CN114036861A - Three-dimensional temperature field prediction method based on infoGAN - Google Patents

Three-dimensional temperature field prediction method based on infoGAN Download PDF

Info

Publication number
CN114036861A
CN114036861A CN202111239698.9A CN202111239698A CN114036861A CN 114036861 A CN114036861 A CN 114036861A CN 202111239698 A CN202111239698 A CN 202111239698A CN 114036861 A CN114036861 A CN 114036861A
Authority
CN
China
Prior art keywords
temperature field
infogan
network
discriminator
heating furnace
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
CN202111239698.9A
Other languages
Chinese (zh)
Other versions
CN114036861B (en
Inventor
李涛
孙全胜
王津申
王艳丽
李洪涛
郭拂娟
李梦瑶
武姝洁
欧阳彤彬
荆瑞静
高丽岩
景大尉
梁江卫
曹德成
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.)
China Petroleum and Chemical Corp
Original Assignee
China Petroleum and Chemical Corp
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 China Petroleum and Chemical Corp filed Critical China Petroleum and Chemical Corp
Priority to CN202111239698.9A priority Critical patent/CN114036861B/en
Publication of CN114036861A publication Critical patent/CN114036861A/en
Application granted granted Critical
Publication of CN114036861B publication Critical patent/CN114036861B/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]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Abstract

The invention discloses a three-dimensional temperature field prediction method based on infoGAN, which comprises the following steps: acquiring working condition data, and performing CFD (computational fluid dynamics) simulation calculation to obtain three-dimensional temperature field data of the industrial heating furnace, wherein a sample obtained from the three-dimensional temperature field data is a true sample; dividing the data into a training set and a testing set after preprocessing; training an infoGAN network, wherein the parameters of each layer of a discriminator and a generator are updated by comparing the difference of Wasserstein distances between the distribution of a false sample and a true sample, and respectively solving the target loss functions of the discriminator and the generator, so that the generator and the discriminator continuously play games with each other and are alternately trained; and testing the performance of the trained info GAN network by using a test set, and realizing the prediction of the three-dimensional temperature field of the industrial heating furnace by using the finally trained info GAN network. The method can directly predict the internal temperature field of the heating furnace from the working condition data of the inlet of the heating furnace, and saves the calculation time and the calculation cost.

Description

Three-dimensional temperature field prediction method based on infoGAN
Technical Field
The invention relates to a temperature field prediction method for a set model, which is based on limited computational fluid mechanics simulation data serving as a training data set and is combined with a deep learning method to realize prediction from model inlet working condition data to a model internal temperature field.
Background
CFD is an abbreviation of Computational Fluid Dynamics (CFD), a product of a combination of recent Fluid Dynamics, numerical mathematics and computer science, and has great vitality. CFD solves problems related to fluid flow using methods of numerical analysis. The working condition data of the temperature, the flow and the like of the industrial heating furnace can be used as the basis for CFD calculation, so that the temperature calculation values of a series of points in the industrial heating furnace can be obtained. The CFD can calculate the three-dimensional temperature field data of the industrial heating furnace which is very detailed and abundant. However, the CFD has a large calculation amount and slow convergence, and a real-time temperature field is difficult to obtain.
GAN (generic adaptive Nets, generating confrontational networks) is a deep learning model that passes through (at least) two modules in a framework: the mutual game learning of the generator and the arbiter yields a reasonably good output. The generator can generate specific data from the input random noise z, and the discriminator can judge the authenticity of the data, namely distinguish whether the data is the data generated by the generator or the real data. The generator's optimization aims at generating data distribution closer to the real data, making it indistinguishable by the arbiter. The optimization goal of the discriminator is to enhance the discrimination ability and discriminate the generated data from the real data as much as possible. Along with the increase of the network iteration times, the generator and the discriminator are mutually confronted, and finally the generator and the discriminator with better effect are obtained. GAN cannot account for the connection between the input noise z and the semantic features of a particular output sample.
The method has the advantages that the GAN and an information theory can be effectively combined, mutual information relation constraint of hidden variable coding c and generated data is introduced, and therefore the purpose of controlling the generated sample can be achieved by modeling the corresponding relation of the hidden coding and the generated sample. The input noise z is controlled in a mutual information mode, and finally the effect of controlling the generation of samples by using the implicit coding c can be achieved.
Disclosure of Invention
Aiming at the prior art, the invention provides a three-dimensional temperature field prediction method based on infoGAN, CFD simulation data proxy modeling based on data drive is adopted, the network is improved by combining the WGAN-GP principle, and input working condition data can be reconstructed into an internal three-dimensional temperature field of a corresponding model in a finally trained network model.
The three-dimensional temperature field prediction method based on the infoGAN establishes a proper model by using the working condition data at the inlet of the heating furnace and the CFD temperature field in the heating furnace correspondingly calculated, can directly predict the temperature field in the heating furnace from the working condition data at the inlet of the heating furnace through the model without CFD calculation, saves the calculation time and the calculation cost, and lays a foundation for the industrial problem applied to a larger temperature field scene.
In order to solve the technical problem, the invention provides a three-dimensional temperature field prediction method based on infoGAN, which comprises the following steps:
step one, establishing a data set, comprising: acquiring working condition data: the working condition data comprises the fuel flow speed at the inlet of the industrial heating furnace in m/s; temperature at the inlet of the industrial furnace, in K; the wall heat conductivity coefficient of the industrial heating furnace is expressed in m × K;
utilizing CFD simulation calculation to obtain three-dimensional temperature field data of the industrial heating furnace, wherein the three-dimensional temperature field data comprises three-dimensional coordinates and corresponding temperature values of sampling points inside the industrial heating furnace, and the three-dimensional temperature field data is a true sample;
step two, data preprocessing: carrying out normalization pretreatment on the working condition data obtained in the step one by using a formula (1):
Figure BDA0003318777620000021
in the formula (1), eiOperating condition data obtained in step one, ciAs a result of calculation using the formula (1), i is 1, 2, 3, e1Means fuel flow velocity at the inlet of the industrial furnace in m/s, c1Is e1The result after normalization is dimensionless; e.g. of the type2Denotes the temperature at the inlet of the industrial furnace in units K, c2Is e2The result after normalization is dimensionless; e.g. of the type3Represents the wall thermal conductivity of the industrial heating furnace, with the unit of m × K, c3Is e3The result after normalization is dimensionless; mean, max and min are respectively the mean, the maximum and the minimum in the range of the corresponding working condition data training samples; the working condition data after normalization processing and step-CFD simulation are carried outThe three-dimensional temperature field data obtained by calculation are in one-to-one correspondence to form a plurality of groups, a% of the data are randomly selected as a training set, and the rest (100-a)% is used as a test set;
step three, training an infoGAN network: putting the training set into an infoGAN network for training, comprising:
3-1) initializing the weight parameters of the infoGAN network generator (G) and the discriminator (D);
3-2) the input of the generator of infoGAN is a random noise vector z and a code c, wherein the random noise vector z is set to be a normal gaussian noise; the flow speed c of the fuel at the inlet of the industrial heating furnace in the working condition data after the normalization processing1Temperature c at the inlet of the industrial furnace2And wall surface thermal conductivity c of industrial heating furnace3Setting the three dimensions of the code c respectively, namely c is 3 multiplied by 1;
3-3) the output of the generator of the infoGAN network is a false sample, and the false sample is used as the input of the discriminator of the infoGAN network;
3-4) the discriminator of the infoGAN network generates a reconstructed code c' with the same structure as the code c, compares the difference of Wasserstein distances between false samples and true samples, calculates the target loss function of the discriminator by using the formula (2), calculates the gradient of the target loss function of the discriminator, and reversely propagates the gradient to each layer of the discriminator of the infoGAN network for updating the parameters of each layer of the discriminator;
3-5) then, calculating mutual information between the code c and the false sample by substituting the reconstructed code c' into the formula (4) by comparing the difference of Wasserstein distances between the false sample and the true sample distribution, then calculating the target loss function of the generator by using the formula (2), solving the gradient of the target loss function of the generator, and reversely propagating the gradient back to each layer of the infoGAN network generator so as to update the parameters of each layer of the generator;
Figure BDA0003318777620000031
in formula (2): v (D, G) is a loss function of GAN:
Figure BDA0003318777620000032
in equations (2) and (3), x represents the true sample entry of the discriminator, i.e., the temperature field data calculated by the step-one CFD simulation, and the distribution of x obeys PxG (-) represents temperature field data obtained after the temperature field data passes through the infoGAN network generator, D (-) represents a result obtained by the infoGAN network discriminator, and E (-) calculates an expected value; beta and lambda are self-defined hyper-parameter weights respectively, and the initial values of beta and lambda are both 1;
Figure BDA0003318777620000033
is a gradient penalty constraint, wherein
Figure BDA0003318777620000034
Figure BDA0003318777620000035
Figure BDA0003318777620000036
Representative pair
Figure BDA0003318777620000039
In the expectation that the position of the target is not changed,
Figure BDA0003318777620000037
representation solution about
Figure BDA00033187776200000310
Gradient of (1) | · | | non-conducting phosphor2A 2-norm representing a computational matrix; i (-) is mutual information:
Figure BDA0003318777620000038
3-6) repeating the steps 3-3) to 3-5), so that the generator and the discriminator continuously play games with each other and alternately train until the training reaches a specified period number or stops artificially; thus obtaining a trained infoGAN network;
step four, testing the performance of the trained infoGAN network by using a test set, and determining whether the trained infoGAN network is further optimized according to a test result, so that the average relative error between the three-dimensional temperature field data generated by the trained infoGAN network and the three-dimensional temperature field data calculated by the CFD simulation in the step one is less than 5%;
and step five, predicting the three-dimensional temperature field of the industrial heating furnace by using the trained infoGAN network tested in the step four.
Further, the three-dimensional temperature field prediction method of the present invention includes:
in the first step, the three-dimensional temperature field data of the industrial heating furnace obtained by CFD simulation calculation means that: the integral and differential terms in the control equation of the fluid mechanics are approximately expressed into a discrete algebraic form to form an algebraic equation set, then the discrete algebraic equation set is solved through a computer to obtain three-dimensional simulation temperature field data, and each group of temperature field data comprises three-dimensional coordinates and corresponding temperature values of sampling points in the industrial heating furnace. The process of the CFD simulation temperature field is as follows: selecting a physical model according to the internal structure of the industrial heating furnace, wherein the selected physical model is a standard k-epsilon model in a turbulence two-equation model; then, calculating a domain and a boundary condition; dividing a computing grid; and finally, CFD simulation calculation is realized.
In step 3-4), updating parameters of each layer of the discriminator to ensure that V in the formula (2)1And (D, G) the calculation result is maximized as much as possible, namely, the discriminator can judge the authenticity of the input data more accurately.
In step 3-5), updating parameters of each layer of the discriminator to enable the calculation result V of the formula (2)1And (D, G) minimizing as much as possible, namely enabling the three-dimensional temperature field data generated again by the generator to be more approximate to the three-dimensional temperature field data obtained by CFD simulation calculation.
In the fourth step, a basic neural network optimization method is adopted for the optimization strategy for further optimizing the trained infoGAN network. And in the optimization process, the numerical values of the self-defined hyperparametric weights beta and lambda are adjusted according to a grid search algorithm.
Compared with the prior art, the invention has the beneficial effects that:
in the traditional research, a CFD simulation numerical experiment is generally used for calculating the three-dimensional temperature field of the industrial heating furnace, but CFD calculation data is huge, so that the complete three-dimensional temperature field data can be obtained only by spending a large amount of time, and a real-time temperature field is difficult to obtain. The invention establishes an internal mapping relation model between the working condition data and the temperature field data based on the infoGAN neural network. Once modeling of a certain industrial heating furnace is completed, working condition data are input on a trained model, three-dimensional temperature field data of the industrial heating furnace with the quality equivalent to that of a CFD result can be obtained more quickly, time cost for calculating the temperature field is greatly saved, and a real-time temperature field can be obtained to provide data support for more downstream research tasks.
Drawings
FIG. 1 is an infoGAN network structure;
FIG. 2 is a three-dimensional model of a simulated heating furnace according to an exemplary embodiment of the present invention;
3-1 and 3-2 are computational domain specific geometries of simulation examples;
FIG. 4-1 is a CFD simulation calculation result of a simulation example;
fig. 4-2 is a network prediction result of the simulation example.
Detailed Description
The invention will be further described with reference to the following figures and specific examples, which are not intended to limit the invention in any way.
The design idea of the three-dimensional temperature field prediction method based on the infoGAN provided by the invention is as follows: and training the infoGAN by using the simulated temperature field calculated by the CFD and corresponding working condition data as input. Any group of tag values in the range is input (each group of tag values comprises the velocity of the fuel at the inlet of the industrial heating furnace, the temperature and the thermal conductivity coefficient of the inner wall surface of the heating furnace), a group of corresponding temperature field data can be obtained (each group of temperature field data comprises the three-dimensional coordinates and the corresponding temperature values of the sampling points in the industrial heating furnace), and Wassertein distance (Earth-mover) is used when the similarity of distribution is measured, so that the problems of gradient disappearance and mode collapse are solved.
The invention provides an infoGAN-based three-dimensional temperature field prediction method, which comprises the following steps:
step one, establishing a data set, comprising: acquiring working condition data: the working condition data comprises the fuel flow speed at the inlet of the industrial heating furnace in m/s; temperature at the inlet of the industrial furnace, in K; the wall heat conductivity coefficient of the industrial heating furnace is expressed in m × K;
utilizing CFD simulation calculation to obtain three-dimensional temperature field data of the industrial heating furnace, wherein the data comprises three-dimensional coordinates and corresponding temperature values of sampling points inside the industrial heating furnace, and the three-dimensional temperature field data are true samples;
step two, data preprocessing: carrying out normalization pretreatment on the working condition data obtained in the step one by using a formula (1):
Figure BDA0003318777620000051
in the formula (1), eiOperating condition data obtained in step one, ciAs a result of calculation using the formula (1), i is 1, 2, 3, e1Means fuel flow velocity at the inlet of the industrial furnace in m/s, c1Is e1The result after normalization is dimensionless; e.g. of the type2Denotes the temperature at the inlet of the industrial furnace in units K, c2Is e2The result after normalization is dimensionless; e.g. of the type3Represents the wall thermal conductivity of the industrial heating furnace, with the unit of m × K, c3Is e3The result after normalization is dimensionless; mean, max and min are respectively the mean, the maximum and the minimum in the range of the corresponding working condition data training samples; the working condition data after normalization processing and three-dimensional temperature field data obtained by CFD simulation calculation in the first step are in one-to-one correspondence to form a plurality of groups, a% of the data is randomly selected to be used as a training set, and the rest (100-a)% is used as a test set;
step three, training an infoGAN network: fig. 1 shows an infoGAN network, where a training set is put into the infoGAN network for training, including:
3-1) initializing the weight parameters of the infoGAN network generator G and the discriminator D;
3-2) the input of the generator of infoGAN is a random noise vector z and a code c, wherein the random noise vector z is set to be a normal gaussian noise; the flow speed c of the fuel at the inlet of the industrial heating furnace in the working condition data after the normalization processing1Temperature at the inlet of the industrial furnace c2 and wall thermal conductivity of the industrial furnace c3Set to three dimensions of code c, i.e., c is 3 x 1 size, respectively.
3-3) the generator of the infoGAN generates "false" temperature field data under the influence of z and c and inputs it into the discriminator, i.e. the output of the generator of the infoGAN network is a false sample, which is taken as the input of the discriminator of the infoGAN network.
3-4) the discriminator of the infoGAN network generates a reconstructed code c' with the same structure as the code c, calculates a value between 0 and 1 by comparing the difference of Wasserstein distance between false sample and true sample distribution, the larger the value is, the higher the probability of similarity between false sample data and true sample data is represented, then calculates the target loss function of the discriminator by using the formula (2), and calculates the gradient of the target loss function of the discriminator, and reversely propagates the gradient to each layer of the discriminator of the infoGAN network for updating the parameters of each layer of the discriminator, so that the discriminator can more accurately judge the authenticity of the input data.
3-5) then, calculating mutual information between the code c and the false sample by substituting the reconstructed code c' into formula (4) by comparing the difference of Wasserstein distances between the false sample and the true sample distribution, then calculating the target loss function of the generator by using formula (2), solving the gradient of the target loss function of the generator, and reversely propagating the gradient back to each layer of the infoGAN network generator to update the parameters of each layer of the generator, so that the temperature field data generated again by the generator can be more approximate to the real temperature field distribution. The ideal goal is to make it impossible for the discriminator to distinguish between the temperature field data produced by the generator and the true temperature field data.
Specifically, the training target loss function of the final generator and the arbiter can be expressed as shown in equation (2):
Figure BDA0003318777620000052
in formula (2): v (D, G) is a loss function of GAN:
Figure BDA0003318777620000061
in equations (2) and (3), x represents the true sample entry of the discriminator, i.e., the temperature field data calculated by the step-one CFD simulation, and the distribution of x obeys PxG (-) represents temperature field data obtained after the temperature field data passes through the infoGAN network generator, D (-) represents a result obtained by the infoGAN network discriminator, and E (-) calculates an expected value; final optimization function V1The meaning of the rest symbols in (D, G) is that beta and lambda are respectively self-defined hyper-parameter weight, and the initial values of beta and lambda are both 1;
Figure BDA0003318777620000062
is a gradient penalty constraint, wherein
Figure BDA0003318777620000063
ε～Uniform[0，1]，
Figure BDA0003318777620000064
Representative pair
Figure BDA0003318777620000067
In the expectation that the position of the target is not changed,
Figure BDA0003318777620000065
representation solution about
Figure BDA0003318777620000068
Gradient of (1) | · | | non-conducting phosphor2Representing the 2-norm of the computation matrix. I (-) is mutual information:
Figure BDA0003318777620000066
3-6) repeating the steps 3-3) to 3-5), so that the generator and the discriminator continuously play games with each other, and alternately training and updating network parameters until the training reaches the specified period number or stops artificially; thus obtaining a trained infoGAN network;
and step four, testing the performance of the trained infoGAN network by using a test set. Because the corresponding relations between the working condition data and the temperature fields of different industrial heating furnaces are not completely the same, the network needs to be further optimized according to the test result, and the optimization strategy is a basic neural network optimization method, such as adjusting the learning rate, adjusting the weight coefficients of different constraint terms in the loss function, and the like, until the average relative error between the three-dimensional temperature field data generated by the trained info GAN network and the three-dimensional temperature field data calculated by the CFD simulation in the first step is less than 5%.
After training and optimization debugging are completed, the infoGAN completes the modeling of the mapping relation between each index change and the temperature field change in the working condition data by learning the relation between the code c and the training sample, so that the effect of controlling the code c to control the output result can be achieved, and the function of predicting the working condition data to the three-dimensional temperature field is also achieved. Once modeling of a certain industrial heating furnace is completed, working condition data are input on a trained model, three-dimensional temperature field data of the industrial heating furnace with the quality equivalent to that of a CFD result can be obtained more quickly, time cost for calculating the temperature field is greatly saved, and a real-time temperature field can be obtained to provide data support for more downstream research tasks.
Simulation calculation example:
the geometric model designed by the simulation algorithm is shown in fig. 2. The coordinate system and the calculation domain are shown in fig. 2. The origin O of the coordinate system is located at the center of the inlet, and the x-axis is directed from the inlet to the outlet. The inlet of the calculation domain is rectangular, and the outlet is square (side length is 10 cm); the calculation domain channel is expanded in the vertical direction and then becomes a straight section, and the calculation domain channel can be obtained by stretching the section in the vertical direction in the horizontal direction. The specific geometry is shown in fig. 3-1 and 3-2.
The fuel speed at the simulated inlet is 0.5m/s, the temperature is 600K, the heat conduction coefficient of the inner wall surface of the furnace is 0.242, and the temperature of the designed external environment is 300K. Due to the heat exchange between the wall surface and the external environment, the temperature of the gas in the channel is reduced, and the closer to the wall surface, the lower the temperature of the gas is. The temperature range of the air flow in the calculation area is 300K-600K. The results of the numerical simulation, which is performed for hours using CFD, are the coordinates and temperature values of 50933 nodes inside the model, where the results for several nodes are shown in the following table:
node numbering X axis coordinate Y-axis coordinate Z-axis coordinate Temperature of
1 16.6835 1.5470 5.6668 653.4119
2 16.6680 1.5635 5.6481 653.9653
3 16.6655 1.5361 5.6654 653.3988
4 16.6680 1.5350 5.6481 653.6810
5 0.0281 1.4843 0.6517 894.0535
6 0.0257 1.5083 0.6523 838.3954
7 0.0274 11.4887 0.6205 891.9551
The data comprises 370 groups of working condition data and corresponding CFD data in total, 300 groups of the working condition data and the corresponding CFD data are randomly selected as training sets, and the rest 70 groups of the working condition data and the corresponding CFD data are selected as test sets. And training the model according to the steps. After the training is finished, 70 groups of test data are put into the trained deep learning model, the temperature data of 50933 nodes in each group of furnaces can be obtained at the speed of about 1 s/group, and the prediction average error is 4.96%. Fig. 4-1 shows the CFD simulation calculation result of the simulation example, and fig. 4-2 shows the network prediction result of the simulation example using the method of the present invention. Compared with the calculation amount of CFD of hours, the calculation resource is saved, and the time consumption is greatly reduced.
While the present invention has been described with reference to the accompanying drawings, the present invention is not limited to the above-described embodiments, which are illustrative only and not restrictive, and various modifications which do not depart from the spirit of the present invention and which are intended to be covered by the claims of the present invention may be made by those skilled in the art.

Claims (7)

1. A three-dimensional temperature field prediction method based on infoGAN is characterized by comprising the following steps:
step one, establishing a data set, comprising:
acquiring working condition data: the working condition data comprises the fuel flow speed at the inlet of the industrial heating furnace in m/s; temperature at the inlet of the industrial furnace, in K; the wall heat conductivity coefficient of the industrial heating furnace is expressed in m × K;
utilizing CFD simulation calculation to obtain three-dimensional temperature field data of the industrial heating furnace, wherein the three-dimensional temperature field data comprises three-dimensional coordinates and corresponding temperature values of sampling points inside the industrial heating furnace, and the three-dimensional temperature field data is a true sample;
step two, data preprocessing:
carrying out normalization pretreatment on the working condition data obtained in the step one by using a formula (1):
Figure FDA0003318777610000011
in the formula (1), eiOperating condition data obtained in step one, ciAs a result of calculation using the formula (1), i is 1, 2, 3, e1Means fuel flow velocity at the inlet of the industrial furnace in m/s, c1Is e1The result after normalization is dimensionless; e.g. of the type2Indicating the temperature at the inlet of the industrial furnace in unitsK，c2Is e2The result after normalization is dimensionless; e.g. of the type3Represents the wall thermal conductivity of the industrial heating furnace, with the unit of m × K, c3Is e3The result after normalization is dimensionless; mean, max and min are respectively the mean, the maximum and the minimum in the range of the corresponding working condition data training samples; the working condition data after normalization processing and three-dimensional temperature field data obtained by CFD simulation calculation in the first step are in one-to-one correspondence to form a plurality of groups, a% of the data is randomly selected to be used as a training set, and the rest (100-a)% is used as a test set;
step three, training an infoGAN network:
putting the training set into an infoGAN network for training, comprising:
3-1) initializing the weight parameters of the infoGAN network generator (G) and the discriminator (D);
3-2) the input of the generator of infoGAN is a random noise vector z and a code c, wherein the random noise vector z is set to be a normal gaussian noise; the flow speed c of the fuel at the inlet of the industrial heating furnace in the working condition data after the normalization processing1Temperature c at the inlet of the industrial furnace2And wall surface thermal conductivity c of industrial heating furnace3Setting the three dimensions of the code c respectively, namely c is 3 multiplied by 1;
3-3) the output of the generator of the infoGAN network is a false sample, and the false sample is used as the input of the discriminator of the infoGAN network;
3-4) the discriminator of the infoGAN network generates a reconstructed code c' with the same structure as the code c, compares the difference of Wasserstein distances between false samples and true samples, calculates the target loss function of the discriminator by using the formula (2), calculates the gradient of the target loss function of the discriminator, and reversely propagates the gradient to each layer of the discriminator of the infoGAN network for updating the parameters of each layer of the discriminator;
3-5) then, calculating mutual information between the code c and the false sample by substituting the reconstructed code c' into the formula (4) by comparing the difference of Wasserstein distances between the false sample and the true sample distribution, then calculating the target loss function of the generator by using the formula (2), solving the gradient of the target loss function of the generator, and reversely propagating the gradient back to each layer of the infoGAN network generator so as to update the parameters of each layer of the generator;
Figure FDA0003318777610000021
in formula (2): v (D, G) is a loss function of GAN:
Figure FDA0003318777610000022
in equations (2) and (3), x represents the true sample entry of the discriminator, i.e., the temperature field data calculated by the step-one CFD simulation, and the distribution of x obeys PxG (-) represents temperature field data obtained after the temperature field data passes through the infoGAN network generator, D (-) represents a result obtained by the infoGAN network discriminator, and E (-) calculates an expected value; beta and lambda are self-defined hyper-parameter weights respectively, and the initial values of beta and lambda are both 1;
Figure FDA0003318777610000023
is a gradient penalty constraint, wherein
Figure FDA0003318777610000024
Figure FDA0003318777610000025
ε～Uniform[0，1]，
Figure FDA0003318777610000026
Representative pair
Figure FDA0003318777610000029
In the expectation that the position of the target is not changed,
Figure FDA0003318777610000027
representation solution about
Figure FDA00033187776100000210
Gradient of (1) | · | | non-conducting phosphor2A 2-norm representing a computational matrix; i (-) is mutual information:
Figure FDA0003318777610000028
3-6) repeating the steps 3-3) to 3-5), so that the generator and the discriminator continuously play games with each other and alternately train until the training reaches a specified period number or stops artificially; thus obtaining a trained infoGAN network;
step four, testing the performance of the trained infoGAN network by using a test set, and determining whether the trained infoGAN network is further optimized according to a test result, so that the average relative error between the three-dimensional temperature field data generated by the trained infoGAN network and the three-dimensional temperature field data calculated by the CFD simulation in the step one is less than 5%;
and step five, predicting the three-dimensional temperature field of the industrial heating furnace by using the trained infoGAN network tested in the step four.
2. The method for predicting the three-dimensional temperature field according to claim 1, wherein in the first step, the step of obtaining the three-dimensional temperature field data of the industrial heating furnace by using CFD simulation calculation comprises the following steps: the integral and differential terms in the control equation of the fluid mechanics are approximately expressed into a discrete algebraic form to form an algebraic equation set, then the discrete algebraic equation set is solved through a computer to obtain three-dimensional simulation temperature field data, and each group of temperature field data comprises three-dimensional coordinates and corresponding temperature values of sampling points in the industrial heating furnace.
3. The three-dimensional temperature field prediction method of claim 2, wherein the process of CFD simulating the temperature field is: selecting a physical model according to the internal structure of the industrial heating furnace, wherein the selected physical model is a standard k-epsilon model in a turbulence two-equation model; then, calculating a domain and a boundary condition; dividing a computing grid; and finally, CFD simulation calculation is realized.
4. The method according to claim 1, wherein in step 3-4), the parameters of each layer of the discriminator are updated so that V in equation (2)1And (D, G) the calculation result is maximized as much as possible, namely, the discriminator can judge the authenticity of the input data more accurately.
5. The method according to claim 1, wherein in step 3-5), the parameters of each layer of the discriminator are updated so that the result V calculated by equation (2)1And (D, G) minimizing as much as possible, namely enabling the three-dimensional temperature field data generated again by the generator to be more approximate to the three-dimensional temperature field data obtained by CFD simulation calculation.
6. The three-dimensional temperature field prediction method of claim 1, wherein in step four, a basic neural network optimization method is adopted for an optimization strategy for further optimizing the trained infoGAN network.
7. The method of claim 6, wherein the values of the customized hyperparametric weights β and λ are adjusted during the optimization according to a grid search algorithm.
CN202111239698.9A 2021-10-25 2021-10-25 Three-dimensional temperature field prediction method based on infoGAN Active CN114036861B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111239698.9A CN114036861B (en) 2021-10-25 2021-10-25 Three-dimensional temperature field prediction method based on infoGAN

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111239698.9A CN114036861B (en) 2021-10-25 2021-10-25 Three-dimensional temperature field prediction method based on infoGAN

Publications (2)

Publication Number Publication Date
CN114036861A true CN114036861A (en) 2022-02-11
CN114036861B CN114036861B (en) 2022-09-27

Family

ID=80141891

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111239698.9A Active CN114036861B (en) 2021-10-25 2021-10-25 Three-dimensional temperature field prediction method based on infoGAN

Country Status (1)

Country Link
CN (1) CN114036861B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114722732A (en) * 2022-06-09 2022-07-08 华中科技大学 Method for predicting temperature field of fuel tank of hypersonic aircraft based on point cloud network
CN115342680A (en) * 2022-08-01 2022-11-15 无锡雪浪数制科技有限公司 Intelligent method for identifying abnormal state of indirect air cooling system
CN115688982A (en) * 2022-10-11 2023-02-03 华能江苏综合能源服务有限公司 Building photovoltaic data completion method based on WGAN and whale optimization algorithm
CN117476125A (en) * 2023-12-27 2024-01-30 豆黄金食品有限公司 Dried beancurd stick raffinate recovery data processing system based on data analysis

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110909508A (en) * 2019-10-29 2020-03-24 中国石油化工股份有限公司 Heating furnace temperature field real-time prediction method based on convolution long and short term memory network
CN112528569A (en) * 2020-10-17 2021-03-19 中国石油化工股份有限公司 Industrial heating furnace temperature field prediction method based on digital twinning

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110909508A (en) * 2019-10-29 2020-03-24 中国石油化工股份有限公司 Heating furnace temperature field real-time prediction method based on convolution long and short term memory network
CN112528569A (en) * 2020-10-17 2021-03-19 中国石油化工股份有限公司 Industrial heating furnace temperature field prediction method based on digital twinning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
曹正彬等: "基于TWSVM和模糊的木质采暖地板蓄热温度预测模型", 《林业科学》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114722732A (en) * 2022-06-09 2022-07-08 华中科技大学 Method for predicting temperature field of fuel tank of hypersonic aircraft based on point cloud network
CN115342680A (en) * 2022-08-01 2022-11-15 无锡雪浪数制科技有限公司 Intelligent method for identifying abnormal state of indirect air cooling system
CN115688982A (en) * 2022-10-11 2023-02-03 华能江苏综合能源服务有限公司 Building photovoltaic data completion method based on WGAN and whale optimization algorithm
CN115688982B (en) * 2022-10-11 2024-01-30 华能江苏综合能源服务有限公司 Building photovoltaic data complement method based on WGAN and whale optimization algorithm
CN117476125A (en) * 2023-12-27 2024-01-30 豆黄金食品有限公司 Dried beancurd stick raffinate recovery data processing system based on data analysis
CN117476125B (en) * 2023-12-27 2024-04-05 豆黄金食品有限公司 Dried beancurd stick raffinate recovery data processing system based on data analysis

Also Published As

Publication number Publication date
CN114036861B (en) 2022-09-27

Similar Documents

Publication Publication Date Title
CN114036861B (en) Three-dimensional temperature field prediction method based on infoGAN
CN106920006B (en) Subway station air conditioning system energy consumption prediction method based on ISOA-LSSVM
CN104765916B (en) A kind of Dynamics Performance of High Speed Trains parameter optimization method
CN111191316B (en) Response surface-based building natural ventilation performance optimization model and optimization method
CN102129259B (en) Neural network proportion integration (PI)-based intelligent temperature control system and method for sand dust environment test wind tunnel
Samadianfard Gene expression programming analysis of implicit Colebrook–White equation in turbulent flow friction factor calculation
CN108197432A (en) A kind of gene regulatory network reconstructing method based on gene expression data
CN110286586A (en) A kind of MR damper hybrid modeling method
CN104408518A (en) Method of learning and optimizing neural network based on particle swarm optimization algorithm
CN110348059A (en) A kind of channel flow field reconstructing method based on structured grid
CN105554873A (en) Wireless sensor network positioning algorithm based on PSO-GA-RBF-HOP
CN112528569A (en) Industrial heating furnace temperature field prediction method based on digital twinning
Zou et al. Bare-bones teaching-learning-based optimization
CN112231964B (en) Gas leakage source autonomous searching and positioning method based on deep reinforcement learning
CN110442974A (en) Horse shoe flame regenerator chamber of glass kiln performance optimization method and device
Lin et al. Evolutionary architectural search for generative adversarial networks
CN116186978A (en) Data and physical dual-drive temperature field prediction method for complex geometric area of aircraft
Xi et al. Air combat maneuver trajectory prediction model of target based on chaotic theory and IGA-VNN
CN105930574A (en) Genetic and neural network algorithm based design method for air intake port model of internal combustion engine
Bouzbida et al. Robust kernel clustering algorithm for nonlinear system identification
Chen et al. Relevance vector machines-based time series prediction for incomplete training dataset: Two comparative approaches
CN111291465A (en) Ship air conditioner thermal comfort temperature prediction method based on genetic algorithm and neural network
CN116595845A (en) Structure optimization method of spiral fin tube type heat exchanger
Ben-Nakhi et al. Inter-model comparison of CFD and neural network analysis of natural convection heat transfer in a partitioned enclosure
CN115310727A (en) Building cooling, heating and power load prediction method and system based on transfer learning

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