An improved method, medium and program product for topology optimization of initial layout with extraction of mainstream features under complex heat dissipation conditions

Abstract

The present invention discloses an improved method, medium, and program product for topological optimization of an initial layout that extracts mainstream features under complex heat dissipation conditions. The method belongs to the interdisciplinary field of thermal management and structural topology optimization. The method is oriented to the design of liquid-cooled heat dissipation structures under high Reynolds number turbulence and conjugate heat transfer conditions. By constructing a two-dimensional array-type parameterized initial layout, generating high-fidelity sample data based on CFD simulation, training a neural network prediction model with an SE attention mechanism, combining the modified objective function under the flow field freezing assumption with the ETO optimization algorithm to screen initialization schemes, and then implementing topological optimization evolution based on the variable density method on the basis of the optimized initial state. The present invention can improve the final effect of topological optimization while reducing computational complexity, achieve the coordinated optimization of heat dissipation efficiency and flow performance, and provide an effective means for the conjugate heat transfer design of liquid-cooled heat dissipation structures under complex conditions of high Reynolds number turbulence.

Description

Technical Field

[0001] The present invention belongs to the interdisciplinary technical field of thermal management and structural topology optimization, and relates to liquid-cooled conjugate heat transfer optimization technology under high Reynolds number turbulence, and in particular to an improved method, medium and program product for topology optimization of an initialization layout that extracts mainstream features under complex heat dissipation conditions. Background Art

[0002] Currently, heat dissipation strategies for electronic devices primarily include passive methods such as natural convection and phase change cooling, as well as active methods such as forced air cooling and forced liquid cooling. Compared to these alternatives, liquid cooling heat sinks, which utilize a highly conductive liquid medium, offer greater compactness, higher heat transfer rates, and lower costs, making them a popular choice for high-heat flux scenarios.

[0003] The structural parameters and geometry of the cooling channels in liquid cooling systems are the primary factors influencing the flow state, heat and mass transfer characteristics, and the distribution of related physical state variables such as temperature and flow rate. Existing technologies use bionic or fractal methods to improve microchannel structures. However, such methods rely on design experience and trial-and-error strategies and lack universal applicability. Furthermore, while some cooling channels employ optimization algorithms to improve performance and design efficiency, their essence is still to change the distribution of the physical field by optimizing the structure and to seek the optimal solution based on changes in performance parameters. This iterative calculation process still requires a significant amount of time and resources.

[0004] In this context, topology optimization (TO) methods are considered an excellent choice for targeted cooling channel design due to their high degree of design freedom, low empirical dependence, and strong physical orientation. The construction and basic principles of the TO model are based on an optimization algorithm. Given loads, control equations, boundary constraints, and optimization variables, the algorithm seeks the optimal structural configuration of the material position and density values ​​of the mesh nodes within the design domain, thereby maximizing or minimizing the objective function value. The TO method, which uses mesh elements as the optimization object for cooling channel design, has a very high degree of design freedom and allows for simultaneous optimization of size, shape, and topology from a global perspective across the entire design domain.

[0005] However, applying the TO method to fluid-solid conjugate heat transfer problems involving high-Reynolds-number turbulence still faces challenges. Most existing research on fluid topology optimization focuses on laminar flows at low Reynolds numbers. For high-Reynolds-number turbulence, due to the extreme complexity of the flow mechanisms, direct topology optimization can easily lead to strong numerical oscillations, convergence difficulties, and even physically distorted optimization results. To circumvent these issues, existing technologies typically adopt two simplification approaches: one is to use simplified models (such as the Darcy model) to represent turbulence, but this sacrifices the accuracy of the physical model and limits the performance ceiling of the final optimization results; the other is to stabilize the optimization process through complex penalty coefficient adjustments and interpolation method settings, but this requires extensive debugging and is time-consuming and labor-intensive. Furthermore, some attempts to use preset structures (such as three-periodic minimal surfaces) to fill the intermediate density region reintroduce reliance on human experience, and their performance adaptability under varying operating conditions is questionable.

[0006] In the topology optimization process of conjugate heat transfer, in order to balance computing resources and results, some technical solutions attempt to use artificial intelligence methods, hoping to directly generate results under new models through a large amount of existing data, or to construct a reasonable initial layout by exploring various biomimetic methods, hoping to generate better results. However, the current topology optimization of turbulent conjugate heat transfer under high Reynolds numbers is itself a difficult problem. There are very few available databases, and most of them are composed of low-fidelity optimization, which limits its optimization space. Traditional biomimetic methods generally have their limitations. The initial layout they provide has considerable human factors, requiring multiple trial and error. Moreover, relatively good results can only be achieved when the biomimetic object and the research object are somewhat similar, and they lack good universality.

[0007] In summary, under high heat flux and high Reynolds number flow conditions, how to effectively identify and guide the dominant structural characteristics of heat dissipation, improve the physical rationality and computational stability of the topology optimization process, and thus obtain a better heat transfer-flow synergistic structure within limited computing resources, is a technical problem that needs to be urgently solved in the current complex heat dissipation topology optimization design. Summary of the Invention

[0008] (1) Purpose of the invention

[0009] To address the aforementioned shortcomings and deficiencies of the prior art, the present invention aims to provide an improved method, medium, and program product for topology optimization of an initial layout that extracts mainstream features under complex heat dissipation conditions. Combining the advantages of artificial intelligence methods and advanced optimization algorithms, this method first directly improves the initial flow field by constructing a universal array layout framework. The optimized initial layout is then used to guide the topological direction of turbulent conjugate heat transfer in complex flow conditions. By integrating an attention-based neural network (Squeeze-and-Excitation Net, SE-Net) with an exponential-trigonometric optimization algorithm (ETO), a framework for multi-objective topology optimization of the initial layout of turbulent thermal fluid systems is established. This framework effectively avoids the extensive trial-and-error costs and intense numerical challenges of the optimization process, achieving remarkably good results with a low mesh count and without requiring extensive continuous changes in topological parameters during the topology process. This invention is of great significance for the innovation of traditional turbulent conjugate heat transfer optimization methods and provides a feasible path and methodological reference for the selection and establishment of benchmark models for multi-objective topology optimization and their application in conjugate heat transfer optimization.

[0010] (2) Technical solution

[0011] In order to achieve the purpose of the invention and solve the technical problems, the present invention adopts the following technical solutions:

[0012] The first invention object of the present invention is to provide an improved method for topology optimization of the initial layout under complex heat dissipation conditions by extracting mainstream features, which is used to perform conjugate heat transfer topology optimization design on liquid-cooled heat dissipation structures under high Reynolds number turbulent conditions, improve the heat dissipation efficiency and flow uniformity of the cooling channel structure, and realize multi-objective collaborative optimization design of thermal-fluid performance, wherein the mainstream features refer to the structural paths or regions that mainly affect the thermal-fluid performance of the system in the areas with high heat flux density and pressure drop sensitivity under complex flow conditions. Different from the traditional bionic layout, the present invention only uses a small amount of initial solid domains to reshape the initial flow field to guide the optimization direction of the topology. The method used is composed of a conjugate heat transfer topology optimization algorithm based on the SE-Net reduced-order proxy model, the ETO optimization algorithm, the turbulent conjugate heat transfer topology optimization algorithm, and the target of relative standard deviation correction under the assumption of flow field freezing. The method comprises at least the following steps when implemented:

[0013] S100. Constructing a parametric structure array layout:

[0014] Within the design domain of the liquid cooling chamber of the liquid cooling plate radiator, an initial two-dimensional array layout is constructed, consisting of multiple structural units arranged along the main flow direction and the lateral direction. One or more of the radius, spacing, and arrangement of the structural units are set as design variables to establish an adjustable parameter space.

[0015] S200. Generate a high-fidelity training sample dataset:

[0016] Random sampling is performed within the adjustable parameter space to generate a predetermined number of initial layout samples with different geometric parameter combinations. For each sample, computational fluid dynamics (CFD) numerical simulation is performed under high Reynolds number conditions to solve the flow and temperature fields under preset high Reynolds number turbulence and conjugate heat transfer boundary conditions. Macroscopic performance indicators are extracted, including at least the average or maximum temperature to characterize heat dissipation performance and the inlet and outlet pressure drops to characterize flow performance. A high-fidelity training dataset containing the initial layout geometric parameter combinations and their corresponding macroscopic performance indicators is constructed.

[0017] S300. Build and train mainstream feature prediction models:

[0018] A convolutional neural network (SE-Net) reduced-order proxy model with an SE attention mechanism is constructed and trained based on a high-fidelity training dataset. Through feature squeezing, excitation, and impact calibration operations, a mapping and prediction relationship between the initial layout geometric parameter combination and the macro-performance indicator output under high Reynolds number conditions is established. This enables the model to quickly evaluate the macro-performance of the initial layout without explicitly analyzing the flow field.

[0019] S400. Construct the modified objective function under the assumption of flow field freezing:

[0020] Define a modified objective function under the assumption of flow field freezing to comprehensively evaluate the macroscopic performance of the initial layout to achieve coordinated optimization of heat dissipation performance and pressure drop characteristics. The modified objective function is a multi-objective evaluation function that includes at least the average or maximum temperature, inlet and outlet pressure drop, and structural unit area ratio indicators, and is constructed by combining empirical correction factors and normalized weights.

[0021] S500. ETO global optimization to select the optimal initialization layout:

[0022] Sampling is performed within the adjustable parameter space to generate a large number of potential initial layout samples. The ETO algorithm is used to iteratively optimize the potential initial layout sample space. Each time the candidate solution is evaluated, the trained SE-Net reduced-order proxy model is called for rapid prediction. Through large-scale sample screening, the initial layout that satisfies the optimal solution of the objective function is selected as the initial state of topology optimization.

[0023] S600. Perform topology optimization based on the initial layout guided by the mainstream:

[0024] The optimal initial layout obtained through screening is used as the initial field for topology optimization. A topology optimization method based on density field interpolation is adopted, combined with the conjugate heat transfer control equation, the k-ε turbulence model under high Reynolds number, and the fluid momentum equation corrected by inverse magnetic permeability. Combined with the boundary conditions of the structural field convection-heat conduction dual domain, the topology optimization iterative process is performed under the comprehensive performance objective function, and the density distribution function is optimized to achieve the optimal configuration of materials in the design domain, thereby realizing the global optimal evolution of the topological structure under the thermal-fluid synergy goal.

[0025] The second object of the present invention is to provide a computer program product, including computer instructions, which are used to execute the above-mentioned improved method of topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions of the present invention.

[0026] The third invention object of the present invention is to provide a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the above-mentioned improved method of topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions of the present invention.

[0027] (3) Technical effects

[0028] Compared with the prior art, the improved method, medium and program product of the present invention for topology optimization of the initial layout of the extraction of mainstream features under complex heat dissipation conditions have the following beneficial and significant technical effects: (1) The present invention establishes a framework for the initial layout of multi-objective topology optimization of turbulent thermal fluid systems by constructing a convolutional neural network reduction agent model with SE attention mechanism and an ETO global optimization algorithm, which can effectively avoid a large amount of trial and error costs and strong numerical problems in the optimization process, and obtains quite good results without making a large number of continuous changes to the topological parameters in the topological process. (2) Based on the structural initialization guided by the mainstream, the present invention introduces multi-physics field coupling control equations, material interpolation models, Helmholtz filters and hyperbolic tangent projection equations in the topology optimization process to form a topology optimization framework with full-process synergy. While achieving dual optimization of thermal performance and flow performance, this method effectively suppresses problems such as topological pathology, numerical oscillation and flow separation, and significantly improves the thermal-fluid synergy performance, structural manufacturability and simulation calculation stability of the optimization results. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flow chart of the improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions of the present invention;

[0030] Figure 2 Shown is an algorithm flow chart of the convolutional neural network algorithm in the present invention;

[0031] Figure 3Shown is the algorithm flow chart of the ETO optimization algorithm in the present invention;

[0032] Figure 4 Shown is a topology optimization flow chart of the initial layout in the present invention;

[0033] Figure 5 Schematic diagram of the topology optimization area in Example 2 of the present invention;

[0034] Figure 6 The figure shows the flow field at the inlet Reynolds number of 14000 when the cylinder radius is 0.005 m.

[0035] Figure 7 The figure shows the SE-Net reduced-order proxy model training results for average temperature, where (a) is the comparison of the prediction results of the training set and (b) is the comparison of the prediction results of the test set.

[0036] Figure 8 The figure shows the results of the SE-Net reduced-order proxy model training on import and export pressure drop, where (a) is the comparison of the prediction results of the training set, and (b) is the comparison of the prediction results of the test set;

[0037] Figure 9 The figure shows the ETO optimization results under the modified objective function, where (a) is the target array 1 and (b) is the target array 2.

[0038] Figure 10 Figure 1 shows a comparison of the flow field distribution of different channels under unified thermal-fluid boundary conditions, including: (a) bionic topology optimized channel; (b) topology channel 1 guided by mainstream features; (c) topology channel 2 guided by mainstream features; (d) regular rectangular channel; (e) cylindrical array channel; (f) square column array channel.

[0039] Figure 11 Shown are the temperature field distribution diagrams under different channels, including: (a) topological flow channel 1 based on mainstream feature guidance; (b) topological flow channel 2 based on mainstream feature guidance; (c) square column array flow channel; (d) cylindrical array flow channel; (e) bionic topological optimization flow channel; (f) regular rectangular flow channel. DETAILED DESCRIPTION

[0040] The present invention aims to provide an improved method, medium and program product for topological optimization of the initial layout under complex heat dissipation conditions by extracting mainstream features, which is used to perform conjugate heat transfer topological optimization design of liquid-cooled heat dissipation structures under high Reynolds number turbulent conditions, improve the heat dissipation efficiency and flow uniformity of the cooling channel structure, and achieve multi-objective thermal-fluid performance collaborative optimization design, wherein the mainstream features refer to the structural paths or regions that mainly affect the thermal-fluid performance of the system in the area with high heat flux density and pressure drop sensitivity under complex flow conditions. In order to make the purpose, technical solutions and advantages of the implementation of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below in conjunction with the drawings in the embodiments of the present invention.

[0041] Example 1: Topology Optimization Improvement Method

[0042] As a specific example, Figure 1 As shown, the improved topology optimization method for the initialization layout with extraction of mainstream features under complex heat dissipation conditions provided by the embodiment of the present invention mainly includes six key steps during implementation: parameterized array construction, high-fidelity data set generation, SE-Net reduced-order proxy model training, modified objective function design, ETO global optimization, and final topology optimization. The synergistic effect of these steps ensures that a heat dissipation structure with excellent performance is obtained while ensuring computational efficiency. Specifically:

[0043] S100. Constructing a parametric structure array layout:

[0044] Within the design domain of the liquid-cooling cavity of the liquid-cooling plate radiator, an initial two-dimensional array layout is constructed, consisting of multiple structural units arranged along the mainstream and lateral directions. One or more of the radius, spacing, and arrangement of the structural units are set as design variables to establish an adjustable parameter space.

[0045] Preferably, the design domain is a symmetrical or asymmetrical structure, the structural unit is a cylinder or an equivalent geometric body, and the two-dimensional array-type initial layout is an orthogonal or non-orthogonal array consisting of multiple rows and columns of structural units. By adjusting the design variables of the structural units, the velocity distribution and temperature gradient of the initial flow field are affected, so as to guide the gradient direction of the objective function in the early stage of topology optimization and provide guiding initial conditions for subsequent topology optimization.

[0046] S200. Generate a high-fidelity training sample dataset:

[0047] Random sampling is performed within the adjustable parameter space to generate a predetermined number of initial layout samples with different geometric parameter combinations. CFD numerical simulation is performed on each sample under high Reynolds number conditions to solve its flow and temperature fields under preset high Reynolds number turbulence and conjugate heat transfer boundary conditions. Macroscopic performance indicators are extracted, including at least the average or maximum temperature representing the heat dissipation performance and the inlet and outlet pressure drops representing the flow performance. A high-fidelity training dataset containing the initial layout geometric parameter combinations and their corresponding macroscopic performance indicators is constructed.

[0048] In an embodiment of the present invention, generating a high-fidelity training sample dataset includes at least the following sub-steps:

[0049] S201. Perform super Latin cube sampling:

[0050] The hyper-Latin cube sampling method is used to perform random sampling in the adjustable parameter space to generate several groups of training samples and test samples. Each group of samples corresponds to an initial layout geometric parameter combination configuration.

[0051] S202. Set CFD numerical simulation boundary conditions:

[0052] For each initial layout sample, a corresponding 2D or 3D computational domain model is constructed, and unified high Reynolds number turbulence and conjugate heat transfer boundary conditions are configured, including at least the inlet high Reynolds number, inlet temperature, outlet pressure, heat source intensity, and characteristic length.

[0053] S203. Perform multi-physics coupled numerical solution:

[0054] Under the preset boundary conditions, CFD numerical simulations at high Reynolds numbers are performed for each initial layout sample. The k-ε turbulence model is used to describe the turbulent flow characteristics at high Reynolds numbers. The conjugate heat transfer coupling effect between the solid domain and the fluid domain is considered, and the velocity field and pressure field distribution in the fluid domain as well as the temperature field distribution in the solid domain and the fluid domain are calculated.

[0055] S204. Extract key macro-performance indicators:

[0056] Extracting macro-performance indicators from the numerical solution results, including at least the average or maximum temperature indicator representing the heat dissipation performance and the inlet and outlet pressure drop indicator representing the flow performance, to form a structured data record containing a one-to-one correspondence between the initial layout geometric parameter combination configuration and the macro-performance indicator output;

[0057] S205. Constructing a high-fidelity training dataset:

[0058] The initial layout geometric parameter combination is used as the input feature vector and the corresponding macro performance index is used as the output label vector. A high-fidelity training dataset containing the input-output mapping relationship is constructed, and part of the data is divided into training data and the rest into test data.

[0059] S300. Build and train mainstream feature prediction models:

[0060] Based on a high-fidelity training dataset, a convolutional neural network (SE-Net) reduced-order proxy model with an SE attention mechanism is constructed and trained. Through feature squeezing, excitation, and impact calibration operations, a mapping and prediction relationship between the initial layout geometric parameter combination and the macro-performance indicator output under high Reynolds number conditions is established, enabling the model to quickly evaluate the macro-performance performance of the initial layout without explicitly analyzing the flow field.

[0061] In the embodiment of the present invention, Figure 2 As shown in Figure 2, the establishment and training of the SE-Net reduced-order proxy model includes:

[0062] S301. Building the SE-Net Architecture:

[0063] Construct a convolutional neural network structure with SE attention mechanism, including input layer, convolution operation layer, feature channel SE attention branch, backbone activation branch and output layer. The input layer receives the feature input tensor encoded by the initial layout geometric parameters, extracts local spatial features through the first layer of convolution operation, and feeds them into the parallel SE attention channel path and backbone feature path.

[0064] S302. SE attention mechanism branch construction:

[0065] The SE attention path performs squeeze and excitation operations in sequence. The squeeze operation squeezes the input initial layout geometric parameter features through global average pooling, compresses the spatial dimension information of each feature channel into a scalar value, and generates a compact feature representation with a channel dimension descriptor; the excitation operation uses a gating mechanism composed of two fully connected layers to excite the squeezed feature descriptor. The first fully connected layer performs dimensionality reduction through the ReLU activation function, and the second fully connected layer generates a channel weight coefficient between 0 and 1 through the sigmoid activation function. The weight coefficient reflects the relative importance of each feature channel for the predicted target.

[0066] S303. Backbone feature extraction and fusion:

[0067] The convolution operation in the backbone path is followed by a ReLU activation function, which is stacked repeatedly several times to extract deep features. This is then element-wise weighted fusion with the output of the attention branch to achieve feature calibration based on importance weights, highlighting feature channels that contribute most to the prediction task and suppressing unimportant feature information. The fused feature map is then fed into a multi-layer fully connected network.

[0068] S304. Deep Network Training and Optimization:

[0069] The fused output is sequentially passed through a multi-layer fully connected network, using the ReLU activation function between layers and adding a dropout layer in the middle (e.g., dropping 5% of the neurons) to prevent overfitting. The final output layer is a continuous-valued output node used to predict the macro-performance indicator response of the initial layout. The SE-Net network is trained end-to-end using the backpropagation algorithm, with the optimization goal of minimizing the mean squared error between the predicted value and the true value. The network parameters are continuously updated using the gradient descent method until convergence.

[0070] S305. Model Validation and Generalization Evaluation:

[0071] The performance of the trained SE-Net reduced-order proxy model is verified using an independent test dataset, and the root mean square error (RMSE) and determination coefficient (R) of relevant macro-performance indicators are statistically analyzed. 2 , and evaluate the prediction stability of the model in each region of the sample space to ensure that it has good generalization ability.

[0072] S400. Construct the modified objective function under the assumption of flow field freezing:

[0073] A modified objective function under the flow field freezing assumption is defined to comprehensively evaluate the macroscopic performance of the initial layout to achieve coordinated optimization of heat dissipation performance and pressure drop characteristics. The modified objective function is a multi-objective evaluation function that includes at least the average or maximum temperature, inlet and outlet pressure drop, and structural unit area ratio indicators, and is constructed in combination with empirical correction factors and normalized weights.

[0074] In the embodiment of the present invention, the correction objective function is an objective function corrected relative to the sample standard under the frozen flow field freezing assumption, and its mathematical expression is:

[0075]

[0076] Among them, Π is the comprehensive performance objective function, are the average or maximum temperature and inlet and outlet pressure drops under the current initial layout sample, are the temperature and pressure drop benchmark values ​​of the reference sample, w i 、 w jare the weight coefficients of temperature and pressure drop targets respectively and w i + w j =1, S i For the i The area of ​​a structural unit, n is the number of structural units, S all The maximum total area allowed for the structural unit; b 1 is the temperature difference correction factor and , are the average temperatures of the solid region and the fluid region, respectively; M is the normalized standard deviation of the current sample structure parameter and , x i For the i The size parameters of each structural unit (such as radius), is the average value of all structural unit sizes, L max is the maximum allowable size of the structural unit.

[0077] S500. ETO global optimization to select the optimal initialization layout:

[0078] Sampling is performed in the parameter space to generate a large number of potential initial layout samples. The ETO algorithm is used to iteratively optimize the potential initial layout sample space. Each time the quality of the candidate solution is evaluated, the trained SE-Net reduced-order proxy model is called for rapid prediction. Through large-scale sample screening, the initial layout that meets the optimal solution of the objective function is selected as the initial state of topology optimization.

[0079] As a preferred method, the ETO algorithm adopts a global search strategy based on a combination of exponential functions and trigonometric functions. It performs iterative optimization in a large-scale potential initial layout sample space, uses an exponential decay factor to control the dynamic adjustment of the search range, and combines the periodic characteristics of trigonometric functions to achieve a comprehensive exploration of the solution space. During the iterative process, the SE-Net reduced-order proxy model is called for rapid performance evaluation to reduce the number of CFD simulation calls and achieve efficient iterative optimization.

[0080] In the embodiment of the present invention, Figure 3 As shown in FIG, the global optimization and screening of the optimal initialization layout based on the ETO algorithm includes at least the following sub-steps:

[0081] S501. Initialize candidate solutions:

[0082] Generate a large number of potential initial layout samples in a uniform or random manner in the parameter space as the initial candidate solution population. Each candidate solution corresponds to a set of initial layout geometric parameter combinations. At the same time, initialize the key control parameters of the ETO algorithm, including at least the maximum number of iterations T. max , the current number of iterations t (initially set to 0), the exploration factor CM (a dynamic parameter that controls the balance between global exploration and local exploitation), the convergence control parameter CEi (a specific iteration threshold used to trigger the update of the search space), and the phase switching parameter Ti (the iterative dividing point that distinguishes the exploration phase from the exploitation phase);

[0083] S502. Create initial parameters and calculate fitness and optimal values:

[0084] The SE-Net reduced-order proxy model is called to quickly predict the performance of each candidate solution, obtain the corresponding macro performance index, calculate the fitness value of each candidate solution using the modified objective function, and identify the solution with the best fitness value in the current population as the global optimal solution F best , initialize position update control parameters α 1 (the first parameter in the exploration phase, controlling the intensity of large-scale search), α 2 (the second parameter in the exploration phase, which adjusts the diversity of search directions), α 3 (development phase parameter, controlling the local fine search range) is used for subsequent exponential trigonometric transformation operations;

[0085] S503. Determine the convergence conditions and perform CM update:

[0086] Check whether the current number of iterations t reaches the maximum number of iterations T max Limit, if t≥T max Then jump to step S507 to return to the best solution. If t <T max The exploration factor CM is dynamically updated according to the current iteration progress, and the algorithm can smoothly transition from global exploration to local development through adaptive adjustment of the CM value.

[0087] S504. Execution condition judgment and parameter calculation branch:

[0088] First, determine whether the current number of iterations t is equal to the convergence control parameter CEi. If t=CEi, reconstruct or perturb to update the search space range, and then continue to perform subsequent judgments; then determine whether the exploration factor CM is greater than 1. If CM>1, enter the exploration phase and calculate the position update parameter α 1 and α 2. It is used for large-scale global search and exploration of new solution space. If CM≤1, it enters the development phase and calculates the updated parameter α3, which is used for refined local search and solution improvement near the current optimal solution.

[0089] S505. Implement a phased position update strategy:

[0090] Execute the corresponding phase of position update according to the judgment result of step S504: In the exploration phase, if t < Ti, use parameters α 1 and α 2 to combine with the current best solution F best and sequentially perform the first and second exploration phase updates of the individual positions to achieve a large-scale exploration of the solution space through the combination of exponential and trigonometric functions; in the development phase, if t < Ti, use parameter α 3 to sequentially perform the first and second development phase updates of the individual positions to achieve a deep optimization of the solution through local fine search around F best .

[0091] S506. Calculate the fitness and update the optimal solution:

[0092] Re-call the SE-Net reduced-order surrogate model for performance prediction and fitness evaluation of all candidate solutions after position update, compare the fitness value of the new solution with the fitness of the current global optimal solution F best . If a better solution is found, update F best to the new optimal solution, and record the corresponding geometric parameter combination configuration at the same time;

[0093] S507. Execute the stop condition judgment and return the best solution:

[0094] Repeat the iterative loop of steps S503~S506 until the preset convergence criterion is met, including the current iteration number t reaching the maximum iteration number T max , the fitness improvement amplitude of the global optimal solution F best is continuously less than the preset threshold and other termination conditions for several generations, and finally return the geometric parameter combination configuration corresponding to F best with the optimal objective function value as the initial layout of the topology optimization.

[0095] S600. Execute the topology optimization based on the mainstream-guided initial layout:

[0096] Use the obtained optimal initial layout as the initial field of the topology optimization, adopt the topology optimization method based on density field interpolation, combine the conjugate heat transfer control equation, the k-ε turbulence model at high Reynolds numbers, and the fluid momentum equation corrected by inverse magnetic permeability, and combine the boundary conditions of the convection-conduction double domain of the structure field to execute the topology optimization iteration process, optimize the density distribution function to achieve the optimal configuration of materials within the design domain, and realize the global optimal evolution of the topology structure under the thermal-fluid cooperation target.

[0097] In the embodiments of the present invention, as Figure 4 shown, the topology optimization of the initial layout includes the following sub-steps:

[0098] S601. Constructing the initial density field for topology optimization:

[0099] The optimal initial layout obtained by the ETO algorithm is used as the initial density field distribution input for topology optimization. The geometric parameter combination configuration of the structural unit is mapped to the initial material density value of the corresponding grid unit in the topological design domain. A continuous material density distribution function ρ(x) is constructed, where ρ∈[0,1], where ρ(x)=0 corresponds to a completely fluid region, ρ(x)=1 corresponds to a completely solid region, and the intermediate value represents a porous medium or transition region. The density value near the cylindrical boundary is smoothly transitioned by a distance function.

[0100] S602. Constructing a multi-physics coupling analysis framework:

[0101] A multi-physics analysis framework for complex thermal-fluid coupling at high Reynolds numbers is established, including: the Navier-Stokes momentum conservation equations for solving the velocity and pressure distributions of the fluid in the design domain; the turbulent transport equations based on the standard k-ε model, which are used to describe the energy changes and dissipation processes of turbulent disturbances at high Reynolds numbers; and the convection-heat conduction coupling equations based on energy conservation, which are used to describe the temperature field distributions in the solid and fluid domains and the heat exchange process between the two.

[0102] S603. Constructing a Thermal-Fluid Synergistic Topology Optimization Model:

[0103] With the goal of achieving optimal coordination between heat dissipation and flow properties, a mathematical model for collaborative thermal-fluid topology optimization was constructed, with the material density function ρ(x) as the design variable. Constraints included at least material volume fraction restrictions, fluid domain connectivity constraints, and fluid domain boundary integrity constraints. A mapping relationship between physical properties (such as thermal conductivity and flow resistance) and design variables was established based on the Solid Isotropic Material with Penalization (SIMP) (Rational Approximation of Material Properties, RAMP) model to ensure that the structural-fluid properties were divisible during the topological evolution process and suitable for gradient iteration algorithms.

[0104] S604. Boundary Condition Setting and Load Application:

[0105] Set unified conjugate boundary conditions, including at least the inlet velocity or pressure boundary, the outlet free pressure boundary, and the structure wall adiabatic boundary or temperature boundary, and apply a uniform volume heat source term Q in the specified area inside the structure to simulate the heating element under actual working conditions;

[0106] S605. Define the comprehensive performance objective function and perform sensitivity analysis and gradient calculation:

[0107] First, define the comprehensive performance objective function, whose mathematical expression is: , where F is the comprehensive performance objective function, are the average temperature and inlet and outlet pressure drops of the current topology, a 、 b are the corresponding weight coefficients and a + b =1, are the average temperature and inlet and outlet pressure drops of the first step when the initial density field is 0.5; considering the influence of the penalty factor on the sensitivity calculation in the SIMP and RAMP interpolation models, the adjoint method is used to perform topological sensitivity analysis. The gradient information of topological optimization is obtained by solving the adjoint temperature field equation and the adjoint flow field equation, and the chain derivative rule is used to calculate the gradient of the objective function F with respect to the material density ρ(x) , used to guide the optimization direction of material distribution in the iterative step, where T is temperature and u is speed;

[0108] S606. Implement density filtering and projection processing:

[0109] Density filtering is performed on the design variables to eliminate checkerboard phenomena and grid dependency problems. A Helmholtz filter ( ,in r is the filter radius, γ is the design variable for material density, is the design variable after filtering, Ω is the design domain, is the design domain boundary, n is the boundary normal vector) to smooth the material density distribution, and then apply the hyperbolic tangent projection function ( ,in is the density value after projection processing, is the material density value after filtering, β and η are the projection parameter and threshold respectively) the filtered density value is projected to the interval of 0-1, and a gradual transition from fuzzy boundary to clear boundary is achieved by adjusting the projection parameter and threshold;

[0110] S607. Update design variables and check constraints:

[0111] Based on the sensitivity analysis results, a gradient-based mathematical programming algorithm, such as the Method of Moving Asymptotes (MMA), is used to update the material density distribution design variables. The iterative formula is: , where α is the step size parameter, are the material densities at iteration steps k and k+1, respectively. In each iteration, the constraint conditions are checked and the design variables that violate the constraints are corrected. The penalty function method is used to deal with inequality constraints, such as the Lagrange multiplier method. The algorithm formula is: ,in are the Lagrange multipliers at iteration steps k and k+1 respectively, μ is the penalty parameter, is the inequality constraint function;

[0112] S608. Determine convergence and output optimization results:

[0113] Convergence judgment criteria are set, including a threshold value for the rate of change of the objective function, a maximum number of iterations T, or a density gradient convergence criterion. If the convergence criteria are met, the optimization process is terminated and the final topology optimization result is output. If convergence is not achieved, the process returns to step S602 and continues iterative optimization, ultimately obtaining a material distribution topology structure that achieves the optimal synergy between heat dissipation performance and flow performance under given constraints.

[0114] S609. Topology Extraction and Post-Processing:

[0115] The final density distribution results were subjected to threshold segmentation processing to form clear topological boundaries and extract a manufacturable cooling channel geometry. CFD simulations were performed based on the resulting structure to verify its temperature distribution, pressure drop characteristics, and mainstream uniformity, and to evaluate its thermal-fluid synergy performance under complex working conditions with high Reynolds numbers.

[0116] Preferably, in step S602, a Darcy-type penalty term for artificial porous media related to material density is introduced into the momentum conservation equation, and is corrected by the inverse magnetic permeability function to reflect the change in flow resistance under different material densities; in the convection-heat conduction coupling equation group, the heat conduction and convection heat transfer mechanisms are defined in the solid domain and the fluid domain respectively, and the heat capacity, density and thermal conductivity are interpolated by the material density function to realize continuous regulation of the structural thermal physical parameters.

[0117] Specifically, the fluid flow governing equations include:

[0118] Continuity equation (conservation of mass): , where u is the fluid velocity vector.

[0119] Momentum conservation equation (introducing porous media resistance term):

[0120]

[0121] in, ρ is the fluid density, I is the unit tensor, μ is the molecular viscosity, μ Tis the turbulent viscosity coefficient, α ( γ ) is the design variable related to material density γ the associated Darcy damping coefficient;

[0122] Turbulent kinetic energy k transport equation:

[0123]

[0124] Where k is the turbulent kinetic energy, σ k is the Prandtl number for turbulent kinetic energy (standard value is 1.0), P k is the turbulent kinetic energy generation term, ε is the turbulent dissipation rate, α f ( γ ) is the Darcy-type artificial drag coefficient for the turbulent kinetic energy equation;

[0125] Turbulent dissipation rate ε Transport equation:

[0126]

[0127] Where, σ ε is the Prandtl number for turbulent dissipation rate (standard value is 1.3), C 1ε 、 C 2ε are model constants (standard values ​​are 1.44 and 1.92 respectively), α ε ( γ ) is the artificial drag coefficient for the turbulence dissipation rate equation;

[0128] The calculation formula of turbulent viscosity is: , C μ is the turbulence model constant (standard value is 0.09);

[0129] Turbulent kinetic energy generation term: , “:” represents a tensor double dot product operation.

[0130] In addition, to achieve continuous expression of structure-fluid properties, a modified form of the inverse magnetic permeability of artificial porous media is introduced into the above momentum and turbulence equations. The interpolation expression of the inverse magnetic permeability parameter is as follows:

[0131]

[0132] Where, γ is the material density field, α(·) represents the inverse magnetic permeability of artificial porous media in solid / fluid domains of different equations, α fmax 、 α kmax 、 α εmax are the maximum drag coefficients of the solid region in the momentum equation, turbulent kinetic energy equation, and turbulent dissipation rate equation, respectively. α fmin 、 α kmin 、 α εmin are the minimum resistance coefficients of the fluid region in the corresponding equations, γ is the material density design variable (range 0-1), p α is the penalty factor (usually 0.01).

[0133] In terms of heat conduction, the conjugate heat transfer model is used to describe the heat conduction and convection heat transfer mechanisms between solid and fluid regions. Its governing equations are as follows:

[0134] Energy conservation equation (conjugate heat transfer model):

[0135]

[0136] in, T is the temperature field, Q is the volume heat source term, ρ ( γ ), c ( γ ), k ( γ ) are density design variables γ Related density, specific heat capacity and thermal conductivity, material property interpolation functions:

[0137] , ,

[0138] in: ρ f is the fluid density, ρ s is the solid density, p 1 is the penalty factor for density interpolation; k f is the thermal conductivity of the fluid, k s is the solid thermal conductivity, p 2 is the penalty factor for thermal conductivity interpolation; c f is the specific heat capacity of the fluid, c s is the specific heat capacity of the solid, p3 is the penalty factor for specific heat capacity interpolation ( p 1, p 2, p 3 usually takes values ​​from 1 to 5);

[0139] The thermal boundary conditions are set as follows:

[0140] The two expressions represent the inlet temperature boundary and the outlet and wall adiabatic boundary, Γ in , Γ out , Γ w Represent the inlet, outlet and wall boundaries respectively, T0 is the inlet temperature, and n is the boundary normal vector.

[0141] Further preferably, in step S600, the optimized initial layout is used to guide the direction of topology optimization, and the target value is continuously reduced in the conventional weighted evaluation function by co-optimizing the mainstream velocity field distribution and the temperature field distribution, thereby continuously updating the topology until it is stable.

[0142] In summary, Example 1 realizes a full-process solution from initial flow field guidance to final structure optimization by constructing key technical links such as parameterized array structure, E-Net reduced-order proxy model training, ETO intelligent optimization algorithm and conjugate heat transfer topology optimization, providing an effective method framework and implementation plan for the collaborative optimization of multi-physics fields under complex heat dissipation conditions.

[0143] Example 2: Application Example

[0144] To verify the effectiveness and practicality of the above-mentioned topology optimization improvement method, this Example 2 selects a typical liquid cold plate radiator design scenario as an application case. By constructing a parameterized array in an asymmetric design domain and combining the SE-Net network with the ETO algorithm, the initial layout design and topology optimization iterations are completed to verify the superior thermal-fluid synergy performance of the present invention.

[0145] Based on the framework of the improved topology optimization method proposed in Example 1, this embodiment takes an asymmetric liquid cooling radiator as a specific application object, and its design domain Ω is asymmetric, with one inlet and one outlet, such as Figure 5 As shown. It is known that the inlet temperature is 293.15K, the outlet pressure P0=0Pa, the inlet width L is set to 6 mm, and the heat source is 10^7 W / m 3 , the inlet Reynolds number is 14000.

[0146] First, a 5×10 cylinder array was constructed in the geometric area. The radius of each cylinder ranged from 0.001 m to 0.005 m, and the interval between cylinders was 0.02 m. Figure 6The flow field diagram is shown for an inlet Reynolds number of 14000 when each cylinder in the cylinder array has a radius of 0.005 meters. Using the Super Latin Cube sampling method, the radius of each cylinder is randomly set between 0.001 meters and 0.005 meters. 300 sets of data are constructed as a training set, and 50 sets of data are constructed using the same method as a test set.

[0147] Then, a SE attention mechanism neural network (SE-Net) is constructed. The SE attention mechanism neural network consists of squeezing, excitation, and impact calibration. The algorithm flow chart is as follows Figure 2 By training the SE neural network, the errors of the training set and test set for average temperature and inlet and outlet pressure drop are obtained respectively. The following figures are average temperature ( Figure 7 ) and inlet and outlet pressure drops ( Figure 8 ) of the training set and test set errors. Figure 7 As shown, Figure (a) shows the average temperature prediction results of 300 samples in the training set, and Figure (b) shows the average temperature prediction results of 50 samples in the test set. The RMSE of the average temperature is 0.032 and 0.264 respectively, and the determination coefficient R 2 They are 0.9977 and 0.902 respectively. The closer R² is to 1, the better the model fit is. Figure 8 As shown in Figure (a), the inlet and outlet pressure drop prediction results of 300 samples in the training set are shown in Figure (b), and the inlet and outlet pressure drop prediction results of 50 samples in the test set are shown in Figure (b). The RMSE of the inlet and outlet pressure drops are 2.67 and 12.21 respectively, and the determination coefficient R 2 The above results show that the SE-Net reduced-order surrogate model achieves high prediction accuracy on both the training set and the test set, and can effectively replace the computationally expensive CFD simulation for rapid performance evaluation.

[0148] After the model training is completed, hundreds of thousands of potential samples are randomly generated through the super Latin cube sampling method to construct the initial layout, and the ETO algorithm is used to obtain the optimal solution in the evaluation function. The ETO algorithm flow chart is as follows Figure 3 As shown. The mathematical expression of the modified objective function Π is:

[0149]

[0150] in, are the average or maximum temperature and inlet and outlet pressure drops under the current initial layout sample, are the temperature and pressure drop benchmark values ​​of the reference sample, w i 、 w j are the weight coefficients of temperature and pressure drop targets respectively andw i + w j =1, b 1 is the temperature difference correction factor and ; M is the normalized standard deviation of the current sample structure parameter, defined In this embodiment, L max Set to 0.005 meters, are the average temperatures of the solid and the liquid, respectively, where b 1 The estimated value is 0.08. S i 、 S all They are the total area of ​​the current array and the maximum array area respectively. are the temperature and pressure drop reference values ​​of the reference sample, respectively. They represent the design domain mean temperature and inlet and outlet pressure drops when the array cylinder radius is 0.008 m. w i 、 w j Represents the weight factors, which are 0.6 and 0.4 respectively.

[0151] By optimizing the SE-Net network and the ETO algorithm under the modified objective function, two target array layouts are obtained, such as Figure 9 As shown in the figure, the two target arrays exhibit obvious non-uniform cylinder radius distribution characteristics: at the flow field inlet area, the cylinder radius is relatively small, which is conducive to reducing flow resistance and inlet pressure drop; in the middle area, the cylinder size changes in a gradient, which ensures sufficient heat transfer area while avoiding excessive flow separation; near the outlet, the cylinder layout is carefully optimized to improve flow uniformity. Both array layouts avoid the large-scale flow separation and pressure drop concentration problems commonly seen in traditional uniform arrays. The non-uniform cylinder radius distribution achieves the coordinated optimization of heat dissipation performance and flow performance, providing an initial layout foundation with good mainstream feature guidance for subsequent topology optimization.

[0152] Using the above two layouts as the initial layout, topology optimization was performed under the comprehensive performance target evaluation function, and the results were comprehensively compared with the benchmark structure. The comparison objects included six typical flow channel layouts, including a bionic topological structure with an initial density set to 0.5, a regular array structure (including cylindrical and square column array flow channels), two groups of topological flow channels guided by mainstream features, and a regular rectangular flow channel. A uniform volume heat source (Q = 10^7 W / m 3), the inlet Reynolds number is 14000, and all adiabatic walls are avoided. The boundary conditions between different models are the same. The k-ε turbulence physics model and the conjugate heat transfer physics model are used to evaluate the hydraulic and heat transfer performance of the final design. The flow field characteristics are as follows Figure 10 As shown, the temperature field characteristics are as follows Figure 11 shown.

[0153] When the Reynolds number is as high as 14000, the pressure drops of the two groups of topological flow channels optimized by this method are 1467Pa and 1530Pa, which is nearly 80% lower than the relative pressure drop of the square array flow channel and nearly 73% lower than the circular flow channel, indicating that the topological flow channel has a good effect on improving the pressure drop. Figure 10 It can be seen that the rectangular and circular array channels generate numerous small vortices at the array walls, increasing flow dissipation. However, the three topological flow channels and rectangular channels do not cause the frequent flow separation and reunion of the circular and square array channels, significantly reducing pressure loss. Furthermore, the relatively smooth boundaries of the topological channels make the fluid flow smoother during the turning process, further reducing the generation of vortices and further reducing pressure drop.

[0154] When the Reynolds number is as high as 14000, the highest temperatures of the two groups of topological regions optimized by this method are 325K and 330K, and the highest temperature of other flow channels is as low as 341K, indicating that the topological flow channels optimized by this method have a significant effect on improving hot spots. Figure 11 It is striking that the hotspot distributions of different flow channels differ. The hotspot of the topological flow channels optimized by this method is located in the upper left corner, due to the attenuated bifurcated flow in the upper left corner. The hotspots of the square and cylindrical array channels are located on the right, due to the uneven flow caused by complex flow separation and convergence. The hotspot of the rectangular channel is located in the upper right corner, also due to uneven flow. The hotspot of the flow channel optimized by traditional topology optimization is located in the lower right corner, due to the refraction of flow toward the center.

[0155] In summary, the topology optimization model proposed in the present invention that integrates the SE-Net reduced-order proxy model and the ETO optimization algorithm can effectively solve the topological pathological problem of the liquid cooling plate radiator topology optimization method under complex geometric and high Reynolds number turbulent conditions. By constructing a small-area array field layout that is different from the traditional bionic mode, the topological results can be significantly improved, which provides an important reference for synergistically improving the heat transfer performance and flow performance of the liquid cooling plate radiator under turbulent conditions, and provides a feasible idea for more complex flow conditions. Without a large number of grids and tedious trial and error of continuous changes in topological parameters, better results can be obtained by guiding the optimization direction of the gradient algorithm while saving a lot of time.

[0156] The above embodiments fully and effectively achieve the objectives of the present invention. Those skilled in the art will appreciate that the present invention includes, but is not limited to, the contents described in the accompanying drawings and the above specific embodiments. Although the present invention has been described with reference to the embodiments currently considered to be the most practical and preferred, it should be understood that the present invention is not limited to the disclosed embodiments, and any modifications that do not deviate from the functional and structural principles of the present invention are intended to be included within the scope of the claims.

Claims

1. An improved method for topology optimization of initialization layout under complex heat dissipation conditions by extracting mainstream features, characterized in that: At least the following steps are included: S100. Constructing an initial two-dimensional array layout within the design domain of the liquid cooling chamber of the liquid cooling plate radiator, comprising a plurality of structural units arranged along the main flow direction and the lateral direction, setting one or more of the radius, spacing, and arrangement of the structural units as design variables to establish an adjustable parameter space; S200. Randomly sample within the adjustable parameter space to generate a predetermined number of initial layout samples with different geometric parameter combinations. For each sample, perform CFD numerical simulation under high Reynolds number conditions and extract macroscopic performance indicators to construct a high-fidelity training dataset containing geometric parameter combinations and their corresponding macroscopic performance indicators. S300. Build and train an SE-Net reduced-order surrogate model based on a high-fidelity training dataset. Establish a predictive mapping relationship between the initial layout geometry parameter combination and the macro-performance output under high Reynolds number conditions. This enables the model to rapidly evaluate the macro-performance of the initial layout without explicitly analyzing the flow field. S400. Define a modified objective function under the assumption of flow field freezing to comprehensively evaluate the heat dissipation performance and pressure drop characteristics of the initial layout. This function includes at least the average or maximum temperature, inlet and outlet pressure drop, and structural unit area ratio. This function is constructed by combining empirical correction factors and normalized weights. S500. Global optimization based on the ETO algorithm to select the optimal initial layout: A large number of potential initial layout samples are sampled within the adjustable parameter space. The ETO algorithm is then used to iteratively optimize the potential initial layout sample space. During the iterative evaluation of candidate solutions, the SE-Net reduced-order surrogate model is used for rapid prediction, ultimately selecting the initial layout that satisfies the optimal solution to the objective function. S600. Initial Layout Topology Optimization: Using the selected initial layout as the starting field for topology optimization, a topology optimization model based on density field interpolation is constructed. The k-ε turbulence model, the inverse permeability-corrected momentum equation, and the governing conjugate heat transfer equations are introduced, and density function evolution is performed to achieve the optimal configuration of materials in the design domain.

2. The improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions according to claim 1 is characterized in that: In step S100, the design domain is a symmetrical or asymmetrical structure, the structural unit is a cylinder or an equivalent geometric body, and the two-dimensional array type initial layout is an orthogonal or non-orthogonal array composed of multiple rows and columns of structural units. The velocity distribution and temperature gradient of the initial flow field are affected by adjusting the design variables of the structural units.

3. The improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions according to claim 1 is characterized in that: In step S200, the generation of a high-fidelity training sample dataset includes at least the following sub-steps: S201. Using the Super Latin Cube sampling method, random sampling is performed within the adjustable parameter space to generate several sets of training and test samples, each set of samples corresponding to an initial layout geometric parameter combination configuration; S202. Construct a corresponding 2D or 3D computational domain model for each initial layout sample, configuring unified high Reynolds number turbulence and conjugate heat transfer boundary conditions, including at least the inlet high Reynolds number, inlet temperature, outlet pressure, heat source intensity, and characteristic length. S203. Perform CFD numerical simulations at high Reynolds numbers for each initial layout sample under pre-set boundary conditions. The k-ε turbulence model is used to describe the turbulent flow characteristics at high Reynolds numbers. The conjugate heat transfer coupling effect between the solid and fluid domains is also considered. The velocity and pressure field distributions within the fluid domain, as well as the temperature field distributions in both the solid and fluid domains, are simultaneously calculated. S204. Extract macro-performance indicators from the numerical solution results, including at least the average or maximum temperature indicator representing heat dissipation performance and the inlet and outlet pressure drop indicator representing flow performance, to form a structured data record containing a one-to-one correspondence between the initial layout geometric parameter combination configuration and the macro-performance indicator output; S205. Use the initial layout geometric parameter combination as the input feature vector and the corresponding macro-performance indicator as the output label vector to construct a high-fidelity training dataset containing the input-output mapping relationship. Part of the data is divided into training data and the rest is divided into test data.

4. The improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions according to claim 1 is characterized in that: In step S300, the establishment and training of the SE-Net reduced-order proxy model includes at least the following sub-steps: S301. Construct a convolutional neural network architecture with SE attention, including an input layer, a convolution operation layer, a feature channel SE attention branch, a backbone activation branch, and an output layer. The input layer receives a feature input tensor encoded with the initial layout geometry parameters. The first convolution operation extracts local spatial features and feeds them into the parallel SE attention channel path and backbone feature path. S302. The SE attention path sequentially performs squeeze and excitation operations. The squeeze operation squeezes the initial layout geometry parameter features using global average pooling to generate a compact feature representation with channel-dimensional descriptors. The excitation operation excites the squeezed feature descriptors using a gating mechanism consisting of two fully connected layers. The first fully connected layer uses the ReLU activation function for dimensionality reduction, and the second fully connected layer uses the sigmoid activation function to generate channel weight coefficients between 0 and 1. S303. The convolution operation in the backbone path is followed by a ReLU activation function, which is stacked several times to extract deep features. Then, the output of the attention branch is fused element-wise with weighted fusion to achieve feature calibration based on importance weights, highlighting feature channels that contribute most to the prediction task and suppressing unimportant feature information. The fused feature map is then fed into a multi-layer fully connected network. S304. The fused output is sequentially passed through a multi-layer fully connected network, using ReLU activation functions between layers and adding dropout layers in between to prevent overfitting. The output layer is a continuous-valued output node used to predict the initial layout macro-performance indicator response. The SE-Net network is trained end-to-end using the backpropagation algorithm, with the optimization goal of minimizing the mean squared error between the predicted and true values. The network parameters are continuously updated using gradient descent until convergence. S305. Use an independent test dataset to verify the performance of the trained SE-Net reduced-order proxy model and calculate the root mean square error (RMSE) and coefficient of determination (R) of relevant macro-performance indicators. 2 , and evaluate the prediction stability of the model in each region of the sample space to ensure that it has good generalization ability.

5. The improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions according to claim 1 is characterized in that: In step S400, the correction objective function is the objective function corrected relative to the sample standard under the assumption of flow field freezing, and its mathematical expression is: Among them, Π is the comprehensive performance objective function, are the average or maximum temperature and inlet and outlet pressure drops under the current initial layout sample, are the temperature and pressure drop benchmark values ​​of the reference sample, w i 、 w j are the weight coefficients of temperature and pressure drop targets respectively and w i + w j =1, S i For the i The area of ​​a structural unit, n is the number of structural units, S all The maximum total area allowed for the structural unit; b 1 is the temperature difference correction factor and , are the average temperatures of the solid region and the fluid region, respectively; M is the normalized standard deviation of the current sample structure parameter and , x i For the i The size parameters of the structural units, is the average value of all structural unit sizes, L max is the maximum allowable size of the structural unit.

6. The improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions according to claim 1 is characterized in that: In step S500, the ETO algorithm adopts a global search strategy based on a combination of exponential and trigonometric functions. It performs iterative optimization in a large-scale potential initial layout sample space, uses an exponential decay factor to control the dynamic adjustment of the search range, and combines the periodic characteristics of trigonometric functions to achieve a comprehensive exploration of the solution space. During the iteration process, the SE-Net reduced-order proxy model is called for rapid performance evaluation.

7. The improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions according to claim 1 or 6, characterized in that: In step S500, performing global optimization based on the ETO algorithm to select the optimal initialization layout includes at least the following sub-steps: S501. Generate a large number of potential initial layout samples in a uniform or random manner within the parameter space as the initial candidate solution population. Each candidate solution corresponds to a set of initial layout geometric parameter combinations. At the same time, initialize the key control parameters of the ETO algorithm, including at least the maximum number of iterations T. max , current iteration number t, exploration factor CM, convergence control parameter CEi and stage switching parameter Ti; S502. Call the SE-Net reduced-order proxy model to quickly predict the performance of each candidate solution, obtain the corresponding macro-performance index, calculate the fitness value of each candidate solution using the modified objective function, and identify the solution with the best fitness value in the current population as the global optimal solution F best , initialize position update control parameters α 1. α 2. α 3 is used for subsequent exponential trigonometric transformation operations; S503. Check whether the current number of iterations t reaches the maximum number of iterations T max Limit, if t≥T max Then jump to step S507 to return to the best solution. If t <T max The exploration factor CM is dynamically updated, and the algorithm can smoothly transition from global exploration to local development through adaptive adjustment of the CM value. S504. First, determine whether the current iteration count t is equal to the convergence control parameter CEi. If t = CEi, update the search space range and continue with the subsequent determination. Next, determine whether the exploration factor CM is greater than 1. If CM>1, enter the exploration phase and calculate the position update parameter. α 1 and α 2. If CM≤1, enter the development phase and calculate the updated parameter α3; S505. Perform position updates for the corresponding stage according to the judgment result of step S504: In the exploration stage, if t < Ti, use the parameters α 1 and α 2 to combine with the current best solution F best Update the individual positions in the first and second exploration stages in sequence, and achieve large-scale solution space exploration through the combination of exponential function and trigonometric function; In the development stage, if t < Ti, use the parameter α 3 to update the individual position in the first and second development stages in sequence, and realize the depth optimization of the solution through local fine search around F best ; S506. Re-call the SE-Net reduced-order proxy model for all candidate solutions after the position update to perform performance prediction and fitness evaluation, and compare the fitness value of the new solution with the current global optimal solution F best If a better solution is found, update F best is the new optimal solution; S507. Repeat the iterative loop of steps S503 to S506 until the preset convergence criterion is met, and finally return the F with the optimal objective function value. best The corresponding geometric parameter combination configuration.

8. The improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions according to claim 1 is characterized in that: In step S600, the initial layout topology optimization includes at least the following sub-steps: S601. Using the optimal initial layout obtained by the ETO algorithm as the initial density field distribution input for topology optimization, the geometric parameter configuration of the structural unit is mapped to the initial material density values ​​of the corresponding mesh units in the topology design domain, and a continuous material density distribution function ρ(x) is constructed, where ρ∈[0,1]. S602. Develop a multiphysics analysis framework for complex thermal-fluid coupling at high Reynolds numbers, including: Navier-Stokes momentum conservation equations for velocity and pressure distributions in the design domain; turbulent transport equations based on the standard k-ε model; and coupled convection-heat conduction equations based on energy conservation. S603. To achieve optimal synergy between heat dissipation and flow properties, construct a mathematical model for thermal-fluid synergistic topology optimization using the material density function ρ(x) as the design variable. Constraints include at least material volume fraction restrictions, fluid domain connectivity constraints, and fluid domain boundary integrity constraints. A mapping relationship between physical properties and design variables is established using SIMP and RAMP interpolation models. S604. Set uniform conjugate boundary conditions, including at least an inlet velocity or pressure boundary, an outlet free pressure boundary, and a structural wall adiabatic or temperature boundary. Apply a uniform volume heat source term Q to a specified region within the structure to simulate the heating element under actual operating conditions. S605. Define comprehensive performance objective function ,in are the average temperature and inlet and outlet pressure drops of the current topology, a 、 b are the corresponding weight coefficients and a + b =1, are the average temperature and inlet and outlet pressure drops of the first step when the initial density field is 0.5; considering the influence of the penalty factor on the sensitivity calculation in the SIMP and RAMP interpolation models, the adjoint method is used to perform topological sensitivity analysis. The gradient information of topological optimization is obtained by solving the adjoint temperature field equation and the adjoint flow field equation, and the gradient of the objective function with respect to the material density ρ(x) is calculated using the chain derivative rule; S606. Density filtering is performed on the design variables. The material density distribution is smoothed using a Helmholtz filter. The filtered density values ​​are projected to the range 0-1 using a hyperbolic tangent projection function. A gradual transition from fuzzy boundaries to sharp boundaries is achieved by adjusting the projection parameters and thresholds. S607. Based on the sensitivity analysis results, a gradient-based mathematical programming algorithm is used to update the material density distribution design variables. Constraints are verified for compliance at each iteration, and design variables that violate the constraints are corrected. Inequality constraints are handled using a penalty function method. S608. Set convergence criteria. If convergence criteria are met, terminate the optimization process and output the final topology optimization results. If convergence is not achieved, return to step S602 and continue iterative optimization to ultimately obtain the optimal material distribution topology that achieves the best combination of heat dissipation and flow properties under given constraints. S609. Perform threshold segmentation on the final density distribution result to form a clear topological boundary and extract the manufacturable cooling channel geometry.

9. The improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions according to claim 8 is characterized in that: In step S602, a Darcy-type penalty term for artificial porous media related to material density is introduced into the momentum conservation equation and corrected by the inverse magnetic permeability function; in the convection-heat conduction coupling equation group, the heat conduction and convection heat transfer mechanisms are defined in the solid domain and the fluid domain respectively, and the heat capacity, density and thermal conductivity are interpolated by the material density function.

10. The improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions according to claim 1, 8 or 9, characterized in that: In step S600, the optimal initial layout is used to guide the direction of topology optimization. By co-optimizing the mainstream velocity field distribution and the temperature field distribution, the target value is continuously reduced in the weighted objective function, so as to continuously update the topology until it is stable.

11. A computer program product comprising computer instructions, characterized in that The computer instructions are used to execute the improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions as described in any one of claims 1 to 10.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the improved method for topology optimization of the initialization layout for extracting mainstream features under complex heat dissipation conditions as described in any one of claims 1 to 10 is implemented.

Images