Intelligent heat exchanger design method based on deep reinforcement learning

Abstract

The invention provides a heat exchanger intelligent design method based on deep reinforcement learning, and relates to the technical field of data processing. Establishing a flow model for simulating the flow path, speed distribution and pressure change of the fluid in the heat exchanger; according to the geometrical shape of the heat exchanger, the physical property of a heat exchanger material and fluid flowing in the heat exchanger, a heat exchange model is established by combining the thermodynamic principle, and the heat exchange efficiency under different fluid flowing conditions is calculated; the flow model and the heat exchange model are coupled, so that a coupling performance evaluation model for evaluating the overall performance of the heat exchanger under the specific working condition is obtained; and determining a state space, an action space and a reward function according to the flow heat exchange intensity coupling performance evaluation model. The design efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an intelligent design method for a heat exchanger based on deep reinforcement learning. Background Art

[0002] Traditional heat exchanger design methods mostly rely on empirical formulas and experimental verification, which not only has a long design cycle, but also makes it difficult to achieve optimal performance under complex and changing working conditions. In recent years, with the rapid development of computing technology, design methods based on numerical simulation have gradually become mainstream, but there are still some shortcomings in processing large-scale data and optimizing decisions.

[0003] In the field of heat exchanger design, deep reinforcement learning technology has great application potential, especially when dealing with complex systems with multivariable, nonlinear and time-varying characteristics. However, in practical applications, deep reinforcement learning methods still face some key challenges in data processing.

[0004] For example, reinforcement learning models require a large amount of sample data during training. However, in the field of heat exchanger design, high-quality experimental or simulation data is often difficult to obtain, which limits the application scope and performance improvement of reinforcement learning models. Summary of the invention

[0005] The present invention provides a heat exchanger intelligent design method based on deep reinforcement learning, which improves the design efficiency and optimizes the performance of the heat exchanger.

[0006] In order to solve the above technical problems, the technical solution of the present invention is as follows: A heat exchanger intelligent design method based on deep reinforcement learning, the method comprising: Determine heat exchanger geometry, including plate and shell and tube, and record key dimensions, including plate thickness, tube diameter, and tube spacing; Obtain the physical properties of heat exchanger materials, including thermal conductivity, density, and specific heat capacity; For the fluid flowing in the heat exchanger, obtain its thermodynamic and fluid mechanics properties, including viscosity, density, specific heat capacity, thermal conductivity and phase change characteristics; According to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, a flow model is established to simulate the flow path, velocity distribution and pressure change of the fluid inside the heat exchanger; According to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, combined with the principles of thermodynamics, a heat exchange model is established to calculate the heat exchange efficiency under different fluid flow conditions; The flow model and the heat transfer model are coupled to obtain a coupled performance evaluation model for evaluating the overall performance of the heat exchanger under specific operating conditions; Determine the state space, action space and reward function according to the flow heat transfer intensity coupling performance evaluation model; According to the state space, action space and reward function, a reinforcement learning model is constructed using the reinforcement learning algorithm, and the reinforcement learning model is trained to obtain the optimal strategy network; The operating parameters of the design condition are input into the strategy network to obtain the final structural design parameters.

[0007] Furthermore, according to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, a flow model is established to simulate the flow path, velocity distribution and pressure change of the fluid inside the heat exchanger, including: Receive and process the geometric and positional parameter data of the heat exchanger, including geometric dimensions, pipe layout and fin structure; Receive physical properties of fluids and materials, including density, viscosity, and thermal conductivity; Based on the input geometry data, a virtual representation of the heat exchanger is constructed in a simulation environment; Set initial conditions, such as initial velocity, pressure, and temperature distribution of the fluid, as the starting point of the simulation; Automatically identify and set the boundary conditions of the flow model, including the fluid properties of the inlet and outlet, the wall conditions and the thermal boundary conditions, to obtain the flow model; According to the simulation requirements, the Navier-Stokes equations and the finite volume solution method are selected to run the flow simulation. The flow equations are solved through iterative calculations to obtain the flow state of the fluid inside the heat exchanger.

[0008] Furthermore, according to the simulation requirements, the Navier-Stokes equations and the finite volume solver are selected to run the flow simulation. The flow equations are solved through iterative calculations to obtain the flow state of the fluid inside the heat exchanger, including: Obtain the geometric model data of the heat exchanger, including its shape, size and internal structure; set the physical parameters of the simulation, including the density, viscosity and thermal conductivity of the fluid; define the boundary conditions of the simulation, including the inlet flow rate, temperature, outlet pressure, and thermal boundary conditions of the wall; Generate a computational grid based on the geometry of the heat exchanger, which is used to discretize the Navier-Stokes equations; The finite volume method is used to discretize the Navier-Stokes equations and transform the continuous partial differential equations into discrete algebraic equations. According to the discrete algebraic equations, the flow state information of the fluid inside the heat exchanger is extracted, including velocity distribution, pressure distribution and temperature distribution.

[0009] Furthermore, a computational grid is generated according to the geometry of the heat exchanger. The computational grid is used to discretize the Navier-Stokes equations, including: Obtain all geometric data of the heat exchanger according to its geometric shape, including the diameter, length, arrangement of the pipe, the size, spacing, thickness of the fins, and the outer and inner walls of the heat exchanger; Generate computational mesh based on all geometric data of the heat exchanger; The geometric space inside the heat exchanger is divided into grid cells, and the physical quantities in each grid cell are regarded as constants; The steps to convert the Navier-Stokes equations from continuous differential form to the corresponding numerical form include: Each control volume contains a certain amount of fluid. The boundaries of the control volume are defined by mesh nodes. The Navier-Stokes equations are integrated within each control volume. The integrated equations are expressed as: ; in, is the velocity field, is the viscosity of the fluid, is an external force; is the volume region of the integration, representing a region in a three-dimensional space; is the velocity field About Time The partial derivative of is the unit normal vector; is the surface area of ​​the integration, representing the boundary surface; is the area of ​​the surface element; is the velocity field The gradient of is the dot product of the velocity gradient tensor with itself; is the gradient operator, which represents the rate of change in space.

[0010] Furthermore, the finite volume method is used to discretize the Navier-Stokes equations, transforming the continuous partial differential equations into discrete algebraic equations, including: The computational domain is divided into multiple control volumes, and the conservation quantity on each control volume is integrated to obtain its discretized algebraic equation; in two-dimensional or three-dimensional space, non-overlapping grids are used to divide the computational domain. In the two-dimensional case, the entire region is divided into multiple rectangular or quadrilateral control volumes; for each control volume, assume that its volume is V i , the surface is S i , each control volume has a normal vector on its outer boundary n i, and each control volume has a scalar value at its center; assuming that a conserved quantity is discretized, within the control volume, the conservation equation has the following form: ; in, is a conserved quantity, including velocity or temperature or mass; is the mass flux, including momentum flux or energy flux, etc.; is the source term, including body forces or heat sources; For a conserved quantity , integrate it in time and space, and apply Gauss's theorem to convert the volume integral of the flux into a surface integral; Time stepping using explicit time discretization methods; For the momentum equation of the fluid, the mass flux Expressed as the amount of fluid flowing over a surface; The fluxes are calculated based on the boundary conditions of the flow problem; the conservation equations for each control volume are discretized to obtain a system of algebraic equations.

[0011] Furthermore, Gauss's theorem is applied to convert the volume integral of the flux into the surface integral, including: For a conserved quantity , integrate it in time and space, and apply Gauss's theorem to convert the volume integral of the flux into the surface integral, and we get: ; in, is the control volume, is the outer boundary surface of the control volume, is the outward normal vector.

[0012] Furthermore, the explicit time discretization method is used for time stepping, and the corresponding calculation formula is: ; in, is the current time step The conservation of time, is the value of the next time step, is the time step.

[0013] Furthermore, the mass flux The calculation formula is: Assuming the calculated mass flux ,in, and represent the density and velocity of the fluid respectively; the upwind difference method is used to calculate the mass flux on the boundary, and the following approximation is obtained: ; in, By controlling the volume Mass flux on the surface; is the control volume Density inside; is the control volume The velocity vector inside; is the control volume Density inside; is the control volume The velocity vector inside.

[0014] Furthermore, the calculation process of the algebraic equation system includes: Assume that for a conserved quantity , and its discretized form in the control volume is: ; in, is the time step The value of the moment, is the flux, which represents the flow through the control volume and its neighbors of traffic, is the source term, which means that The volume source term in the control volume is solved by explicit method to obtain the conservation quantities of all control volumes in the whole computational domain. is the time step The value of the moment.

[0015] Furthermore, the flow model and the heat transfer model are coupled to obtain a coupled performance evaluation model for evaluating the overall performance of the heat exchanger under specific working conditions, including: Establish a data exchange interface so that the flow model and heat transfer model can recognize and receive the data output by each other; Determine the common parameters in the flow model and heat transfer model, including the temperature, pressure, and velocity of the fluid; Integrate the flow model and heat transfer model into one simulation environment so that the flow model and heat transfer model can run on the same time and space scale; Since the flow and heat transfer processes influence each other, in the coupling process, the final solution is found through iterative solution. In each iterative step, the flow model calculates the flow state of the fluid and transfers the flow state to the heat transfer model; the heat transfer model calculates the heat exchange according to the flow state and feeds the result back to the flow model. This is repeated until the predetermined number of iterations is reached to obtain the final coupling performance evaluation model.

[0016] The above solution of the present invention includes at least the following beneficial effects: Through an automated and intelligent design process, the present invention can significantly reduce the manual calculation and trial-and-error time required in traditional design methods, thereby greatly improving the design efficiency of the heat exchanger. During the training process, the deep reinforcement learning model can automatically explore and optimize the design parameters of the heat exchanger to maximize the reward function, thereby ensuring that the designed heat exchanger has excellent overall performance under specific working conditions. Since the reinforcement learning model has powerful self-learning and adaptive capabilities, the present invention can easily cope with various complex and changeable working conditions and provide a flexible and efficient solution for heat exchanger design. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a heat exchanger intelligent design method based on deep reinforcement learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0019] like Figure 1 As shown, an embodiment of the present invention proposes a heat exchanger intelligent design method based on deep reinforcement learning, the method comprising: Determine heat exchanger geometry, including plate and shell and tube, and record key dimensions, including plate thickness, tube diameter, and tube spacing; Obtain the physical properties of heat exchanger materials, including thermal conductivity, density, and specific heat capacity; For the fluid flowing in the heat exchanger, obtain its thermodynamic and fluid mechanics properties, including viscosity, density, specific heat capacity, thermal conductivity and phase change characteristics; According to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, a flow model is established to simulate the flow path, velocity distribution and pressure change of the fluid inside the heat exchanger; According to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, combined with the principles of thermodynamics, a heat exchange model is established to calculate the heat exchange efficiency under different fluid flow conditions; The flow model and the heat transfer model are coupled to obtain a coupled performance evaluation model for evaluating the overall performance of the heat exchanger under specific operating conditions; Determine the state space, action space and reward function according to the flow heat transfer intensity coupling performance evaluation model; According to the state space, action space and reward function, a reinforcement learning model is constructed using the reinforcement learning algorithm, and the reinforcement learning model is trained to obtain the optimal strategy network; The operating parameters of the design condition are input into the strategy network to obtain the final structural design parameters.

[0020] In an embodiment of the present invention, the traditional heat exchanger design method usually requires a lot of manual calculations and experimental verifications, which is time-consuming and inefficient. The present invention realizes the automation and intelligence of the design process by introducing deep reinforcement learning technology, thereby greatly improving the design efficiency and shortening the product development cycle. The deep reinforcement learning model can continuously try and error and learn optimization during the training process, and automatically find the best design parameter combination. This enables the present invention to design a heat exchanger with better performance under specific working conditions, thereby improving the heat exchange efficiency and overall performance of the product. The reinforcement learning model constructed by the present invention has a strong adaptive ability and can automatically adjust the design strategy according to different working conditions. This enables the method to flexibly respond to various complex and changeable design requirements and enhance the market competitiveness of the product. By reducing the number of physical tests and prototypes, as well as optimizing material selection and structural design, the present invention can effectively reduce production costs while ensuring design quality.

[0021] Determine the geometry of the heat exchanger, including plate and shell and tube, and record key dimensions, including plate thickness, tube diameter, and tube spacing; obtain the physical properties of the heat exchanger material, including thermal conductivity, density, and specific heat capacity; for the fluid flowing in the heat exchanger, obtain its thermodynamic and fluid dynamic properties, including viscosity, density, specific heat capacity, thermal conductivity, and phase change characteristics, including: Suppose you choose to design a shell and tube heat exchanger. The key dimensions include: Pipe diameter: For example, select a pipe diameter of 25 mm, which is a common size suitable for a variety of fluid and heat exchange scenarios.

[0022] Tube spacing: set to 30mm to ensure sufficient fluid passage and heat exchange area.

[0023] Plate thickness (for plate heat exchangers): If designing a plate heat exchanger, the plate thickness may be selected to be 0.5mm to provide good heat transfer performance and structural strength.

[0024] Assume that carbon steel is chosen as the main material for the heat exchanger. Its physical properties may include: Thermal conductivity: The thermal conductivity of carbon steel is usually between 45-50 W / (m·K), which indicates that it has good heat conduction capabilities. Density: The density of carbon steel is about 7850 kg / m³, which is necessary for calculating the weight and structural strength of the heat exchanger. Specific heat capacity: The specific heat capacity of carbon steel is about 0.46 kJ / (kg·K), which helps to understand the temperature change of the material during heat absorption or heat release. Obtain the thermodynamic and fluid dynamic properties of the fluid. In this step, collect the relevant properties of the fluid flowing in the heat exchanger.

[0025] Furthermore, according to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, a flow model is established to simulate the flow path, velocity distribution and pressure change of the fluid inside the heat exchanger, including: Receive and process the geometric and positional parameter data of the heat exchanger, including geometric dimensions, pipe layout and fin structure; Receive physical properties of fluids and materials, including density, viscosity, and thermal conductivity; Based on the input geometry data, a virtual representation of the heat exchanger is constructed in a simulation environment; Set initial conditions, such as initial velocity, pressure, and temperature distribution of the fluid, as the starting point of the simulation; Automatically identify and set the boundary conditions of the flow model, including the fluid properties of the inlet and outlet, the wall conditions and the thermal boundary conditions, to obtain the flow model; According to the simulation requirements, the Navier-Stokes equations and the finite volume solution method are selected to run the flow simulation. The flow equations are solved through iterative calculations to obtain the flow state of the fluid inside the heat exchanger.

[0026] In an embodiment of the present invention, by establishing a sophisticated flow model, the designer can more accurately predict the flow behavior of the fluid inside the heat exchanger, including the flow path, velocity distribution and pressure change. This helps to discover and correct potential problems in the design stage, thereby ensuring that the performance of the heat exchanger in actual operation meets expectations. The flow model can reveal the dynamic characteristics of the fluid in the heat exchanger, such as vortex, dead zone and backflow. By identifying these unfavorable flow modes, the designer can adjust the geometry or parameters of the heat exchanger in a targeted manner to optimize the flow state of the fluid and improve the heat exchange efficiency. Traditional heat exchanger design often requires a large number of physical tests to verify the performance. By establishing an accurate flow model, the operating state of the heat exchanger can be simulated on a computer, thereby reducing the reliance on physical tests. This not only reduces the test cost, but also shortens the product development cycle. By simulating and evaluating the performance of the heat exchanger under various working conditions through the flow model, potential failure modes or performance bottlenecks can be discovered in the early stages of the design. This helps to take improvement measures in a timely manner to improve the reliability and durability of the product. The flow model can respond quickly to design changes. When adjustments are needed to the heat exchanger geometry, materials or fluid parameters, the model can be quickly recalculated to reflect the new flow regime.

[0027] Receive and process the shape and position parameter data of the heat exchanger. The implementation process is as follows: Receive detailed data of the heat exchanger's geometric dimensions (such as length, width, height, pipe diameter, etc.), pipe layout (such as pipe spacing, arrangement, etc.) and fin structure (such as fin thickness, spacing, etc.) through the user interface or data file; check the integrity and rationality of the input data to ensure that all data are within the valid range and meet the design requirements of the heat exchanger; perform necessary unit conversion and format adjustment on the input data; save the processed shape and position parameter data in a suitable data structure, such as an array, list or object, so that it can be accessed at any time during the simulation process.

[0028] Receive the physical parameters of fluid and material. The implementation process is as follows: Receive the physical properties of the fluid (such as water, oil, etc.) and heat exchanger materials (such as carbon steel, stainless steel, etc.) through the user interface or database, including density, viscosity, thermal conductivity, etc. Verify that the input physical properties are accurate and ensure that all parameters use a unified unit system. Store the verified physical properties in a data structure so that these values ​​can be quickly accessed during simulation calculations.

[0029] Based on the input geometry data, a virtual representation of the heat exchanger is constructed in the simulation environment as follows: Using computer graphics or computational geometry techniques, a three-dimensional model or two-dimensional mesh representation of the heat exchanger is constructed in a simulation environment based on the input geometric size and shape data; if a numerical simulation method such as the finite volume method (FVM) is used, a computational mesh needs to be generated. This can be done through an automatic mesh generation algorithm or professional software to ensure the quality and adaptability of the mesh. Check whether the constructed virtual representation accurately reflects the geometric characteristics of the actual heat exchanger, and perform model correction or optimization if necessary.

[0030] According to the design requirements or actual operation scenarios, set the initial velocity, pressure, temperature distribution and other parameters of the fluid. These parameters can be used as the starting point for the simulation calculation. Verify whether the input initial conditions are reasonable and make adjustments as needed to ensure the accuracy and effectiveness of the simulation. Save the verified initial conditions in the simulation environment so that these values ​​can be loaded at the beginning of the simulation.

[0031] Automatically identify and set the boundary conditions of the flow model. The implementation process is as follows: Based on the geometric model and flow characteristics of the heat exchanger, the locations where boundary conditions need to be set, such as inlet, outlet, wall, etc., are automatically identified. According to the identification results, the corresponding fluid characteristics (such as velocity, pressure, temperature, etc.), wall conditions (such as no slip, partial slip, etc.) and thermal boundary conditions (such as constant temperature, constant heat flux, etc.) are set for each boundary position. Check whether the set boundary conditions meet the physical reality and simulation requirements, and adjust and optimize them according to the feedback. Apply the verified boundary conditions to the flow model to ensure that the simulation process can correctly reflect the actual flow of the fluid inside the heat exchanger.

[0032] Furthermore, according to the simulation requirements, the Navier-Stokes equations and the finite volume solver are selected to run the flow simulation. The flow equations are solved through iterative calculations to obtain the flow state of the fluid inside the heat exchanger, including: Obtain the geometric model data of the heat exchanger, including its shape, size and internal structure; set the physical parameters of the simulation, including the density, viscosity and thermal conductivity of the fluid; define the boundary conditions of the simulation, including the inlet flow rate, temperature, outlet pressure, and thermal boundary conditions of the wall; Generate a computational grid based on the geometry of the heat exchanger, which is used to discretize the Navier-Stokes equations; The finite volume method is used to discretize the Navier-Stokes equations and transform the continuous partial differential equations into discrete algebraic equations. According to the discrete algebraic equations, the flow state information of the fluid inside the heat exchanger is extracted, including velocity distribution, pressure distribution and temperature distribution.

[0033] In an embodiment of the present invention, by using the Navier-Stokes equations and the finite volume solution method, the flow behavior of the fluid in a complex geometric shape can be accurately simulated, so that the performance of the heat exchanger can be accurately predicted in the design stage. The simulation results provide detailed information on the fluid flow state, including velocity distribution, pressure distribution, and temperature distribution. This information helps to identify unfavorable areas in the flow, such as dead zones, vortices, or high pressure drop areas, thereby guiding designers to optimize the heat exchanger structure to improve fluid flow and heat exchange efficiency. Traditional heat exchanger design usually requires a large number of physical tests to verify performance. Through numerical simulation methods, these tests can be performed virtually on a computer, thereby significantly reducing the number of physical tests and related costs. This not only speeds up the product development cycle, but also reduces the overall development cost. Numerical simulation technology allows designers to explore a variety of different design options in a short period of time and evaluate their performance. This provides a powerful tool for product innovation, enabling designers to customize high-performance heat exchanger products according to customer needs or specific application scenarios.

[0034] Furthermore, a computational grid is generated according to the geometry of the heat exchanger. The computational grid is used to discretize the Navier-Stokes equations, including: According to the geometric shape of the heat exchanger, all its geometric data are obtained, including the diameter, length, arrangement of the pipe, the size, spacing, thickness of the fins, and the outer and inner walls of the heat exchanger; according to all the geometric data of the heat exchanger, a calculation grid is generated; the geometric space inside the heat exchanger is divided into grid cells, and the physical quantities in each grid cell are regarded as constants, including: Use measuring tools (such as calipers, laser rangefinders, etc.) or design drawings to obtain detailed geometric data of the heat exchanger; record the diameter, length, arrangement (such as parallel, staggered, etc.) of the pipe; measure the size (such as length, width), spacing and thickness of the fins; determine the size and shape of the outer and inner walls of the heat exchanger. Organize the collected data into a standard format, such as a spreadsheet or CAD file, to ensure the accuracy and completeness of the data; select a suitable mesh generation tool, such as professional CFD (computational fluid dynamics) pre-processing software, based on the complexity and calculation requirements of the heat exchanger; import the organized geometric data into the mesh generation tool, and simplify or repair the geometric model as needed to ensure smooth mesh generation; define the density, type and distribution of the mesh to adapt to the geometric characteristics and flow characteristics of the heat exchanger, and set finer meshes in key areas (such as areas with complex fluid flow or high heat transfer efficiency) to improve the accuracy of the simulation; run the mesh generation tool according to the settings The computational grid is automatically generated based on the parameters of the simulation, and the quality of the generated grid is checked to ensure that there are no deformed or overly distorted grid cells to ensure the stability and accuracy of the calculation. In the generated grid, each grid cell represents a control volume, which is the basic calculation unit in fluid dynamics simulation. The boundary of the control volume is defined by the nodes and edges of the grid. In each control volume, continuous physical quantities (such as velocity, pressure, temperature, etc.) are discretized into constants representing the average value in the volume. This discretization method allows complex partial differential equations to be converted into algebraic equations that are easier to solve. The physical quantity data of each grid cell is stored in a suitable data structure for efficient access and update during the simulation process.

[0035] The steps to convert the Navier-Stokes equations from continuous differential form to the corresponding numerical form include: Each control volume contains a certain amount of fluid. The boundaries of the control volume are defined by mesh nodes. The Navier-Stokes equations are integrated within each control volume. The integrated equations are expressed as: ; in, is the velocity field, is the viscosity of the fluid, is an external force; is the volume region of the integration, representing a region in a three-dimensional space; is the velocity field About Time The partial derivative of is the unit normal vector; is the surface area of ​​the integration, representing the boundary surface; is the area of ​​the surface element; is the velocity field The gradient of is the dot product of the velocity gradient tensor with itself; is the gradient operator, which represents the rate of change in space.

[0036] In an embodiment of the present invention, by obtaining all geometric data of the heat exchanger in detail, including the diameter, length, arrangement of the pipe, the size, spacing, thickness of the fins, and the outer and inner walls of the heat exchanger, a highly accurate geometric model of the heat exchanger can be established. This makes it possible to simulate the flow and heat transfer of heat exchangers with complex internal structures, thereby more accurately predicting their performance. The generated computational grid can reasonably divide the geometric space inside the heat exchanger into grid units, and the physical quantities in each grid unit are regarded as constants. This discretization processing method not only simplifies complex flow problems, but also makes the numerical solution process more efficient, while optimizing the utilization of computing resources. By integrating the Navier-Stokes equations in each control volume, the continuous fluid dynamics problem can be accurately converted into a discrete numerical problem. This conversion retains the main physical properties of the original equation, such as the inertia, viscosity and external force of the fluid, so that the flow state of the fluid inside the heat exchanger can be more realistically reflected. With the help of accurate numerical simulation results, designers can gain an in-depth understanding of the flow details of the fluid in the heat exchanger, such as velocity distribution, pressure changes and heat transfer efficiency. Traditional heat exchanger design often requires multiple physical tests to verify performance. However, through numerical simulation methods, these tests can be performed virtually on the computer, which significantly shortens the R&D cycle and reduces test costs. At the same time, numerical simulation also allows multiple schemes to be compared and optimized during the design phase, further improving the efficiency and accuracy of the design.

[0037] Furthermore, the finite volume method is used to discretize the Navier-Stokes equations, transforming the continuous partial differential equations into discrete algebraic equations, including: The computational domain is divided into multiple control volumes, and the conservation quantity on each control volume is integrated to obtain its discretized algebraic equation; in two-dimensional or three-dimensional space, non-overlapping grids are used to divide the computational domain. In the two-dimensional case, the entire region is divided into multiple rectangular or quadrilateral control volumes; for each control volume, assume that its volume is V i , the surface is S i , each control volume has a normal vector on its outer boundary n i , and each control volume has a scalar value at its center; assuming that a conserved quantity is discretized, within the control volume, the conservation equation has the following form: ; in, is a conserved quantity, including velocity or temperature or mass; is the mass flux, including momentum flux or energy flux, etc.; is the source term, including body forces or heat sources; For a conserved quantity , integrate it in time and space, and apply Gauss's theorem to convert the volume integral of the flux into a surface integral; Time stepping using explicit time discretization methods; For the momentum equation of the fluid, the mass flux Expressed as the amount of fluid flowing over a surface; The fluxes are calculated based on the boundary conditions of the flow problem; the conservation equations for each control volume are discretized to obtain a system of algebraic equations.

[0038] In an embodiment of the present invention, the finite volume method integrates the physical quantities in the control volume based on the conservation principle, thereby ensuring the numerical stability and convergence in the discretization process. This means that in the solution process, the numerical solution can gradually approach the true solution, reducing the calculation error and uncertainty. Since the finite volume method strictly satisfies the conservation equations in each control volume, it can accurately capture the local changes in the fluid flow process, such as velocity gradients, pressure mutations, etc. This guarantee of local conservation makes the simulation results more realistic and reliable. The finite volume method is applicable to grids of various shapes, including structured grids and unstructured grids. This makes the method highly flexible and versatile when dealing with heat exchangers with complex geometries. At the same time, it can also easily cope with different boundary conditions and physical parameter settings. By discretizing the Navier-Stokes equations and converting continuous partial differential equations into discrete algebraic equations, the difficulty of solution and the complexity of calculation can be significantly reduced. This means that under the same computing resources, the finite volume method can give simulation results faster and improve the computing efficiency.

[0039] Furthermore, Gauss's theorem is applied to convert the volume integral of the flux into the surface integral, including: For a conserved quantity , integrate it in time and space, and apply Gauss's theorem to convert the volume integral of the flux into the surface integral, and we get: ; in, is the control volume, is the outer boundary surface of the control volume, is the outward normal vector.

[0040] In an embodiment of the present invention, by converting the volume integral into the surface integral, the complexity and amount of calculation in the calculation can be significantly reduced. In numerical simulation, the integration of the surface area is usually easier to implement than the integration of the volume, because the dimension of the surface area is lower than that of the volume, thereby simplifying the process of numerical integration. The calculation of the surface integral usually involves fewer grid points and calculation elements, so it is faster and has lower memory usage than the volume integral. This means that with the same computing resources, larger or more sophisticated simulation problems can be processed, thereby improving the overall computing efficiency. Gauss's theorem is one of the basic principles in physics, which ensures that the conservation properties of the fluid within the control volume (such as mass, momentum, energy, etc.) are strictly observed during the discretization process. By applying Gauss's theorem, it can be ensured that the numerical simulation results are physically reasonable and reliable. Converting continuous partial differential equations into discrete algebraic equations based on control volumes, combined with the application of Gauss's theorem, helps to enhance the stability of numerical solutions.

[0041] Furthermore, the explicit time discretization method is used for time stepping, and the corresponding calculation formula is: ; in, is the current time step The conservation of time, is the value of the next time step, is the time step. The mass flux The calculation formula is: Assuming the calculated mass flux ,in, and represent the density and velocity of the fluid respectively; the upwind difference method is used to calculate the mass flux on the boundary, and the following approximation is obtained: ; in, By controlling the volume Mass flux on the surface; is the control volume Density inside; is the control volume The velocity vector inside; is the control volume Density inside; is the control volume The velocity vector inside.

[0042] In an embodiment of the present invention, the explicit time discretization method allows the solution of the next time step to be directly calculated at each time step without solving a complex linear system. This method is simple and fast to calculate, and can significantly improve the calculation efficiency, especially when dealing with large-scale problems. The algorithm structure of the explicit method is relatively simple and easy to program. In addition, since the calculation of each time step does not depend on the results of other time steps, the explicit method is easier to parallelize, thereby further accelerating the calculation process. Compared with the implicit method, the explicit time discretization method usually does not need to store a large number of coefficient matrices or intermediate variables, thereby reducing memory usage. This enables larger-scale problems to be processed in a computing environment with limited resources. Under a suitable time step, the explicit method can provide a stable numerical solution. By reasonably selecting the time step Δt, the numerical stability during the simulation process can be ensured, and the divergence or non-physical behavior of the solution can be avoided. As a boundary processing method, the upwind difference method can flexibly process the mass flux of the fluid at the boundary. It selects the upstream physical quantity to calculate the flux according to the flow direction of the fluid, thereby more accurately simulating the behavior of the fluid at the boundary.

[0043] Furthermore, the calculation process of the algebraic equation system includes: Assume that for a conserved quantity , and its discretized form in the control volume is: ; in, is the time step The value of the moment, is the flux, which represents the flow through the control volume and its neighbors of traffic, is the source term, which means that The volume source term in the control volume is solved by explicit method to obtain the conservation quantities of all control volumes in the whole computational domain. is the time step The value of the moment.

[0044] In an embodiment of the present invention, the explicit method allows the solution of the next time step to be directly calculated without solving a complex linear system or iterative process, thereby significantly improving the computational efficiency. Since the explicit method does not need to store a large number of coefficient matrices or intermediate variables, it takes up less memory, which is particularly important for large-scale computing or limited resources. The algorithm structure of the explicit method is relatively simple and easy to program. In addition, the calculation of each time step is independent, which makes parallelization easy and can further accelerate the calculation process. The explicit method has high flexibility in dealing with complex boundary conditions and source terms, and can more easily adapt to different physical models and scenarios. At an appropriate time step, the explicit method can provide a stable numerical solution. By reasonably selecting the time step, the numerical stability during the simulation process can be ensured to avoid the divergence or non-physical behavior of the solution.

[0045] According to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, combined with the principles of thermodynamics, a heat exchange model is established to calculate the heat exchange efficiency under different fluid flow conditions, including: The heat exchanger consists of two isolated channels, one for the flow of two fluids (e.g., hot fluid and cold fluid); the fluids flow in one dimension within their respective channels; the walls of the heat exchanger are uniform; the physical properties of the fluids (e.g., specific heat capacity, density, etc.) remain constant within their respective operating temperature ranges, and the logarithmic mean temperature difference (LMTD) method is used to estimate the temperature difference between the fluids.

[0046] For each fluid, calculate its heat capacity ratio, which will be used to determine how much heat the fluid can absorb or release. Calculate the average temperature difference between the two fluids using the LMTD method. LMTD takes into account the temperature variation of the fluids at the inlet and outlet of the heat exchanger and provides a single value that represents the temperature difference across the heat exchanger.

[0047] The total heat exchange is calculated using the thermal conductivity of the heat exchanger material, the surface area of ​​the heat exchanger (calculated based on the geometric dimensions), and the LMTD. The heat exchange efficiency can be defined as the ratio of the actual amount of heat transferred to the maximum amount of heat that could be transferred theoretically. In this model, the actual amount of heat transferred can be compared to the maximum possible amount of heat calculated based on the fluid inlet temperature difference and the minimum heat capacity ratio, and the calculated total heat exchange and heat exchange efficiency are output.

[0048] Furthermore, the flow model and the heat transfer model are coupled to obtain a coupled performance evaluation model for evaluating the overall performance of the heat exchanger under specific working conditions, including: Establish a data exchange interface so that the flow model and heat transfer model can recognize and receive the data output by each other; Determine the common parameters in the flow model and heat transfer model, including the temperature, pressure, and velocity of the fluid; Integrate the flow model and heat transfer model into one simulation environment so that the flow model and heat transfer model can run on the same time and space scale; Since the flow and heat transfer processes influence each other, in the coupling process, the final solution is found through iterative solution. In each iterative step, the flow model calculates the flow state of the fluid and transfers the flow state to the heat transfer model; the heat transfer model calculates the heat exchange according to the flow state and feeds the result back to the flow model. This is repeated until the predetermined number of iterations is reached to obtain the final coupling performance evaluation model.

[0049] In an embodiment of the present invention, by coupling the flow model and the heat exchange model, the actual physical process in the heat exchanger can be simulated more accurately. This method that comprehensively considers the mutual influence of flow and heat exchange can more accurately predict the performance of the heat exchanger, including heat exchange efficiency, fluid flow characteristics, etc. The coupled performance evaluation model can provide a comprehensive performance analysis for the design of the heat exchanger. Designers can optimize the structure, material selection, fluid flow path, etc. of the heat exchanger based on the simulation results to improve overall performance and reduce costs. Traditional heat exchanger performance evaluation usually relies on a large number of experimental tests. The coupled performance evaluation model can simulate the performance of the heat exchanger under various working conditions on a computer, thereby reducing the need for physical experiments and saving time, resources and costs. By quickly simulating and evaluating the performance of different design schemes, companies can find the best solution more quickly, accelerate the development process of new products, and improve market competitiveness.

[0050] The above steps are achieved by following these steps: Define the data format and determine the data format transmitted between the flow model and the heat transfer model. This includes determining the data type (such as floating point numbers, vectors, matrices, etc.), units (such as meters, seconds, degrees Celsius, etc.), and data structures (such as arrays, objects, etc.). In the flow model and heat transfer model, create interface functions for sending and receiving data respectively. These functions should be able to handle data format conversion, error checking, and synchronous or asynchronous data transmission.

[0051] Data transmission between the flow model and the heat transfer model is achieved through file exchange, memory sharing, network communication (such as using TCP / IP protocol), etc. The choice of which method depends on the complexity of the model, the distribution of computing resources, and the real-time requirements of the simulation. Before integration, unit test the interface to ensure that the data can be correctly transferred between the two models and there is no data loss or format error.

[0052] Analyze the input and output of the flow model and the heat transfer model, and identify the common parameters, such as the temperature, pressure, and velocity of the fluid. Define clear physical meanings, units, and value ranges for each common parameter. Ensure that these definitions are consistent in the two models. In the data exchange interface, establish a mapping relationship between the common parameters of the flow model and the heat transfer model. In this way, when one model updates a parameter, the other model can immediately perceive the change. Design a mechanism to ensure that the common parameters are synchronized between the two models during the simulation process, which requires data exchange and update at the end of each simulation step.

[0053] Select a simulation platform that supports the integration of flow model and heat transfer model, such as ANSYS Workbench, COMSOL Multiphysics or other professional CFD (computational fluid dynamics) and heat transfer simulation software. Import the established flow model and heat transfer model into the simulation platform. This may require converting the model into a specific format supported by the simulation platform. In the simulation platform, set the simulation time step, spatial resolution, and other relevant simulation parameters to ensure that the flow model and heat transfer model run on the same time and space scales. Use the tools or scripts provided by the simulation platform to connect the flow model and the heat transfer model to form an overall coupled model. Ensure that the data exchange interface can work properly and that the two models can recognize and receive each other's data. Set initial conditions for the flow model and the heat transfer model, such as the initial temperature, pressure, velocity distribution, etc. of the fluid. In each iteration step, first run the flow model to calculate the flow state of the fluid, including the velocity field, pressure field, etc. The flow state data calculated by the flow model is passed to the heat transfer model through the data exchange interface.

[0054] Based on the received flow state data, the heat exchange model calculates the heat exchange conditions, including temperature distribution, heat flow, etc. The heat exchange data calculated by the heat exchange model is fed back to the flow model, and the boundary conditions or physical parameters of the flow model (such as temperature-dependent fluid viscosity) are updated. Check whether the predetermined number of iterations has been reached or the convergence conditions have been met. If the conditions are met, the iteration is stopped; otherwise, it is returned to continue the iteration. After the iteration is completed, the final coupling performance evaluation results are output, including the overall heat exchange efficiency of the heat exchanger, fluid flow characteristics, and other performance indicators of interest.

[0055] The following is the detailed implementation process of each step: The process of determining the state space, action space and reward function is as follows: Define the state space, which is the set of all possible values ​​that describe the current state of the system; in the flow-heat transfer intensity coupling performance evaluation model, the state space can include the real-time values ​​of key physical quantities such as fluid temperature, pressure, and velocity, as well as the structural parameters of the heat exchanger (such as pipe diameter, fin spacing, etc.). These state variables will be used to describe the complete state of the heat exchanger at a certain moment.

[0056] Define the action space. The action space is the set of all possible actions that the agent can take. In this scenario, the action space can be defined as the adjustment of the structural design parameters of the heat exchanger, such as changing the pipe diameter, adjusting the fin spacing, etc. Each action corresponds to a specific structural adjustment aimed at optimizing the performance of the heat exchanger.

[0057] Define a reward function, which is used to evaluate the performance of the agent after taking an action in a certain state. In the flow-heat transfer intensity coupling performance evaluation model, the reward function can be defined based on performance indicators such as heat transfer efficiency, pressure drop, and fluid flow stability. For example, if an action leads to an increase in heat transfer efficiency, a positive reward is given; conversely, if the performance decreases, a negative reward is given.

[0058] The reinforcement learning algorithm is used to build a reinforcement learning model. The implementation process is as follows: According to the characteristics of the problem (such as the dimensions of the state space and action space, whether it is continuous, etc.), select the appropriate reinforcement learning algorithm, such as Q-learning, Deep Q-Network (DQN), Proximal Policy Optimization (PPO), etc. Use a deep learning framework (such as TensorFlow or PyTorch) to build a neural network model, which will be used to approximate the value function or policy function; for the value function method (such as DQN), the network outputs the state or action state value; for the policy gradient method (such as PPO), the network outputs the action probability distribution; randomly initialize the weights and biases of the neural network, or use a pre-trained model for initialization.

[0059] Perform reinforcement learning model training in an interactive environment to obtain the optimal strategy network. The implementation process is as follows: Create a simulator or actual system as an interactive environment to simulate the operation of the heat exchanger under various working conditions; the environment should be able to provide state information and return new states and rewards based on the agent's actions; in each training iteration, the agent selects an action based on the current strategy and applies it to the environment; the environment executes the action and returns a new state and reward, and the agent uses the new state and reward to update its policy network or value function network, repeating this process until the stopping condition is met (such as reaching the maximum number of iterations, the policy performance is stable, etc.). During the training process, the performance of the policy is evaluated regularly, and the parameters of the policy network with the best performance are recorded. After the training is completed, the policy network that achieves the best performance is saved for subsequent use.

[0060] The working condition parameters of the design condition are input into the strategy network to obtain the final structural design parameters. The implementation process is as follows: According to the actual application scenario, determine the design operating parameters of the heat exchanger, such as inlet fluid temperature, flow rate, target heat transfer efficiency, etc.; input the design operating parameters as states into the trained optimal strategy network; the strategy network will output the optimized structural design parameter adjustments under the current operating conditions, apply these adjustments to the initial structural design, obtain the final structural design parameters, and use simulation or experimental methods to verify the performance of the heat exchanger based on the structural design parameters output by the strategy network.

[0061] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A heat exchanger intelligent design method based on deep reinforcement learning, characterized in that: The method comprises: Determine heat exchanger geometry, including plate and shell and tube, and record key dimensions, including plate thickness, tube diameter, and tube spacing; Obtain the physical properties of heat exchanger materials, including thermal conductivity, density, and specific heat capacity; For the fluid flowing in the heat exchanger, obtain its thermodynamic and fluid mechanics properties, including viscosity, density, specific heat capacity, thermal conductivity and phase change characteristics; According to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, a flow model is established to simulate the flow path, velocity distribution and pressure change of the fluid inside the heat exchanger; According to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, combined with the principles of thermodynamics, a heat exchange model is established to calculate the heat exchange efficiency under different fluid flow conditions; The flow model and the heat transfer model are coupled to obtain a coupled performance evaluation model for evaluating the overall performance of the heat exchanger under specific operating conditions; Determine the state space, action space and reward function according to the flow heat transfer intensity coupling performance evaluation model; According to the state space, action space and reward function, the reinforcement learning algorithm is used to construct a reinforcement learning model, and the reinforcement learning model is trained to obtain the optimal strategy network; the working condition parameters of the design condition are input into the strategy network to obtain the final structural design parameters.

2. The heat exchanger intelligent design method based on deep reinforcement learning according to claim 1 is characterized in that: According to the geometric shape of the heat exchanger, the physical properties of the heat exchanger material and the fluid flowing in the heat exchanger, a flow model is established to simulate the flow path, velocity distribution and pressure change of the fluid inside the heat exchanger, including: Receive and process the geometric and positional parameter data of the heat exchanger, including geometric dimensions, pipe layout and fin structure; Receive physical properties of fluids and materials, including density, viscosity, and thermal conductivity; Based on the input geometry data, a virtual representation of the heat exchanger is constructed in a simulation environment; Set initial conditions, such as initial velocity, pressure, and temperature distribution of the fluid, as the starting point of the simulation; Automatically identify and set the boundary conditions of the flow model, including the fluid properties of the inlet and outlet, the wall conditions and the thermal boundary conditions, to obtain the flow model; According to the simulation requirements, the Navier-Stokes equations and the finite volume solution method are selected to run the flow simulation. The flow equations are solved through iterative calculations to obtain the flow state of the fluid inside the heat exchanger.

3. The heat exchanger intelligent design method based on deep reinforcement learning according to claim 2 is characterized in that: Select the Navier-Stokes equations and the finite volume solver according to the simulation requirements, run the flow simulation, solve the flow equations through iterative calculations, and obtain the flow state of the fluid inside the heat exchanger, including: Obtain the geometric model data of the heat exchanger, including its shape, size and internal structure; set the physical parameters of the simulation, including the density, viscosity and thermal conductivity of the fluid; define the boundary conditions of the simulation, including the inlet flow rate, temperature, outlet pressure, and thermal boundary conditions of the wall; Generate a computational grid based on the geometry of the heat exchanger, which is used to discretize the Navier-Stokes equations; The finite volume method is used to discretize the Navier-Stokes equations and transform the continuous partial differential equations into discrete algebraic equations. According to the discrete algebraic equations, the flow state information of the fluid inside the heat exchanger is extracted, including velocity distribution, pressure distribution and temperature distribution.

4. The heat exchanger intelligent design method based on deep reinforcement learning according to claim 3 is characterized in that: Based on the geometry of the heat exchanger, a computational mesh is generated. The computational mesh is used to discretize the Navier-Stokes equations, including: Obtain all geometric data of the heat exchanger according to its geometric shape, including the diameter, length, arrangement of the pipe, the size, spacing, thickness of the fins, and the outer and inner walls of the heat exchanger; Generate computational mesh based on all geometric data of the heat exchanger; The geometric space inside the heat exchanger is divided into grid cells, and the physical quantities in each grid cell are regarded as constants; The steps to convert the Navier-Stokes equations from continuous differential form to the corresponding numerical form include: Each control volume contains a certain amount of fluid. The boundaries of the control volume are defined by mesh nodes. The Navier-Stokes equations are integrated within each control volume. The integrated equations are expressed as: ； in, is the velocity field, is the viscosity of the fluid, is an external force; is the volume region of the integration, representing a region in a three-dimensional space; is the velocity field About Time The partial derivative of is the unit normal vector; is the surface area of ​​the integration, representing the boundary surface; is the area of ​​the surface element; is the velocity field The gradient of is the dot product of the velocity gradient tensor with itself; is the gradient operator, which represents the rate of change in space.

5. The heat exchanger intelligent design method based on deep reinforcement learning according to claim 4 is characterized in that: The finite volume method is used to discretize the Navier-Stokes equations and transform the continuous partial differential equations into discrete algebraic equations, including: The computational domain is divided into multiple control volumes, and the conservation quantity on each control volume is integrated to obtain its discretized algebraic equation; in two-dimensional or three-dimensional space, non-overlapping grids are used to divide the computational domain. In the two-dimensional case, the entire region is divided into multiple rectangular or quadrilateral control volumes; for each control volume, assume that its volume is V i , the surface is S i , each control volume has a normal vector on its outer boundary n i , and each control volume has a scalar value at its center; assuming that a conserved quantity is discretized, within the control volume, the conservation equation has the following form: ； in, is a conserved quantity, including velocity or temperature or mass; is the mass flux, including momentum flux or energy flux, etc.; is the source term, including body forces or heat sources; For a conserved quantity , integrate it in time and space, and apply Gauss's theorem to convert the volume integral of the flux into a surface integral; Time stepping using explicit time discretization methods; For the momentum equation of the fluid, the mass flux Expressed as the amount of fluid flowing over a surface; The fluxes are calculated based on the boundary conditions of the flow problem; the conservation equations for each control volume are discretized to obtain a system of algebraic equations.

6. The heat exchanger intelligent design method based on deep reinforcement learning according to claim 5 is characterized in that: Apply Gauss's theorem to convert the volume integral of the flux into a surface integral, including: For a conserved quantity , integrate it in time and space, and apply Gauss's theorem to convert the volume integral of the flux into the surface integral, and we get: ； in, is the control volume, is the outer boundary surface of the control volume, is the outward normal vector.

7. The heat exchanger intelligent design method based on deep reinforcement learning according to claim 6 is characterized in that: The corresponding calculation formula for time stepping using the explicit time discretization method is: ； in, is the current time step The conservation of time, is the value of the next time step, is the time step.

8. The heat exchanger intelligent design method based on deep reinforcement learning according to claim 7 is characterized in that: Mass flux The calculation formula is: Assuming the calculated mass flux ,in, and represent the density and velocity of the fluid respectively; the upwind difference method is used to calculate the mass flux on the boundary, and the following approximation is obtained: ； in, By controlling the volume Mass flux on the surface; is the control volume Density inside; is the control volume The velocity vector inside; is the control volume Density inside; is the control volume The velocity vector inside.

9. The heat exchanger intelligent design method based on deep reinforcement learning according to claim 8, characterized in that: The calculation process of the algebraic equation system includes: Assume that for a conserved quantity , and its discretized form in the control volume is: ； in, is the time step The value of the moment, is the flux, which represents the flow through the control volume and its neighbors of traffic, is the source term, which means that The volume source term in the control volume is solved by explicit method to obtain the conservation quantities of all control volumes in the whole computational domain. is the time step The value of the moment.

10. The heat exchanger intelligent design method based on deep reinforcement learning according to claim 9, characterized in that: The flow model and the heat transfer model are coupled to obtain a coupled performance evaluation model for evaluating the overall performance of the heat exchanger under specific operating conditions, including: Establish a data exchange interface so that the flow model and heat transfer model can recognize and receive the data output by each other; Determine the common parameters in the flow model and heat transfer model, including the temperature, pressure, and velocity of the fluid; Integrate the flow model and heat transfer model into one simulation environment so that the flow model and heat transfer model can run on the same time and space scale; Since the flow and heat transfer processes influence each other, in the coupling process, the final solution is found through iterative solution. In each iterative step, the flow model calculates the flow state of the fluid and transfers the flow state to the heat transfer model; the heat transfer model calculates the heat exchange according to the flow state and feeds the result back to the flow model. This is repeated until the predetermined number of iterations is reached to obtain the final coupling performance evaluation model.

Images