CN116343936A - Combustion chemical reaction calculation acceleration method based on deep neural network - Google Patents
Combustion chemical reaction calculation acceleration method based on deep neural network Download PDFInfo
- Publication number
- CN116343936A CN116343936A CN202310524037.3A CN202310524037A CN116343936A CN 116343936 A CN116343936 A CN 116343936A CN 202310524037 A CN202310524037 A CN 202310524037A CN 116343936 A CN116343936 A CN 116343936A
- Authority
- CN
- China
- Prior art keywords
- neural network
- deep neural
- chemical reaction
- combustion
- calculation
- 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.)
- Pending
Links
- 238000006243 chemical reaction Methods 0.000 title claims abstract description 68
- 238000004364 calculation method Methods 0.000 title claims abstract description 53
- 238000002485 combustion reaction Methods 0.000 title claims abstract description 47
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 37
- 238000000034 method Methods 0.000 title claims abstract description 35
- 230000001133 acceleration Effects 0.000 title claims abstract description 21
- 238000005070 sampling Methods 0.000 claims abstract description 33
- 238000003062 neural network model Methods 0.000 claims abstract description 24
- 238000004088 simulation Methods 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 16
- 239000012530 fluid Substances 0.000 claims abstract description 14
- 238000010606 normalization Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 9
- 230000008878 coupling Effects 0.000 claims abstract description 7
- 238000010168 coupling process Methods 0.000 claims abstract description 7
- 238000005859 coupling reaction Methods 0.000 claims abstract description 7
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims description 18
- 239000000126 substance Substances 0.000 claims description 13
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 8
- 238000001228 spectrum Methods 0.000 claims description 5
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- 229910052757 nitrogen Inorganic materials 0.000 claims description 4
- 239000001301 oxygen Substances 0.000 claims description 4
- 229910052760 oxygen Inorganic materials 0.000 claims description 4
- 230000000704 physical effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 238000012216 screening Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 2
- 238000011426 transformation method Methods 0.000 claims description 2
- 238000006467 substitution reaction Methods 0.000 abstract description 8
- 239000000446 fuel Substances 0.000 description 6
- 239000004215 Carbon black (E152) Substances 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 229930195733 hydrocarbon Natural products 0.000 description 5
- 150000002430 hydrocarbons Chemical class 0.000 description 5
- 230000007246 mechanism Effects 0.000 description 5
- 230000008901 benefit Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000001419 dependent effect Effects 0.000 description 2
- BASFCYQUMIYNBI-UHFFFAOYSA-N platinum Chemical compound [Pt] BASFCYQUMIYNBI-UHFFFAOYSA-N 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- VGGSQFUCUMXWEO-UHFFFAOYSA-N Ethene Chemical compound C=C VGGSQFUCUMXWEO-UHFFFAOYSA-N 0.000 description 1
- 239000005977 Ethylene Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 229910052697 platinum Inorganic materials 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/10—Analysis or design of chemical reactions, syntheses or processes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
- G06F17/12—Simultaneous equations, e.g. systems of linear equations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
- G06F17/13—Differential equations
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/70—Machine learning, data mining or chemometrics
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Algebra (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Operations Research (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Crystallography & Structural Chemistry (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Analytical Chemistry (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Medical Informatics (AREA)
- Fluid Mechanics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a combustion chemical reaction calculation acceleration method based on a deep neural network, which relates to the field of combustion numerical simulation and comprises the following steps of: global manifold sampling, the sampling being performed within a predetermined thermal chemistry phase space; preprocessing the sampled data, wherein the preprocessing comprises Box-Cox conversion and zero mean normalization; training and storing a deep neural network model; and coupling the pre-trained deep neural network model with a computational fluid dynamics program to realize the combustion numerical simulation calculation. According to the invention, through the chemical reaction substitution model based on the deep neural network, the strong rigidity in the combustion chemical reaction is removed, the solution of the chemical reaction rigidity differential equation set with a larger time step is realized, and the chemical reaction multi-grid parallel calculation can be realized through the GPU, so that the calculation efficiency is greatly improved on the premise of ensuring the precision, and the method is efficient and convenient.
Description
Technical Field
The invention relates to the field of combustion numerical simulation, in particular to a combustion chemical reaction calculation acceleration method based on a deep neural network.
Background
The aero-engine, the scramjet engine and the internal combustion engine are key equipment power of national defense war industry and national economy, and the numerical simulation is a key tool for scientific research and industrial design. Combustion is a typical multi-scale problem and numerical modeling suffers from complex chemical reactions and flow coupling of the actual fuel. In the combustion numerical simulation related to a detailed chemical mechanism, the number of components is huge, the dimension is high, the chemical reaction rigidity is high, the calculation time of 70-90% is generally occupied by solving the chemical reaction, and the acceleration of the chemical reaction operation is the bottleneck for realizing the high-efficiency numerical simulation to be broken through.
In the combustion value of hydrocarbon fuel, the calculation time is mostly occupied by solving the chemical reaction, and the conventional numerical simulation needs a far smaller Yu Kulang number of time steps to evolve the reaction source term due to the strong combustion chemical rigidity. The combustion chemical source item substitution model based on the machine learning algorithm is not constrained by chemical rigidity, chemical reaction items can be predicted in a larger time step, and compared with various meter building methods, the method has the advantage of low memory expense. However, the existing neural network substitution model has larger constraint on training data, is seriously dependent on specific problem scenes, and needs to obtain samples required by training by performing a large number of numerical simulations on similar scenes. It is often also necessary to enhance the sample by adding additional perturbations, which are generally more dependent on the particular problem.
Accordingly, those skilled in the art have been working to develop a combustion chemistry reaction calculation acceleration method based on deep neural networks.
Disclosure of Invention
In view of the above-mentioned drawbacks of the prior art, the present invention aims to solve the technical problem of removing the strong rigidity in the combustion chemical reaction, realizing the parallel calculation of the multi-grid chemical reaction, and effectively realizing the acceleration of the chemical reaction calculation in the combustion numerical simulation of hydrocarbon fuel under the premise of ensuring the accuracy.
In order to achieve the above object, the present invention provides a combustion chemical reaction calculation acceleration method based on a deep neural network, which is characterized in that the method comprises the following steps:
s101: global manifold sampling, said sampling being performed within a predetermined thermal chemistry phase space;
s103: preprocessing the sampled data, wherein the preprocessing comprises Box-Cox conversion and zero mean normalization;
s105: training and storing a deep neural network model;
s107: and coupling the pre-trained deep neural network model with a computational fluid dynamics program to realize the simulation calculation of the combustion numerical value.
Further, in the step S101, the thermochemical phase space is a vector space composed of thermochemical states including: temperature, pressure and mass fraction of the components.
Further, each component in the thermochemical phase space comprises methane, oxygen and nitrogen, and the mass fraction of each component is 1:1:3.76, the initial temperature in the thermochemical phase space is 300K and the initial pressure in the thermochemical phase space is 1atm.
Further, the global manifold sampling includes the steps of:
s1011: performing zero-dimensional manifold sampling on the low-dimensional sub-manifold in the thermochemical space to obtain a zero-dimensional manifold data set;
s1012: randomly collecting a large number of samples in the thermochemical phase space according to logarithmic scale to obtain a multi-scale sampling data set;
s1013: taking the boundary of each dimension change rate in the zero-dimensional manifold data set as a reference, removing samples with time change rates exceeding the boundary range in the multi-scale sampling data set, and screening to obtain sample data conforming to actual physical properties;
s1014: carrying out evolution, label generation and merging processing on the sample data to generate a data set; the evolution generates a state quantity after a 1E-3s time step from a preset current state quantity, and the generation tag generates a state quantity after a 1E-6s time step from the preset current state quantity; the merging comprises merging the data obtained by evolution as input data and merging the data obtained by generating the tag as target data.
Further, in the step S103, the Box-Cox transformation adopts the following transformation method:
where x is an argument and λ is a variable hyper-parameter.
Further, in the step S103, the zero-mean normalization means: subtracting the average value of the samples from all the characteristic values of the input samples, dividing the average value by the standard deviation of the samples to make the average value of the samples zero and the variance 1.
Further, in the step S105, the deep neural network is a fully connected deep neural network, the training data is the generated data set in the global manifold sampling, and the network model parameters of the deep neural network after the training are saved as a binary format file.
Further, in the step S107, the computational fluid dynamics program includes EBI-DNS, and the chemical reaction calculation portion of the computational fluid dynamics program is replaced with the deep neural network model.
Further, the neural network model supports deployment on a CPU and/or a GPU and supports parallel computation, a computation domain is a matrix area of 1.5cm×1.5cm, and a computation grid is n=512×512.
Further, the radius of the ignition source is 0.02cm, the energy of the ignition source in the thermal chemical phase space is w/m3, the ignition lasts for 0.2ms, and the initial speed field and the initial vortex quantity field are generated by the following modes:
where E (k) is the energy spectrum, k is the frequency, average speed u rms =3m/s, constant k e ＝418.67。
In the preferred embodiment of the present invention, compared with the prior art, the present invention achieves the following beneficial effects:
1. according to the invention, through the chemical reaction substitution model based on the deep neural network, the strong rigidity in the combustion chemical reaction is removed, the solution of the chemical reaction rigidity differential equation set in a larger time step is realized, and the chemical reaction multi-grid parallel calculation can be realized through the GPU, so that the calculation efficiency is greatly improved on the premise of ensuring the precision.
2. The neural network-based chemical reaction substitution model provided by the invention has the advantages that a single model can be used for numerical simulation of laminar flow, jet flow flame and turbulent flow flame without fine adjustment, plug and play is realized, and the method is efficient and convenient.
3. The global manifold sampling provided by the invention performs random sampling on the whole chemical phase space, and does not depend on specific example chemical reaction prior setting and flow field information.
The conception, specific structure, and technical effects of the present invention will be further described with reference to the accompanying drawings to fully understand the objects, features, and effects of the present invention.
Drawings
FIG. 1 is a flow chart of a method of deep neural network based combustion chemistry computational acceleration in accordance with a preferred embodiment of the present invention;
FIG. 2 is a schematic diagram of an initial velocity field of a deep neural network based combustion chemistry computational acceleration method in accordance with a preferred embodiment of the present invention;
FIG. 3 is a schematic diagram of an initial vortex flow field of a deep neural network based combustion chemistry computational acceleration method in accordance with a preferred embodiment of the present invention;
FIG. 4 is a schematic diagram of turbulent flame temperature cloud after 1ms of advancement of a deep neural network based combustion chemistry computational acceleration method in accordance with a preferred embodiment of the present invention;
FIG. 5 is a schematic diagram of turbulent flame temperature cloud after 1ms of advancement of a deep neural network based combustion chemistry computational acceleration method in accordance with a preferred embodiment of the present invention;
FIG. 6 is a schematic diagram of the calculation time consumption after 1ms of the advanced combustion chemistry calculation acceleration method based on the deep neural network according to a preferred embodiment of the present invention.
Detailed Description
The following description of the preferred embodiments of the present invention refers to the accompanying drawings, which make the technical contents thereof more clear and easy to understand. The present invention may be embodied in many different forms of embodiments and the scope of the present invention is not limited to only the embodiments described herein.
In the drawings, like structural elements are referred to by like reference numerals and components having similar structure or function are referred to by like reference numerals. The dimensions and thickness of each component shown in the drawings are arbitrarily shown, and the present invention is not limited to the dimensions and thickness of each component. The thickness of the components is exaggerated in some places in the drawings for clarity of illustration.
As shown in fig. 1, the method for accelerating the combustion chemical reaction calculation based on the deep neural network provided by the embodiment of the invention comprises the following steps:
s101: global manifold sampling, which is performed within a predetermined thermal chemistry phase space.
The thermochemical phase space is a vector space composed of thermochemical states including: temperature, pressure and mass fraction of each component, each component in the thermochemical phase space comprises methane, oxygen and nitrogen, and the mass fraction of each component is 1:1:3.76, an initial temperature in the thermochemical phase space is 300K, and an initial pressure in the thermochemical phase space is 1atm. The radius of the ignition source in the thermal chemical phase space is 0.02cm, the energy of the ignition source is w/m3, the ignition lasts for 0.2ms, and the initial speed field and the initial vortex flow field are generated by the following modes:
where E (k) is the energy spectrum, k is the frequency, average speed u rms =3m/s, constant k e ＝418.67。
When global manifold sampling is performed, the method comprises the following steps:
s1011: performing zero-dimensional manifold sampling on the low-dimensional sub-manifold in the thermochemical space to obtain a zero-dimensional manifold data set;
s1012: randomly collecting a large number of samples in the thermochemical phase space according to logarithmic scale to obtain a multi-scale sampling data set;
s1013: taking the boundary of each dimension change rate in the zero-dimensional manifold data set as a reference, removing samples with time change rates exceeding the boundary range in the multi-scale sampling data set, and screening to obtain sample data conforming to actual physical properties;
s1014: carrying out evolution, label generation and merging processing on the sample data to generate a data set; the evolution generates a state quantity after a 1E-3s time step from a preset current state quantity, and the generated tag generates a state quantity after a 1E-6s time step from the preset current state quantity; the merging comprises merging the data obtained by evolution as input data and merging the data obtained by generating the tag as target data.
S103: the sampled data is preprocessed, including Box-Cox transforms and zero-mean normalization.
The Box-Cox conversion adopts the following conversion method:
where x is an argument and λ is a variable hyper-parameter.
When zero-mean normalization processing is performed, the mean value of the samples is subtracted from all the characteristic values of the input samples, and then the standard deviation of the samples is divided, so that the mean value of the samples is zero, and the variance is 1.
S105: training and saving the deep neural network model.
The deep neural network is a fully-connected deep neural network, training data adopts a data set generated in global manifold sampling, and network model parameters of the deep neural network are saved as a binary format file after training is finished so as to be directly imported and used in the follow-up.
S107: and coupling the pre-trained deep neural network model with a computational fluid dynamics program to realize the simulation calculation of the combustion numerical value.
In the specific combustion numerical simulation calculation, the pre-trained deep neural network model is coupled with a computational fluid dynamics program, such as EBI-DNS, and the chemical reaction calculation part of the computational fluid dynamics program is replaced by the deep neural network model, wherein the neural network model supports deployment on a CPU and/or a GPU and supports parallel calculation, a calculation domain adopts a matrix area of 1.5cm×1.5cm, and a calculation grid is n=512×512.
The invention removes the strong rigidity in the combustion chemical reaction by the general substitution model based on the deep neural network from the perspective of removing the strong rigidity in the combustion chemical reaction, the single model can be used for numerical simulation of laminar flow, jet flame and turbulent flame without fine adjustment, the multi-grid parallel calculation of the chemical reaction can be realized by the GPU, the solution of the chemical reaction rigidity differential equation set with larger time step can be realized, the multi-grid chemical reaction parallel calculation can be realized, the chemical reaction calculation acceleration in the combustion numerical simulation of hydrocarbon fuel can be effectively realized on the premise of ensuring the accuracy, the calculation efficiency is greatly improved, the plug and play is realized, and the efficiency is high and convenient.
In addition, the global manifold sampling provided by the invention performs random sampling on the whole chemical phase space, and does not depend on specific case chemical reaction prior setting and flow field information; the neural network algorithm provided can be suitable for a plurality of combustion test examples, and has the advantages of good universality, high accuracy and the like.
The present invention will be described in detail with reference to preferred embodiments thereof.
As shown in fig. 2, a preferred embodiment of the present invention relates to a combustion chemistry reaction calculation acceleration method based on a neural network, comprising: the calculation grid takes n=512×512 for a matrix area with a calculation field of 1.5cm×1.5 cm. The chemical reaction mechanism of common hydrocarbon fuels such as methane, ethylene, etc. is chosen so that the methane combustion number mimics the common methane DRM19 mechanism of 21 component 84 reactions. In the initial state, the calculation area is filled with mixed gas of methane, oxygen and nitrogen, and the ratio of the amounts of substances is 1:1:3.76. the initial temperature is 300K, the initial pressure is 1atm, a coordinate system is established by taking the center of the reaction vessel as an origin, an ignition source is positioned at (0.75 cm ), the radius is 0.02cm, the energy of the ignition source is w/m3, and the ignition lasts for 0.2ms. The initial velocity field and the initial vorticity field are generated through a Passot-Pouquet isotropic kinetic energy spectrum, specifically:
where E (k) is the energy spectrum, k is the frequency, u rms Average speed u rms =3m/s, constant k e ＝418.67。
The initial velocity field is shown in FIG. 3, the initial vorticity field is shown in FIG. 4, and the turbulent flame temperature cloud after 1ms of propulsion is shown in FIG. 5.
The combustion chemical reaction substitution model based on the deep neural network, which is established in the preferred embodiment, comprises the following steps:
and 4, coupling the pre-trained deep neural network model with a Computational Fluid Dynamics (CFD) program, wherein the chemical reaction part is calculated by the neural network model, and simulating combustion.
The method based on the neural network model and the existing classical differential equation set solver CVODE of the present embodiment respectively calculate chemical reactions in the combustion numerical simulation, and compare the required calculation time consumption. The two methods are used as solvers of chemical reactions respectively on the premise of ensuring numerical simulation precision, and after 1ms of operation from the ignition of combustion simulation, the calculation time of the neural network model on the CPU, the calculation time of the neural network model on the GPU and the time of the CVODE on the CPU in the overall simulation are compared.
The thermochemical phase space in the above preferred embodiment is a vector space composed of thermochemical states including: temperature, pressure and mass fraction of the components. The global manifold sampling forms a low-dimensional sub-manifold in a thermochemical space by giving an initial state space consisting of temperature, pressure and equivalence ratio, and the initial state vector of the global manifold sampling evolves to a chemical steady state, and the global manifold sampling comprises the following specific steps: first, the way to randomly select data on the low-dimensional manifold is called zero-dimensional manifold sampling, and the data set obtained by sampling is a zero-dimensional manifold data set, denoted as D MF . Secondly, multi-scale sampling is carried out in the whole thermochemical phase space, namely a large number of samples are randomly collected according to logarithmic scale, a multi-scale sampling data set is obtained and is marked as D MS . Then in manifold data set D MF Removing the multi-scale data set D by taking the boundary of each dimension change rate as a reference MS Samples with a time change rate exceeding the boundary range are screened out, so that a batch of samples conforming to the actual physical properties are screened out.
For evolution in the steps, the state quantity after 1E-3s time steps is generated by using chemical kinetics open source software Canera at a given current state quantity, and for generating a label, the state quantity after 1E-6s time steps is generated by using chemical kinetics open source software Canera at a given current state quantity; the data obtained through merging evolution is input data, and the data obtained through merging labels is target data.
When data preprocessing is carried out, the following conversion mode is adopted for Box-Cox conversion:
where x is an argument and λ is a variable hyper-parameter.
When the data preprocessing is carried out, zero mean normalization is also included, namely, the mean value of the samples is subtracted from all characteristic values of the input samples, and the characteristic values are divided by the standard deviation of the samples, so that the mean value of the samples is zero, and the variance is 1.
When training the deep neural network, training the full-connection deep neural network by using a deep learning open source framework pytorch, wherein training data adopts the data set of the merging step, and the trained network model parameters are saved as a binary format file so as to be used in practical application later.
When the deep neural network model is coupled with widely used computational fluid dynamics programs, based on the existing mature computational fluid dynamics program EBI-DNS in the combustion field, the chemical reaction calculation part is changed into the deep neural network operation, so that the deep neural network coupling fluid is realized, and the combustion numerical simulation is realized.
Chemical reaction mechanism selection methane combustion values the primitive components, primitive reactions and reaction rate coefficients a/B/Ea in the methane DRM19 mechanism simulating the usual 21 component 84 reaction are described in table 1 below.
TABLE 1 primitive components and primitive reactions in component 84 reactions
The general substitution model for chemical reaction based on the deep neural network, which is provided by the preferred embodiment, removes the strong rigidity in the combustion chemical reaction, the single model can be used for laminar flow, jet flow flame and turbulent flame numerical simulation without fine adjustment, the chemical reaction multi-grid parallel calculation can be realized through a GPU, the multi-grid chemical reaction parallel calculation can be realized, the GPU model involved in the preferred embodiment is NVIDIATesla V100-SXM2-32GB, the CPU model is Intel (R) Xeon (R) Platinum 8260CPU@2.40GHz, and the test is to compare the acceleration effect of adopting CPU-GPU parallel calculation and adopting only CPU for parallel calculation. The calculation time after 1ms of the advance of the preferred embodiment is shown in fig. 6, and the chemical reaction in the combustion simulation is calculated by using a CVODE and a Deep Neural Network (DNN), wherein DNN-CPU refers to the operation of the DNN model on the CPU, and DNN-GPU refers to the operation of the DNN model on the GPU. The preferred embodiment shows that on the premise of ensuring the calculation accuracy, the neural network model is matched with the CPU to realize about 10 times of chemical reaction calculation acceleration, the GPU is matched with the parallel calculation to realize about 30 times of calculation ratio, the calculation efficiency can be remarkably improved by the chemical reaction acceleration algorithm based on the neural network model, the chemical reaction calculation acceleration in the combustion numerical simulation of hydrocarbon fuel can be effectively realized on the premise of ensuring the accuracy, the calculation efficiency is greatly improved, and the method is plug-and-play and efficient and convenient.
The foregoing describes in detail preferred embodiments of the present invention. It should be understood that numerous modifications and variations can be made in accordance with the concepts of the invention without requiring creative effort by one of ordinary skill in the art. Therefore, all technical solutions which can be obtained by logic analysis, reasoning or limited experiments based on the prior art by the person skilled in the art according to the inventive concept shall be within the scope of protection defined by the claims.
Claims (10)
1. A combustion chemistry calculation acceleration method based on a deep neural network, characterized in that the method comprises the following steps:
s101: global manifold sampling, said sampling being performed within a predetermined thermal chemistry phase space;
s103: preprocessing the sampled data, wherein the preprocessing comprises Box-Cox conversion and zero mean normalization;
s105: training and storing a deep neural network model;
s107: and coupling the pre-trained deep neural network model with a computational fluid dynamics program to realize the simulation calculation of the combustion numerical value.
2. The method of claim 1, wherein in step S101, the thermochemical phase space is a vector space composed of thermochemical states including: temperature, pressure and mass fraction of the components.
3. The method of claim 2, wherein the components in the thermochemical phase space comprise methane, oxygen, and nitrogen, the components having a mass fraction of 1:1:3.76, the initial temperature in the thermochemical phase space is 300K and the initial pressure in the thermochemical phase space is 1atm.
4. The method of claim 3, wherein the global manifold sampling comprises the steps of:
s1011: performing zero-dimensional manifold sampling on the low-dimensional sub-manifold in the thermochemical space to obtain a zero-dimensional manifold data set;
s1012: randomly collecting a large number of samples in the thermochemical phase space according to logarithmic scale to obtain a multi-scale sampling data set;
s1013: taking the boundary of each dimension change rate in the zero-dimensional manifold data set as a reference, removing samples with time change rates exceeding the boundary range in the multi-scale sampling data set, and screening to obtain sample data conforming to actual physical properties;
s1014: carrying out evolution, label generation and merging processing on the sample data to generate a data set; the evolution generates a state quantity after a 1E-3s time step from a preset current state quantity, and the generation tag generates a state quantity after a 1E-6s time step from the preset current state quantity; the merging comprises merging the data obtained by evolution as input data and merging the data obtained by generating the tag as target data.
5. The method of claim 1, wherein in the step S103, the Box-Cox transformation uses the following transformation method:
where x is an argument and λ is a variable hyper-parameter.
6. The method of claim 5, wherein in the step S103, the zero-mean normalization means: subtracting the average value of the samples from all the characteristic values of the input samples, dividing the average value by the standard deviation of the samples to make the average value of the samples zero and the variance 1.
7. The method of claim 1, wherein in the step S105, the deep neural network is a fully connected deep neural network, training data is obtained by using the generated data set in the global manifold sample, and network model parameters of the deep neural network after training are saved as a binary format file.
8. The method of claim 1, wherein in the step S107, the computational fluid dynamics program comprises EBI-DNS, and the chemical reaction calculation portion of the computational fluid dynamics program is replaced with the deep neural network model.
9. The method of claim 8, wherein the neural network model supports deployment on a CPU and/or GPU and supports parallel computing with a computing domain of 1.5cm x 1.5cm matrix area and a computing grid of N = 512 x 512.
10. A method as claimed in claim 9, wherein the thermal chemical phase space has an ignition source radius of 0.02cm, an ignition source energy of w/m3, and an ignition duration of 0.2ms, and wherein the initial velocity field and initial vortex field are generated by:
where E (k) is the energy spectrum, k is the frequency, average speed u rms =3m/s, constant k e ＝418.67。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310524037.3A CN116343936A (en) | 2023-05-10 | 2023-05-10 | Combustion chemical reaction calculation acceleration method based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310524037.3A CN116343936A (en) | 2023-05-10 | 2023-05-10 | Combustion chemical reaction calculation acceleration method based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116343936A true CN116343936A (en) | 2023-06-27 |
Family
ID=86876167
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310524037.3A Pending CN116343936A (en) | 2023-05-10 | 2023-05-10 | Combustion chemical reaction calculation acceleration method based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116343936A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117912585A (en) * | 2024-03-20 | 2024-04-19 | 中国人民解放军战略支援部队航天工程大学 | Optimization method for combustion chemical reaction based on deep artificial neural network |
-
2023
- 2023-05-10 CN CN202310524037.3A patent/CN116343936A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117912585A (en) * | 2024-03-20 | 2024-04-19 | 中国人民解放军战略支援部队航天工程大学 | Optimization method for combustion chemical reaction based on deep artificial neural network |
| CN117912585B (en) * | 2024-03-20 | 2024-06-25 | 中国人民解放军战略支援部队航天工程大学 | Optimization method for combustion chemical reaction based on deep artificial neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Själander et al. | EPIC: An energy-efficient, high-performance GPGPU computing research infrastructure | |
| Bellomo et al. | Modeling in applied sciences: A kinetic theory approach | |
| Saqib et al. | Pipelined decision tree classification accelerator implementation in FPGA (DT-CAIF) | |
| Liu et al. | Hardware acceleration of fully quantized bert for efficient natural language processing | |
| Xiong et al. | Efficient parallel implementation of the lattice Boltzmann method on large clusters of graphic processing units | |
| CN116343936A (en) | Combustion chemical reaction calculation acceleration method based on deep neural network | |
| Singh et al. | Quantum artificial intelligence for the science of climate change | |
| Rybakin et al. | Accelerated solution of problems of combustion gas dynamics on GPUs | |
| Rościszewski et al. | Optimizing throughput of Seq2Seq model training on the IPU platform for AI-accelerated CFD simulations | |
| Patel et al. | Artificial intelligence is aiding the search for energy materials | |
| Ugliengo | The rise of computer modeling in prebiotic chemistry. Comment on" Prebiotic chemistry and origins of life research with atomistic computer simulations" by A. Pérez-Villa et al. | |
| Cantwell | Future directions in turbulence research and the role of organized motion | |
| Wang et al. | Integrated materials design and informatics platform within the materials genome framework | |
| Kashkovsky | DSMC investigations of reentry vehicle aerothermodynamics on GPU | |
| Sankaran | GPU-accelerated software library for unsteady flamelet modeling of turbulent combustion with complex chemical kinetics | |
| Scherding et al. | Data-driven reduction of lookup libraries for the simulation of non-equilibrium effects in hypersonic flows | |
| Rojek et al. | Chemical Mixing Simulations with Integrated AI Accelerator | |
| Sarsembayev et al. | USING THE CUDA TECHNOLOGY TO SPEED UP COMPUTATIONS IN PROBLEMS OF CHEMICAL KINETICS | |
| Melnikova et al. | Periodic variations of an autowave structure in two-dimensional system of parabolic equations | |
| Chen et al. | eSSpMV: An embedded-FPGA-based hardware accelerator for symmetric sparse matrix-vector multiplication | |
| Yu et al. | Application of higher-order FV-WENO scheme to the interaction between shock wave and bubble | |
| Mao et al. | An integrated framework for accelerating reactive flow simulation using GPU and machine learning models | |
| Dziekan et al. | University of Warsaw Lagrangian Cloud Model (UWLCM) 2.0: adaptation of a mixed Eulerian–Lagrangian numerical model for heterogeneous computing clusters | |
| Liu et al. | Evaluating a New Attention Framework Based on Matrix Blocking for Attention Models on FPGAs | |
| Ribeiro et al. | A Data-Driven Extension to the Evolution-Variable Manifold Model in High-Speed Hydrogen Combustion |
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 |