A machine learning based method, apparatus, and medium for predicting channel flow heat transfer

Abstract

The application discloses a channel flow heat transfer prediction method and device based on machine learning and a medium, relates to the technical field of parameter prediction, generates supplementary data based on an existing flow heat transfer formula, effectively covers the working conditions with few experimental data, determines key dimensionless parameters affecting flow resistance and heat transfer coefficients, enables the model to focus on the most critical factors affecting the flow heat transfer relationship, improves the model's ability to grasp the nature of the flow heat transfer phenomenon, and can more accurately predict the flow heat transfer coefficient and friction resistance coefficient, effectively captures the complex nonlinear relationship between the input parameters and the flow heat transfer coefficient or the friction resistance coefficient by constructing a neural network model, and uses a sparse regression method to fit the experimental data, the relational expression prediction data and a large amount of prediction data generated by the neural network model into an explicit mathematical expression, so that the problem of narrow prediction range and poor universality of the existing relational expression can be solved, and the prediction accuracy of the channel flow heat transfer relationship is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of parameter prediction technology, and specifically to a method, device, and medium for predicting channel flow heat transfer based on machine learning. Background Technology

[0002] Flow heat transfer is widely used in many industrial fields such as chemical, nuclear energy, petroleum, and environmental protection. In many industrial processes, such as air conditioning, refrigerators, boilers, heat exchangers and nuclear power plants, efficient heat transfer technology is required to improve energy utilization efficiency. Flow heat transfer relationships are needed in industrial design to realize the design and process optimization of heat exchange equipment, improve the efficiency and quality of industrial production, and at the same time achieve energy conservation and environmental protection.

[0003] Therefore, to improve heat exchange efficiency in industrial equipment design, it is necessary to predict the flow resistance and heat transfer characteristics of the device. However, there are numerous existing flow and heat transfer equations, each with a limited scope of application, which imposes significant limitations in practice. Furthermore, the high-dimensional parameter space makes it difficult to design empirical formulas, resulting in decreased prediction accuracy and failing to meet the needs of industrial applications. Traditional prediction methods face the problem of large amounts of data and difficulty in fitting complex nonlinear relationships. Therefore, it is necessary to utilize artificial intelligence methods to obtain a unified equation applicable to different heat exchange channels and operating conditions. Summary of the Invention

[0004] The technical problem this invention aims to solve is that existing flow heat transfer relationships have limited applicability and significant limitations in application, leading to decreased prediction accuracy and difficulty in meeting industrial application requirements. The goal is to provide a machine learning-based method, device, and medium for predicting channel flow heat transfer. This involves collecting experimental data and empirical relationships to generate supplementary data to cover data-scarce operating conditions. A deep neural network is constructed, taking operating parameters as input and outputting predicted values ​​of the flow heat transfer coefficient or friction resistance coefficient, and training it using loss functions and regularization techniques. Finally, using experimental data, calculated data from empirical relationships, and a large amount of prediction data generated by the neural network, a sparse regression method is used to fit the explicit mathematical expression, obtaining a unified relationship for optimizing the prediction of flow heat transfer coefficients or friction resistance coefficients for heat exchange channels of different sizes across a wide range of operating conditions, thereby improving the generalization ability and prediction accuracy of the neural network prediction model.

[0005] This invention is achieved through the following technical solution:

[0006] The first aspect of this invention provides a channel flow heat transfer prediction method based on machine learning, comprising the following specific steps:

[0007] Collect experimental data on fluid flow and heat transfer in heat exchange channels of different sizes and structures, and collect empirical relationships between flow resistance and heat transfer coefficients formed by multiple parameter conditions of existing heat exchange channels of different sizes and structures.

[0008] The collected data is preprocessed, and key dimensionless parameters are determined based on the preprocessed data;

[0009] Supplementary data within the applicable working conditions is generated based on empirical relationships to obtain a training set;

[0010] Build a neural network model and input the training set into the neural network model for training;

[0011] Fluid flow and heat transfer prediction data are generated based on the trained neural network model; wherein, the fluid flow and heat transfer prediction data is data on operating conditions that are not covered by experimental data and physical models.

[0012] The boundaries of flow and heat transfer mechanisms are identified, and experimental data, calculated data from empirical relationships, and fluid flow and heat transfer prediction data are classified to achieve regional fitting. A sparse regression method is used to fit the experimental data, calculated data from empirical relationships, and fluid flow and heat transfer prediction data into explicit mathematical expressions. Further, the preprocessing of the collected data specifically includes:

[0013] The dimensions and data formats of experimental data on fluid flow and heat transfer in heat exchange channels of different sizes and structures were standardized.

[0014] Data cleaning was performed on the experimental data of fluid flow and heat transfer in heat exchange channels of different sizes and structures after unification.

[0015] Based on the experimental data of fluid flow and heat transfer in heat exchange channels of different sizes and structures after data cleaning, key dimensionless parameters are extracted.

[0016] Furthermore, the key dimensionless parameters include: dimensionless equivalent diameter, dimensionless pressure, dimensionless temperature, Reynolds number, Weber number, and Jacobian number; wherein,

[0017] The dimensionless equivalent diameter value is used to reflect the effect of channel size on the flow resistance coefficient and heat transfer coefficient;

[0018] Dimensionless pressure is used to reflect the effect of system pressure on fluid properties;

[0019] Dimensionless temperature is used to reflect the effect of temperature on the physical properties of fluids;

[0020] The Reynolds number is used to reflect the influence of fluid flow state;

[0021] The Weber number is used to reflect the ratio of inertial force to surface tension effect;

[0022] Jacob's number is used to reflect the degree of liquid film subcooling.

[0023] Furthermore, the step of inputting the training set into the neural network model for training includes:

[0024] Define a loss function and minimize the loss function to optimize the parameters of the neural network model;

[0025] During the training phase, a regularization layer is added, and an L2 regularization term is added to the loss function to penalize the weights.

[0026] Periodically evaluate the performance of the validation set extracted from the covered experimental data and monitor changes in validation loss;

[0027] If the validation loss increases over several consecutive cycles, training is stopped, and the model parameters are rolled back to the point where the validation loss is lowest, which is then used as the final trained neural network model.

[0028] Furthermore, the definition of the loss function specifically includes:

[0029] We select mean squared error as the basic loss and construct the basic loss function.

[0030] The output layer is designed to constrain physical consistency loss, and predictions that exceed a set threshold loss are penalized.

[0031] The penalty includes a negative prediction parameter and a parameter that does not satisfy the laws of conservation of mass and energy. The penalty decreases the heat transfer coefficient as the Reynolds number Re increases.

[0032] Further, the fitting is performed by region, specifically including:

[0033] Design a classifier;

[0034] Fluid heat transfer coefficients are divided into single-phase and two-phase regions.

[0035] When satisfied It is either a single-phase region or a two-phase region; among them, x e Gas content;

[0036] The frictional resistance coefficient is divided into laminar flow region, transition region and turbulent flow region;

[0037] Among them, when the Reynolds number Re is less than the discrimination criterion Re c The flow is in the laminar region when the Reynolds number Re is greater than 4000, and in the turbulent region otherwise, it is in the transition region.

[0038] Furthermore, experimental data, calculated data from empirical relationships, and predicted data from fluid flow and heat transfer are fitted into explicit mathematical expressions, specifically including:

[0039] Based on the key dimensionless parameters of the data, candidate basis functions are generated.

[0040] Each candidate basis function is normalized to eliminate dimensional differences;

[0041] Regularization methods are used to identify the contribution values ​​of basis functions to the prediction target, and basis functions are selected based on the contribution values;

[0042] The coefficients are solved based on the selected basis functions to obtain an explicit mathematical expression. Further, the candidate basis functions include:

[0043] Power functions are used to describe the nonlinear effects of parameters;

[0044] Exponential functions are used to characterize the decay / growth process of physical quantities;

[0045] Logarithmic functions are used to characterize weak nonlinear relationships between parameters;

[0046] The product function is used to characterize the synergistic effect of parameters;

[0047] The ratio function is used to characterize the competitive balance mechanism of various parameters.

[0048] Furthermore, the process of solving for the coefficients is as follows:

[0049] A genetic algorithm is used to globally search for non-zero basis functions, and the coefficients are fine-tuned using the L-BFGS-B algorithm to meet physical constraints.

[0050] We used Bootstrap to sample hundreds of times and calculated the confidence intervals of the coefficients.

[0051] Furthermore, the explicit mathematical expression includes:

[0052]

[0053]

[0054] in, h This is the predicted value of the flow heat transfer coefficient. λ This is the predicted value of the frictional resistance coefficient. D The diameter is a dimensionless equivalent diameter. P For dimensionless pressure, T The temperature is dimensionless. Re Let Reynolds number be 1. Ja For Jacob's number, f For the function mapping relationship, △ T sub This indicates the degree of liquid film subcooling.

[0055] A second aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a channel flow heat transfer prediction method based on machine learning.

[0056] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a machine learning-based method for predicting channel flow heat transfer.

[0057] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0058] By generating supplementary data based on existing flow and heat transfer relationships, it is possible to effectively cover operating conditions where experimental data is scarce, making the data more comprehensive and reducing the model's prediction bias. Through standardized dimensions and data cleaning, data consistency and accuracy are ensured, eliminating potential unit differences between different data sources. By identifying key dimensionless parameters affecting flow resistance and heat transfer coefficients, the model can focus on the most critical factors influencing flow and heat transfer relationships, improving its ability to grasp the essence of flow and heat transfer phenomena, and thus enabling more accurate predictions of flow and heat transfer coefficients and friction resistance coefficients. Constructing a neural network model effectively captures the complex nonlinear relationship between input parameters and flow and heat transfer coefficients or friction resistance coefficients. Simultaneously, a loss function is designed to penalize predictions that violate physical laws, improving the model's extrapolation ability in sparse data regions, enhancing its reliability and practicality. Using sparse regression to fit experimental data, calculated data from empirical relationships, and a large amount of prediction data generated by the neural network model into explicit mathematical expressions, a unified relationship for heat exchange channels of different sizes across a wide operating range is obtained. This solves the problems of narrow prediction range and poor universality of existing relationships, effectively improving the prediction accuracy and reliability of channel flow and heat transfer relationships. Attached Figure Description

[0059] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0060] Figure 1 This is a flowchart illustrating the prediction method in an embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0062] As one possible implementation method, such as Figure 1 As shown, this embodiment provides a channel flow heat transfer prediction method based on machine learning, including the following specific steps:

[0063] Collect experimental data on fluid flow and heat transfer in heat exchange channels of different sizes and structures, and collect empirical relationships between flow resistance and heat transfer coefficients formed by multiple parameter conditions of existing heat exchange channels of different sizes and structures.

[0064] The collected data is preprocessed, and key dimensionless parameters are determined based on the preprocessed data;

[0065] Supplementary data within the applicable working conditions is generated based on empirical relationships to obtain a training set;

[0066] Build a neural network model and input the training set into the neural network model for training;

[0067] Fluid flow and heat transfer prediction data are generated based on the trained neural network model; wherein, the fluid flow and heat transfer prediction data is data on operating conditions that are not covered by experimental data and physical models.

[0068] The boundaries of flow and heat transfer mechanisms are identified, and experimental data, calculated data of empirical relationships, and fluid flow and heat transfer prediction data are classified to achieve regional fitting. The sparse regression method is used to fit the experimental data, calculated data of empirical relationships, and fluid flow and heat transfer prediction data into explicit mathematical expressions.

[0069] This embodiment achieves efficient prediction of fluid flow and heat transfer relationships within heat exchange channels of different sizes and structures through the following steps. First, a large amount of experimental data on fluid flow and heat transfer within heat exchange channels of different sizes and structures, as well as existing empirical relational calculation data, are collected. Next, the collected data is preprocessed, including unifying dimensions and data cleaning, to ensure data quality and determine key dimensionless parameters. To compensate for the deficiencies in experimental data under certain operating conditions, supplementary data is generated based on a physical model to cover conditions where data is scarce. Then, a neural network model is constructed, and the processed data is input into the model for training, enabling the model to learn the complex relationship between input parameters and the flow heat transfer coefficient or friction resistance coefficient. After training, the model is used to generate a large amount of prediction data, and this data, along with the experimental data and empirical relational calculation data, is fitted into an explicit mathematical expression, thereby obtaining a unified relational expression for heat exchange channels of different sizes over a wide range of operating conditions. This prediction method not only improves the accuracy and reliability of predicting channel flow heat transfer relationships but also provides strong technical support for engineering design and practical applications, enabling engineers to quickly and accurately calculate the flow heat transfer coefficient or friction resistance coefficient under different operating conditions, thus improving the efficiency of engineering design and analysis.

[0070] In some possible implementations, the collected data is preprocessed, specifically including:

[0071] The dimensions and data formats of experimental data on fluid flow and heat transfer in heat exchange channels of different sizes and structures were standardized.

[0072] Data cleaning was performed on the experimental data of fluid flow and heat transfer in heat exchange channels of different sizes and structures after unification.

[0073] Based on the experimental data of fluid flow and heat transfer in heat exchange channels of different sizes and structures after data cleaning, key dimensionless parameters are extracted.

[0074] In some possible implementations, key dimensionless parameters include: dimensionless equivalent diameter, dimensionless pressure, dimensionless temperature, Reynolds number, Weber number, and Jacobian number; wherein,

[0075] The dimensionless equivalent diameter value is used to reflect the effect of channel size on the flow resistance coefficient and heat transfer coefficient;

[0076] Dimensionless pressure is used to reflect the effect of system pressure on fluid properties;

[0077] Dimensionless temperature is used to reflect the effect of temperature on the physical properties of fluids;

[0078] The Reynolds number is used to reflect the influence of fluid flow state;

[0079] The Weber number is used to reflect the ratio of inertial force to surface tension effect;

[0080] Jacob's number is used to reflect the degree of liquid film subcooling.

[0081] It is understandable that fluid properties include parameters such as specific heat capacity, viscosity, thermal conductivity, and density. For example, dimensionless pressure is used to reflect the effect of system pressure on fluid properties such as specific heat capacity, while dimensionless temperature is used to reflect the effect of temperature on fluid properties such as viscosity and thermal conductivity.

[0082] In some possible implementations, the neural network model includes:

[0083] Input layer: Input operating parameters, mainly including dimensionless equivalent diameter. D Dimensionless pressure P Dimensionless temperature T Reynolds number ( Re ), Weber number ( We ), Jacob's number ( Ja )wait;

[0084] The hidden layers consist of more than 30 fully connected layers, using the ReLU activation function;

[0085] The output layer uses a linear activation function.

[0086] In some possible implementations, the training set is fed into the neural network model for training, including:

[0087] Define the loss function;

[0088] Add a regularization layer during the training phase and add an L2 regularization term to the loss function to penalize excessively large weights;

[0089] Periodically evaluate the performance of the validation set extracted from the covered experimental data and monitor changes in validation loss;

[0090] If the validation loss increases over several consecutive cycles, training is stopped, and the model parameters are rolled back to the point where the validation loss is lowest, which is then used as the final trained neural network model.

[0091] In some possible implementations, the loss function includes a base loss and a physical consistency loss. The base loss is the mean squared error (MSE) or mean absolute error (MAE). The physical consistency loss refers to the penalty imposed on predictions that violate physical laws, such as penalizing negative prediction parameters and parameters that do not satisfy the laws of conservation of mass and energy, penalizing the tendency of the heat transfer coefficient to decrease as Re increases.

[0092] In some possible implementations, the neural network model DNN is regularized, such as using L2 regularization, combined with the addition of a Dropout layer to drop probabilities. pAn option of 0.2-0.5 can be used to prevent overfitting. During the testing phase, Dropout should be disabled, and the weights multiplied by 1 / (1- p To maintain expected values, batch normalization is used during training to avoid over-regularization. This normalization standardizes the input for each layer by subtracting the mean and then dividing by the standard deviation. For training strategies, early stopping is employed to prevent overfitting. This is done by evaluating the validation set performance at regular intervals. If the validation loss does not improve over several consecutive epochs, training is stopped and the model is rolled back to its optimal parameters. An adaptive learning rate is used as the optimizer, with an initial learning rate of 1e. -4 This is combined with learning rate decay.

[0093] In some possible implementations, region-specific fitting specifically includes:

[0094] Design a classifier;

[0095] Fluid heat transfer coefficients are divided into single-phase and two-phase regions.

[0096] When satisfied It is either a single-phase region or a two-phase region; among them, x e Gas content;

[0097] The frictional resistance coefficient is divided into laminar flow region, transition region and turbulent flow region;

[0098] Among them, when the Reynolds number Re is less than the discrimination criterion Re c The flow is in the laminar region when the Reynolds number Re is greater than 4000, and in the turbulent region otherwise, it is in the transition region.

[0099] The criteria for determining Re differ across channels. The criterion for determining laminar flow is the Reynolds number Re. c .

[0100] In closed channels such as circular tubes, the criterion for distinguishing between the laminar flow region and the transition region is Re. c The value is 2000; in open channels such as bar bundle channels, the criterion for distinguishing between the laminar flow region and the transition region is Re. c =1000; when Re is less than the discrimination criterion Re c At that time, it is in the laminar flow region.

[0101] Criteria for distinguishing between the transition zone and the turbulent zone Re coThe value of Re is 4000; when Re is greater than 4000, it is the turbulent region; the rest is the transition region. In some possible implementations, the predicted data is fitted into an explicit mathematical expression, specifically including: using a sparse regression method, first generating candidate basis functions, then standardizing each basis function, then selecting regularization parameters based on regularized feature selection, filtering out important terms based on non-zero coefficients, then evaluating model performance, and checking whether the filtered model is overfitted or underfitted. Through the above steps, a general explicit mathematical expression with a wide range of applications can be obtained.

[0102] In some possible implementations, taking the heat transfer coefficient as an example, the mathematical expression and the relationship between the basis functions are shown as follows:

[0103] ,in, This refers to the predicted value of the flow heat transfer coefficient. , , Refers to different types of basis functions. , , It refers to the coefficients of different basis functions.

[0104] In some possible implementations, candidate basis functions include:

[0105] Power functions are used to describe the nonlinear effects of parameters;

[0106] Exponential functions are used to characterize the decay / growth process of physical quantities;

[0107] Logarithmic functions are used to characterize weak nonlinear relationships between parameters;

[0108] The product function is used to characterize the synergistic effect of parameters;

[0109] The ratio function is used to characterize the competitive balance mechanism of various parameters.

[0110] In one possible implementation, the process of solving for the coefficients is as follows:

[0111] A genetic algorithm is used to globally search for non-zero basis functions, and the coefficients are fine-tuned using the L-BFGS-B algorithm to meet physical constraints.

[0112] We used Bootstrap to sample hundreds of times and calculated the confidence intervals of the coefficients.

[0113] In some possible implementations, Lasso or elastic networks are used for regularization-based feature selection. Lasso regression uses cross-validation to select the optimal regularization strength and extracts the basis functions corresponding to non-zero coefficients as importance terms.

[0114] The elastic network combines L1 and L2 regularization and optimizes the hybrid parameters through cross-validation.

[0115] In some possible implementations, explicit mathematical expressions include:

[0116] ;

[0117] ;

[0118] in, h This is the predicted value of the flow heat transfer coefficient. λ This is the predicted value of the frictional resistance coefficient. D The diameter is a dimensionless equivalent diameter. P For dimensionless pressure, T The temperature is dimensionless. Re Let Reynolds number be 1. Ja For Jacob's number, f For the function mapping relationship, △ T sub This represents the liquid film subcooling. The mathematical expression is a piecewise function, and the flow heat transfer coefficient can be divided into single-phase and two-phase regions; the friction drag coefficient is divided into laminar, transition, and turbulent regions.

[0119] The explicit mathematical expressions mentioned above can be encapsulated as library functions or embedded as models into heat exchanger design software or nuclear reactor system analysis software.

[0120] As one possible implementation, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a channel flow heat transfer prediction method based on machine learning.

[0121] As one possible implementation, this embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a machine learning-based channel flow heat transfer prediction method.

[0122] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting heat transfer in channel flow based on machine learning, characterized in that, The specific steps include the following: Collect experimental data on fluid flow and heat transfer in heat exchange channels of different sizes and structures, and collect empirical relationships between flow resistance and heat transfer coefficients formed by multiple parameter conditions of existing heat exchange channels of different sizes and structures. The collected data is preprocessed, and key dimensionless parameters are determined based on the preprocessed data; Supplementary data within the applicable working conditions is generated based on empirical relationships to obtain a training set; Build a neural network model and input the training set into the neural network model for training; Fluid flow and heat transfer prediction data are generated based on the trained neural network model; wherein, the fluid flow and heat transfer prediction data is data on operating conditions that are not covered by experimental data and physical models. The boundaries of flow and heat transfer mechanisms are identified, and experimental data, calculated data of empirical relationships, and fluid flow and heat transfer prediction data are classified to achieve regional fitting. The sparse regression method is used to fit the experimental data, calculated data of empirical relationships, and fluid flow and heat transfer prediction data into explicit mathematical expressions.

2. The channel flow heat transfer prediction method based on machine learning according to claim 1, characterized in that, The preprocessing of the collected data specifically includes: The dimensions and data formats of experimental data on fluid flow and heat transfer in heat exchange channels of different sizes and structures were standardized. Data cleaning was performed on the experimental data of fluid flow and heat transfer in heat exchange channels of different sizes and structures after unification. Based on the experimental data of fluid flow and heat transfer in heat exchange channels of different sizes and structures after data cleaning, key dimensionless parameters are extracted.

3. The channel flow heat transfer prediction method based on machine learning according to claim 2, characterized in that, The key dimensionless parameters include: dimensionless equivalent diameter, dimensionless pressure, dimensionless temperature, Reynolds number, Weber number, and Jacobian number; among which... The dimensionless equivalent diameter value is used to reflect the effect of channel size on the flow resistance coefficient and heat transfer coefficient; Dimensionless pressure is used to reflect the effect of system pressure on fluid properties; Dimensionless temperature is used to reflect the effect of temperature on the physical properties of fluids; The Reynolds number is used to reflect the influence of fluid flow state; The Weber number is used to reflect the ratio of inertial force to surface tension effect; Jacob's number is used to reflect the degree of liquid film subcooling.

4. The channel flow heat transfer prediction method based on machine learning according to claim 1, characterized in that, The step of inputting the training set into the neural network model for training includes: Define a loss function and minimize the loss function to optimize the parameters of the neural network model; During the training phase, a regularization layer is added, and an L2 regularization term is added to the loss function to penalize the weights. Periodically evaluate the performance of the validation set extracted from the covered experimental data and monitor changes in validation loss; If the validation loss increases over several consecutive cycles, training is stopped, and the model parameters are rolled back to the point where the validation loss is lowest, which is then used as the final trained neural network model.

5. The channel flow heat transfer prediction method based on machine learning according to claim 4, characterized in that, The loss function is defined as follows: We select mean squared error as the basic loss and construct the basic loss function. The output layer is designed to constrain physical consistency loss, and predictions that exceed a set threshold loss are penalized. The penalty includes a negative prediction parameter and a parameter that does not satisfy the laws of conservation of mass and energy. The penalty decreases the heat transfer coefficient as the Reynolds number Re increases.

6. The channel flow heat transfer prediction method based on machine learning according to claim 1, characterized in that, Regional fitting, specifically including: Design a classifier; Fluid heat transfer coefficients are divided into single-phase and two-phase regions. When satisfied It is either a single-phase region or a two-phase region; among them, x e Gas content; The frictional resistance coefficient is divided into laminar flow region, transition region and turbulent flow region; Among them, when the Reynolds number Re is less than the discrimination criterion Re c The flow is in the laminar region when the Reynolds number Re is greater than 4000, and in the turbulent region otherwise, it is in the transition region.

7. The channel flow heat transfer prediction method based on machine learning according to claim 1, characterized in that, The experimental data, calculated data from empirical relationships, and predicted data from fluid flow and heat transfer are fitted into explicit mathematical expressions, specifically including: Based on key dimensionless parameters from experimental data, computational data from empirical relationships, and fluid flow and heat transfer prediction data, candidate basis functions are generated. Each candidate basis function is normalized to eliminate dimensional differences; Regularization methods are used to identify the contribution values ​​of basis functions to the prediction target, and basis functions are selected based on the contribution values; The coefficients are solved based on the selected basis functions to obtain an explicit mathematical expression.

8. The channel flow heat transfer prediction method based on machine learning according to claim 7, characterized in that, The candidate basis functions include: Power functions are used to describe the nonlinear effects of parameters; Exponential functions are used to characterize the decay / growth process of physical quantities; Logarithmic functions are used to characterize weak nonlinear relationships between parameters; The product function is used to characterize the synergistic effect of parameters; The ratio function is used to characterize the competitive balance mechanism of various parameters.

9. The channel flow heat transfer prediction method based on machine learning according to claim 7, characterized in that, The process of solving for the coefficients is as follows: A genetic algorithm is used to globally search for non-zero basis functions, and the coefficients are fine-tuned using the L-BFGS-B algorithm to meet physical constraints. We used Bootstrap to sample hundreds of times and calculated the confidence intervals of the coefficients.

10. The channel flow heat transfer prediction method based on machine learning according to claim 7, characterized in that, The explicit mathematical expressions include: in, h This is the predicted value of the flow heat transfer coefficient. λ This is the predicted value of the frictional resistance coefficient. D The diameter is a dimensionless equivalent diameter. P For dimensionless pressure, T The temperature is dimensionless. Re Let Reynolds number be 1. Ja For Jacob's number, f For the function mapping relationship, △ T sub This indicates the degree of liquid film subcooling.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the machine learning-based channel flow heat transfer prediction method as described in any one of claims 1 to 10.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the machine learning-based channel flow heat transfer prediction method as described in any one of claims 1 to 10.

Images