A Multiphysics Intelligent Computation Method for Rectangular Flow Channels

Abstract

This invention belongs to the field of nuclear reactor engineering technology, specifically relating to a multiphysics intelligent calculation method for rectangular flow channels. It includes the following steps: Step 1: Determine the input-output matching data pairs related to the two-phase flow and heat transfer process in the rectangular channel; Step 2: Obtain the input-output matching data pairs related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm, and establish a data pool; Step 3: Construct a loss function, and after training with machine learning, perform supervised regression learning on the output matching data to obtain an intelligent calculation model for the two-phase parameters of the rectangular channel; Step 4: Calculate the input parameters related to the two-phase flow and heat transfer process in the rectangular channel to obtain the corresponding output parameters, realizing rapid parameter calculation in the two-phase flow and heat transfer process of the rectangular channel. The beneficial effects are: it can well approximate the input-output characteristics and effectively ensure the calculation accuracy of the intelligent calculation model.

Description

Technical Field

[0001] This invention belongs to the field of nuclear reactor engineering technology, specifically relating to a multiphysics intelligent calculation method for rectangular flow channels. Background Technology

[0002] In various industrial equipment of nuclear power systems, gas-liquid two-phase flow phenomena are widespread. Studying these phenomena helps improve the safety and operational reliability of the system. As one of the most complex two-phase flow phenomena, gas-liquid two-phase flow processes are accompanied by heat and mass transfer and phase changes between the two phases, making the study of gas-liquid two-phase flow extremely challenging. Currently, a large amount of research has been conducted on the multiphysics characteristics of flow fields and temperature fields in rectangular channels. Most of these studies are based on experimental methods to fundamentally study the development and changes of flow patterns, phase distribution characteristics, pressure drop changes, and temperature changes within the channels. However, experimental research is time-consuming, labor-intensive, and costly, and also places high demands on the use of test benches and measurement personnel. On the other hand, in terms of numerical simulation, the two-phase flow and heat transfer processes in rectangular channels are mainly modeled and analyzed in detail using high-precision three-dimensional CFD fluid software or multi-channel thermal-hydraulic software for reactor cores, based on experimental data. When the channel structure is extended to the whole reactor core level, the consumption of computing resources is high, the calculation speed is slow, and the calculation efficiency is not high. Even on ultra-large-scale high-performance servers, multiphysics calculations of rectangular channels in reactor cores require a significant amount of time. Therefore, it is necessary to study a multiphysics intelligent prediction method for rectangular flow channels that consumes less computational resources and can achieve fast calculation. Summary of the Invention

[0003] The purpose of this invention is to provide a multiphysics intelligent calculation method for rectangular flow channels, which consumes very few computing resources and can realize rapid prediction and analysis of parameters in the two-phase flow and heat transfer process in the rectangular channel of the reactor core.

[0004] The technical solution of the present invention is as follows: A multiphysics intelligent calculation method for rectangular flow channels, comprising the following steps:

[0005] Step 1: Analyze the key influencing factors of multiphysics characteristics in the two-phase flow and heat transfer process in a rectangular channel, and determine the input-output matching data pairs related to the two-phase flow and heat transfer process in a rectangular channel;

[0006] Step 2: Calculate using high-precision 3D CFD fluid software or core multi-channel thermal-hydraulic software, or obtain input-output matching data pairs related to the rectangular channel two-phase flow and heat transfer process required by the deep learning algorithm from the rectangular channel two-phase flow and heat transfer test data, and establish a data pool;

[0007] Step 3: Based on the input-output matching data pairs related to the two-phase flow and heat transfer process in the rectangular channel, construct a loss function, and after training with machine learning, perform supervised regression learning on the output matching data to obtain an intelligent calculation model for the two-phase parameters of the rectangular channel.

[0008] Step 4: Calculate the input parameters related to the two-phase flow and heat transfer process in the rectangular channel using the intelligent calculation model for two-phase parameters in the rectangular channel, and obtain the corresponding output parameters to achieve rapid calculation of parameters in the two-phase flow and heat transfer process in the rectangular channel.

[0009] Step 1 specifically includes the following:

[0010] Step 10: By analyzing the key influencing factors of the multiphysics characteristics in the two-phase flow and heat transfer process in the rectangular channel, select the input parameters related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm;

[0011] Step 11: By analyzing the multiphysics characteristics of the two-phase flow and heat transfer process in the rectangular channel, select the output parameters related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm;

[0012] Step 12: Based on the multiphysics calculation requirements, determine the input-output matching data pairs required by the deep learning algorithm that are related to the two-phase flow and heat transfer process in the rectangular channel.

[0013] The deep learning algorithms mentioned in step 10 include Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM).

[0014] The input parameters related to the two-phase flow and heat transfer process in the rectangular channel mentioned in step 10 include the gas-phase reduced velocity j. g Liquid phase reduced velocity j f The channel cross-sectional dimensions include radial height H and width G, channel inlet and outlet fluid pressure P, channel inlet and outlet fluid temperature T, channel wall roughness ε, channel inlet fluid flow rate W, channel heating power Q or heat flux density q”, fluid physical properties, and one or more combinations of motion degree of freedom parameters.

[0015] The output parameters related to the two-phase flow and heat transfer process in the rectangular channel mentioned in step 11 include those at measurement points z along different length directions. i Position of measuring point x in the width direction i and the position of the measuring point L in the axial direction i The two-phase interface concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, bubble Souter mean diameter, and one or more combinations of two-phase temperature, pressure, flow rate, and velocity.

[0016] The data set mentioned in step 12 includes input parameters related to the two-phase flow and heat transfer process in the rectangular channel, including the gas-phase reduced velocity j. g Liquid phase reduced velocity j f The channel cross-sectional dimensions include radial height H and width G, channel inlet and outlet fluid pressure P, channel inlet and outlet fluid temperature T, channel wall roughness ε, channel inlet fluid flow rate W, channel heating power Q or heat flux density q”, fluid physical property parameters, and one or more combinations of motion degree of freedom parameters, as well as the corresponding output parameters, two-phase interface concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, bubble Souter mean diameter parameters, and one or more combinations of two-phase temperature, pressure, flow rate, and velocity parameters.

[0017] Step 2 includes the following steps:

[0018] Step 20: Divide the rectangular channel into multiple spatial grids in the radial and axial directions, respectively;

[0019] Step 21: Calculate the initial multiphysics state parameters of the two-phase flow and heat transfer process for each spatial grid using high-precision 3D CFD fluid software or core multi-channel thermal-hydraulic software, or measure the corresponding parameters through experimental means to obtain the two-phase parameters characterizing the multiphysics characteristics of the two-phase flow and heat transfer process, including interface parameters such as two-phase interface concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, and bubble Sotter mean diameter, as well as phase parameters such as two-phase temperature, pressure, flow rate, and velocity.

[0020] Step 22: Select the phase parameters of the two-phase flow and heat transfer process in the rectangular channel of the spatial grid where the test measurement point is located, establish the input-output matching data pairs related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm, and form the data pool required by the multiphysics field.

[0021] One expression of the loss function in step 3 is represented in the form of mean squared error, as shown below:

[0022]

[0023] Step 3 involves building a deep learning neural network structure based on input-output matching data pairs related to the two-phase flow and heat transfer process in a rectangular channel, constructing loss functions corresponding to different physical fields, and performing supervised regression learning on the output matching data after training through machine learning. Based on the K-fold cross-validation method, learning stops when the error of the calculation result is less than a set value; otherwise, learning continues to relearn, thereby obtaining an intelligent calculation model for the two-phase parameters of the rectangular channel.

[0024] The deep learning neural network mentioned in step 3 includes one or more of the following structures: fully connected neural network, long short-term memory neural network, or recurrent neural network. The combination is selected according to the multiphysics calculation requirements.

[0025] In step 3, the input-output matching data related to the two-phase flow and heat transfer process in the rectangular channel are processed to eliminate the influence of different order of magnitude of phase parameters.

[0026] In step 3, the neuron connection weights and biases of the neural network are used as adjustable optimization parameters for deep machine learning. This is used to successively adjust the loss function of partial spatial sample data until the loss function of all spatial sample data has been adjusted.

[0027] The beneficial effects of this invention are as follows:

[0028] 1) The computational accuracy of the deep learning-based two-phase parameter intelligent calculation model for rectangular flow channels established by this invention can closely approximate the input-output characteristics, effectively ensuring the computational accuracy of the intelligent calculation model.

[0029] 2) The intelligent computing model obtained by this invention has good generalization ability on the measured data. When the deep learning algorithm is consistent, the intelligent computing model can be quickly corrected and optimized by relearning a large amount of measured data.

[0030] 3) The intelligent computing model obtained by using this invention consumes relatively few computing resources and can, to a certain extent, replace two-phase flow test measurement or simulation methods, thereby reducing test or simulation costs. Attached Figure Description

[0031] Figure 1 Here is a flowchart of a multiphysics intelligent calculation method for a rectangular flow channel provided by the present invention;

[0032] Figure 2 This is a top view of a rectangular flow channel;

[0033] Figure 3 This is a schematic diagram of a neural network structure. Detailed Implementation

[0034] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] like Figure 1 As shown, a multiphysics intelligent computing method for rectangular flow channels includes the following steps:

[0036] Step 1: Analyze the key influencing factors of multiphysics characteristics in the two-phase flow and heat transfer process in a rectangular channel, and determine the input-output matching data pairs related to the two-phase flow and heat transfer process in a rectangular channel;

[0037] Step 1 specifically includes the following steps:

[0038] Step 10: By analyzing the key influencing factors of the multiphysics characteristics in the two-phase flow and heat transfer process in the rectangular channel, select the input parameters related to the two-phase flow and heat transfer process in the rectangular channel required by deep learning algorithms (such as convolutional neural network CNN, recurrent neural network RNN, long short-term memory neural network LSTM, etc.).

[0039] The input parameters related to the two-phase flow and heat transfer process in the rectangular channel include the gas-phase reduced velocity j. g Liquid phase reduced velocity j f One or more combinations of parameters, such as channel cross-sectional dimensions (radial height H and width G), channel inlet and outlet fluid pressure P, channel inlet and outlet fluid temperature T, channel wall roughness ε, channel inlet fluid flow rate W, channel heating power Q or heat flux density q”, fluid physical property parameters (such as fluid density ρ, etc.), and motion degree of freedom parameters (such as swing angle α, position, etc.).

[0040] Step 11: By analyzing the multi-physics characteristics of the two-phase flow and heat transfer process in the rectangular channel (for different physical fields such as flow field, temperature field, and pressure field of the rectangular channel, determine the physical factors related to different physical fields, so as to further analyze the parameters of the corresponding physical factors), select the output parameters related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm.

[0041] The output parameters related to the two-phase flow and heat transfer process in the rectangular channel include those at measurement points z along different length directions. i Position of measuring point x in the width direction i and the position of the measuring point in the axial direction L i Interface parameters such as (e.g., starting from the channel inlet position), including the two-phase interface concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, bubble Souter mean diameter, etc., and one or more combinations of phase parameters such as two-phase temperature, pressure, flow rate, velocity, etc.

[0042] Step 12: Based on the multiphysics calculation requirements, determine the input-output matching data pairs required by the deep learning algorithm related to the two-phase flow and heat transfer process in the rectangular channel. These data pairs include input parameters related to the two-phase flow and heat transfer process in the rectangular channel (including the gas-phase reduced velocity j). g Liquid phase reduced velocity j fThe parameters include one or more combinations of the following: channel cross-sectional dimensions (radial height H and width G), channel inlet and outlet fluid pressure P, channel inlet and outlet fluid temperature T, channel wall roughness ε, channel inlet fluid flow rate W, channel heating power Q or heat flux density q", fluid physical properties (such as fluid density ρ), and motion degrees of freedom parameters (such as swing angle α, position, etc.). Also, what are the corresponding output parameters (measurement point positions z at different length directions)? i Position of measuring point x in the width direction i and the position of the measuring point in the axial direction L i (Taking the channel inlet position as the starting point) What parameters are included in one or more combinations of two-phase interface parameters such as concentration at the two-phase interface, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, and bubble Souter mean diameter?

[0043] Step 2: Calculate using high-precision 3D CFD fluid software or core multi-channel thermal-hydraulic software, or obtain input-output matching data pairs related to the rectangular channel two-phase flow and heat transfer process required by the deep learning algorithm from the rectangular channel two-phase flow and heat transfer test data (different combinations of the above input parameters and output parameters are determined according to different physical fields), and establish a data pool;

[0044] Step 2 specifically includes the following steps:

[0045] Step 20: Divide the rectangular channel into multiple spatial grids in the radial and axial directions, respectively;

[0046] Step 21: Calculate the initial multiphysics state parameters of the two-phase flow and heat transfer process for each spatial grid using high-precision 3D CFD fluid software or core multichannel thermal-hydraulic software (such as CTF), or measure the corresponding parameters through experimental means (such as obtaining flow rate from a flow meter; obtaining pressure from a differential pressure gauge; obtaining interface concentration and velocity from a probe; obtaining the void fraction from a cavitation meter; obtaining temperature from a thermometer, etc.) to obtain two-phase parameters characterizing the multiphysics characteristics of the two-phase flow and heat transfer process, including interface parameters such as two-phase interface concentration, void fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, and bubble Sauter mean diameter, as well as phase parameters such as two-phase temperature, pressure, flow rate, and velocity.

[0047] Step 22: Select phase parameters (interfacial concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord distribution, cross-sectional gas phase distribution, bubble Sauter mean diameter, etc., and phase parameters such as two-phase temperature, pressure, flow rate, and velocity) in the rectangular channel two-phase flow and heat transfer process of the spatial grid where the test measurement point is located. Establish input-output matching data pairs related to the rectangular channel two-phase flow and heat transfer process required by the deep learning algorithm (different combinations of the above input and output parameters are determined according to different physical fields, i.e., input combinations correspond to output combinations), forming the data pool required by the multiphysics field.

[0048] Step 3: Based on deep learning algorithms and input-output matching data pairs related to the two-phase flow and heat transfer process in rectangular channels (different combinations of the above input and output parameters are determined according to different physical fields, i.e., input combinations correspond to output combinations), construct a deep learning neural network structure (e.g., ...). Figure 3 (As shown). Construct the corresponding loss function, and after training with machine learning, perform supervised regression learning on the output matching data to obtain an intelligent calculation model for the two-phase parameters of the rectangular channel;

[0049] Loss functions can be expressed in various forms, one of which can be represented by the mean squared error, as shown below:

[0050]

[0051] In the formula, y i This is a predicted value; The true value is N; the number of data points is N.

[0052] Step 3 specifically includes the following steps: Based on the input-output matching data pairs related to the two-phase flow and heat transfer process in a rectangular channel using deep learning algorithms, a deep learning neural network structure is built, loss functions corresponding to different physical fields are constructed, and supervised regression learning is performed on the output matching data after training through machine learning. Based on the K-fold cross-validation method, when the error of the calculation result is less than the set value, the learning stops; otherwise, the learning continues to be re-learned, thereby obtaining an intelligent calculation model for the two-phase parameters of the rectangular channel.

[0053] In one possible implementation, in step 3, the deep learning neural network includes one or more structures of neural networks such as fully connected neural networks, long short-term memory neural networks, or recurrent neural networks, and the combination is selected according to the multiphysics calculation requirements.

[0054] In one possible implementation, in step 3, the input-output matching data related to the two-phase flow and heat transfer process in the rectangular channel are processed to be dimensionless to eliminate the influence of different magnitudes of phase parameters.

[0055] In one possible implementation, in step 3, for multiphysics calculation analysis, parameters of different categories or changing trends can be classified. This involves training and learning the changing trends on the same order of magnitude in the same neural network, with the classifier determined by the user based on experience.

[0056] The process includes:

[0057] 1. Confirm the trend or type of parameter change.

[0058] 2. Place parameters with similar magnitudes of change or parameters of the same type in the same neural network. Construct different loss function terms, and during deep learning using the loss function, continuously and adaptively adjust the weights of each part of the loss function until the loss function approaches its minimum value.

[0059] Loss functions can be expressed in various forms, one of which can be represented by the mean squared error, as shown below:

[0060]

[0061] In one possible implementation, in step 3, different loss function terms can be trained in parallel on different GPUs.

[0062] In one possible implementation, in step 3, the neuron connection weights and biases of the neural network are used as adjustable optimization parameters for deep machine learning to successively adjust the loss function of partial spatial sample data until the loss function of all spatial sample data has been adjusted.

[0063] Step 4: Calculate the input parameters related to the two-phase flow and heat transfer process in the rectangular channel using the intelligent calculation model for two-phase parameters in the rectangular channel, and obtain the corresponding output parameters to achieve rapid calculation of parameters in the two-phase flow and heat transfer process in the rectangular channel.

[0064] Example:

[0065] This embodiment addresses a smooth rectangular channel under static conditions with a given geometric structure, such as... Figure 2 As shown, under the conditions of normal temperature and pressure, known inlet flow rate, fixed radial measuring point position, and 3 measuring point positions in the axial direction, a two-phase parameter intelligent calculation model for a rectangular flow channel under a certain specific working condition is established.

[0066] A multiphysics intelligent computing method for rectangular flow channels includes the following steps:

[0067] Step 1: By analyzing the key influencing factors of the multiphysics characteristics in the two-phase flow and heat transfer process in the rectangular channel, select the input parameters related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm; by analyzing the multiphysics characteristics in the two-phase flow and heat transfer process in the rectangular channel, select the output parameters related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm; based on this, and according to the multiphysics calculation requirements, determine the input-output matching data pairs related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm.

[0068] In this embodiment, the key influencing factors of the flow characteristics during the two-phase flow and heat transfer process in the rectangular channel include, but are not limited to, the gas phase reduced velocity j. g Liquid phase reduced velocity j f The initial state parameters of the flow process and the geometric information of the channel include the fluid pressure P at the inlet and outlet of the channel, the fluid temperature T at the inlet and outlet of the channel, the fluid flow rate W at the inlet of the channel, the heating power Q or heat flux density q” of the channel, the wall roughness ε of the channel, and fluid physical properties (such as fluid density ρ).

[0069] Select the computational input parameters required by the deep learning algorithm related to the two-phase flow and heat transfer process in the rectangular channel, including but not limited to the gas-phase reduced velocity j. g Liquid phase reduced velocity j f One or more combinations of parameters.

[0070] Select the computational output parameters required by the deep learning algorithm related to the two-phase flow and heat transfer process in the rectangular channel, including but not limited to the measurement point positions z in different length directions. i Position of measuring point x in the width direction i and the position of the measuring point in the axial direction L i Interface parameters such as (e.g., starting from the channel inlet position), including the two-phase interface concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, bubble Souter mean diameter, etc., and one or more combinations of phase parameters such as two-phase temperature, pressure, flow rate, velocity, etc.

[0071] Step 2: Using high-precision 3D CFD fluid dynamics software or multi-channel core thermal-hydraulic software, or from rectangular channel two-phase flow and heat transfer experimental data, obtain input-output matching data pairs related to the rectangular channel two-phase flow and heat transfer process required by the deep learning algorithm, and establish a data pool. This specifically includes the following steps:

[0072] The rectangular channel is divided into multiple spatial grids in the radial and axial directions. Initial multiphysics state parameters of the two-phase flow and heat transfer process in each spatial grid are calculated using high-precision 3D CFD fluid dynamics software or multi-channel thermal-hydraulic software for the reactor core. Alternatively, corresponding parameters are measured experimentally (e.g., flow rate from a flow meter; pressure from a differential pressure gauge; interface concentration and velocity from a probe; void fraction from a cavitation meter; temperature from a thermometer), to obtain two-phase parameters characterizing the multiphysics properties of the two-phase flow and heat transfer process. Phase parameters in the two-phase flow and heat transfer process of the rectangular channel within the spatial grid where the experimental measurement points are located are selected. Input-output matching data pairs related to the two-phase flow and heat transfer process in the rectangular channel, required by the deep learning algorithm, are established to form the data pool needed for the multiphysics process.

[0073] In this embodiment, different gas phase conversion velocities j are selected. g Liquid phase reduced velocity j f The data sets, including the channel axial measurement point lengths L1, L2, and L3, are used to form a data set under different gas-liquid phase reduced velocity conditions. Through high-precision three-dimensional CFD fluid software or core multi-channel thermal-hydraulic software, or through two-phase flow and heat transfer experiments, parameters in the two-phase flow and heat transfer process in the rectangular channel are obtained. These parameters include interface parameters such as two-phase interface concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, and bubble Souter mean diameter, as well as phase parameters such as two-phase temperature, pressure, flow rate, and velocity. This allows for the acquisition of input-output matching data pairs related to the two-phase and heat transfer flow process in the rectangular channel required by deep learning algorithms.

[0074] Step 3: Based on the input-output matching data pairs related to the two-phase flow and heat transfer process in a rectangular channel, a deep learning neural network structure is built, a corresponding loss function is constructed, and supervised regression learning is performed on the output matching data after machine learning training to obtain an intelligent calculation model of the two-phase parameters in the rectangular channel.

[0075] The establishment of data learning and intelligent computing models can be determined based on actual multiphysics computing needs and application methods. For example, the gas phase reduced velocity j can be selected. g Liquid phase reduced velocity j f The combination of parameters is used as the calculation input parameters; for pressure field or velocity field calculations, different length direction measuring point positions z are selected. i Position of measuring point x in the width direction i and the position of the measuring point L in the axial direction i The combination of pressure and velocity (starting from the inlet of the channel) is used as the calculation output parameter; a sufficient number of feature data are extracted from the data pool as the input-output matching data pairs related to the two-phase flow and heat transfer process in the rectangular channel required for the learning of the two-phase parameter intelligent calculation model of the rectangular channel.

[0076] Based on this, a fully connected neural network can be used. Hyperparameters such as the number of layers, neurons, network weights, and biases can be set, and the activation function for hidden layer nodes can be set to sigmoid, etc. A deep neural network structure can be designed and established. Supervised regression learning can be performed by constructing loss functions for stress and speed. When the loss function decreases to a stable minimum value, the neural network intelligent computing model can be obtained. Based on K-fold cross-validation, learning stops when the calculation error is less than a set value; otherwise, learning continues until the intelligent computing model can be solidified.

[0077] Step 4: Combining new physical field calculation requirements, such as temperature field, the input parameters related to the two-phase flow and heat transfer process in the rectangular channel are calculated using the intelligent calculation model of two-phase parameters in the rectangular channel to obtain the corresponding output parameters, thereby realizing the rapid calculation of parameters in the two-phase flow and heat transfer process in the rectangular channel.

[0078] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A multiphysics intelligent calculation method for rectangular flow channels, characterized in that, Includes the following steps: Step 1: Analyze the key influencing factors of multiphysics characteristics in the two-phase flow and heat transfer process in a rectangular channel, and determine the input-output matching data pairs related to the two-phase flow and heat transfer process in a rectangular channel; Step 2: Calculate using high-precision 3D CFD fluid software or core multi-channel thermal-hydraulic software, or obtain input-output matching data pairs related to the rectangular channel two-phase flow and heat transfer process required by the deep learning algorithm from the rectangular channel two-phase flow and heat transfer test data, and establish a data pool; Step 3: Based on the input-output matching data pairs related to the two-phase flow and heat transfer process in the rectangular channel, construct a loss function, and after training with machine learning, perform supervised regression learning on the output matching data to obtain an intelligent calculation model for the two-phase parameters of the rectangular channel. Step 4: Calculate the input parameters related to the two-phase flow and heat transfer process in the rectangular channel using the intelligent calculation model for two-phase parameters in the rectangular channel, and obtain the corresponding output parameters to achieve rapid calculation of parameters in the two-phase flow and heat transfer process in the rectangular channel.

2. The multiphysics intelligent calculation method for rectangular flow channels as described in claim 1, characterized in that, Step 1 specifically includes the following: Step 10: By analyzing the key influencing factors of the multiphysics characteristics in the two-phase flow and heat transfer process in the rectangular channel, select the input parameters related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm; Step 11: By analyzing the multiphysics characteristics of the two-phase flow and heat transfer process in the rectangular channel, select the output parameters related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm; Step 12: Based on the multiphysics calculation requirements, determine the input-output matching data pairs required by the deep learning algorithm that are related to the two-phase flow and heat transfer process in the rectangular channel.

3. The intelligent multiphysics calculation method for rectangular flow channels as described in claim 2, characterized in that: The deep learning algorithms mentioned in step 10 include Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM).

4. The multiphysics intelligent calculation method for rectangular flow channels as described in claim 2, characterized in that: The input parameters related to the two-phase flow and heat transfer process in the rectangular channel mentioned in step 10 include the gas-phase reduced velocity j. g Liquid phase reduced velocity j f The channel cross-sectional dimensions include radial height H and width G, channel inlet and outlet fluid pressure P, channel inlet and outlet fluid temperature T, channel wall roughness ε, channel inlet fluid flow rate W, channel heating power Q or heat flux density q”, fluid physical properties, and one or more combinations of motion degree of freedom parameters.

5. The multiphysics intelligent calculation method for rectangular flow channels as described in claim 2, characterized in that: The output parameters related to the two-phase flow and heat transfer process in the rectangular channel mentioned in step 11 include those at measurement points z along different length directions. i Position of measuring point x in the width direction i and the position of the measuring point L in the axial direction i The two-phase interface concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, bubble Souter mean diameter, and one or more combinations of two-phase temperature, pressure, flow rate, and velocity.

6. The intelligent multiphysics calculation method for rectangular flow channels as described in claim 2, characterized in that: The data set mentioned in step 12 includes input parameters related to the two-phase flow and heat transfer process in the rectangular channel, including the gas-phase reduced velocity j. g Liquid phase reduced velocity j f The channel cross-sectional dimensions include radial height H and width G, channel inlet and outlet fluid pressure P, channel inlet and outlet fluid temperature T, channel wall roughness ε, channel inlet fluid flow rate W, channel heating power Q or heat flux density q″, fluid physical property parameters, and one or more combinations of motion degree of freedom parameters, as well as the corresponding output parameters, including two-phase interface concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, bubble Souter mean diameter parameters, and one or more combinations of two-phase temperature, pressure, flow rate, and velocity parameters.

7. The intelligent multiphysics calculation method for rectangular flow channels as described in claim 1, characterized in that, Step 2 includes the following steps: Step 20: Divide the rectangular channel into multiple spatial grids in the radial and axial directions, respectively; Step 21: Calculate the initial multiphysics state parameters of the two-phase flow and heat transfer process for each spatial grid using high-precision 3D CFD fluid software or core multi-channel thermal-hydraulic software, or measure the corresponding parameters through experimental means to obtain the two-phase parameters characterizing the multiphysics characteristics of the two-phase flow and heat transfer process, including interface parameters such as two-phase interface concentration, cavitation fraction, gas phase velocity, gas phase frequency, gas phase chord length distribution, cross-sectional gas phase distribution, and bubble Sotter mean diameter, as well as phase parameters such as two-phase temperature, pressure, flow rate, and velocity. Step 22: Select the phase parameters of the two-phase flow and heat transfer process in the rectangular channel of the spatial grid where the test measurement point is located, establish the input-output matching data pairs related to the two-phase flow and heat transfer process in the rectangular channel required by the deep learning algorithm, and form the data pool required by the multiphysics field.

8. The intelligent multiphysics calculation method for rectangular flow channels as described in claim 1, characterized in that: One expression of the loss function in step 3 is represented in the form of mean squared error, as shown below:

9. The intelligent multiphysics calculation method for rectangular flow channels as described in claim 1, characterized in that: Step 3 involves building a deep learning neural network structure based on input-output matching data pairs related to the two-phase flow and heat transfer process in a rectangular channel, constructing loss functions corresponding to different physical fields, and performing supervised regression learning on the output matching data after training through machine learning. Based on the K-fold cross-validation method, learning stops when the error of the calculation result is less than a set value; otherwise, learning continues to relearn, thereby obtaining an intelligent calculation model for the two-phase parameters of the rectangular channel.

10. The multiphysics intelligent calculation method for rectangular flow channels as described in claim 9, characterized in that: The deep learning neural network mentioned in step 3 includes one or more of the following structures: fully connected neural network, long short-term memory neural network, or recurrent neural network. The combination is selected according to the multiphysics calculation requirements.

11. The multiphysics intelligent calculation method for rectangular flow channels as described in claim 9, characterized in that: In step 3, the input-output matching data related to the two-phase flow and heat transfer process in the rectangular channel are processed to eliminate the influence of different order of magnitude of phase parameters.

12. The multiphysics intelligent calculation method for rectangular flow channels as described in claim 9, characterized in that: In step 3, the neuron connection weights and biases of the neural network are used as adjustable optimization parameters for deep machine learning. This is used to successively adjust the loss function of partial spatial sample data until the loss function of all spatial sample data has been adjusted.

Images