of Military Science
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by National Defense Technology Innovation Institute PLA Academy of Military Science filed Critical National Defense Technology Innovation Institute PLA Academy of Military Science
Priority to CN202110068649.7A priority Critical patent/CN112733275B/en
Publication of CN112733275A publication Critical patent/CN112733275A/en
Application granted granted Critical
Publication of CN112733275B publication Critical patent/CN112733275B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/15—Vehicle, aircraft or watercraft design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/08—Thermal analysis or thermal optimisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Automation & Control Theory (AREA)
  • Medical Informatics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Investigating Or Analyzing Materials Using Thermal Means (AREA)

Abstract

The invention discloses a satellite component thermal layout temperature field prediction method based on semi-supervised learning, which comprises the following steps: building a structural model of the satellite assembly layout; generating a plurality of training data sets comprising satellite component layouts and temperature field distributions corresponding to the satellite component layouts, and generating a plurality of test data sets comprising satellite component layouts and temperature field distributions corresponding to the satellite component layouts; constructing a supervised learning loss function with embedded physical knowledge; training the deep learning model; and inputting the satellite component layout of the temperature field distribution to be calculated into the trained deep learning model, and acquiring the temperature field distribution corresponding to the satellite component layout. The method utilizes a large number of training samples without labels, a small number of training samples with labels and a supervised learning loss function embedded with physical knowledge to train the deep learning model, can realize the rapid calculation of the temperature field distribution of the satellite component, and has the advantages of less required calculation resources and workload, high precision and high efficiency.

Description

Satellite assembly thermal layout temperature field prediction method based on semi-supervised learning
Technical Field
The invention relates to the technical field of satellite layout design, in particular to a satellite component thermal layout temperature field prediction method based on semi-supervised learning.
Background
From cell-phone, TV, the car in daily life to satellite, tank, unmanned aerial vehicle etc. in the national defense industry, electronic equipment is very common, and electronic equipment is in the course of the work, because electronic equipment's power loss, inevitably can produce the temperature rise that heat dissipation caused electronic equipment. With the improvement of the function requirement of modern electronic equipment and the progress of modern industrial manufacturing process, the internal integration level of the electronic equipment is higher and higher, and the internal structure of the corresponding satellite component is more and more complicated. Because the satellite operates in the outer space, when the satellite assembly on the satellite has a problem, the maintenance and the replacement cannot be carried out in time; in order to ensure that the satellite can run efficiently for a long time to perform tasks, the satellite assembly is required to have better service performance and longer service life. However, the high integration level and complexity of the satellite assembly easily cause the internal temperature of the satellite assembly to rise sharply during normal operation, which has important influence on the service life, safety, reliability and the like of electronic components inside the satellite assembly, thereby affecting the service performance and service life of the whole satellite assembly. Therefore, the electronic components in the satellite assembly need to be reasonably distributed to control the temperature field distribution of the satellite assembly, so that the heat load of the satellite assembly is reduced, and the service performance and the service life of the satellite assembly are improved.
With the development of computer technology, computer simulation technology is generally adopted to complete the thermal design of the satellite component so as to improve the thermal design efficiency of the satellite component. And by using a computer simulation technology and repeated testing, the positions of components in the satellite assembly are continuously adjusted, and finally, the satellite assembly layout with reasonable temperature field distribution is found. When the computer simulation technology is used for satellite component thermal analysis, the temperature field calculation needs to be completed through a mathematical method, and the currently common calculation methods mainly include an analytic method and a numerical solution method. The analytic method is limited by the solving difficulty of a high-order differential equation, and an accurate temperature field model is often difficult to obtain. The numerical solution method comprises a finite difference method, a finite volume method and a finite element method, which can theoretically obtain any required precision, however, the higher precision is accompanied by a large amount of matrix calculation, and the temperature field distribution calculation needs to be carried out again when the satellite component layout is changed once, so that the test efficiency is low and the test period is long.
In order to solve the technical problem existing when the satellite component thermal analysis is carried out by utilizing a computer simulation technology to complete temperature field calculation, a depth agent model is constructed by utilizing a depth learning method to realize the prediction of a satellite component layout temperature field at present in view of the universal approximation property of the depth learning method. However, in the training process of the deep proxy model, a large number of labeled training samples are required, that is, the training samples of different satellite component layouts should have corresponding temperature field distributions as labels, and true value samples of the satellite component layouts and the temperature field distributions are difficult to obtain.
Disclosure of Invention
In order to solve part or all of the technical problems in the prior art, the invention provides a satellite component thermal layout temperature field prediction method based on semi-supervised learning.
Therefore, the invention discloses a satellite component thermal layout temperature field prediction method based on semi-supervised learning, which comprises the following steps:
approximately describing the structure of the satellite assembly, and establishing a structural model of the satellite assembly layout;
generating a plurality of training data sets comprising satellite component layouts and temperature field distributions corresponding to the satellite component layouts, and generating a plurality of test data sets comprising satellite component layouts and temperature field distributions corresponding to the satellite component layouts;
constructing a supervised learning loss function with embedded physical knowledge;
based on a supervised learning loss function of embedded physical knowledge, training a deep learning model by utilizing a training data set and a test data set to fit a mapping relation between the satellite component layout and the temperature field distribution;
and inputting the satellite component layout of the temperature field distribution to be calculated into the trained deep learning model, and acquiring the temperature field distribution corresponding to the satellite component layout.
In some alternative embodiments, the structural model of the satellite component layout is built in the following way:
setting the satellite component layout as a square area, and dividing the square area into n1×n1A grid, wherein a small hole with a set length is arranged on one of the four edges of the square area as a heat dissipation hole, and the temperature of the heat dissipation hole area is constant and is T0；
The electronic components are set to be of a square structure, one electronic component can be placed on one grid position, and different electronic components are placed on different grid positions according to the specific position of each electronic component in the layout of the satellite assembly.
In some alternative embodiments, generating the training dataset and the testing dataset comprises:
randomly selecting a plurality of grids in the square area, placing electronic components on the selected grids to obtain a satellite component layout, and repeating the random generation process for multiple times to obtain N satellite component layouts;
randomly extracting M satellite component layouts from the N satellite component layouts, determining temperature field distributions corresponding to the M satellite component layouts, and dividing the M satellite component layouts into M satellite component layouts1A first group of satellite component layouts and a second group of layouts comprising M2A second group of satellite component layouts, generating N-M satellite component layouts and M1Generating M from a training data set comprising a satellite component layout and a temperature field distribution corresponding to the satellite component layout2Temperature including satellite component layout and corresponding to satellite component layoutField distributed test data set, N, M, M1And M2Are all positive integers, M ═ M1+M2。
In some alternative embodiments, the supervised learning loss function that constructs the embedded physical knowledge is:
L＝w1Llaplace+w2Loutside+w3Llabel
wherein, w1、w2And w3Weight parameters respectively representing a heat conduction loss function based on a heat conduction equation, a boundary loss function based on a boundary condition, and an L1 norm loss function, LlaplaceRepresenting the heat conduction loss function based on the heat conduction equation, LoutsideRepresenting boundary loss functions based on boundary conditions, LlabelRepresenting the L1 norm loss function.
In some alternative embodiments, the heat transfer loss function based on the heat transfer equation is:
Figure BDA0002905029480000031
wherein the content of the first and second substances,
Figure BDA0002905029480000032
temperature output from a heat source whose position coordinates are x and y in a square grid area, l represents a side length of the square grid area, n represents the number of grids,
Figure BDA0002905029480000033
heat source intensity corresponding to heat source with Input (x, y) position coordinates of x and y representing normalization operation
Figure BDA0002905029480000034
As the system input obtained by the normalization process, m represents a normalization constant.
In some alternative embodiments, the boundary loss function based on the boundary condition is:
Loutside＝|Toutput|Ω|
wherein, Toutput|ΩRepresenting the temperature output at the louvered area.
In some alternative embodiments, the L1 norm loss function is:
Figure BDA0002905029480000035
xlabelrepresenting a labeled training sample, ToutputWhich is indicative of the temperature of the output of the heat source,
Figure BDA0002905029480000036
indicating the corresponding true temperature of the sample.
In some optional embodiments, the deep learning model adopts a neural network model with a main structure as a feature pyramid network.
The technical scheme of the invention has the following main advantages:
according to the satellite component thermal layout temperature field prediction method based on semi-supervised learning, the distribution of the satellite component temperature field is calculated and determined by adopting the deep learning model, so that the problems of large workload and low efficiency caused by adopting a traditional numerical solution can be solved; meanwhile, on the basis of the deep learning model, the training of the deep learning model is completed by utilizing a large number of training samples without labels and a small number of training samples with labels based on the supervised learning loss function of embedded physical knowledge, so that the problem that the existing deep learning model needs a large number of labeled samples during training can be solved, the computing resources and the computing workload are effectively saved, the working efficiency is improved, and the computing precision of the distribution of the thermal layout temperature field of the satellite component can be ensured.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flowchart of a method for predicting a thermal layout temperature field of a satellite assembly based on semi-supervised learning according to an embodiment of the present invention;
fig. 2 is a schematic structural model diagram of a satellite component layout according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the specific embodiments of the present invention and the accompanying drawings. It is to be understood that the described embodiments are merely a few embodiments of the invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
The technical scheme provided by the embodiment of the invention is described in detail below with reference to the accompanying drawings.
As shown in fig. 1, an embodiment of the present invention provides a method for predicting a thermal layout temperature field of a satellite assembly based on semi-supervised learning, the method including the following steps:
s1, approximately describing the structure of the satellite assembly, and establishing a structural model of the satellite assembly layout;
s2, generating a plurality of training data sets comprising satellite component layouts and a plurality of training data sets comprising the satellite component layouts and temperature field distributions corresponding to the satellite component layouts, and generating a plurality of test data sets comprising the satellite component layouts and the temperature field distributions corresponding to the satellite component layouts;
s3, constructing a supervised learning loss function of embedded physical knowledge;
s4, training a deep learning model by utilizing a training data set and a test data set based on a supervised learning loss function of embedded physical knowledge to fit the mapping relation between the satellite component layout and the temperature field distribution;
and S5, inputting the satellite component layout of the temperature field distribution to be calculated into the trained deep learning model, and acquiring the temperature field distribution corresponding to the satellite component layout.
The following is a detailed description of the steps and principles of the method for predicting the thermal layout temperature field of a satellite component based on semi-supervised learning according to an embodiment of the present invention.
S1, approximately describing the structure of the satellite assembly, and establishing a structural model of the satellite assembly layout;
specifically, based on the structural characteristics and layout characteristics of the satellite components, the structure of the satellite components is approximately described, and a structural model of the satellite component layout is established.
In one embodiment of the present invention, the structural features and layout features of the satellite components are: the satellite assembly layout is a set region layout, a set number of electronic components with different sizes are distributed in the set region, and the electronic components continuously generate heat to dissipate when working normally, namely, the electronic components are regarded as heat sources; the satellite assembly adopts the natural heat dissipation mode to dispel the heat, has seted up the louvre of setting for the size on satellite assembly overall arrangement upside, and the louvre department temperature is invariable, and except the louvre, satellite assembly cloth overall arrangement is adiabatic all around.
As shown in fig. 2, based on the structural features and the layout features of the satellite components, a structural model of the satellite component layout is established in the following manner:
setting the layout of the satellite components as a square area, and meshing the square area to divide the square area into n1×n1A grid, wherein a small hole with a set length is arranged on one of the four sides of the square layout region as a heat dissipation hole, and the temperature of the heat dissipation hole region is constant and is T0(ii) a The electronic components are set to be of a square structure, one electronic component can be placed on one grid position, and different electronic components are placed on different grid positions according to the specific position of each electronic component in the layout of the satellite assembly.
In an embodiment of the invention, for different satellite component layouts, the structural models corresponding to the different satellite component layouts can be obtained by using the method.
S2, generating a plurality of training data sets comprising satellite component layouts and a plurality of training data sets comprising the satellite component layouts and temperature field distributions corresponding to the satellite component layouts, and generating a plurality of test data sets comprising the satellite component layouts and the temperature field distributions corresponding to the satellite component layouts;
specifically, on the basis of the structure model of the satellite component layout established above, a plurality of grids in a square area are randomly selected, electronic components are placed on the selected grids to obtain a satellite component layout, the random generation process is repeated for multiple times to obtain N satellite component layouts, M satellite component layouts are randomly extracted from the obtained N satellite component layouts, the temperature field distributions corresponding to the M satellite component layouts are calculated by using a finite difference method or a finite element method or a fenics resolver, and the M satellite component layouts are divided into M satellite component layouts1A first group of satellite component layouts and a second group of layouts comprising M2A second group of satellite component layouts, generating N-M satellite component layouts and M1Generating M from a training data set comprising a satellite component layout and a temperature field distribution corresponding to the satellite component layout2A test data set comprising a satellite component layout and a temperature field distribution corresponding to the satellite component layout, N, M, M1And M2Are all positive integers, M ═ M1+M2。
In one embodiment of the present invention, NM; optionally, N ═ 50M.
The more the training data sets and the testing data sets, the higher the prediction accuracy of the deep learning model after training, but the more the training data sets and the testing data sets, the larger the corresponding calculation workload. To this end, in an embodiment of the present invention, 50000 satellite component layouts are randomly obtained, 1000 satellite component layouts are randomly extracted from the 50000 satellite component layouts, temperature field distributions corresponding to the 1000 satellite component layouts are determined, the 1000 satellite component layouts are divided into a first group including 10 satellite component layouts and a second group including 990 satellite component layouts, 49000 training data sets including the satellite component layouts and the 10 temperature field distributions corresponding to the satellite component layouts and the satellite component layouts are generated, and 990 test data sets including the satellite component layouts and the temperature field distributions corresponding to the satellite component layouts are generated.
S3, constructing a supervised learning loss function of embedded physical knowledge;
setting the distribution of the temperature field of the satellite component not to change along with time, and solving the distribution of the temperature field of the satellite component through a heat conduction equation shown in a formula 1;
Figure BDA0002905029480000061
in the formula, x and y represent coordinates of a point in a two-dimensional plane, T represents a temperature at the point, k represents a heat transfer coefficient,
Figure BDA0002905029480000062
indicating the intensity of the heat source at that point; for a square grid area corresponding to the satellite component layout, x and y represent the location coordinates of the grid.
Because the temperature field of the satellite component can be solved through the heat conduction equation, in one embodiment of the invention, the heat conduction loss function of the deep learning model is constructed based on the heat conduction equation, and the supervised learning loss function embedded with physical knowledge is obtained so as to carry out targeted guidance on the training process of the deep learning model.
Specifically, for the heat conduction equation shown in equation 1, it is set:
Figure BDA0002905029480000063
the problem of solving the heat transfer equation can be converted to minimize S0To a problem of (a). In one embodiment of the present invention, the heat transfer coefficient k is 1, which is the intensity of the heat source
Figure BDA0002905029480000064
Dividing by m as the normalization process yields the system Input (x, y), then equation 2 can be further derived as:
Figure BDA0002905029480000071
in the formula, Tx,yTemperature, T, at points with location coordinates x and yx+1,yTemperature, T, at points with location coordinates x +1 and yx-1,yRepresenting the temperature, T, at points with location coordinates x-1 and yx,y+1Temperature, T, at points with location coordinates x and y +1x,y-1Representing the temperature at points with location coordinates x and y-1, and m represents a normalization constant used to normalize the neural network inputs.
Since the satellite component layout is set as a square grid area, assuming that the side length of the square grid area is l, the grid number is n, and n is n1×n1And then Δ x ═ Δ y ═ l/x.
Further, equation 3 can be derived as:
Figure BDA0002905029480000072
the normalization operation is performed on equation 4, and the following results are obtained:
Figure BDA0002905029480000073
further, the heat conduction loss function can be expressed as:
Figure BDA0002905029480000074
in the formula, the first step is that,
Figure BDA0002905029480000075
denotes the normalization operation, T0Indicating the temperature at the heat sink hole.
Setting the temperature of the heat source output to
Figure BDA0002905029480000081
The thermal conduction loss function can be rewritten as:
Figure BDA0002905029480000082
from the derivation process described above, a heat conduction loss function based on the heat conduction equation can be constructed.
Further, since certain boundary conditions are required to solve the temperature field distribution of the satellite assembly, a boundary loss function based on the boundary conditions is required to be constructed. In one embodiment of the invention, the temperature at the radiating holes is constant to T due to the layout of the satellite assembly0That is, the following formula 8 is required to be satisfied at the louver region Ω:
Figure BDA0002905029480000085
Toutput|Ωrepresenting the temperature output at the louvered area.
For this reason, a boundary loss function based on the boundary condition is constructed as follows:
Loutside＝|Toutput|Ωequation 9
Other boundary conditions, such as a first type of boundary condition (Dirichlet boundary condition), a second type of boundary condition (Neumann boundary condition), and a third type of boundary condition (Robin boundary condition), can also be satisfied due to the solution of the temperature field distribution; in one embodiment of the invention, except for heat dissipation at the heat dissipation holes, other boundary conditions are realized by copying boundaries, namely, the internal and external temperatures at the layout boundaries of the satellite components are the same.
Further, in an embodiment of the present invention, prior physical knowledge is embedded into a loss function of a deep learning model to perform targeted guidance on a training process of the deep learning model, and a small number of labeled samples, that is, training samples including a satellite component layout and temperature field distribution corresponding to the satellite component layout, are introduced, and the labeled supervised loss function is used as a regularization term to be combined with a loss function of embedded physical knowledge.
For the supervised learning component, L1 loss was used as the corresponding loss function to assist the unsupervised training process.
Specifically, the L1 norm loss function is constructed as:
Figure BDA0002905029480000083
in the formula, the first step is that,
Figure BDA0002905029480000084
representing the true temperature, x, to which the sample correspondslabelRepresenting a labeled training sample.
Further, on the basis of the heat conduction loss function based on the heat conduction equation, the boundary loss function based on the boundary condition, and the L1 norm loss function constructed as described above, the supervised learning loss function with embedded physical knowledge is:
L＝w1Llaplace+w2Loutside+w3Llabelequation 11
In the formula, w1、w2And w3Weight parameters of a heat conduction loss function based on a heat conduction equation, a boundary loss function based on a boundary condition, and an L1 norm loss function are respectively expressed.
S4, training a deep learning model by utilizing a training data set and a test data set based on a supervised learning loss function of embedded physical knowledge to fit the mapping relation between the satellite component layout and the temperature field distribution;
specifically, in an embodiment of the present invention, the deep learning model adopts a neural network model whose main structure is a Feature Pyramid Network (FPN), the Feature Pyramid network uses a residual error network (ResNet) as a framework and is divided into a bottom-up path, a top-down path and a middle connection portion, Feature maps of different scales output by each layer are fused and mapped, and distribution and output of the satellite component layout temperature field can be obtained.
In an embodiment of the present invention, the input of the deep learning model is n1×n1And (3) the satellite component layout under the grid scale, wherein the value of each grid indicates whether a heatable electronic component exists at the grid position.
Because a large amount of labeled training data sets are needed in training of the deep learning model based on supervised learning, the temperature field distribution obtained by prediction of the deep learning model based on unsupervised learning generally generates certain transformation, so that the predicted maximum temperature generally has a certain difference with the actual temperature, and the prediction error is large.
In one embodiment of the invention, a large number of samples comprising satellite component layout and a small number of samples comprising satellite component layout and temperature field distribution corresponding to the satellite component layout are randomly generated to serve as training data sets, a small number of samples comprising satellite component layout and temperature field distribution corresponding to the satellite component layout are generated to serve as testing data sets, a deep learning model is trained by adopting error back propagation based on a constructed supervised learning loss function with embedded physical knowledge, and the deep learning model is stopped when training reaches a set training frequency and is stored.
In an embodiment of the present invention, when the total number of the training data set and the test data set is 50000, the number of training iterations may be set to 50.
S5, inputting the layout of the satellite component with the temperature field distribution to be calculated into the trained deep learning model, and acquiring the temperature field distribution corresponding to the layout of the satellite component;
further, after the deep learning model is trained and stored, the stored deep learning model is loaded, and the satellite component layout of the temperature field distribution to be calculated is input into the deep learning model to obtain the temperature field distribution corresponding to the satellite component layout, so that the optimization design of the satellite component layout is assisted.
Therefore, the satellite component thermal layout temperature field prediction method based on semi-supervised learning provided by the embodiment of the invention realizes the calculation and determination of the satellite component temperature field distribution by adopting the deep learning model, and can avoid the problems of large workload and low efficiency caused by adopting a traditional numerical solution; meanwhile, on the basis of the deep learning model, the training of the deep learning model is completed by utilizing a large number of training samples without labels and a small number of training samples with labels based on the supervised learning loss function of embedded physical knowledge, so that the problem that the existing deep learning model needs a large number of labeled samples during training can be solved, the computing resources and the computing workload are effectively saved, the working efficiency is improved, and the computing precision of the distribution of the thermal layout temperature field of the satellite component can be ensured.
It is noted that, in this document, relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. In addition, "front", "rear", "left", "right", "upper" and "lower" in this document are referred to the placement states shown in the drawings.
Finally, it should be noted that: the above examples are only for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (8)

1. A satellite component thermal layout temperature field prediction method based on semi-supervised learning is characterized by comprising the following steps:
approximately describing the structure of the satellite assembly, and establishing a structural model of the satellite assembly layout;
generating a plurality of training data sets comprising satellite component layouts and temperature field distributions corresponding to the satellite component layouts, and generating a plurality of test data sets comprising satellite component layouts and temperature field distributions corresponding to the satellite component layouts;
constructing a supervised learning loss function with embedded physical knowledge;
based on a supervised learning loss function of embedded physical knowledge, training a deep learning model by utilizing a training data set and a test data set to fit a mapping relation between the satellite component layout and the temperature field distribution;
and inputting the satellite component layout of the temperature field distribution to be calculated into the trained deep learning model, and acquiring the temperature field distribution corresponding to the satellite component layout.
2. The semi-supervised learning based satellite component thermal layout temperature field prediction method according to claim 1, wherein the structural model of the satellite component layout is established in the following way:
setting the satellite component layout as a square area, and dividing the square area into n1×n1A grid, wherein a small hole with a set length is arranged on one of the four edges of the square area as a heat dissipation hole, and the temperature of the heat dissipation hole area is constant and is T0；
The electronic components are set to be of a square structure, one electronic component can be placed on one grid position, and different electronic components are placed on different grid positions according to the specific position of each electronic component in the layout of the satellite assembly.
3. The semi-supervised learning based satellite component thermal layout temperature field prediction method of claim 2, wherein generating a training data set and a test data set comprises:
randomly selecting a plurality of grids in the square area, placing electronic components on the selected grids to obtain a satellite component layout, and repeating the random generation process for multiple times to obtain N satellite component layouts;
randomly extracting M satellite component layouts from the N satellite component layouts, determining temperature field distributions corresponding to the M satellite component layouts, and dividing the M satellite component layouts into M satellite component layouts1A first group of satellite component layouts and a second group of layouts comprising M2A second group of satellite component layouts, generating N-M satellite component layouts and M1Generating M from a training data set comprising a satellite component layout and a temperature field distribution corresponding to the satellite component layout2A test data set comprising a satellite component layout and a temperature field distribution corresponding to the satellite component layout, N, M, M1And M2Are all positive integers, M ═ M1+M2。
4. The semi-supervised learning based satellite component thermal layout temperature field prediction method of claim 3, wherein the supervised learning loss function for building embedded physical knowledge is as follows:
L＝w1Llaplace+w2Loutside+w3Llabel
wherein, w1、w2And w3Weight parameters respectively representing a heat conduction loss function based on a heat conduction equation, a boundary loss function based on a boundary condition, and an L1 norm loss function, LlaplaceRepresenting the heat conduction loss function based on the heat conduction equation, LoutsideRepresenting boundary loss functions based on boundary conditions, LlabelRepresenting the L1 norm loss function.
5. The semi-supervised learning based satellite component thermal layout temperature field prediction method of claim 4, wherein the heat conduction loss function based on the heat conduction equation is as follows:
Figure FDA0002905029470000021
wherein the content of the first and second substances,
Figure FDA0002905029470000022
temperature output from a heat source whose position coordinates are x and y in a square grid area, l represents a side length of the square grid area, n represents the number of grids,
Figure FDA0002905029470000023
heat source intensity corresponding to heat source with Input (x, y) position coordinates of x and y representing normalization operation
Figure FDA0002905029470000024
As the system input obtained by the normalization process, m represents a normalization constant.
6. The semi-supervised learning based satellite component thermal layout temperature field prediction method of claim 5, wherein the boundary loss function based on the boundary condition is:
Loutside＝|Toutput|Ω|
wherein, Toutput|ΩRepresenting the temperature output at the louvered area.
7. The semi-supervised learning based satellite component thermal layout temperature field prediction method of claim 6, wherein the L1 norm loss function is:
Figure FDA0002905029470000025
xlabelrepresenting a labeled training sample, ToutputWhich is indicative of the temperature of the output of the heat source,
Figure FDA0002905029470000026
indicating the corresponding true temperature of the sample.
8. The semi-supervised learning based satellite component thermal layout temperature field prediction method of any one of claims 1-7, wherein the deep learning model adopts a neural network model with a main structure as a feature pyramid network.
CN202110068649.7A 2021-01-19 2021-01-19 Satellite component thermal layout temperature field prediction method based on semi-supervised learning Active CN112733275B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110068649.7A CN112733275B (en) 2021-01-19 2021-01-19 Satellite component thermal layout temperature field prediction method based on semi-supervised learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110068649.7A CN112733275B (en) 2021-01-19 2021-01-19 Satellite component thermal layout temperature field prediction method based on semi-supervised learning

Publications (2)

Publication Number Publication Date
CN112733275A true CN112733275A (en) 2021-04-30
CN112733275B CN112733275B (en) 2023-07-25

Family

ID=75592400

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110068649.7A Active CN112733275B (en) 2021-01-19 2021-01-19 Satellite component thermal layout temperature field prediction method based on semi-supervised learning

Country Status (1)

Country Link
CN (1) CN112733275B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113609763A (en) * 2021-07-27 2021-11-05 中国人民解放军军事科学院国防科技创新研究院 Uncertainty-based satellite component layout temperature field prediction method
CN114781207A (en) * 2022-03-29 2022-07-22 中国人民解放军军事科学院国防科技创新研究院 Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning
CN115630565A (en) * 2022-09-28 2023-01-20 中国人民解放军军事科学院国防科技创新研究院 Temperature field reconstruction method based on embedded physical knowledge neural network

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111222648A (en) * 2020-01-15 2020-06-02 深圳前海微众银行股份有限公司 Semi-supervised machine learning optimization method, device, equipment and storage medium
CN111507032A (en) * 2020-03-31 2020-08-07 中国人民解放军军事科学院国防科技创新研究院 Deep learning technology-based component layout optimization design method for predicting temperature distribution
CN111782459A (en) * 2020-06-03 2020-10-16 东南大学 Traveling wave tube internal temperature prediction method based on tube external temperature image and CNN
CN111783331A (en) * 2020-06-04 2020-10-16 中国人民解放军军事科学院国防科技创新研究院 Satellite component temperature field determination method based on transfer learning
CN111783337A (en) * 2020-06-28 2020-10-16 中国人民解放军军事科学院国防科技创新研究院 Hyper-parameter optimization method of satellite thermal layout agent model

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111222648A (en) * 2020-01-15 2020-06-02 深圳前海微众银行股份有限公司 Semi-supervised machine learning optimization method, device, equipment and storage medium
CN111507032A (en) * 2020-03-31 2020-08-07 中国人民解放军军事科学院国防科技创新研究院 Deep learning technology-based component layout optimization design method for predicting temperature distribution
CN111782459A (en) * 2020-06-03 2020-10-16 东南大学 Traveling wave tube internal temperature prediction method based on tube external temperature image and CNN
CN111783331A (en) * 2020-06-04 2020-10-16 中国人民解放军军事科学院国防科技创新研究院 Satellite component temperature field determination method based on transfer learning
CN111783337A (en) * 2020-06-28 2020-10-16 中国人民解放军军事科学院国防科技创新研究院 Hyper-parameter optimization method of satellite thermal layout agent model

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
匡秋明;于廷照;: "AI技术与卫星资料应用研究现状分析", 气象科技进展, no. 03 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113609763A (en) * 2021-07-27 2021-11-05 中国人民解放军军事科学院国防科技创新研究院 Uncertainty-based satellite component layout temperature field prediction method
CN113609763B (en) * 2021-07-27 2022-11-11 中国人民解放军军事科学院国防科技创新研究院 Uncertainty-based satellite component layout temperature field prediction method
CN114781207A (en) * 2022-03-29 2022-07-22 中国人民解放军军事科学院国防科技创新研究院 Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning
CN114781207B (en) * 2022-03-29 2024-04-12 中国人民解放军军事科学院国防科技创新研究院 Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning
CN115630565A (en) * 2022-09-28 2023-01-20 中国人民解放军军事科学院国防科技创新研究院 Temperature field reconstruction method based on embedded physical knowledge neural network
CN115630565B (en) * 2022-09-28 2023-12-15 中国人民解放军军事科学院国防科技创新研究院 Temperature field reconstruction method based on embedded physical knowledge neural network

Also Published As

Publication number Publication date
CN112733275B (en) 2023-07-25

Similar Documents

Publication Publication Date Title
CN112257321B (en) Unsupervised learning-based satellite component temperature field determination method
CN112733275A (en) Satellite assembly thermal layout temperature field prediction method based on semi-supervised learning
CN111507032B (en) Deep learning technology-based component layout optimization design method for predicting temperature distribution
Witherden et al. Future directions in computational fluid dynamics
CN113177290B (en) Satellite component temperature field prediction method based on depth agent model normalization
Poole et al. Control point-based aerodynamic shape optimization applied to AIAA ADODG test cases
Deng et al. An efficient 146-line 3D sensitivity analysis code of stress-based topology optimization written in MATLAB
Clark Jr et al. Engineering design exploration using locally optimized covariance kriging
CN111783331B (en) Satellite component temperature field determination method based on transfer learning
Rao et al. A time-parallel approach to strong-constraint four-dimensional variational data assimilation
CN114036787A (en) Satellite component thermal layout temperature field prediction based on physical prior self-supervision learning
Oktay et al. Three-dimensional structural topology optimization of aerial vehicles under aerodynamic loads
US20210295167A1 (en) Generative networks for physics based simulations
Zijing et al. A new hybrid aerodynamic optimization framework based on differential evolution and invasive weed optimization
CN113221326B (en) Satellite component temperature field prediction method based on teacher-student agent model
Bai et al. DEIM reduced order model constructed by hybrid snapshot simulation
WO2022247092A1 (en) Methods and systems for congestion prediction in logic synthesis using graph neural networks
US11023639B2 (en) Systems and methods for providing approximate electronic-structure models from calculated band structure data
CN114781207B (en) Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning
Zhang et al. A graphics processing unit-accelerated meshless method for two-dimensional compressible flows
Banerjee et al. Cmt-bone—a proxy application for compressible multiphase turbulent flows
Vanderwyst et al. Reduced order models for generation of large, high speed aerodynamic databases with jet interactions
Strönisch et al. Flow field prediction on large variable sized 2D point clouds with graph convolution
Raghavan et al. Implicit constraint handling for shape optimisation with pod-morphing
Zhang et al. Wing jig shape optimisation with gradient-assisted metamodel building in a trust-region optimisation framework

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant