CN117875188A - Hydrogen leakage diffusion process modeling method and system - Google Patents

Hydrogen leakage diffusion process modeling method and system Download PDF

Info

Publication number
CN117875188A
CN117875188A CN202410158883.2A CN202410158883A CN117875188A CN 117875188 A CN117875188 A CN 117875188A CN 202410158883 A CN202410158883 A CN 202410158883A CN 117875188 A CN117875188 A CN 117875188A
Authority
CN
China
Prior art keywords
hydrogen leakage
model
leakage diffusion
diffusion process
hydrogen
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
Application number
CN202410158883.2A
Other languages
Chinese (zh)
Inventor
孙波
张辉
李雪芳
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.)
Shandong University
Original Assignee
Shandong University
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 Shandong University filed Critical Shandong University
Priority to CN202410158883.2A priority Critical patent/CN117875188A/en
Publication of CN117875188A publication Critical patent/CN117875188A/en
Pending legal-status Critical Current

Links

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
    • G06F30/27—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
    • 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
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/096—Transfer learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/08—Fluids
    • 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
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Computer Hardware Design (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Geometry (AREA)
  • Molecular Biology (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Fluid Mechanics (AREA)
  • Algebra (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Examining Or Testing Airtightness (AREA)

Abstract

The invention provides a modeling method and a system for a hydrogen leakage diffusion process, which relate to the field of hydrogen safety, and specifically comprise the following steps: acquiring space-time coordinates of a limited space where a hydrogen leakage diffusion process to be modeled is located; inputting the space-time coordinates into a trained physical constraint data driving model, and recovering a physical field with complete hydrogen leakage diffusion process in a limited space to obtain hydrogen leakage diffusion concentration, volume concentration, speed and temperature in the limited space under the space-time coordinates as a modeling result; the invention follows the distribution rule of training data samples, follows the physical law described by partial differential equation, constructs a hydrogen leakage diffusion model which is reasonable in physics, accurate in mathematics, stable and efficient in calculation, solves the problem of disjoint between the traditional numerical simulation modeling method and the actual diffusion condition, avoids the difficult problem that the conventional data driving method cannot fuse the domain knowledge, and gets rid of the dependence on massive training data.

Description

Hydrogen leakage diffusion process modeling method and system
Technical Field
The invention belongs to the field of hydrogen safety, and particularly relates to a modeling method and a system for a hydrogen leakage diffusion process.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
From the global aspect, the establishment of a truly sustainable low-carbon novel energy system is a mainstream development trend; in this context, hydrogen energy is given important life and significance as a green energy carrier and energy storage medium, and the development of the hydrogen energy industry is attracting attention.
Along with the rapid development of hydrogen energy, the hydrogen safety problem is increasingly prominent, and becomes one of key factors restricting the further popularization and application of the hydrogen energy; the hydrogen has the characteristics of low density, fast diffusion and hydrogen embrittlement effect on a plurality of materials, and is extremely easy to locally gather once leakage occurs in a limited space, so that the explosion risk is increased; in addition, the combustion range is wide, the combustion heat value is high, the combustion speed is high, and the explosion energy is high, so that the consequences are very serious; the hydrogen leakage is often the starting event of a hydrogen safety accident and is also the input condition of the analysis of the subsequent blasting behavior, so that the establishment of a hydrogen leakage diffusion model and the exploration of an evolution rule are basic contents in the field of hydrogen safety; the method is helpful to define the evolution process and the consequences of the hydrogen safety accident in the process of developing the hydrogen energy and the application technology thereof, thereby forming an emergency treatment measure for the hydrogen safety accident and promoting the development and the large-scale application of the hydrogen energy technology.
However, hydrogen leak diffusion is an extremely complex physical process involving both gas dynamics and hydrodynamics; through years of research, a large number of hydrogen leakage diffusion physical models based on a numerical simulation modeling method are developed, and are mostly described or described in the form of partial differential equations, for example, a prediction method and a prediction system of a liquid hydrogen leakage diffusion range of Chinese patent No. 113128755A are used for predicting the liquid hydrogen leakage diffusion range by measuring the temperature T of a liquid hydrogen leakage part and the liquid hydrogen leakage mass m, constructing an instantaneous evaporation diffusion model and a continuous evaporation diffusion model and predicting the liquid hydrogen leakage diffusion range; chinese patent CN116958521A discloses a method and a system for analyzing the liquid hydrogen diffusion state in real time, shooting temperature-change white fog images of a liquid hydrogen leakage site according to a set time interval, monitoring surrounding related atmospheric environmental factors, and constructing a hydrogen concentration calculation model to realize effective prediction of liquid hydrogen diffusion; the modeling methods face the problems of overlarge errors caused by simple physical models, overlarge solving complexity caused by complex physical models, overlarge solving deviation caused by missing or inaccurate measurement of physical model parameters and initial value data, and the like; therefore, the traditional numerical simulation modeling method is difficult to accurately model the hydrogen leakage diffusion process, difficult to reflect the distribution rule of field actual measurement data and incapable of meeting the actual field requirements.
In summary, the modeling of hydrogen leakage diffusion is still limited to a numerical simulation method at present, and a hydrogen leakage diffusion accurate modeling method which can meet physical constraints and can be fused with on-site sparse observation data is lacking.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a modeling method and a system for a hydrogen leakage diffusion process in a limited space, which follow the distribution rule of training data samples, follow the physical law described by partial differential equations, construct a hydrogen leakage diffusion model which is reasonable in physics, accurate in mathematics, stable and efficient in calculation, overcome the problem of disjoint between the traditional numerical simulation modeling method and the actual diffusion condition, avoid the difficult problem that the conventional data driving method cannot be used for fusing field knowledge, and break away from the dependence on massive training data.
To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:
the first aspect of the invention provides a hydrogen leakage diffusion process modeling method.
A hydrogen leak diffusion process modeling method comprising:
acquiring space-time coordinates of a limited space where a hydrogen leakage diffusion process to be modeled is located;
inputting the space-time coordinates into a trained physical constraint data driving model, and recovering a physical field with complete hydrogen leakage diffusion process in a limited space to obtain hydrogen leakage diffusion concentration, volume concentration, speed and temperature in the limited space under the space-time coordinates as a modeling result;
the physical constraint data driving model is constructed based on a neural network, high-fidelity data obtained through numerical simulation and sparse observation data obtained through field actual measurement are used as training data, and a partial differential equation followed by a hydrogen leakage diffusion process in a limited space is embedded into a model training process in a physical constraint mode, so that a final trained physical constraint data driving model is obtained.
Further, the physical constraint data driven model includes three sub-models: the agent sub-model, the difference sub-model and the recovery sub-model have the same network structure.
Furthermore, the agent sub-model takes high-fidelity data obtained by numerical simulation as a training set, embeds a hydrogen leakage diffusion partial differential equation as physical constraint, and realizes self-supervision so as to learn the diffusion physical rule of hydrogen leakage in a limited space.
Furthermore, the high-fidelity data obtained by the numerical simulation is based on computational fluid dynamics CFD technology, and the problem of hydrogen leakage diffusion in the limited space is subjected to numerical simulation to obtain the high-fidelity data.
Furthermore, the difference sub-model takes sparse observation data obtained through field actual measurement as a training set, and minimizes the difference between high-fidelity data and observation data by transferring and learning the deviation between the agent sub-model and the real environment.
Further, the sparse observation data obtained through field actual measurement are based on a sensor technology, concentration data in the hydrogen leakage diffusion process in the limited space are collected, and the sparse observation data are obtained.
Further, the recovery sub-model transfers the trained network parameters of the difference sub-model to the recovery sub-model, and is directly used for recovering the complete physical field in the hydrogen leakage diffusion process in the limited space.
A second aspect of the present invention provides a hydrogen leak diffusion process modeling system.
A hydrogen leakage diffusion process modeling system, comprising an acquisition module and a modeling module:
an acquisition module configured to: acquiring space-time coordinates of a limited space where a hydrogen leakage diffusion process to be modeled is located;
a modeling module configured to: inputting the space-time coordinates into a trained physical constraint data driving model, and recovering a physical field with complete hydrogen leakage diffusion process in a limited space to obtain hydrogen leakage diffusion concentration, volume concentration, speed and temperature in the limited space under the space-time coordinates as a modeling result;
the physical constraint data driving model is constructed based on a neural network, high-fidelity data obtained through numerical simulation and sparse observation data obtained through field actual measurement are used as training data, and a partial differential equation followed by a hydrogen leakage diffusion process in a limited space is embedded into a model training process in a physical constraint mode, so that a final trained physical constraint data driving model is obtained.
A third aspect of the present invention provides a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements steps in a hydrogen leakage diffusion process modeling method according to the first aspect of the present invention.
A fourth aspect of the present invention provides an electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, the processor implementing the steps in a hydrogen leakage diffusion process modeling method according to the first aspect of the present invention when the program is executed.
The one or more of the above technical solutions have the following beneficial effects:
aiming at the hydrogen leakage diffusion process in a limited space, the invention provides a modeling method and a system for the hydrogen leakage diffusion process, which follow the distribution rule of training data samples and follow the physical law described by partial differential equations, are constructed based on a neural network, take high-fidelity data obtained by numerical simulation and sparse observation data obtained by field actual measurement as training data, embed the partial differential equation followed by the hydrogen leakage diffusion process in the limited space into a model training process in a physical constraint mode, construct a physically reasonable, mathematically accurate, computationally stable and efficient hydrogen leakage diffusion model, overcome the problem of disjoint of the traditional numerical simulation modeling method and actual diffusion conditions, avoid the problem that the conventional data driving method cannot fuse field knowledge, and break away from the dependence on massive training data.
Additional aspects of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
Fig. 1 is a flow chart of a method of a first embodiment.
Fig. 2 is a schematic diagram of a first embodiment.
Fig. 3 is a schematic diagram of a proxy sub-model of the first embodiment.
Fig. 4 is a schematic diagram of a first embodiment difference submodel.
FIG. 5 is a schematic diagram of a first embodiment recovery sub-model.
Detailed Description
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present application. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Example 1
In one embodiment of the present disclosure, a modeling method for a hydrogen leakage diffusion process is provided, as shown in fig. 2, including the following steps:
step S1: acquiring space-time coordinates of a limited space where a hydrogen leakage diffusion process to be modeled is located;
step S2: inputting the space-time coordinates into a trained physical constraint data driving model, and recovering a physical field with complete hydrogen leakage diffusion process in a limited space to obtain hydrogen leakage diffusion concentration, volume concentration, speed and temperature in the limited space under the space-time coordinates as a modeling result;
the physical constraint data driving model is constructed based on a neural network, high-fidelity data obtained through numerical simulation and sparse observation data obtained through field actual measurement are used as training data, and a partial differential equation followed by a hydrogen leakage diffusion process in a limited space is embedded into a model training process in a physical constraint mode, so that a final trained physical constraint data driving model is obtained.
The following describes in detail the implementation procedure of a modeling method for hydrogen leakage diffusion process in this embodiment from the viewpoint of construction and training of a physical constraint data-driven model.
The problem of hydrogen leakage and diffusion in a limited space relates to aerodynamics and hydrodynamics, and is extremely complex; the diffusion concentration distribution of hydrogen in a limited space is influenced by factors such as leakage position, release rate, leakage duration and the like, and the evolution paths of leakage diffusion are also different due to boundary conditions and environmental changes, so that accurate modeling of the leakage diffusion process of hydrogen in the limited space is difficult.
In order to solve the above-mentioned problems, the present embodiment provides a physical constraint data-driven model for evolving and modeling a hydrogen leakage diffusion process, as shown in fig. 2:
first, partial differential equations fusing domain knowledge are generalized to physical constraints for the generation of the following high-fidelity data and knowledge embedding (training of the model).
And then, generating high-fidelity numerical simulation data based on the physical constraint, and acquiring hydrogen concentration sparse actual measurement data with space distribution conditions through a sensor to form a training set.
Then, constructing a proxy sub-model and a difference sub-model based on a neural network, wherein the proxy sub-model is used for approximating high-fidelity data behaviors and capturing the variation trend of the hydrogen concentration in a limited space by encoding physical constraints in a loss function; in order to learn the deviation between the proxy sub-model and the environment of the real hydrogen leak, the measured sparse data of the hydrogen leak diffusion is used to train a difference sub-model based on transfer learning.
And finally, fusing the agent sub-model and the difference sub-model, constructing a recovery sub-model, and recovering the complete hydrogen leakage diffusion physical field in the limited space to realize the rapid and accurate evolution of the diffusion process.
The following description will be made respectively:
1. physical constraints
Assume that: the hydrogen leaked into the limited space is taken as ideal gas, an ideal gas state equation is followed, the hydrogen and air mixture does not react chemically in the flowing process, and the hydrogen leakage rate is unchanged in the leakage process.
Based on the assumption, the leakage diffusion problem of hydrogen in a limited space can be attributed to the single-phase multicomponent convection diffusion problem without chemical reaction; therefore, the problem of hydrogen leakage diffusion in a limited space needs to follow several partial differential equations, namely a continuity equation, a momentum conservation equation, an energy conservation equation and a component mass conservation equation, which are also called control equations followed by the hydrogen leakage diffusion process in the limited space.
Continuity equation:
wherein ρ is the density of the mixture in kg/m 3 The method comprises the steps of carrying out a first treatment on the surface of the t is time, and the unit is s; x is x j J is the j-th direction, and j=1, 2,3; u (u) j The unit is m/s for the speed in the j-th direction.
Momentum conservation equation:
wherein μ is the dynamic viscosity of the mixture of hydrogen and air in Pa.s; g is the acceleration of gravity in m/s 2 The method comprises the steps of carrying out a first treatment on the surface of the p is absolute pressure in Pa; ρ a Is the density of air in kg/m 3 。
Energy conservation equation:
wherein, c p The specific heat capacity is expressed as J/(kg.K); t is temperature, in K; k is the heat transfer coefficient in W/(m.K); s is S T Is a viscous dissipation term.
Component mass conservation equation:
wherein, c s Is the mass fraction of component s; d (D) s Is a binary diffusion coefficient.
The physical process of hydrogen leakage diffusion in the limited space follows the partial differential equation, including the continuity equation, the momentum conservation equation, the energy conservation equation and the component mass conservation equation.
The traditional hydrogen leakage diffusion modeling method based on numerical simulation needs to consider the partial differential equation and a proper turbulence model equation, and adopts special software such as Fluent and the like to solve so as to obtain a hydrogen leakage diffusion physical model.
The data-driven method is at the other extreme, and is completely dependent on field actual measurement or simulation data for training, so that no physical constraint exists to guide the process of model construction.
2. Training data
In this embodiment, two types of data are used as training data, which are high-fidelity data and field actual measurement data, respectively.
On the one hand, based on the existing mature computational fluid dynamics CFD technology, numerical simulation is carried out on the problem of hydrogen leakage and diffusion in a limited space so as to obtain high-fidelity data; the CFD-based numerical simulation method can accurately simulate the concentration distribution of the hydrogen leakage diffusion process in the limited space under the given boundary condition, so that the simulation result can be regarded as high-fidelity data; however, in a practical limited space, the boundary conditions cannot be accurately obtained; this results in high-fidelity data not being able to accurately model the evolution path of hydrogen leakage diffusion in the real world, although high-fidelity data may reflect the physical process of hydrogen leakage diffusion to some extent.
On the other hand, based on the mature sensor technology at present, concentration data in the hydrogen leakage diffusion process can be acquired in a limited space so as to acquire sparse observation data.
In this embodiment, the sparse observation data adopted does not make explicit limitation on the data amount, that is, the modeling method provided in this embodiment does not completely depend on data driving, so that a large amount of field actual measurement data is not required to complete the supervised learning task; it should be noted that this embodiment aims to implement accurate modeling of the hydrogen leakage diffusion evolution in a limited space, and therefore, in addition to the hydrogen concentration data collected by the sensor, the corresponding spatial position (three-dimensional coordinates) and time of each hydrogen concentration data are also required as part of sparse observation data.
3. Physical constraint data driven model
The embodiment provides a physical constraint data driving model for hydrogen leakage diffusion in a limited space, which consists of three sub-models based on transfer learning, takes high-fidelity data and sparse observation data as training data, and allows a control equation followed by the hydrogen leakage diffusion process in the limited space to be embedded into the model in a physical constraint mode; in this embodiment, the network structure of the physical constraint data driving model adopts a conventional fully-linked neural network.
(1) Agent sub-model
The agent sub-model based on the fully-linked neural network is used for approximating CFD numerical simulation high-assurance data; meanwhile, a control equation (namely a partial differential equation) of physical constraint is embedded into the proxy sub-model in the form of a loss function so as to force the full-link neural network to follow the physical rule of hydrogen leakage diffusion in a limited space, and the network structure of the proxy sub-model is shown in figure 3.
High-fidelity data assuming hydrogen leak diffusion in a confined space can be expressed asWherein N is the total number of high-fidelity data, and the agent sub-model uses space-time coordinates +.>For input, the hydrogen concentration +.>Thus, the input dimension of the proxy sub-model is 4; in order to fully consider other physical quantities changing with time and space in the leakage diffusion process, namely density rho, temperature T and speed u, the output dimension of the agent submodel is set to be 4; the hidden layer network structure can be flexibly designed according to actual requirements; let->Representing a fully linked neural network, then the functional relationship represented between its inputs and outputs is:
in the method, in the process of the invention,w s and b s Weight and bias parameters representing a fully linked neural network, a s Parameters such as gravitational acceleration, specific heat capacity, heat transfer coefficient and the like are represented;
in the present embodiment, the loss function L of the proxy sub-model s The method consists of two items, namely a data driving loss item and a physical constraint loss item, and the two items are shown in the following formula:
in the method, in the process of the invention,is a data drive loss term,/->The coefficient is the coefficient of the term, and the concentration evolution result of the agent submodel is enabled to be continuously approximate to high-fidelity data of hydrogen concentration in the leakage diffusion process in the training process, namely, the evolution rule of hydrogen leakage diffusion is learned from the high-fidelity data;
is a physical constraint loss term,/->Is a coefficient of the term that allows the agent sub-model to conform to the equation (1-4) in terms of the temporal and spatial evolution of hydrogen leak diffusion.
Data driven penalty termHaving the form of an average absolute error, expressed by the formula:
wherein,represents the hydrogen concentration of the proxy sub-model output, +.>And (3) representing the hydrogen concentration in the high-fidelity data, wherein N is the total number of the high-fidelity data.
Physical constraint loss termExpressed by the formula:
in the method, in the process of the invention,loss term representing the j-th dimension in the e-th control equation,/and/or>Is the corresponding weight parameter.
Specifically:
the automatic differentiation technology in the deep learning is mature and widely applied, and the physical constraint data driving model provided by the embodiment can be simply and efficiently realized; that is, equations (9) - (12) do not require manual derivation when embedded as physical constraints in the proxy sub-model, and the respective partial derivatives can be quickly calculated during training by means of automatic differentiation techniques.
In this embodiment, the training optimizer of the proxy sub-model supports Adam and AdaBound etc. adaptive algorithms and uses a small batch gradient descent to minimize losses:
(2) Differential submodel
The difference submodel based on the full-link neural network is used for learning the deviation between the agent submodel and the real environment, and hydrogen concentration data is used for training the difference submodel based on transfer learning at a plurality of sparsely distributed points; as shown in fig. 1, when training the differential sub-model, network parameters (including weight parameters and bias parameters) of the proxy sub-model need to be loaded to the differential sub-model, and only the weight and bias of the last layer of the proxy sub-model are retrained, while the weights and bias of other layers are fixed, as shown in fig. 4; the network structure of the difference submodel is consistent with that of the agent submodel, space-time coordinates in a limited space are used as input, and output is the change condition of the hydrogen leakage diffusion physical field under the corresponding coordinates, namely the hydrogen leakage diffusion concentration, the volume concentration, the speed and the temperature.
Assume that sparse observations can be expressed asThe difference submodel takes space-time coordinatesFor the input, the output is the hydrogen concentration +.>The relationship between the space-time coordinates and the hydrogen concentration constructed by the difference submodel can be expressed as:
wherein w is d 、b d Network parameters representing the difference sub-model, which are at initialization time the network parameters w of the proxy sub-model s * And b s *；a d A parameter indicating leakage and diffusion of hydrogen in a limited space, and a is at the time of initialization s *。
In the present embodiment, the loss function L of the difference submodel d The method also comprises two items, namely a data driving loss item and a physical constraint loss item, and the two items are shown in the following formula:
in the method, in the process of the invention,is a data drive loss term,/->The coefficient is the coefficient of the term, and the concentration evolution result of the difference submodel is enabled to be continuously approximate to the measured data of the hydrogen concentration in the leakage diffusion process in the training process, namely, the evolution rule of the hydrogen leakage diffusion is learned from the actual sparse observation data on site; />Is a physical constraint loss term,/->Is a coefficient of the term that allows the space-time evolution of the differential submodel in hydrogen leak diffusion to conform to the formula (1-4); />And->Is expressed in a form similar to the agent submodel +.>And->And will not be described in detail, see fig. 3.
In this embodiment, the training process of the difference sub-model may be regarded as fine tuning after pre-training the proxy sub-model, where the initialized network parameters originate from the proxy sub-model, and only optimize and update the weights and offsets of the last layer of network of the fully-linked neural network; the training optimizer of the difference sub-model supports Adam and AdaBound etc. adaptive algorithms and employs small batch gradient descent to minimize losses:
wherein m represents the network layer number of the difference submodel;representing the weight and bias parameters of the last layer.
(3) Restoring sub-model
In this embodiment, the recovery sub-model is a combination of the proxy sub-model and the differential sub-model, and aims to recover a physical field with complete hydrogen leakage diffusion process in the limited space, namely to realize hydrogen leakage diffusion evolution, and the network structure of the recovery sub-model is the same as that of the proxy sub-model and the differential sub-model; in the embodiment, the recovery sub-model is not required to be retrained in theory, and migration of network parameters is carried out from the trained difference sub-model to obtain a final recovery sub-model, wherein the input of the final recovery sub-model is space-time coordinates, and the output of the final recovery sub-model is hydrogen leakage diffusion concentration, volume concentration, speed and temperature in a limited space under the corresponding space-time coordinates; the functional relationship constructed by the recovery sub-model is expressed as:
the parameters of the recovery sub-model originate from the parameter migration of the proxy sub-model and the difference sub-model, as shown in fig. 5.
The embodiment provides a physical constraint data driving modeling method and system for hydrogen leakage diffusion, which solve the problem that the traditional numerical simulation modeling method is disjointed from the actual diffusion condition, avoid the difficulty that the conventional data driving method cannot fuse the knowledge in the field, and break away from the dependence on massive training data.
The modeling method and the modeling system provided by the embodiment not only follow the distribution rule of sparse measurement data samples, but also follow the physical law described by the control equation, so that a hydrogen leakage diffusion evolution model with more generalization capability can be obtained by learning fewer data samples.
Example two
In one embodiment of the present disclosure, a hydrogen leakage diffusion process modeling system is provided, including an acquisition module and a modeling module:
an acquisition module configured to: acquiring space-time coordinates of a limited space where a hydrogen leakage diffusion process to be modeled is located;
a modeling module configured to: inputting the space-time coordinates into a trained physical constraint data driving model, and recovering a physical field with complete hydrogen leakage diffusion process in a limited space to obtain hydrogen leakage diffusion concentration, volume concentration, speed and temperature in the limited space under the space-time coordinates as a modeling result;
the physical constraint data driving model is constructed based on a neural network, high-fidelity data obtained through numerical simulation and sparse observation data obtained through field actual measurement are used as training data, and a partial differential equation followed by a hydrogen leakage diffusion process in a limited space is embedded into a model training process in a physical constraint mode, so that a final trained physical constraint data driving model is obtained.
Example III
An object of the present embodiment is to provide a computer-readable storage medium.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs steps in a hydrogen leak diffusion process modeling method according to one embodiment of the present disclosure.
Example IV
An object of the present embodiment is to provide an electronic apparatus.
An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, the processor implementing the steps in a hydrogen leak diffusion process modeling method according to embodiment one of the present disclosure when the program is executed.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. A method of modeling a hydrogen leak diffusion process, comprising:
acquiring space-time coordinates of a limited space where a hydrogen leakage diffusion process to be modeled is located;
inputting the space-time coordinates into a trained physical constraint data driving model, and recovering a physical field with complete hydrogen leakage diffusion process in a limited space to obtain hydrogen leakage diffusion concentration, volume concentration, speed and temperature in the limited space under the space-time coordinates as a modeling result;
the physical constraint data driving model is constructed based on a neural network, high-fidelity data obtained through numerical simulation and sparse observation data obtained through field actual measurement are used as training data, and a partial differential equation followed by a hydrogen leakage diffusion process in a limited space is embedded into a model training process in a physical constraint mode, so that a final trained physical constraint data driving model is obtained.
2. A method of modeling a hydrogen leak diffusion process as defined in claim 1, wherein said physical constraint data driven model comprises three sub-models: the agent sub-model, the difference sub-model and the recovery sub-model have the same network structure.
3. The modeling method of hydrogen leakage diffusion process according to claim 2, wherein the proxy sub-model uses high-fidelity data obtained by numerical simulation as a training set, embeds a partial differential equation of hydrogen leakage diffusion as physical constraint, and realizes self-supervision to learn the diffusion physical rule of hydrogen leakage in a limited space.
4. The modeling method of hydrogen leakage diffusion process according to claim 1, wherein the high-fidelity data obtained by the numerical simulation is based on computational fluid dynamics CFD technology, and the numerical simulation is performed on the problem of hydrogen leakage diffusion in the limited space to obtain the high-fidelity data.
5. The modeling method of hydrogen leakage diffusion process according to claim 2, wherein the difference submodel uses sparse observed data obtained by field measurement as a training set, and shifts the deviation between the learning agent submodel and the real environment to minimize the difference between high-fidelity data and observed data.
6. The modeling method for hydrogen leakage diffusion process according to claim 1, wherein the sparse observation data obtained by field measurement is based on sensor technology, and concentration data in the hydrogen leakage diffusion process in the limited space is collected to obtain the sparse observation data.
7. The modeling method of hydrogen leakage diffusion process according to claim 2, wherein the recovery sub-model transfers the trained network parameters of the differential sub-model to the recovery sub-model, and is directly used for recovering the complete physical field of the hydrogen leakage diffusion process in the limited space.
8. A hydrogen leakage diffusion process modeling system, comprising an acquisition module and a modeling module:
an acquisition module configured to: acquiring space-time coordinates of a limited space where a hydrogen leakage diffusion process to be modeled is located;
a modeling module configured to: inputting the space-time coordinates into a trained physical constraint data driving model, and recovering a physical field with complete hydrogen leakage diffusion process in a limited space to obtain hydrogen leakage diffusion concentration, volume concentration, speed and temperature in the limited space under the space-time coordinates as a modeling result;
the physical constraint data driving model is constructed based on a neural network, high-fidelity data obtained through numerical simulation and sparse observation data obtained through field actual measurement are used as training data, and a partial differential equation followed by a hydrogen leakage diffusion process in a limited space is embedded into a model training process in a physical constraint mode, so that a final trained physical constraint data driving model is obtained.
9. An electronic device, comprising:
a memory for non-transitory storage of computer readable instructions;
a processor for executing the computer readable instructions;
wherein the computer readable instructions, when executed by the processor, perform the method of any of the preceding claims 1-7.
10. A storage medium, characterized by non-transitory storing computer readable instructions, wherein the computer readable instructions, when executed by a computer, perform the method of any of claims 1-7.
CN202410158883.2A 2024-02-04 2024-02-04 Hydrogen leakage diffusion process modeling method and system Pending CN117875188A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410158883.2A CN117875188A (en) 2024-02-04 2024-02-04 Hydrogen leakage diffusion process modeling method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410158883.2A CN117875188A (en) 2024-02-04 2024-02-04 Hydrogen leakage diffusion process modeling method and system

Publications (1)

Publication Number Publication Date
CN117875188A true CN117875188A (en) 2024-04-12

Family

ID=90583041

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410158883.2A Pending CN117875188A (en) 2024-02-04 2024-02-04 Hydrogen leakage diffusion process modeling method and system

Country Status (1)

Country Link
CN (1) CN117875188A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118211489A (en) * 2024-04-19 2024-06-18 中国石油大学(华东) Case library-based intelligent rapid prediction method for consequences of leakage accidents of hydrogen station
CN118428242A (en) * 2024-07-03 2024-08-02 德燃(浙江)动力科技有限公司 Model training method and hydrogen leakage prediction method for hydrogen fuel cell automobile

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118211489A (en) * 2024-04-19 2024-06-18 中国石油大学(华东) Case library-based intelligent rapid prediction method for consequences of leakage accidents of hydrogen station
CN118428242A (en) * 2024-07-03 2024-08-02 德燃(浙江)动力科技有限公司 Model training method and hydrogen leakage prediction method for hydrogen fuel cell automobile

Similar Documents

Publication Publication Date Title
Li et al. Digital twin in aerospace industry: A gentle introduction
Wang et al. Hybrid physics-based and data-driven models for smart manufacturing: Modelling, simulation, and explainability
Patil et al. Real time prediction of internal temperature of heat generating bodies using neural network
CN117875188A (en) Hydrogen leakage diffusion process modeling method and system
CN107609313B (en) Reliability analysis method for passive safety system for ship
Sotgiu et al. A turbulent heat flux prediction framework based on tensor representation theory and machine learning
CN111664823B (en) Method for detecting thickness of scale layer of voltage-sharing electrode based on difference of medium heat conduction coefficients
Li et al. Deep learning for multifidelity aerodynamic distribution modeling from experimental and simulation data
CN114638048A (en) Three-dimensional spray pipe flow field rapid prediction and sensitivity parameter analysis method and device
Liu et al. Partially implicit finite difference scheme for calculating dynamic pressure in a terrain-following coordinate non-hydrostatic ocean model
Moya et al. Physics perception in sloshing scenes with guaranteed thermodynamic consistency
Peng et al. Multi-fidelity nonlinear unsteady aerodynamic modeling and uncertainty estimation based on Hierarchical Kriging
CN116011125A (en) Response prediction method for uncertain axle coupling system
Wen et al. Cost reduction for data acquisition based on data fusion: Reconstructing the surface temperature of a turbine blade
Hollenbeck et al. A digital twin framework for environmental sensing with suas
Tao et al. A digital twin-based fault diagnostic method for subsea control systems
Lin et al. Towards efficient and accurate CO2 sequestration optimization: Integrating hierarchical spatio-temporal information into deep-learning-based surrogate models
Serrano et al. Infinity: Neural field modeling for reynolds-averaged navier-stokes equations
Liu et al. Remaining useful life prediction combining temporal convolutional network with nonlinear target function
Xiang et al. Adaptive GN block-based model for seismic response prediction of train-bridge coupled systems
Guo et al. Steady-state characteristics prediction of marine towing cable with BPNN
Wang et al. The prediction of external flow field and hydrodynamic force with limited data using deep neural network
Xu et al. A finite element level set method based on adaptive octree meshes for thermal free‐surface flows
Joseph et al. Mesh Based Neural Networks for Estimating High Fidelity CFD from Low Fidelity Input
Harada et al. Application of Transfer Learning for Multi-Fidelity Regression using Physics-Informed Neural Network on an Airfoil

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