CN114781207A - Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning - Google Patents
Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning Download PDFInfo
- Publication number
- CN114781207A CN114781207A CN202210318240.0A CN202210318240A CN114781207A CN 114781207 A CN114781207 A CN 114781207A CN 202210318240 A CN202210318240 A CN 202210318240A CN 114781207 A CN114781207 A CN 114781207A
- Authority
- CN
- China
- Prior art keywords
- training data
- temperature field
- model
- component layout
- uncertainty
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 205
- 238000013135 deep learning Methods 0.000 claims abstract description 65
- 238000003062 neural network model Methods 0.000 claims abstract description 57
- 230000008569 process Effects 0.000 claims abstract description 15
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 14
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 239000000126 substance Substances 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 claims description 2
- 238000000638 solvent extraction Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 abstract description 27
- 238000013461 design Methods 0.000 description 17
- 238000005457 optimization Methods 0.000 description 16
- 239000003795 chemical substances by application Substances 0.000 description 8
- 238000004088 simulation Methods 0.000 description 8
- 230000009471 action Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000005315 distribution function Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005293 physical law Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- 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
- 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)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a heat source layout temperature field prediction method based on uncertainty and semi-supervised learning, which comprises the following steps of: acquiring first training data and second training data; preprocessing training data; constructing a deep learning neural network model; carrying out random initialization on the deep learning neural network model, training the model by utilizing first training data, and repeating the processes of random initialization and training for multiple times to obtain a plurality of trained models; determining a model to be updated and an auxiliary model; inputting the component layout of the second training data into the auxiliary model, and calculating a predicted temperature field and predicted uncertainty corresponding to the component layout; training a model to be updated by utilizing the first training data, the second training data and the corresponding predicted temperature field and prediction uncertainty; and predicting the temperature field by using the model to be updated. The invention can obtain the deep learning agent model with higher prediction precision by using a small amount of labeled training data, thereby reducing the calculation cost.
Description
Technical Field
The invention relates to the technical field of component layout optimization design, in particular to a heat source layout temperature field prediction method based on uncertainty and semi-supervised learning.
Background
To perform different functions and perform various tasks, satellites often have a large number of integrated components that generate a large amount of heat during normal operation, which can be considered as a heat source. In order to meet the overall function of the satellite, it is usually considered when determining the layout of each component inside the satellite whether the temperature field under the current component layout meets the design requirements, such as whether the maximum temperature is too high, whether the temperature of a specific location is too high, etc., so as to avoid the reliability of the component from being reduced or the component from failing due to the too high temperature. Therefore, how to determine the temperature field corresponding to the component layout is an important issue to be solved when optimizing the component layout design.
Two methods are currently mainly used for how to determine the temperature field corresponding to the component layout. The first method is to model the component layout, and then perform simulation analysis on the component layout by adopting numerical calculation methods such as a finite element method, a finite difference method, a finite volume method and the like to determine a temperature field corresponding to the component layout. The second method is to construct a deep learning agent model, train the agent model by adopting a large amount of sample data with temperature field labels, and then predict the temperature field of the component layout by utilizing the trained deep learning agent model.
However, because the component layout optimization design is a repeated iteration process, the existing finite element-based simulation calculation method determines that the single-time simulation calculation efficiency is not too high, while in the iterative optimization process, the component layout temperature field needs to be analyzed and calculated for many times, a large amount of calculation resources and calculation time are consumed, the calculation cost is high, the optimization efficiency is low, and the required calculation resources and calculation time are increased step by step along with the complexity of the component layout and the calculation accuracy of the component layout temperature field. The existing method for predicting the temperature field of the satellite component layout by using the deep learning agent model needs to train the agent model by using a large amount of sample data with temperature field labels, because the real data of the temperature field corresponding to the satellite component layout is difficult to obtain, each sample data with the temperature field labels needs to be obtained by means of simulation experiments and numerical calculation, more calculation resources and calculation time are consumed, and the problems of high calculation cost and low efficiency of component layout optimization design still exist.
Disclosure of Invention
In order to solve part or all of the technical problems in the prior art, the invention provides a heat source layout temperature field prediction method based on uncertainty and semi-supervised learning.
The technical scheme of the invention is as follows:
a heat source layout temperature field prediction method based on uncertainty and semi-supervised learning is provided, and the method comprises the following steps:
acquiring first training data and second training data, wherein the first training data comprises component layout and a temperature field corresponding to the component layout, and the second training data comprises component layout;
preprocessing the first training data and the second training data;
constructing a deep learning neural network model;
randomly initializing a deep learning neural network model, training the deep learning neural network model by using the first training data to obtain a trained deep learning neural network model, and repeatedly performing random initialization and training processes for multiple times to obtain a plurality of trained deep learning neural network models;
determining a model to be updated and an auxiliary model, wherein the model to be updated is one of the trained deep learning neural network models with the best prediction performance, and the auxiliary model is the other models except the model to be updated;
inputting the component layout of the second training data into the auxiliary model, and calculating a predicted temperature field and a predicted uncertainty corresponding to the input component layout;
training the model to be updated by using the first training data, the second training data and the corresponding predicted temperature field and prediction uncertainty thereof to update model parameters;
and predicting the component layout temperature field by using the trained model to be updated.
In some possible implementations, the obtaining the first training data and the second training data includes:
carrying out grid division on the component layout area, randomly selecting a corresponding number of grids to place components according to the number of the components to obtain a component layout, simulating and calculating a temperature field corresponding to the component layout by using a finite element method to obtain first training data comprising the component layout and the temperature field corresponding to the component layout, and repeating the random obtaining process for multiple times until first training data of a first preset number are obtained;
and carrying out grid division on the component layout area, randomly selecting a corresponding number of grid placement components according to the number of the components to obtain a component layout, obtaining second training data comprising the component layout, and repeating the random obtaining process for multiple times until second training data of a second preset number are obtained.
In some possible implementations, it is set that: partitioning a component layout area into M1×M2The grid carries out component layout;
the preprocessing the first training data and the second training data includes:
the dimension for laying out the components of the first training data and the components of the second training data is M1×M2The matrix of (b) indicates that the matrix element corresponding to the grid position with the component is the component power, and the matrix element corresponding to the grid position without the component is 0.
In some possible implementations, the deep learning neural network model is a U-net convolutional neural network.
In some possible implementations, the training the deep learning neural network model using the first training data includes:
and taking the component layout of the first training data as input, taking a temperature field corresponding to the input component layout as output, and training the deep learning neural network model by a gradient descent method.
In some possible implementations, the predicted temperature field corresponding to the component layout is calculated using the following formula;
wherein the content of the first and second substances,
representing the predicted temperature field for the component layout, Q representing the number of auxiliary models, TqAnd the predicted temperature field of the output of the q-th auxiliary model corresponding to the component layout is represented.
In some possible implementations, the predicted uncertainty corresponding to the component layout is calculated using the following formula;
wherein, TstdRepresenting the prediction uncertainty corresponding to the component layout.
In some possible implementations, the training the model to be updated to update the model parameters by using the first training data, the second training data, and the corresponding predicted temperature field and prediction uncertainty thereof includes:
and taking the component layout of the first training data and the component layout of the second training data as input, constructing a loss function according to the input temperature field corresponding to the component layout of the first training data, the input predicted temperature field and prediction uncertainty corresponding to the component layout of the second training data and the predicted temperature field output by the model to be updated, and training the model to be updated by a gradient descent method based on the constructed loss function.
In some possible implementations, in training the model to be updated, the loss function is set to:
wherein α and β represent preset weight coefficients, Ω represents the number of elements of the temperature field, N represents the number of first training data input to the model to be updated, sum (x) represents the sum of all elements in x,
and
a component layout and a temperature field representing the preprocessed ith first training data,
a predicted temperature field corresponding to the component layout of the ith first training data output by the model to be updated, N' represents the number of second training data input into the model to be updated,
a component layout representing the preprocessed jth second training data,
a predicted value of a temperature field, W, corresponding to the component layout of the jth second training data representing the model output to be updatedjAn uncertainty weight matrix corresponding to the component layout representing the jth second training data,
and a predicted temperature field corresponding to the component layout representing the jth second training data.
In some of the possible implementations of the present invention,uncertainty weight matrix WjDetermined using the following formula:
wherein, Tstd_jThe prediction uncertainty corresponding to the component layout of the jth second training data is represented, and γ represents an uncertainty threshold.
The technical scheme of the invention has the following main advantages:
the heat source layout temperature field prediction method based on uncertainty and semi-supervised learning can obtain a deep learning agent model with higher prediction precision by using a small amount of labeled training data, obviously reduces the requirement of model training on the labeled training data, effectively reduces the calculation time and calculation resources required for obtaining the training data, reduces the component layout optimization design cost, and improves the optimization efficiency.
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 embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a method for predicting a heat source layout temperature field based on uncertainty and semi-supervised learning in accordance with an embodiment of the present invention;
fig. 2 is a schematic diagram of a training process of a model to be updated according to an embodiment of the present 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 explained in detail in the following with the accompanying drawings.
Referring to fig. 1, an embodiment of the present invention provides a method for predicting a heat source layout temperature field based on uncertainty and semi-supervised learning, the method including the following steps:
step S1, acquiring first training data and second training data, wherein the first training data comprises component layout and a temperature field corresponding to the component layout, and the second training data comprises component layout;
step S2, preprocessing the first training data and the second training data;
step S3, constructing a deep learning neural network model;
step S4, randomly initializing the deep learning neural network model, training the deep learning neural network model by using first training data to obtain a trained deep learning neural network model, and repeatedly performing random initialization and training for multiple times to obtain multiple trained deep learning neural network models;
step S5, determining a model to be updated and an auxiliary model, wherein the model to be updated is one of the plurality of trained deep learning neural network models with the best prediction performance, and the auxiliary model is other than the model to be updated;
step S6, inputting the component layout of the second training data into an auxiliary model, and calculating a predicted temperature field and a predicted uncertainty corresponding to the input component layout;
step S7, training the model to be updated by utilizing the first training data, the second training data and the corresponding predicted temperature field and prediction uncertainty thereof to update the model parameters;
and step S8, predicting the component layout temperature field by using the trained model to be updated.
The uncertainty and semi-supervised learning based heat source layout temperature field prediction method provided by the embodiment of the invention firstly trains a plurality of deep learning neural network models by using the labeled training data, then performs integrated prediction of the unlabeled training data by using the trained deep learning neural network models, determines the predicted temperature field and the prediction uncertainty corresponding to the unlabeled training data, trains the model with the best prediction performance in the trained deep learning neural network models again by using the labeled training data, the unlabeled training data, the predicted temperature field and the prediction uncertainty corresponding to the labeled training data to update the model parameters, can obtain a deep learning agent model with higher prediction accuracy by using a small amount of labeled training data, obviously reduces the requirement of model training on the labeled training data, effectively reduces the calculation time and calculation resources required for obtaining the training data, the optimization design cost of the component layout is reduced, and the optimization efficiency is improved.
In the case of the component layout optimization design problem considering the temperature field index, each component can be simplified to be a heat source. The component layout optimization design may be regarded as a heat source layout optimization design. When a certain number of components need to be placed in a certain layout area, the purpose of the component layout optimization design is to enhance the heat conduction efficiency of the layout area by optimizing the positions of the components to minimize the maximum temperature of the layout area. Considering a heat source layout optimization design problem in a two-dimensional plane, a mathematical model of the heat source layout optimization design problem can be expressed as:
wherein X represents the layout scheme of the heat source, (X)i,yi) Position coordinates representing the ith heat source, NsDenotes the number of heat sources, T denotes the temperature field of the layout area, k denotes the thermal conductivity of the layout area, (x, y) denotes the position coordinates of any point of the layout area, [ phi ] (x, y) denotes the intensity distribution function of the heat sources, T0Representing the temperature at the isothermal boundary, n representing the normal direction at the boundary, h representing the distance between the object on the boundary and the surrounding fluidThe surface heat transfer coefficient of (2).
The intensity distribution function phi (x, y) of the heat source is determined by the position of the heat source, and is specifically expressed as:
wherein phiiRepresenting the intensity of the ith heat source, ΓiIndicating the layout area covered by the heat source.
The following specifically describes the steps and principles of the uncertainty and semi-supervised learning based heat source layout temperature field prediction method provided by an embodiment of the present invention, taking optimization design of component layout in a two-dimensional plane as an example:
in step S1, first training data and second training data are acquired.
In an embodiment of the present invention, the first training data and the second training data may be obtained by:
grid division is carried out on the component layout area, and a component layout X is obtained by randomly selecting a corresponding number of grid placement components according to the number of the componentsmSimulating and calculating the temperature field T corresponding to the component layout by using a finite element methodmObtaining a first training data (X) comprising a component layout and a temperature field corresponding to the component layoutm,Tm) Repeating the random acquisition process for a plurality of times until a first preset number M of first training data is obtained (X)m,Tm)|m＝1,2,…,M}；
Grid division is carried out on the component layout area, and a component layout X 'is obtained by randomly selecting a corresponding number of grid placement components according to the number of the components'm′Obtaining a second training data (X ') comprising the layout of the components'm′) Repeating the random acquisition process for multiple times until a second preset number M ' of second training data { (X { ' is obtained 'm′)|m′＝1,2,…,M′}。
The specific amount of the first training data and the specific amount of the second training data may be set according to training efficiency and precision requirements, for example, M may be 1000 to 5000 or other values, and M' may be 20000 to 50000 or other values. In order to improve the prediction accuracy of the finally obtained model and reduce the calculation cost, the number of the second training data is larger than that of the first training data.
In an embodiment of the present invention, the component layout area is determined according to an actual layout requirement, and the number of grids divided by the component layout area is also determined according to an actual situation. For example, when the component layout region is a square component layout region, the square component layout region may be divided into M1×M2Grids, each grid can be used for placing at most one heat source component, when the number of the heat source components is NsWhen the heat source component is arranged, the number of the grids selected for placing the heat source component is also NsAnd (4) respectively.
The grid number of the component layout area division can be determined according to the resolution and the precision required by the component layout design, and when the required resolution and precision are higher, the more the grid number is divided, namely the finer the grid division is. For example, M may be1×M2Set to 200 x 200.
Further, when calculating the temperature field corresponding to the component layout by using the finite element method, the component layout region may be divided into N1×N2And carrying out simulation calculation on the finite element calculation grids to obtain a temperature field corresponding to the component layout. The specific numerical value of the finite element calculation grid can be determined according to the resolution and precision required by the component layout design, and when the required resolution and precision are higher, the more finite element calculation grids are divided during the simulation calculation, but correspondingly, the more time and resources are required by the simulation calculation. Optionally, to facilitate data processing, N1×N2May be equal to M1×M2。
Step S2, pre-processing the first training data and the second training data.
In particular, to facilitate training subsequently built deep-learning neural network models with training data to divide component layout regions into M1×M2Performing component layout on the grid as an example, for the first training data and the second training dataAnd preprocessing the training data, including:
the dimension for laying out the components of the first training data and the components of the second training data is M1×M2The matrix of (b) indicates that the matrix element corresponding to the grid position with the component is the component power, and the matrix element corresponding to the grid position without the component is 0.
Further, to facilitate data processing, the component layout area is divided into N1×N2For example, the finite element computing mesh performs temperature field simulation calculation, and preprocesses the first training data and the second training data, and may further include:
dimension is N1×N2Temperature field matrix TmThe values of (2) are normalized to ensure that the matrix elements all belong to [0,1 ]]。
And step S3, constructing a deep learning neural network model.
In one embodiment of the invention, a U-Net convolution neural network can be used as a deep learning neural network model to fit the mapping relation between the component layout and the temperature field. The deep learning neural network model is trained by utilizing the acquired training data, so that the deep learning neural network model can learn the physical law in the training data, a neural network with strong generalization capability is obtained, and the rapid prediction of the temperature field corresponding to the component layout can be realized. In essence, the deep learning neural network model is a proxy model.
And step S4, performing random initialization on the deep learning neural network model, training the deep learning neural network model by using the first training data to obtain a trained deep learning neural network model, and repeating the random initialization and training processes for multiple times to obtain a plurality of trained deep learning neural network models.
In order to obtain a proxy model with higher prediction accuracy by using a small amount of labeled training data, in an embodiment of the invention, a plurality of deep learning proxy models, namely a plurality of trained deep learning neural network models, are obtained by using labeled training data through a random initialization and independent training mode.
Specifically, when a deep learning agent model is obtained, the deep learning neural network model is randomly initialized to determine initial parameters of the model, and then the initialized deep learning neural network model is trained by using first training data, so that a trained deep learning neural network model is obtained.
In an embodiment of the present invention, training a deep learning neural network model by using first training data includes:
and taking the component layout of the first training data as input, taking a temperature field corresponding to the input component layout as output, and training the deep learning neural network model by a gradient descent method.
Optionally, when the deep neural network model is trained, the corresponding loss function may be set according to an actual situation, for example, Mean Absolute Error (MAE) may be used as the loss function.
The more the number of the obtained deep learning neural network models is, the higher the prediction accuracy is when the integrated deep learning neural network models are subsequently used for predicting the temperature field, but the longer the corresponding training time and training period are. In one embodiment of the invention, the number of the deep learning neural network models can be set according to the prediction precision and the training time of the actual demand. For example, 4 to 8 can be provided.
And step S5, determining a model to be updated and an auxiliary model, wherein the model to be updated is one of the plurality of trained deep learning neural network models with the best prediction performance, and the auxiliary model is the other models except the model to be updated.
In an embodiment of the present invention, the prediction performance of the deep learning neural network model may be determined by calculating an average Absolute Error (MAE) corresponding to the model, and when the MAE is lower, the prediction performance of the model is better.
Specifically, the best predictive performance one of the trained deep learning neural network models may be determined using:
selecting a certain amount of first training data, inputting the component layout of the selected first training data into a trained deep learning neural network model to obtain a predicted temperature field corresponding to the component layout output by the deep learning neural network model, calculating an average absolute error according to the temperature field of the first training data and the predicted temperature field output by the deep learning neural network model, and determining the deep learning neural network with the minimum average absolute error as the deep learning neural network model with the best prediction performance.
And step S6, inputting the component layout of the second training data into an auxiliary model, and calculating the predicted temperature field and the predicted uncertainty corresponding to the input component layout.
Referring to fig. 2, in fig. 2, there are 5 deep learning neural network models including Unet, Unet2, Unet3, Unet4, and Unet5, and 4 auxiliary models including Unet2, Unet3, Unet4, and Unet 5. In an embodiment of the present invention, the component layout of the second training data is input to each auxiliary model to obtain the temperature field output by each auxiliary model, and the predicted temperature field and the prediction uncertainty corresponding to the component layout are calculated according to the temperature fields output by all the auxiliary models.
In an embodiment of the present invention, the predicted temperature field corresponding to the component layout may be calculated and determined by using the following formula:
wherein the content of the first and second substances,
representing the predicted temperature field corresponding to the component layout, Q representing the number of auxiliary models, TqAnd the predicted temperature field of the output of the q-th auxiliary model corresponding to the component layout is represented.
In one embodiment of the invention, the prediction uncertainty corresponding to the component layout can be calculated and determined by using the following formula;
wherein, TstdRepresenting the prediction uncertainty corresponding to the component layout.
According to the above calculation formula, by predicting the temperature field of the component layout of each second training data by using the plurality of auxiliary models, the predicted temperature field and the prediction uncertainty corresponding to the component layout of each second training data can be determined.
In an embodiment of the present invention, the accuracy of prediction of the temperature field can be further improved by performing the integrated prediction of the component layout of the second training data using the plurality of auxiliary models. And the quality degree of the predicted temperature field can be measured according to the obtained prediction uncertainty, and the smaller the prediction uncertainty is, the higher the reliability of the predicted temperature field is.
And step S7, training the model to be updated by utilizing the first training data, the second training data and the corresponding predicted temperature field and prediction uncertainty thereof to update the model parameters.
In an embodiment of the invention, the model to be updated is trained by utilizing the first training data, the second training data, the corresponding predicted temperature field and the prediction uncertainty, so that the problem of small quantity of the first training data can be solved, and the prediction precision of the model to be updated is further improved.
Referring to fig. 2, in an embodiment of the present invention, training a model to be updated by using first training data, second training data, and corresponding predicted temperature fields and prediction uncertainties to update model parameters may include:
the method comprises the steps of taking component layout of first training data and component layout of second training data as input, constructing a loss function according to a temperature field corresponding to the input component layout of the first training data, a predicted temperature field and prediction uncertainty corresponding to the input component layout of the second training data and a predicted temperature field output by a model to be updated, and training the model to be updated through a gradient descent method based on the constructed loss function.
Optionally, when training the model to be updated, the loss function may be:
wherein alpha and beta represent preset weight coefficients,
for the corresponding loss function part of the first training data, Ω represents the number of elements of the temperature field, N represents the number of first training data input to the model to be updated, sum (x) represents the sum of all the elements in x,
and
a component layout and a temperature field representing the preprocessed ith first training data,
a predicted temperature field corresponding to a component layout of the ith first training data representing an output of the model to be updated,
n' represents the number of second training data input to the model to be updated,
a component layout representing the preprocessed jth second training data,
a predicted value of a temperature field, W, corresponding to the component layout of the jth second training data representing the model output to be updatedjCorresponding to the component layout representing the jth second training dataThe uncertainty weight matrix is then used to determine,
and the predicted temperature field corresponding to the component layout of the jth second training data is represented.
Uncertainty weight matrix WjCan be determined using the following formula:
wherein, Tstd_jThe prediction uncertainty corresponding to the component layout of the jth second training data is represented, and γ represents an uncertainty threshold.
Wherein the predicted temperature field corresponding to the component layout of the jth second training data
Using the predicted temperature field
Determining a prediction uncertainty T corresponding to the component layout of the jth second training datastd_jUsing the prediction uncertainty T described abovestdThe uncertainty threshold y can be set as the actual case may be.
And step S8, predicting the component layout temperature field by using the trained model to be updated.
Specifically, after the training of the model to be updated is completed, the component layout of the temperature field to be calculated is input into the model to be updated, and the predicted value of the temperature field corresponding to the component layout can be obtained, so that the optimal design of the component layout is assisted.
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 all referred to the placement state 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 (10)
1. A heat source layout temperature field prediction method based on uncertainty and semi-supervised learning is characterized by comprising the following steps:
acquiring first training data and second training data, wherein the first training data comprises component layout and a temperature field corresponding to the component layout, and the second training data comprises component layout;
preprocessing the first training data and the second training data;
constructing a deep learning neural network model;
carrying out random initialization on the deep learning neural network model, training the deep learning neural network model by using the first training data to obtain a trained deep learning neural network model, and repeating the random initialization and training processes for multiple times to obtain a plurality of trained deep learning neural network models;
determining a model to be updated and an auxiliary model, wherein the model to be updated is one of the trained deep learning neural network models with the best prediction performance, and the auxiliary model is the other models except the model to be updated;
inputting the component layout of the second training data into the auxiliary model, and calculating a predicted temperature field and a predicted uncertainty corresponding to the input component layout;
training the model to be updated by utilizing the first training data, the second training data and the corresponding predicted temperature field and prediction uncertainty thereof to update model parameters;
and predicting the component layout temperature field by using the trained model to be updated.
2. The uncertainty and semi-supervised learning based heat source layout temperature field prediction method of claim 1, wherein the obtaining first training data and second training data comprises:
carrying out gridding division on the component layout area, randomly selecting a corresponding number of grids to place components according to the number of the components to obtain a component layout, simulating and calculating a temperature field corresponding to the component layout by using a finite element method to obtain first training data comprising the component layout and the temperature field corresponding to the component layout, and repeating the random acquisition process for multiple times until first training data of a first preset number are obtained;
and carrying out grid division on the component layout area, randomly selecting a corresponding number of grids according to the number of the components, placing the components to obtain a component layout, obtaining second training data comprising the component layout, and repeating the random obtaining process for multiple times until second training data of a second preset number are obtained.
3. A heat source layout temperature field prediction method based on uncertainty and semi-supervised learning according to claim 2, characterized by setting: partitioning a component layout area into M1×M2The grid carries out component layout;
the preprocessing the first training data and the second training data includes:
the dimension for laying out the components of the first training data and the components of the second training data is M1×M2The matrix of (d) indicates that the matrix element corresponding to the grid position with the component is the component power, and the matrix element corresponding to the grid position without the component is 0.
4. A heat source layout temperature field prediction method based on uncertainty and semi-supervised learning according to any one of claims 1-3, characterized in that the deep learning neural network model is a U-net convolutional neural network.
5. The uncertainty and semi-supervised learning based heat source layout temperature field prediction method of claim 1, wherein the training of the deep learning neural network model using the first training data comprises:
and taking the component layout of the first training data as input, taking a temperature field corresponding to the input component layout as output, and training the deep learning neural network model by a gradient descent method.
6. The uncertainty and semi-supervised learning based heat source layout temperature field prediction method according to any one of claims 1 to 5, wherein a predicted temperature field corresponding to a component layout is calculated by using the following formula;
wherein the content of the first and second substances,
representing the predicted temperature field for the component layout, Q representing the number of auxiliary models, TqAnd the predicted temperature field of the output of the q-th auxiliary model corresponding to the component layout is represented.
7. A heat source layout temperature field prediction method based on uncertainty and semi-supervised learning as recited in claim 6, wherein the corresponding prediction uncertainty of the component layout is calculated using the following formula;
wherein, TstdRepresenting the prediction uncertainty corresponding to the component layout.
8. The uncertainty and semi-supervised learning based heat source layout temperature field prediction method of any one of claims 1-7, wherein the training of the model to be updated with the first training data, the second training data and their corresponding predicted temperature fields and predicted uncertainties to update model parameters comprises:
taking the component layout of the first training data and the component layout of the second training data as input, constructing a loss function according to the temperature field corresponding to the input component layout of the first training data, the predicted temperature field and the predicted uncertainty corresponding to the input component layout of the second training data and the predicted temperature field output by the model to be updated, and training the model to be updated by a gradient descent method based on the constructed loss function.
9. The uncertainty and semi-supervised learning based heat source layout temperature field prediction method of claim 8, wherein in training the model to be updated, a loss function is set as:
wherein α and β represent preset weight coefficients, Ω represents the number of elements of the temperature field, N represents the number of first training data input to the model to be updated, sum represents the sum of all elements in x,
and
a component layout and a temperature field representing the preprocessed ith first training data,
the predicted temperature field corresponding to the component layout of the ith first training data output by the model to be updated is represented, N' represents the number of the second training data input into the model to be updated,
a component layout representing the preprocessed jth second training data,
a predicted value of a temperature field, W, corresponding to the component layout of the jth second training data representing the model output to be updatedjAn uncertainty weight matrix corresponding to the component layout representing the jth second training data,
and a predicted temperature field corresponding to the component layout representing the jth second training data.
10. A heat source layout temperature field prediction method based on uncertainty and semi-supervised learning as recited in claim 9, wherein the uncertainty weight matrix WjDetermined using the following equation:
wherein, Tstd_jA prediction uncertainty corresponding to the component layout representing the jth second training dataAnd γ represents an uncertainty threshold.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210318240.0A CN114781207B (en) | 2022-03-29 | 2022-03-29 | Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210318240.0A CN114781207B (en) | 2022-03-29 | 2022-03-29 | Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114781207A true CN114781207A (en) | 2022-07-22 |
| CN114781207B CN114781207B (en) | 2024-04-12 |
Family
ID=82425779
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210318240.0A Active CN114781207B (en) | 2022-03-29 | 2022-03-29 | Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114781207B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115730509A (en) * | 2022-09-20 | 2023-03-03 | 中国人民解放军军事科学院国防科技创新研究院 | Spacecraft in-cabin temperature field reconstruction task research benchmark method based on machine learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN112733275A (en) * | 2021-01-19 | 2021-04-30 | 中国人民解放军军事科学院国防科技创新研究院 | Satellite assembly thermal layout temperature field prediction method based on semi-supervised learning |
| CN114036787A (en) * | 2021-10-09 | 2022-02-11 | 中国人民解放军军事科学院国防科技创新研究院 | Satellite component thermal layout temperature field prediction based on physical prior self-supervision learning |
| WO2022042002A1 (en) * | 2020-08-31 | 2022-03-03 | 华为技术有限公司 | Training method for semi-supervised learning model, image processing method, and device |
-
2022
- 2022-03-29 CN CN202210318240.0A patent/CN114781207B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| WO2022042002A1 (en) * | 2020-08-31 | 2022-03-03 | 华为技术有限公司 | Training method for semi-supervised learning model, image processing method, and device |
| CN112733275A (en) * | 2021-01-19 | 2021-04-30 | 中国人民解放军军事科学院国防科技创新研究院 | Satellite assembly thermal layout temperature field prediction method based on semi-supervised learning |
| CN114036787A (en) * | 2021-10-09 | 2022-02-11 | 中国人民解放军军事科学院国防科技创新研究院 | Satellite component thermal layout temperature field prediction based on physical prior self-supervision learning |
Non-Patent Citations (1)
| Title |
|---|
| 孔繁辉;李健;: "深度信念网络优化BP神经网络的交通流预测模型", 管理评论, no. 03, 31 March 2020 (2020-03-31) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115730509A (en) * | 2022-09-20 | 2023-03-03 | 中国人民解放军军事科学院国防科技创新研究院 | Spacecraft in-cabin temperature field reconstruction task research benchmark method based on machine learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114781207B (en) | 2024-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11714937B2 (en) | Estimating physical parameters of a physical system based on a spatial-temporal emulator | |
| Praveen et al. | Low cost PSO using metamodels and inexact pre-evaluation: Application to aerodynamic shape design | |
| Gong et al. | Multi-objective parameter optimization of common land model using adaptive surrogate modeling | |
| CN111507032B (en) | Deep learning technology-based component layout optimization design method for predicting temperature distribution | |
| Gaier et al. | Data-efficient exploration, optimization, and modeling of diverse designs through surrogate-assisted illumination | |
| US20210064802A1 (en) | Method and System for Increasing the Resolution of Physical Gridded Data | |
| CN112257321B (en) | Unsupervised learning-based satellite component temperature field determination method | |
| Rao et al. | A time-parallel approach to strong-constraint four-dimensional variational data assimilation | |
| CN113177290B (en) | Satellite component temperature field prediction method based on depth agent model normalization | |
| Sommerfeld et al. | Confidence regions for spatial excursion sets from repeated random field observations, with an application to climate | |
| CN113792478B (en) | Heat source layout temperature field reliability analysis method based on integrated depth proxy model | |
| CN112733275B (en) | Satellite component thermal layout temperature field prediction method based on semi-supervised learning | |
| CN116822214A (en) | Multi-credibility data fusion method based on heteroscedastic Gaussian process regression | |
| CN112257021A (en) | Selectable kriging spatial interpolation rainfall estimation method | |
| CN113609763B (en) | Uncertainty-based satellite component layout temperature field prediction method | |
| CN114781207B (en) | Heat source layout temperature field prediction method based on uncertainty and semi-supervised learning | |
| CN114692347A (en) | Temperature field proxy model construction method based on neural network architecture search technology | |
| WO2022247092A1 (en) | Methods and systems for congestion prediction in logic synthesis using graph neural networks | |
| Han et al. | CoolVox: Advanced 3D convolutional neural network models for predicting solar radiation on building facades | |
| CN109540089B (en) | Bridge deck elevation fitting method based on Bayes-Kriging model | |
| Raghavan et al. | Implicit constraint handling for shape optimisation with pod-morphing | |
| CN114692348B (en) | Component layout temperature field prediction method based on multi-fidelity deep learning agent model | |
| CN116628863B (en) | Method, device and medium for determining wind resistance coefficient of vehicle | |
| Rao et al. | Parallel solution of DDDAS variational inference problems | |
| CN115577618B (en) | Construction method and prediction method of high-pressure converter valve hall environmental factor prediction model |
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 |