A valve performance simulation and optimization method and system based on hybrid deep learning

Abstract

The present invention discloses a valve performance simulation and optimization method and system based on hybrid deep learning, including: S1. performing finite element simulation analysis on the valve and using multi-physics field thermal-solid coupling solution to obtain the stress, deformation and temperature distribution data of the valve under different working conditions and the maximum thermal stress and maximum displacement of thermal deformation of key components of the valve; S2. after data processing, using the stress, deformation and temperature distribution data of the valve under different working conditions as input parameters and the maximum thermal stress and maximum displacement of thermal deformation as output parameters, training a neural network model and establishing a thermal-structural performance prediction model; S3. based on the prediction results, using a non-dominated genetic algorithm based on a reference point, combined with reference point generation and an adaptive agent model, to perform multi-objective optimization and output the optimal parameter combination to adjust the valve design; the present invention uses a neural network to perform deep learning and multi-objective optimization on the finite element simulation results, thereby improving the valve performance prediction accuracy and optimization efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment design and optimization, and more specifically to a valve performance simulation and optimization method and system based on hybrid deep learning. Background Art

[0002] Currently, existing valve performance estimation mainly includes the following technologies:

[0003] Physical experiments and empirical formulas: Data is obtained through laboratory tests such as pressure tests and flow tests, and performance is estimated using relevant empirical formulas such as the Darcy-Weisbach formula. However, these experiments are costly and time-consuming, and empirical formulas lack the ability to generalize to complex operating conditions.

[0004] Finite element simulation technology (FEA): Use tools such as ANSYS and COMSOL to perform stress analysis on valve structures and simulate material deformation and fatigue life. Computational fluid dynamics (CFD) is used to simulate the flow field distribution inside the valve (such as flow velocity and pressure gradient). However, high-precision simulation requires dense meshing, which is computationally time-consuming and can take several hours to several days for a single simulation.

[0005] Traditional optimization methods: Adjusting valve parameters based on trial and error or response surface methodology lacks the ability to effectively handle multi-objective and nonlinear problems, relies on manual experience to adjust parameters, and is difficult to achieve automated optimization.

[0006] In summary, traditional valve design relies on experience and experiments, repeated trial and error, low efficiency and high cost, and finite element analysis takes a long time to calculate and consumes a lot of resources under high-precision grids, making it difficult to meet the needs of rapid iteration; existing machine learning models lack sufficient high-quality training data support, resulting in limited prediction accuracy and generalization capabilities, and it is difficult to achieve rapid mapping from target performance to geometric parameters, and cannot meet design requirements with specific performance requirements; the coupling between finite element analysis and deep learning models is not tight enough, and the advantages of both are not fully utilized; in terms of optimization, parameter optimization relies on manual parameter adjustment and lacks automation and multi-objective collaborative optimization capabilities.

[0007] Therefore, how to provide a valve performance simulation and optimization method and system based on hybrid deep learning to quickly simulate and optimize the performance of valves under different working conditions and improve the accuracy of valve performance prediction is a problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0008] In view of this, the present invention provides a valve performance simulation and optimization method and system based on hybrid deep learning, which adopts multi-physical field coupling simulation technology to simulate multiple physical fields under the target working conditions of the valve at the same time to capture the dynamic response of the valve; constructs the input parameter and output parameter models in the finite element analysis to predict the performance indicators of the valve under any working conditions, and adopts transfer learning technology to transfer the pre-trained weights of the general valve model to the new type of valve to reduce data requirements; finite element simulation generates high-precision data to train the neural network, and the neural network prediction results are fed back to the optimization module. The optimized design parameters re-trigger the data model to ensure the self-correction capability of the system to solve some of the technical problems mentioned in the background technology.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] A valve performance simulation and optimization method based on hybrid deep learning includes the following steps:

[0011] S1. Perform finite element simulation analysis on the valve using a multi-physics field thermal-solid coupling solution method to obtain the valve's stress, deformation, and temperature distribution data under different operating conditions, as well as the maximum thermal stress and thermal deformation maximum displacement of the valve's key components;

[0012] S2. After processing the simulation data, a neural network model is trained using the valve's stress, deformation, and temperature distribution data under different operating conditions as input parameters, and the maximum thermal stress and maximum thermal deformation displacement as output parameters to establish a thermal-structural performance prediction model.

[0013] S3. Based on the prediction results of the thermal-structural performance prediction model, a reference point-based non-dominated genetic algorithm is used to perform multi-objective optimization in combination with reference point generation and an adaptive surrogate model to output the optimal parameter combination to adjust the valve design.

[0014] Preferably, the specific contents of step S1 include:

[0015] S11. Build a 3D geometric model of the valve and define material properties, including thermal and structural parameters.

[0016] S12. Use tetrahedral and triangular hybrid adaptive meshing to create a three-dimensional geometric model of the valve.

[0017] S13. Use multiphysics coupling to set thermodynamic boundary conditions and structural mechanics boundaries, and configure the solver and convergence criteria.

[0018] S14. Solve and perform post-processing to output the parameters of maximum thermal stress and maximum displacement of thermal deformation.

[0019] Preferably, in step S11, the thermal parameters include thermal conductivity k, specific heat capacity cp, thermal expansion coefficient α; the structural parameters include elastic modulus E, Poisson's ratio ν, input pressure σ in ;

[0020] In step S13, the thermodynamic boundary conditions are specifically: temperature field heat transfer parameters, fluid inlet temperature T in , thermal expansion coefficient α, outer wall convection heat transfer coefficient h conv And the ambient temperature T ref ;

[0021] The structural mechanics boundary is specifically: according to the actual working conditions, the pressure load is applied to set the input pressure σ in , set fixed constraints and impose fluid dynamic boundary conditions;

[0022] The configured solver is the direct coupling solver in the thermal stress module of COMSOL Multiphysics multi-physics simulation; the residual error and the maximum number of iterations are used as the convergence criteria.

[0023] Preferably, in step S2, the thermal-structural performance prediction model structure established is specifically:

[0024] The input layer is the material parameters and operating parameters. The material parameters include thermal conductivity k, thermal expansion coefficient α, elastic modulus E and Poisson's ratio ν; the operating parameters include input temperature T in , outer wall convection heat transfer coefficient h conv and input pressure σ in ;

[0025] The hidden layer uses four convolutional layers with a kernel size of 3×3×3 to extract the three-dimensional temperature and structural features of the valve. The geometric and material parameters are encoded into a 128-dimensional vector. The CNN output is concatenated with the encoder vector and passed through three fully connected layers to generate the prediction result.

[0026] The activation function of the hidden layer uses LeakyReLU, and the output layer is linearly activated;

[0027] Output layer: Output parameters thermal stress and maximum displacement of thermal deformation.

[0028] Preferably, the specific content of training the neural network model is:

[0029] Generate steady-state data through parameterized thermal-solid coupling simulation, perform data enhancement, and add Gaussian noise to the temperature field;

[0030] Set the loss function to minimize the difference between the predicted value and the true value:

[0031] L=λ1|T max,pre -T max,sim∣+λ2∣σ max,pre -σ max,sim ∣

[0032] Among them, T max,pre To predict the maximum temperature, T is the maximum temperature prediction value of the valve in steady state output by the neural network model. max,sim is the maximum temperature of the simulation, is the highest temperature value of the valve obtained by finite element simulation, as the training label, σ max,pre is the predicted maximum thermal stress, is the predicted value of the valve thermal stress peak value output by the neural network model, σ max,sim is the maximum thermal stress simulated, is the maximum thermal stress value obtained by finite element simulation, is used as the training label, λ1 is the temperature error weight coefficient, and λ2 is the stress error weight coefficient.

[0033] Preferably, the specific contents of step S3 include:

[0034] S31. Initialization phase: Define the design variables as material parameters and operating condition parameters. Material parameters include elastic modulus E and Poisson's ratio ν, and operating condition parameters include inlet temperature T. in and input pressure σ in , reference points are generated based on material parameters and working condition parameter variables;

[0035] S32. Proxy model construction:

[0036] The CNN-LSTM hybrid model is trained and validated based on historical finite element data. The input is the design variable and the output is the maximum allowable stress σ of the material. max and the maximum allowable deformation u max ;

[0037] Define the proxy model confidence threshold for dynamic confidence evaluation. If σ max If the variance of the predicted value Var≤10MPa, the finite element verification is skipped, otherwise calibration is performed with the simulation data;

[0038] S33. Iterative optimization using initial full parameter space exploration and later local refined search;

[0039] S34. Dynamically adjust material solutions: A built-in material database records the elastic modulus E and Poisson's ratio ν. Every 10 iterations, the E and ν of the current optimal solution are checked and the closest engineering material is matched from the library while ensuring that the constraints are not violated, that is, the maximum allowable stress and maximum allowable deformation must meet the material's allowable range.

[0040] Preferably, in step S33, the specific content of full parameter space exploration is:

[0041] Initial population generation: randomly generate n sets of design parameters, covering material parameters: elastic modulus E, Poisson's ratio ν and thermal expansion coefficient α, as well as operating parameters: inlet temperature T in and input pressure σ in ;Call the surrogate model to predict performance and select non-dominated solutions as the initial population;

[0042] Adaptive sampling: Generate new parameters in sparse areas to supplement sampling points;

[0043] The specific contents of local refined search are as follows:

[0044] Focus on hotspot areas: select a preset proportion of individuals with the best performance in the current Pareto solution set and perform local search in the defined neighborhood;

[0045] Hybrid optimization strategy:

[0046] Gradient-assisted mutation: Calculate the objective function gradient for continuous variables of input pressure and input temperature;

[0047] Directed crossover: Cross high-temperature, low-stress individuals with low-temperature, high-stress individuals to explore potential equilibrium points.

[0048] A valve performance simulation and optimization system based on hybrid deep learning, based on the aforementioned valve performance simulation and optimization method based on hybrid deep learning, comprising: a finite element simulation module, a thermal-structural performance prediction module, and a multi-objective optimization module;

[0049] Finite element simulation module, used to perform finite element simulation analysis on valves and use multi-physics field thermal-solid coupling solution method to obtain the stress, deformation and temperature distribution data of valves under different working conditions, as well as the maximum thermal stress and maximum displacement of key valve components;

[0050] The thermal-structural performance prediction module is used to train the neural network model after data processing of the simulation data, using the stress, deformation and temperature distribution data of the valve under different working conditions as input parameters and the maximum thermal stress and maximum displacement of thermal deformation as output parameters to establish a thermal-structural performance prediction model;

[0051] The multi-objective optimization module is used to perform multi-objective optimization based on the prediction results of the thermal-structural performance prediction model, using a non-dominated genetic algorithm based on a reference point, combined with reference point generation and an adaptive agent model, to output the optimal parameter combination to adjust the valve design.

[0052] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a valve performance simulation and optimization method based on hybrid deep learning.

[0053] A processing terminal includes a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor executes the computer program, a valve performance simulation and optimization method based on hybrid deep learning is implemented.

[0054] It can be seen from the above technical solutions that compared with the existing technology, the present invention discloses a valve performance simulation and optimization method and system based on hybrid deep learning. By combining finite element analysis with deep learning technology and using neural networks to perform deep learning on finite element simulation results, the accuracy of valve performance prediction can be improved, especially when facing complex working conditions, it can better capture nonlinear relationships; compared with traditional manual optimization or optimization methods based solely on physical models, the optimization module of the present invention can greatly shorten the design cycle and improve optimization efficiency; through the training of deep learning models, the system can handle valve performance predictions under different working conditions and has strong adaptability and flexibility; the system has a user-friendly interface, which is convenient for engineers to get started quickly, reduces manual intervention, and improves work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0056] Figure 1 A schematic diagram of a valve performance simulation and optimization method based on hybrid deep learning provided by the present invention;

[0057] Figure 2 A schematic diagram of the finite element simulation analysis process provided by the present invention;

[0058] Figure 3 This is a schematic diagram of the neural network model structure provided by the present invention. DETAILED DESCRIPTION

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0060] The embodiment of the present invention discloses a valve performance simulation and optimization method based on hybrid deep learning, such as Figure 1 , including the following steps:

[0061] S1. Perform finite element simulation analysis on the valve using a multi-physics field thermal-solid coupling solution method to obtain the valve's stress, deformation, and temperature distribution data under different operating conditions, as well as the maximum thermal stress and thermal deformation maximum displacement of the valve's key components;

[0062] S2. After processing the simulation data, a neural network model is trained using the valve's stress, deformation, and temperature distribution data under different operating conditions as input parameters, and the maximum thermal stress and maximum thermal deformation displacement as output parameters to establish a thermal-structural performance prediction model.

[0063] S3. Based on the prediction results of the thermal-structural performance prediction model, a reference point-based non-dominated genetic algorithm is used to perform multi-objective optimization in combination with reference point generation and an adaptive surrogate model to output the optimal parameter combination to adjust the valve design.

[0064] In order to further implement the above technical solutions, Figure 2 , the specific contents of step S1 include:

[0065] S11. Build a 3D geometric model of the valve and define material properties, including thermal and structural parameters.

[0066] In this embodiment, a three-dimensional geometric model of the valve is established based on CAD and SOLIDWORKS software;

[0067] S12. Use a hybrid adaptive mesh of tetrahedrons and triangles to partition the established 3D geometric model of the valve to improve computational accuracy and stability. Mesh refinement is performed on some thermally sensitive areas and connection areas.

[0068] S13. Use multiphysics coupling to set thermodynamic boundary conditions and structural mechanics boundaries, and configure the solver and convergence criteria.

[0069] S14. Solve and perform post-processing to output the parameters of maximum thermal stress and maximum displacement of thermal deformation.

[0070] In order to further implement the above technical solution, in step S11, the thermal parameters include thermal conductivity k, specific heat capacity cp, thermal expansion coefficient α; the structural parameters include elastic modulus E, Poisson's ratio ν, input pressure σ in ;

[0071] In step S13, the thermodynamic boundary conditions are specifically: temperature field heat transfer parameters, fluid inlet temperature T in , thermal expansion coefficient α, outer wall convection heat transfer coefficient h conv and ambient temperature T ref ;

[0072] In this embodiment, the basis for setting the thermodynamic boundary conditions is:

[0073] Heat conduction equation (steady state): , where T is the temperature field and Q is the internal heat source;

[0074] Thermoelastic equation (steady-state stress): ,in, is the strain tensor, T ref is the reference temperature, σ is the steady-state stress;

[0075] Apply thermal contact boundary conditions;

[0076] The structural mechanics boundary is specifically: according to the actual working conditions, the pressure load is applied to set the input pressure σ in , set fixed constraints and impose fluid dynamic boundary conditions;

[0077] In this embodiment, the basis for setting the structural mechanics boundary is to set the flange fixed constraint ux=uy=uz=0; different pressure loads are applied to the internal flow channel;

[0078] Configure the solver: Use the direct coupling solver in the thermal stress module of COMSOL Multiphysics multi-physics simulation;

[0079] Convergence criterion: residual R<10 -6 , the maximum number of iterations is 200.

[0080] In this embodiment, finite element simulation is used to conduct a detailed analysis of the mechanical properties (stress, deformation, etc.) of the valve under different working conditions. The simulation results provide a data basis for the training of the neural network.

[0081] In order to further implement the above technical solutions, Figure 3 In step S2, the thermal-structural performance prediction model structure established is specifically as follows:

[0082] The input layer is the material parameters and operating parameters. The material parameters include thermal conductivity k, thermal expansion coefficient α, elastic modulus E and Poisson's ratio ν; the operating parameters include input temperature T in , outer wall convection heat transfer coefficient h conv and input pressure σ in ;

[0083] The hidden layer uses four convolutional layers with a kernel size of 3×3×3 to extract the three-dimensional temperature and structural features of the valve. The geometric and material parameters are encoded into 128-dimensional vectors. The CNN output is concatenated with the encoder vector and the prediction result is obtained through three fully connected layers. The number of neurons in the three fully connected layers is 256-128-64.

[0084] The activation function of the hidden layer uses LeakyReLU, α=0.1, and the output layer is linearly activated;

[0085] Output layer: Output parameters thermal stress and maximum displacement of thermal deformation.

[0086] In order to further implement the above technical solution, the specific content of training the neural network model is as follows:

[0087] Generate steady-state data through parameterized thermal-solid coupling simulation (generate 500 sets of steady-state data, covering extreme working conditions, such as T in =500℃, σ in =30MPa, and perform data enhancement to add Gaussian noise to the temperature field (C μ =0, σ=2°C, to improve model robustness);

[0088] Set the loss function to minimize the difference between the predicted value and the true value:

[0089] L=λ1|T max,pre -T max,sim ∣+λ2∣σ max,pre -σ max,sim ∣

[0090] Among them, T max,pre To predict the maximum temperature, T is the maximum temperature prediction value of the valve in steady state output by the neural network model. max,sim is the maximum temperature of the simulation, is the highest temperature value of the valve obtained by finite element simulation, as the training label, σ max,pre is the predicted maximum thermal stress, is the predicted value of the valve thermal stress peak value output by the neural network model, σ max,sim is the maximum thermal stress simulated, is the maximum thermal stress value obtained by finite element simulation, used as the training label, λ1 is the temperature error weight coefficient, set to 0.6, and λ2 is the stress error weight coefficient, set to 0.4.

[0091] In this embodiment, before training the neural network model, the data obtained from the finite element analysis is also preprocessed by cleaning, normalizing, etc., to remove noise and unify the scale to ensure high data quality. Subsequently, multiple data sources are integrated to improve the generalization ability and prediction accuracy of the model. By designing a suitable neural network model and combining the finite element analysis results and experimental data, a high-precision performance prediction model is trained. This model can predict the performance of the valve under different working conditions, including parameters such as flow, pressure, and sealing.

[0092] To further implement the above technical solution, step S3 uses the reference point-based non-dominated genetic algorithm NSGA-III algorithm, combined with reference point generation and adaptive agent model, to solve the valve design optimization problem under high-dimensional objectives (temperature, stress, deformation). The specific contents include:

[0093] S31. Initialization phase: Define the design variables as material parameters and operating condition parameters. Material parameters include elastic modulus E and Poisson's ratio ν, and operating condition parameters include inlet temperature T. in and input pressure σ in , reference points are generated based on material parameters and working condition parameter variables;

[0094] S32. Proxy model construction:

[0095] A CNN-LSTM hybrid model was trained and validated based on historical finite element data (500 sets). The input was the design variables, and the output was the maximum allowable stress σ of the material. max and the maximum allowable deformation u max ; Model verification: test concentrated stress error <5%, deformation error <5%;

[0096] Define the proxy model confidence threshold (η=0.9) for dynamic confidence evaluation. If σ max If the variance of the predicted value Var≤10MPa, the finite element verification is skipped, otherwise calibration is performed with the simulation data;

[0097] S33. Iterative optimization using initial full parameter space exploration and later local refined search;

[0098] S34. Dynamically adjust material solutions: A built-in material database (including stainless steel, titanium alloy, ceramics, etc.) records the elastic modulus E and Poisson's ratio ν. Every 10 iterations, the E and ν of the current optimal solution are checked, and the closest engineering material is matched from the library. At the same time, it ensures that the constraints are not violated, that is, the maximum allowable stress and maximum allowable deformation must meet the material's allowable range.

[0099] In this embodiment, after obtaining the valve performance prediction, an optimization algorithm is used to adjust the valve design to quickly find the optimal design solution, thereby improving the working efficiency and reliability of the valve.

[0100] In order to further implement the above technical solution, the specific content of step S33, full parameter space exploration is as follows:

[0101] Initial population generation: randomly generate n sets of design parameters (50 sets), covering material parameters: elastic modulus E, Poisson's ratio ν and thermal expansion coefficient α, as well as operating parameters: inlet temperature T in and input pressure σ in ;Call the surrogate model to predict performance and select non-dominated solutions (Pareto solutions) as the initial population;

[0102] Adaptive sampling: Generate new parameters in sparse areas to supplement sampling points. Sparse areas are areas not covered by the target space. For example, if the target value (σ max =100MPa,umax =0.1mm) has no solution, then 10 new sets of parameters are generated in this neighborhood;

[0103] The specific contents of local refined search are as follows:

[0104] Focus on hotspot areas: select a preset proportion of individuals with the best performance in the current Pareto solution set and perform local search in the defined neighborhood;

[0105] In this embodiment, the 50% individuals with the best performance in the current Pareto solution set are selected, such as σ max ≤100MPa,u max ≤0.1mm, perform local search in its neighborhood; define the neighborhood as the inlet temperature T in ±20℃, input pressure σ in ±10MPa;

[0106] Hybrid optimization strategy:

[0107] Gradient-assisted mutation: Calculate the objective function gradient for continuous variables of input pressure and input temperature;

[0108] In this embodiment, the input pressure is generated along an input gradient of 5 MPa, and the input temperature is generated along a descending input gradient of 20°C;

[0109] Directed crossover: Cross high-temperature, low-stress individuals with low-temperature, high-stress individuals to explore potential equilibrium points.

[0110] In this embodiment, if the predicted value is directly used during the performance optimization process, one finite element calculation (about 2 hours) can be saved, the proxy model reduces finite element calls by 70%, dynamically matches the actual engineering materials, and avoids the problem that the theoretical optimal solution cannot be manufactured.

[0111] A valve performance simulation and optimization system based on hybrid deep learning is based on a valve performance simulation and optimization method based on hybrid deep learning, including: a finite element simulation module, a thermal-structural performance prediction module and a multi-objective optimization module;

[0112] Finite element simulation module, used to perform finite element simulation analysis on valves and use multi-physics field thermal-solid coupling solution method to obtain the stress, deformation and temperature distribution data of valves under different working conditions, as well as the maximum thermal stress and maximum displacement of key valve components;

[0113] The thermal-structural performance prediction module is used to train the neural network model after data processing of the simulation data, using the stress, deformation and temperature distribution data of the valve under different working conditions as input parameters and the maximum thermal stress and maximum displacement of thermal deformation as output parameters to establish a thermal-structural performance prediction model;

[0114] The multi-objective optimization module is used to perform multi-objective optimization based on the prediction results of the thermal-structural performance prediction model, using a reference point-based non-dominated genetic algorithm combined with reference point generation and an adaptive agent model to output the optimal parameter combination to adjust the valve design.

[0115] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a valve performance simulation and optimization method based on hybrid deep learning.

[0116] A processing terminal includes a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor executes the computer program, a valve performance simulation and optimization method based on hybrid deep learning is implemented.

[0117] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0118] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A valve performance simulation and optimization method based on hybrid deep learning, characterized in that: The following steps are involved: S1. Perform finite element simulation analysis on the valve using a multi-physics field thermal-solid coupling solution method to obtain the valve's stress, deformation, and temperature distribution data under different operating conditions, as well as the maximum thermal stress and thermal deformation maximum displacement of the valve's key components; S2. After processing the simulation data, a neural network model is trained using the valve's stress, deformation, and temperature distribution data under different operating conditions as input parameters, and the maximum thermal stress and maximum thermal deformation displacement as output parameters to establish a thermal-structural performance prediction model. S3. Based on the prediction results of the thermal-structural performance prediction model, a reference-point-based non-dominated genetic algorithm is used to perform multi-objective optimization, combining reference point generation with an adaptive surrogate model, and outputting the optimal parameter combination to adjust the valve design. The specific contents of step S3 include: S31. Initialization phase: Define the design variables as material parameters and operating condition parameters. Material parameters include elastic modulus E and Poisson's ratio ν, and operating condition parameters include inlet temperature T. in and input pressure σ in , reference points are generated based on material parameters and working condition parameter variables; S32. Proxy model construction: The CNN-LSTM hybrid model is trained and validated based on historical finite element data. The input is the design variable and the output is the maximum allowable stress σ of the material. max and the maximum allowable deformation u max ; Define the proxy model confidence threshold for dynamic confidence evaluation. If σ max If the variance of the predicted value Var≤10MPa, the finite element verification is skipped, otherwise calibration is performed with the simulation data; S33. Iterative optimization using initial full parameter space exploration and later local refined search; S34. Dynamically adjust material solutions: A built-in material database records the elastic modulus E and Poisson's ratio ν. Every 10 iterations, the E and ν of the current optimal solution are checked and the closest engineering material is matched from the library while ensuring that the constraints are not violated, that is, the maximum allowable stress and maximum allowable deformation must meet the material's allowable range.

2. A valve performance simulation and optimization method based on hybrid deep learning according to claim 1, characterized in that: The specific contents of step S1 include: S11. Build a 3D geometric model of the valve and define material properties, including thermal and structural parameters. S12. Use tetrahedral and triangular hybrid adaptive meshing to create a three-dimensional geometric model of the valve. S13. Use multiphysics coupling to set thermodynamic boundary conditions and structural mechanics boundaries, and configure the solver and convergence criteria. S14. Solve and perform post-processing to output the parameters of maximum thermal stress and maximum displacement of thermal deformation.

3. A valve performance simulation and optimization method based on hybrid deep learning according to claim 2, characterized in that: In step S11, the thermal parameters include thermal conductivity k, specific heat capacity cp, thermal expansion coefficient α; the structural parameters include elastic modulus E, Poisson's ratio ν, input pressure σ in ; In step S13, the thermodynamic boundary conditions are specifically: temperature field heat transfer parameters, fluid inlet temperature T in , thermal expansion coefficient α, outer wall convection heat transfer coefficient h cov And the ambient temperature T ref ; The structural mechanics boundary is specifically: according to the actual working conditions, the pressure load is applied to set the input pressure σ in , set fixed constraints and impose fluid dynamic boundary conditions; The configured solver is the direct coupling solver in the thermal stress module of COMSOL Multiphysics multi-physics simulation; the residual error and the maximum number of iterations are used as the convergence criteria.

4. The valve performance simulation and optimization method based on hybrid deep learning according to claim 1 is characterized in that: In step S2, the thermal-structural performance prediction model structure established is specifically as follows: The input layer is the material parameters and operating parameters. The material parameters include thermal conductivity k, thermal expansion coefficient α, elastic modulus E and Poisson's ratio ν; the operating parameters include input temperature T in , outer wall convection heat transfer coefficient h conv and input pressure σ in ; The hidden layer uses four convolutional layers with a kernel size of 3×3×3 to extract the three-dimensional temperature and structural features of the valve. The geometric and material parameters are encoded into a 128-dimensional vector. The CNN output is concatenated with the encoder vector and passed through three fully connected layers to generate the prediction result. The activation function of the hidden layer uses LeakyReLU, and the output layer is linearly activated; Output layer: Output parameters thermal stress and maximum displacement of thermal deformation.

5. The valve performance simulation and optimization method based on hybrid deep learning according to claim 1 is characterized in that: The specific content of training the neural network model is: Generate steady-state data through parameterized thermal-solid coupling simulation, perform data enhancement, and add Gaussian noise to the temperature field; Set the loss function to minimize the difference between the predicted value and the true value: L=λ1∣T max,pre -T max,sim ∣+λ2∣σ max,pre -s max,sim ∣ Among them, T max,pre To predict the maximum temperature, T is the maximum temperature prediction value of the valve in steady state output by the neural network model. max,sim is the maximum temperature of the simulation, is the highest temperature value of the valve obtained by finite element simulation, as the training label, σ max,pre is the predicted maximum thermal stress, is the predicted value of the valve thermal stress peak value output by the neural network model, σ max,sim is the simulated maximum thermal stress, is the maximum thermal stress value obtained by finite element simulation, is used as the training label, λ1 is the temperature error weight coefficient, and λ2 is the stress error weight coefficient.

6. The valve performance simulation and optimization method based on hybrid deep learning according to claim 1, characterized in that: Step S33, the specific content of full parameter space exploration is: Initial population generation: randomly generate n sets of design parameters, covering material parameters: elastic modulus E, Poisson's ratio ν and thermal expansion coefficient α, as well as operating parameters: inlet temperature T in and input pressure σ in ;Call the surrogate model to predict performance and select non-dominated solutions as the initial population; Adaptive sampling: Generate new parameters in sparse areas to supplement sampling points; The specific contents of local refined search are as follows: Focus on hotspot areas: select a preset proportion of individuals with the best performance in the current Pareto solution set and perform local search in the defined neighborhood; Hybrid optimization strategy: Gradient-assisted mutation: Calculate the objective function gradient for continuous variables of input pressure and input temperature; Directed crossover: Cross high-temperature, low-stress individuals with low-temperature, high-stress individuals to explore potential equilibrium points.

7. A valve performance simulation and optimization system based on hybrid deep learning, characterized in that: A valve performance simulation and optimization method based on hybrid deep learning according to any one of claims 1 to 6, comprising: a finite element simulation module, a thermal-structural performance prediction module, and a multi-objective optimization module; Finite element simulation module, used to perform finite element simulation analysis on valves and use multi-physics field thermal-solid coupling solution method to obtain the stress, deformation and temperature distribution data of valves under different working conditions, as well as the maximum thermal stress and maximum displacement of key valve components; The thermal-structural performance prediction module is used to train the neural network model after data processing of the simulation data, using the stress, deformation and temperature distribution data of the valve under different working conditions as input parameters and the maximum thermal stress and maximum displacement of thermal deformation as output parameters to establish a thermal-structural performance prediction model; The multi-objective optimization module is used to perform multi-objective optimization based on the prediction results of the thermal-structural performance prediction model, using a non-dominated genetic algorithm based on a reference point, combined with reference point generation and an adaptive agent model, to output the optimal parameter combination to adjust the valve design.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the valve performance simulation and optimization method based on hybrid deep learning as described in any one of claims 1 to 6.

9. A processing terminal comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, it implements the valve performance simulation and optimization method based on hybrid deep learning as described in any one of claims 1 to 6.

Images