A method for constructing a high-pressure fluid model

Abstract

This invention relates to the field of fluid mechanics, specifically to a method for constructing a high-pressure fluid model. The method includes: acquiring multi-scale fluid data, including molecular dynamics data, experimental-scale data, and in-situ detection data; transferring data from the low- and medium-pressure domains to the ultra-high-pressure domain using a pressure-gradient transfer learning network, and outputting predicted values ​​of physical property parameters; receiving in-situ data in real time using embedded hardware and calculating correction factors; dynamically reconstructing the source terms of the fluid control equations based on the correction factors; and triggering a reinforcement learning optimizer to reconstruct the model equations when the average relative error exceeds a threshold. The purpose of this invention is to provide a method, computer equipment, and storage medium for constructing a high-pressure fluid model, solving many problems in the prior art and demonstrating significant advantages in the accuracy, real-time performance, and adaptability of the fluid model, providing strong technical support for research and applications in related fields.

Description

Technical Field

[0001] This invention relates to the field of fluid mechanics, and more specifically to a method for constructing a high-pressure fluid model. Background Technology

[0002] Currently, the construction of high-pressure fluid models has significant application value in fields such as chemical engineering, petroleum, and natural gas. Traditional fluid model construction methods typically rely on experimental data and theoretical derivations; however, these methods often have limitations when facing complex fluid behavior and high-pressure environments. Existing technologies primarily rely on single-scale data, making it difficult to comprehensively reflect the physical properties of fluids under different pressure conditions. Furthermore, existing models exhibit low prediction accuracy under high-pressure conditions, failing to meet the high accuracy and real-time requirements of industrial applications.

[0003] In acquiring multi-scale fluid data, existing technologies often lack effective integration mechanisms, leading to information silos between data at different scales and affecting the overall performance of the model. Meanwhile, traditional model update methods are typically static and cannot respond to changes in fluid state in real time, resulting in insufficient model adaptability. This is especially true in ultra-high pressure domains, where fluid properties change drastically, making accurate prediction difficult using traditional methods.

[0004] Therefore, there is an urgent need for a new method for constructing high-pressure fluid models that can effectively integrate multi-scale fluid data and achieve data transfer and optimization through advanced learning algorithms, thereby improving the model's prediction accuracy and real-time performance. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention aims to provide a method for constructing a high-pressure fluid model, a computer device, and a storage medium, which solves many problems in the existing technology and also shows significant advantages in the accuracy, real-time performance, and adaptability of the fluid model, providing strong technical support for research and application in related fields.

[0006] This invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention discloses a method for constructing a high-pressure fluid model, comprising the following steps:

[0008] S100. Acquire multi-scale fluid data, including molecular dynamics data, experimental-scale data, and in-situ detection data;

[0009] S200. By using a pressure grading transfer learning network, data from the medium and low pressure domains are transferred to the ultra-high pressure domain, and predicted values ​​of physical property parameters are output.

[0010] S300. Receives in-situ data in real time based on embedded hardware and calculates correction factors;

[0011] S400. Dynamically reconstructing the source terms of the fluid control equations based on correction factors;

[0012] S500. When the average relative error exceeds the threshold, the reinforcement learning optimizer is triggered to reconstruct the model equation.

[0013] In conjunction with the first aspect, further, in step S100, a pressure-temperature mapping function is set. Perform data alignment:

[0014]

[0015] in, The original pressure value. This is the original temperature value. For learnable weight matrix, It is the bias vector;

[0016] Pressure-temperature mapping function Training via adversarial generative networks:

[0017]

[0018] in, For generator, For discriminator, This is molecular dynamics data;

[0019] And generate distribution-aligned data .

[0020] In conjunction with the first aspect, further, in step S200, the formula for calculating the predicted values ​​of the physical property parameters is as follows:

[0021]

[0022] in, For the intermolecular distances calculated in the high-pressure model, This represents the measured intermolecular distance in the low-pressure model. Based on low to medium data Pre-trained neural network models; For the high-voltage adapter module, input molecular-scale data. ; For feature splicing operations;

[0023] High voltage adapter module The calculation formula is:

[0024]

[0025] in, These are molecular-scale eigenvectors. For the gated vector, For layer normalization operators, Activate the Gaussian error linear unit.

[0026] In conjunction with the first aspect, further, between steps S200 and S300, step S201 is also included: uncertainty quantification step:

[0027]

[0028] in, For model variance, For data uncertainty, For the sensitivity of density to pressure, For pressure control accuracy.

[0029] In conjunction with the first aspect, further, in step S300, in-situ data is received in real time via embedded hardware. Calculate the correction factor :

[0030]

[0031] in, To predict the material property vector, These are in-situ detection inversion values. The standard deviation of sensor noise. This is the sensitivity coefficient;

[0032]

[0033]

[0034] in, This is the current measurement value. For the observation matrix, For Kalman gain, For sensor noise covariance, To estimate the error covariance.

[0035] In conjunction with the first aspect, further, in step S400, the source terms of the fluid control equations The corrected formula is:

[0036]

[0037] in, For density field, For velocity vector field, Adjust the weights for time;

[0038] Time-corrected weights Optimization through variational inference:

[0039]

[0040]

[0041] in, This is the timing error vector. It is a multi-layer sensing module. The standard deviation is denoted as .

[0042] In conjunction with the first aspect, further, in step S500, when the average relative error At that time, the reinforcement learning optimizer is triggered:

[0043]

[0044] in, For the reward function, Let the action value function be... For the first Error distribution status, For the first The action of modifying the number of network layers in the item. The regularization coefficient is used.

[0045] reward function Defined as:

[0046]

[0047] in, For the amount of error reduction, To calculate resource increments, These are the weighting coefficients. As the attenuation factor, This represents the resource threshold.

[0048] Secondly, the present invention also discloses a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for constructing a high-pressure fluid model as described above.

[0049] Thirdly, the present invention also discloses a computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the method for constructing the high-pressure fluid model as described above.

[0050] The beneficial effects of this invention are:

[0051] The high-pressure fluid model construction method provided by this invention significantly improves the efficiency and accuracy of fluid model construction through a multi-step process design. First, it employs a multi-scale fluid data acquisition approach, covering molecular scale, experimental scale, and in-situ detection data, ensuring the comprehensiveness and accuracy of the model under different pressure conditions. Second, through a pressure-gradient transfer learning network, data from the low- and medium-pressure domains is effectively transferred to the ultra-high-pressure domain, overcoming the predictive limitations of traditional methods under high-pressure environments.

[0052] Furthermore, by receiving in-situ data in real time using embedded hardware and calculating correction factors, the model can dynamically adapt to changes in fluid state, improving its real-time performance and responsiveness. When the model's average relative error exceeds a set threshold, a reinforcement learning optimizer is triggered to reconstruct the model equations, further enhancing the model's accuracy and reliability.

[0053] In summary, the high-pressure fluid model construction method of the present invention not only solves many problems in the prior art, but also shows significant advantages in the accuracy, real-time performance and adaptability of the fluid model, providing strong technical support for research and application in related fields. Attached Figure Description

[0054] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0055] Figure 1 This is a flowchart of the method for constructing the high-pressure fluid model of the present invention.

[0056] Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0057] Figure 3 This is a schematic diagram of a storage medium provided in an embodiment of the present invention. Detailed Implementation

[0058] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0059] Example 1

[0060] Due to insufficient experimental data (>50 GPa), the extrapolation of traditional state equations is distorted, and offline correction models cannot adapt to real-time changes in high-pressure environments (such as pipeline pressure fluctuations). Therefore, as... Figure 1 As shown, this embodiment discloses a method for constructing a high-pressure fluid model. Through cross-domain knowledge transfer via transfer learning and embedded hardware closed-loop correction, a high-precision, low-latency intelligent modeling system is constructed. It includes the following steps:

[0061] S100. Acquire multi-scale fluid data, including molecular dynamics data, experimental-scale data, and in-situ detection data;

[0062] S200. By using a pressure grading transfer learning network, data from the medium and low pressure domains are transferred to the ultra-high pressure domain, and predicted values ​​of physical property parameters are output.

[0063] S300. Receives in-situ data in real time based on embedded hardware and calculates correction factors;

[0064] S400. Dynamically reconstructing the source terms of the fluid control equations based on correction factors;

[0065] S500. When the average relative error exceeds the threshold, the reinforcement learning optimizer is triggered to reconstruct the model equation.

[0066] In this embodiment, S100-S200 solves the initial modeling accuracy, S300-S400 realizes dynamic calibration during operation, and S500 ensures long-term reliability, forming a complete chain of "perception-decision-execution-optimization".

[0067] Example 2

[0068] Step S100: Acquire multi-scale fluid data, including molecular dynamics data, experimental-scale data, and in-situ detection data, and set the pressure-temperature mapping function. Perform data alignment:

[0069]

[0070] in, The original pressure value is in GPa, representing the static pressure exerted on the fluid. Typical values ​​for ultra-high pressure scenarios are 10-100 GPa. The original temperature value is in K, representing the fluid thermodynamic temperature, with a typical supercritical range of 300-1000 K. The natural logarithm of the pressure is used to compress the numerical span of the high-pressure zone; The empirical transformation, which is the temperature raised to the power of 0.7, eliminates the temperature distribution skewness. For dimension 2 A learnable weight matrix of 2 is used to establish a mapping relationship between the molecular scale (Å level) and the engineering scale (m level); For dimension 2 A bias vector of 1 is used to compensate for systematic errors in the experimental equipment.

[0071] Pressure-temperature mapping function Training via adversarial generative networks:

[0072]

[0073] And generate distribution-aligned data .

[0074] in, As a generator, it will generate molecular dynamics data. Mapping to experimental scale data , ; This is a discriminator in a 3-layer convolutional network, outputting true / false probabilities. This embodiment provides distribution-aligned input for transfer learning and data uncertainty for uncertainty quantification. The basis for estimation.

[0075] Example 3

[0076] Step S200: Using a pressure grading transfer learning network, transfer the data from the low- and medium-pressure domain to the ultra-high-pressure domain, and output the predicted values ​​of physical property parameters. The formula for calculating the predicted values ​​of physical property parameters is as follows:

[0077]

[0078] in, For the intermolecular distances calculated in the high-pressure model, This represents the measured intermolecular distance in the low-pressure model. Based on low to medium data A pre-trained 4-layer ResNet neural network model, trained on experimental data of <10GPa; For the high-voltage adapter module, a gating mechanism prevents negative migration; input data is at the molecular scale. The electron density matrix calculated by DFT; For feature splicing operations;

[0079] High voltage adapter module The calculation formula is:

[0080]

[0081] in, This is a molecular-scale feature vector with dimension d=512; For dimension d A 1-gated vector is used to control the intensity of knowledge transfer. A layer normalization operator is used to stabilize the high-pressure characteristic distribution; It is activated by Gaussian error linear units, which is more suitable for nonlinear high-pressure characteristics.

[0082] Gated vectors Determined by the gradient of features with respect to pressure, a physics-inspired adaptive transfer is achieved, allowing high-pressure sensitive features (such as electron density) to automatically obtain high gating values. .

[0083] The following are experimental data from one of the embodiments (80 GPa metallic hydrogen transport modeling):

[0084]

[0085] Experimental data shows that the predicted conductivity at 80 GPa is:

[0086] No adapter used: 38.7% error (phase transition caused the traditional model to fail);

[0087] Using the adapter of this embodiment: error 3.1% (accurate capture of electron delocalization effect);

[0088] The viscosity prediction error at 10 GPa remains at 0.8%.

[0089] Example 4

[0090] Between steps S200 and S300, step S201 is also included: uncertainty quantification step:

[0091]

[0092] in, For model variance, For data uncertainty, For the sensitivity of density to pressure, For pressure control accuracy.

[0093] In one embodiment of a 55 GPa supercritical water transport system, the following scenario parameters are set:

[0094] Target property: density ;

[0095] Work location: , ;

[0096] Equipment: Diamond anvil synchrotron radiation measurement.

[0097] Calculation of uncertainty components:

[0098] Model variance was sampled using Dropout Monte Carlo sampling (100 forward propagations):

[0099]

[0100] The X-ray diffraction calibration certificate provides the data uncertainty.

[0101] Pressure sensitivity ;

[0102] Pressure control accuracy ;

[0103]

[0104] Overall uncertainty

[0105]

[0106] Predicted value Final output .

[0107] Experimental results prove:

[0108]

[0109] The confidence interval width decreased by 42% (0.12 → 0.07).

[0110] During 20 high-pressure fluctuations, the number of emergency shutdowns caused by density misjudgment decreased from 5 to 1.

[0111] This embodiment will incorporate model randomness ( ), data error ( ), parameter sensitivity ( A single formula is used to calculate the standard deviation of the predicted values.

[0112] By embedding error propagation theory into the real-time fluid model control closed loop, the confidence interval coverage is improved by 22%, directly reducing the risk of high-voltage engineering accidents.

[0113] Example 5

[0114] Step S300: Receive in-situ data in real time using embedded hardware, calculate the correction factor, and receive in-situ data in real time using embedded hardware. Calculate the correction factor The correction factor is used to scale the spacetime terms of the mass conservation equation:

[0115]

[0116] in, To predict the material property vector, the following components are formed. ; These are in-situ detection inversion values, derived from diamond anvil Raman spectroscopy inversion values; This represents the standard deviation of sensor noise, typically 0.03-0.05. This is the sensitivity coefficient. ;

[0117]

[0118]

[0119] in, This is the current measurement value; The observation matrix is ​​usually the identity matrix; Kalman gain; Sensor noise covariance, typical value: ; To estimate the error covariance.

[0120] Example 6

[0121] Step S400: Based on the correction factor, dynamically reconstruct the source terms of the fluid control equations. The corrected formula is:

[0122]

[0123] in, This is a density field, with units of... ; Let be a velocity vector field, with velocity as ; Adjust the weights for time and update them through variational inference;

[0124] Time-corrected weights Optimization through variational inference:

[0125]

[0126]

[0127] in, This is the timing error vector; It is a multilayer perceptual module with 2 fully connected layers and 64 hidden neurons; Standard deviation, constraints .

[0128] Example 7

[0129] Further, in step S500: when the average relative error exceeds a threshold, the reinforcement learning optimizer is triggered to reconstruct the model equation. At that time, the reinforcement learning optimizer is triggered:

[0130]

[0131] in, For the reward function; The action value function takes the input state as its input. ,action ; For the first Error distribution status, For the first The action of modifying the number of network layers in the item. This is the regularization coefficient, with a default value of 0. To prevent overfitting;

[0132] reward function Defined as:

[0133]

[0134] in, The amount of error reduction is calculated using the following formula: ; To calculate the incremental increase in computing resources, the equivalent CPU core time; These are the weighting coefficients. As the attenuation factor, Set as a resource threshold .

[0135] Example 8

[0136] Please refer to Figure 2 This document illustrates a schematic diagram of the structure of a computer device provided by some embodiments of this application. The electronic device 10 includes: a processor 100, a memory 101, a bus 102, and a communication interface 103. The processor 100, the communication interface 103, and the memory 101 are connected via the bus 102. The memory 101 stores a computer program that can run on the processor 100. The processor 100 is used to operate according to the instructions to execute the steps of the method described in any one of embodiments 1-7.

[0137] The memory 101 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 103 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.

[0138] Bus 102 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. Memory 101 is used to store programs. After receiving an execution instruction, processor 100 executes the program. The high-pressure fluid model construction method disclosed in any of the foregoing embodiments of this application can be applied to processor 100, or implemented by processor 100.

[0139] Processor 100 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 100 or by instructions in software form. Processor 100 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), an Off-the-shelf Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 101. The processor 100 reads the information in memory 101 and, in conjunction with its hardware, completes the steps of the above method.

[0140] Example 9

[0141] This embodiment also provides a computer-readable storage medium corresponding to the high-pressure fluid model construction method provided in the foregoing embodiments. Please refer to [link / reference]. Figure 3 The computer-readable storage medium shown is an optical disc, on which a computer program (i.e., program product 20) is stored. When the computer program is run by a processor, it executes the power scheduling method provided in any of the foregoing embodiments.

[0142] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0143] The computer-readable storage medium provided in the above embodiments of this application and the method for constructing a high-pressure fluid model provided in the embodiments of this application are based on the same application concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

[0144] It should be noted that in the above text, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0145] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0146] The embodiments of this application have been described above with reference to the accompanying drawings. These are merely specific implementations of this application, but this application is not limited to the specific implementations described above. The specific implementations described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for constructing a high-pressure fluid model, characterized in that, Includes the following steps: S100. Acquire multi-scale fluid data, including molecular dynamics data, experimental-scale data, and in-situ detection data; S200. By using a pressure grading transfer learning network, data from the medium and low pressure domains are transferred to the ultra-high pressure domain, and predicted values ​​of physical property parameters are output. S300. Receives in-situ data in real time based on embedded hardware and calculates correction factors. ; S400. Dynamically reconstructing the source terms of the fluid control equations based on correction factors; S500. When the average relative error exceeds the threshold, the reinforcement learning optimizer is triggered to reconstruct the model equation; In step S100, the pressure-temperature mapping function is set. Perform data alignment: in, The original pressure value. This is the original temperature value; It is a learnable weight matrix used to establish the mapping relationship between the molecular scale and the engineering scale; This serves as a bias vector to compensate for systematic errors in the experimental equipment. Pressure-temperature mapping function Training via adversarial generative networks: in, For generator, For discriminator, For molecular dynamics data, For experimental data; And generate distribution-aligned data It provides distribution-aligned input for transfer learning and data uncertainty for uncertainty quantification. The basis for estimation; In step S200, the formula for calculating the predicted values ​​of physical property parameters is as follows: in, For the intermolecular distances calculated in the high-pressure model, This represents the measured intermolecular distance in the low-pressure model. Based on low to medium data Pre-trained neural network models; For the high-voltage adapter module, input molecular-scale data. ; For feature splicing operations; High voltage adapter module The calculation formula is: in, These are molecular-scale feature vectors; This serves as a gating vector to control the intensity of knowledge transfer. For layer normalization operators, Activate the Gaussian error linear unit; In step S400, the source terms of the fluid control equations The corrected formula is: in, For density field, For velocity vector field, Adjust the weights for time.

2. The method for constructing a high-pressure fluid model according to claim 1, characterized in that, Between steps S200 and S300, step S201 is also included: uncertainty quantification step: in, For model variance, For data uncertainty, For the sensitivity of density to pressure, For pressure control accuracy.

3. The method for constructing a high-pressure fluid model according to claim 1, characterized in that, In step S300, in-situ data is received in real time via embedded hardware. Calculate the correction factor : in, To predict the material property vector, These are in-situ detection inversion values. The standard deviation of sensor noise. This is the sensitivity coefficient; in, This is the current measurement value. For the observation matrix, For Kalman gain, For sensor noise covariance, To estimate the error covariance.

4. The method for constructing a high-pressure fluid model according to claim 1, characterized in that, Time-corrected weights Optimization through variational inference: in, This is the timing error vector. It is a multi-layer sensing module. The standard deviation is denoted as .

5. The method for constructing a high-pressure fluid model according to claim 1, characterized in that, In step S500, when the average relative error At that time, the reinforcement learning optimizer is triggered: in, For the reward function, Let the action value function be... For the first Error distribution status, For the first The action of modifying the number of network layers in the item. The regularization coefficient is used. reward function Defined as: in, For the amount of error reduction, To calculate resource increments, These are the weighting coefficients. As the attenuation factor, This represents the resource threshold.

6. A computer device, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method as described in any one of claims 1-5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by a processor to implement the method for constructing a high-pressure fluid model as described in any one of claims 1-5.

Images