Three-dimensional channel structure design method of microchannel heat exchange module and related device

Abstract

The present invention relates to a three-dimensional channel structure design method for a microchannel heat exchange module and a related device thereof. The method: the heat conduction coefficient and the convection heat transfer coefficient of the microchannel heat exchange module are input into the Rayleigh-Ritz coupling model for dynamic analysis to obtain channel structure parameter data; the temperature field and pressure field of the channel structure parameter data are collected, and the time series feature is extracted through a dual-time-space automatic encoder to obtain a fluid dynamic feature data matrix; the fluid dynamic feature data matrix is ​​input into a recursive window Bayesian network, and the channel optimization objective function is obtained by calculating the heat exchange efficiency weight coefficient, the fluid resistance weight coefficient, and the heat transfer uniformity weight coefficient; according to the channel optimization objective function, the genetic algorithm is used to perform global iterative calculations on the channel main structure, and numerical simulation is used to perform local optimization on the flow separation and confluence structure to obtain a three-dimensional channel structure optimization scheme. The implementation of the present invention effectively improves the optimization accuracy of the channel structure parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of heat exchange modules, and in particular to a three-dimensional channel structure design method of a microchannel heat exchange module and a related device thereof. Background Art

[0002] Microchannel heat exchange technology has become an important way to solve the problem of high heat flux heat dissipation due to its advantages such as high heat exchange efficiency and compact structure. However, the flow and heat transfer process inside the microchannel is extremely complex, and traditional design methods are difficult to accurately predict and optimize the flow and heat transfer characteristics inside the microchannel.

[0003] At present, the design of microchannel heat exchangers mainly relies on empirical formulas and simplified models, lacking an in-depth understanding of microscale flow and heat transfer mechanisms. Especially in the design of three-dimensional channel structures, due to the high coupling of fluid flow and heat transfer processes, as well as the complexity of the flow separation and confluence structure, traditional design methods are difficult to achieve precise control and optimization of heat transfer performance. In addition, existing microchannel heat exchanger design methods generally have problems such as low parameter optimization efficiency and high computational cost. Especially when it is necessary to consider multiple performance indicators such as heat transfer efficiency, flow resistance and heat transfer uniformity at the same time, traditional optimization methods are difficult to find the optimal solution under reasonable computing resources and time costs, which seriously restricts the design efficiency and performance improvement of microchannel heat exchangers. Summary of the invention

[0004] The main purpose of the present invention is to provide a three-dimensional channel structure design method for a microchannel heat exchange module and related devices, so as to effectively improve the optimization accuracy of channel structure parameters.

[0005] To achieve the above object, the present invention provides a three-dimensional channel structure design method for a microchannel heat exchange module, comprising the following steps:

[0006] The heat conduction coefficient and convection heat transfer coefficient of the microchannel heat exchange module are input into the Rayleigh-Ritz coupling model for dynamic analysis to obtain the channel structure parameter data;

[0007] The temperature field and pressure field of the channel structure parameter data are collected, and the time encoder of the dual-time-space autoencoder is used to extract the time series features, and the space encoder of the dual-time-space autoencoder is used to extract the space features, and the time series features and the space features are processed by the co-attention mechanism to obtain the fluid dynamic feature data matrix;

[0008] Inputting the fluid dynamic characteristic data matrix into a recursive window Bayesian network, and obtaining a channel optimization objective function by calculating a heat transfer efficiency weight coefficient, a fluid resistance weight coefficient, and a heat transfer uniformity weight coefficient;

[0009] According to the channel optimization objective function, a genetic algorithm is used to perform global iterative calculations on the channel main structure, and numerical simulation is used to perform local optimization on the flow-dividing and converging structure to obtain a three-dimensional channel structure optimization solution.

[0010] The present invention also provides a three-dimensional channel structure design device for a microchannel heat exchange module, comprising:

[0011] The analysis module is used to input the heat conduction coefficient and convection heat transfer coefficient of the microchannel heat exchange module into the Rayleigh-Ritz coupling model for dynamic analysis to obtain channel structure parameter data;

[0012] An extraction module is used to collect the temperature field and pressure field of the channel structure parameter data, and extract the time series features through the time encoder of the dual-time-space autoencoder, and extract the space features through the space encoder of the dual-time-space autoencoder, and obtain the fluid dynamic feature data matrix after the time series features and the space features are processed by the co-attention mechanism;

[0013] A calculation module, used for inputting the fluid dynamic characteristic data matrix into a recursive window Bayesian network, and obtaining a channel optimization objective function by calculating a heat exchange efficiency weight coefficient, a fluid resistance weight coefficient, and a heat transfer uniformity weight coefficient;

[0014] The optimization module is used to perform global iterative calculation on the main structure of the channel using a genetic algorithm according to the channel optimization objective function, and to perform local optimization on the flow-dividing and converging structure using numerical simulation to obtain a three-dimensional channel structure optimization solution.

[0015] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0016] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0017] In summary, the technical solution provided by the present invention realizes accurate capture of the dynamic characteristics of the fluid in the microchannel by introducing the Rayleigh-Ritz coupling model for dynamic analysis, combined with the dual spatiotemporal autoencoder and the attention mechanism, and effectively improves the optimization accuracy of the channel structure parameters; the recursive window Bayesian network is used to analyze the fluid characteristics, and the channel optimization objective function including heat exchange efficiency, fluid resistance and heat transfer uniformity is established through multi-parameter weight calculation, and the collaborative optimization of multi-objective performance indicators is realized; by combining the genetic algorithm and the SIMPLE numerical solution method, an optimization strategy combining global iterative calculation and local optimization is established, which significantly reduces the consumption of computing resources; by establishing a local optimization model of the diversion and confluence structure and using the gradient descent method for parameter optimization, the problem of uneven fluid distribution in the microchannel is effectively solved; the multi-level optimization method is used to comprehensively optimize the channel structure, and the automation and intelligence of the microchannel heat exchange module structure design are realized, which greatly improves the design efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the steps of a three-dimensional channel structure design method for a microchannel heat exchange module in one embodiment of the present invention;

[0019] Figure 2 It is a structural block diagram of a three-dimensional channel structure design device of a microchannel heat exchange module in one embodiment of the present invention;

[0020] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0021] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0023] Reference Figure 1 This embodiment provides a three-dimensional channel structure design method for a microchannel heat exchange module, comprising the following steps:

[0024] S1, input the heat conduction coefficient and convection heat transfer coefficient of the microchannel heat exchange module into the Rayleigh-Ritz coupling model for dynamic analysis to obtain channel structure parameter data;

[0025] Among them, the microchannel heat exchange module is subjected to characteristic decomposition operation to clarify the boundaries of the functional areas inside the microchannel, and the entire channel is divided into three functional areas: the inlet section, the heat exchange section, and the outlet section. The fluid phase characteristics are analyzed for the boundary data of the functional area. Combined with the state changes of the fluid in the area, the fluid physical property data set is obtained. The fluid physical property data set is a comprehensive data set generated based on the fluid behavior in the area, including physical properties such as density, viscosity, specific heat capacity, thermal conductivity, etc. of the fluid, and combined with the diameter range parameters of the microchannel to ensure the accuracy and practicality of the data analysis. By processing these data, characteristic information reflecting the behavior of the fluid interface is extracted to obtain the characteristic vector of the fluid interface. These characteristic vectors express the heat transfer characteristics between the fluid and the channel wall in a highly abstract manner by calculating the convective heat transfer coefficient and the thermal conductivity coefficient, forming the interface heat transfer characteristic vector. The interface heat transfer characteristic vector is input into the Rayleigh-Ritz coupling model for orthogonal processing, and the complex heat transfer and flow problems are decomposed into a series of independent coupled dynamic equations, thereby reducing the computational complexity and improving the computational efficiency. Based on the coupled dynamic equations, the fluid characteristics inside the microchannel are analyzed to reveal the multidimensional distribution characteristics of the fluid in the microchannel, including velocity distribution, pressure distribution and temperature field distribution. Based on the multidimensional fluid distribution data, the channel cross-sectional area is optimized. The cross-sectional area is adjusted by the optimization algorithm to achieve the goals of minimizing convection resistance, maximizing heat transfer efficiency and optimizing heat transfer uniformity. After the cross-sectional area parameters are optimized, the channel depth is accurately solved in combination with the depth optimization operation. Channel depth is an important geometric parameter that affects the fluid flow path and heat transfer performance. The depth optimization operation must comprehensively consider the fluid distribution characteristics, heat transfer efficiency and manufacturing process limitations to obtain the optimal depth value and channel structure parameter data. The bending angle field equation and the diversion structure field equation are numerically solved based on the channel depth. The bending angle field equation is used to determine the bending shape of the channel in three-dimensional space to ensure that the flow dead zone and pressure loss can be minimized when the fluid flows in the channel; while the diversion structure field equation is used to optimize the diversion and confluence structure inside the channel to improve heat transfer uniformity and heat transfer efficiency.

[0026] S2, collect the temperature field and pressure field of the channel structure parameter data, and extract the temporal features through the time encoder of the dual-temporal and spatial autoencoders, and extract the spatial features through the spatial encoder of the dual-temporal and spatial autoencoders, and obtain the fluid dynamic feature data matrix after processing the temporal features and spatial features with the co-attention mechanism;

[0027] Specifically, according to the channel structure parameter data, the sensor network composed of temperature sensors and pressure sensors performs multi-channel parallel sampling to obtain the original data of temperature field and pressure field. The original sampled data is standardized and normalized to a unified scale to form a normalized time series data set. The normalization process can reduce the difference in data dimension and improve the stability of the deep learning model. The normalized time series data set is input into the dual spatiotemporal autoencoder, and the time series feature is extracted through the time encoder. The structure of the time encoder consists of three one-dimensional convolutional layers, each of which uses the ReLU activation function, so that the dynamic change characteristics in the time dimension can be captured. The role of the one-dimensional convolutional layer is to extract features from the time series data, aggregate the time-related information in the data by sliding windows, and map the dynamic behavior features into time feature maps. The time feature map obtained from the time encoder is input into the space encoder to extract spatial features. The spatial encoder consists of four two-dimensional convolutional layers, each of which is normalized by BatchNormalization and uses LeakyReLU as the activation function. This design can effectively avoid the gradient vanishing problem and enhance the model's ability to extract local spatial features. At this stage, the spatial encoder analyzes the spatial distribution characteristics of the fluid in the channel through a two-dimensional convolution operation, converts it into a spatial feature map, and reveals the distribution law of the temperature field and pressure field in the channel geometry. The spatial feature map is input into the middle layer of the dual spatiotemporal autoencoder for nonlinear transformation to generate a latent feature vector. The structure of the middle layer consists of two fully connected layers, which are used to combine spatial features with temporal features to extract implicit relationships between multi-dimensional features. Through nonlinear transformation of high-dimensional space, feature representation is compressed and refined to make it more expressive, forming a latent feature space. Multi-head attention is calculated on the latent feature vector, and the attention weight matrix is ​​calculated through the multi-head attention mechanism to capture the correlation and importance between different features. The multi-head attention mechanism can focus on different parts of the feature vector at the same time to achieve more comprehensive information integration. By performing linear projection and concatenation operations on the attention weight matrix, the temporal and spatial features are fused to generate a fused feature tensor. The fused feature tensor is input into the feature reconstruction network for upsampling and reconstruction to obtain the final fluid dynamic feature data matrix. The feature reconstruction network contains three transposed convolutional layers, each of which uses the ReLU activation function and residual connection. The feature tensor is gradually enlarged through the transposed convolution operation to restore it to a resolution consistent with the original input data while maintaining high-quality feature expression. The residual connection directly adds the input and output to avoid information loss during the upsampling process, ensuring that the reconstructed fluid dynamic feature data matrix can accurately characterize the fluid behavior and dynamic characteristics inside the channel.

[0028] S3, inputting the fluid dynamic characteristic data matrix into the recursive window Bayesian network, and obtaining the channel optimization objective function by calculating the heat transfer efficiency weight coefficient, the fluid resistance weight coefficient, and the heat transfer uniformity weight coefficient;

[0029] It should be noted that the fluid dynamic characteristic data matrix is ​​input into the Bayesian state estimator in the recursive window Bayesian network, and the state space equation containing the fluid velocity field, temperature field and pressure field is constructed by using the estimator. Combined with the measurement equation, these state space equations are dynamically updated through Kalman filter recursive calculation to generate the state probability distribution. The state probability distribution is weighted by multiple parameters to construct the heat transfer efficiency objective function. By coupling the fluid velocity field and temperature field in the state probability distribution with the parameters related to the heat transfer efficiency, the mathematical expression of the heat transfer efficiency is defined. The maximum a posteriori probability criterion is used to estimate the parameters of the heat transfer efficiency objective function to obtain the initial weight vector of the heat transfer efficiency. By inputting the initial weight vector of the heat transfer efficiency into the weight optimization network and repeatedly training and adjusting it using the optimization iterative algorithm, the heat transfer efficiency weight coefficient is obtained to ensure that it can fully reflect the contribution of the channel structure to the heat transfer performance. The state probability distribution is subjected to fluid resistance analysis to construct a Bayesian probability model containing the Reynolds number and the friction coefficient. Reynolds number and friction coefficient are important parameters for describing fluid resistance in fluid dynamics. By combining these parameters under the Bayesian framework, the posterior probability density function of fluid resistance is defined. Based on the calculation results of the posterior probability density function, the statistical characteristics of fluid resistance can be accurately extracted, and the fluid resistance weight coefficient can be determined accordingly. The calculation of this weight coefficient takes into account the flow pattern, wall friction loss and turbulence characteristics of the fluid in the microchannel, providing important constraints for channel structure optimization. Heat transfer uniformity analysis is performed on the state probability distribution. Heat transfer uniformity is an important indicator for evaluating the heat transfer performance of microchannels. Its analysis process requires the calculation of conditional probability distributions in combination with the variance and skewness coefficient of the temperature field. These conditional probability distributions reflect the heat distribution characteristics of local areas in the microchannel, thereby revealing the uniformity in the heat transfer process. By combining the heat flux density distribution function, the calculation results of the heat transfer uniformity weight are corrected to ensure that it can reflect the actual situation of the heat flux distribution in each area of ​​the channel. Based on these analysis results, the heat transfer uniformity weight coefficient is obtained. A comprehensive penalty function is constructed based on the heat transfer efficiency weight coefficient, the fluid resistance weight coefficient and the heat transfer uniformity weight coefficient. The penalty function consists of three parts: heat transfer efficiency, fluid resistance loss, and heat transfer uniformity. Different performance objectives are balanced in a weighted manner. The heat transfer efficiency term aims to maximize the heat transfer capacity of the channel, while the fluid resistance loss term is used to minimize the negative impact of fluid resistance on system performance. The heat transfer uniformity term ensures the smoothness of heat distribution and reduces local overheating or insufficient cooling. The penalty function is used as the channel optimization objective function in the overall optimization framework. By optimizing and solving it, the optimal three-dimensional channel structure design solution is found under multi-objective constraints to achieve the best balance between performance, efficiency, and uniformity of the microchannel heat exchange module.

[0030] S4, according to the channel optimization objective function, the genetic algorithm is used to perform global iterative calculations on the main channel structure, and numerical simulation is used to perform local optimization on the flow diversion and confluence structure to obtain a three-dimensional channel structure optimization solution.

[0031] Specifically, according to the definition of the optimization objective function, a genetic optimization model of the channel main structure is established. In this process, the value range of the optimization parameters is clarified, including the value range of the channel cross-sectional area, the value range of the channel depth, and the value range of the bending angle. These range constraints constitute the parameter space of the optimization model. The optimization parameter coding sequence is generated using these parameter spaces, and the parameters are discretized in the form of binary or floating-point coding to provide a unified expression for the population operation of the genetic algorithm. Based on the optimization parameter coding sequence, the initial population is randomly generated. The diversity of the initial population directly determines the global search ability of the genetic algorithm, so the population size and parameter distribution need to be reasonably designed. After the initial population is generated, a population fitness evaluation system is constructed. The core of the fitness evaluation system is to evaluate the performance of each individual in the population based on the channel optimization objective function. The fitness function comprehensively considers the objectives such as heat exchange efficiency, fluid resistance and heat transfer uniformity, and determines the fitness of each individual by calculating the weighted penalty value of each objective. After completing the fitness evaluation, the initial population is selected according to the fitness value, and strategies such as roulette or tournament selection are adopted to prioritize individuals with higher fitness to form the parent population. Genetic crossover operations are performed on the parent population to simulate the recombination process in natural inheritance. The offspring population is generated by randomly selecting some genes of the parent individuals and exchanging their positions, thereby improving the population diversity and exploring new solution spaces. After the obtained offspring population is interpreted as the actual channel structure parameters, the fluid control equations are numerically solved by the SIMPLE algorithm to obtain the flow field distribution characteristics corresponding to each body. The SIMPLE algorithm is a fluid dynamics solution method that can efficiently calculate the flow behavior of incompressible fluids. Its core is to gradually approach the convergent solution through pressure-velocity coupling iteration. At this stage, the velocity, pressure and temperature distribution characteristics of the flow field are analyzed to provide accurate data support for the subsequent optimization of the diversion and confluence structure. Based on the flow field distribution characteristics, a local optimization model of the diversion and confluence structure is constructed. The model uses the diversion ratio and confluence angle as the main optimization parameters. By imposing constraints on these parameters, such as the range limit of the diversion ratio and the design rules of the confluence angle, it ensures that the optimization results meet the actual manufacturing and application requirements. The gradient descent method is used to iteratively calculate the local optimization model. By gradually adjusting the parameters to minimize the error of the objective function, the local optimal parameters of the diversion and confluence structure are obtained. The convergence and computational efficiency of the gradient descent method in this process directly determine the effect of local optimization. Therefore, it is necessary to combine intelligent initial value setting and learning rate adjustment strategy to improve the optimization performance. The optimal parameters of the local structure are combined with the global optimization individual parameters for multi-level optimization. Multi-level optimization balances the overall performance of the channel main structure and the diversion and confluence structure by comprehensively considering the coupling relationship between global parameters and local parameters.In this process, a hierarchical optimization strategy is adopted to first fix the local parameters and optimize the global structure, and then fix the global parameters and adjust the local structure until the global and local synergy is achieved. After multiple iterations and fusions, a complete three-dimensional channel structure optimization solution is generated.

[0032] The gene coding sites of the parent population are selected, and the gene crossover site matrix is ​​generated by analyzing the coding structure in the parent gene sequence. The matrix identifies the distribution of sites suitable for crossover operations in the gene sequence. The selection of these crossover sites should fully consider gene diversity and structural characteristics to avoid potential genetic drift and maintain the evolutionary ability of the population. After the construction of the gene crossover site matrix is ​​completed, the parent individuals are paired according to the matrix to form a gene fragment group to be crossed. The pairing principle is based on the natural selection mechanism of the genetic algorithm, which gives priority to combining individuals with high fitness, while retaining a certain probability to select individuals with low fitness to enhance the diversity and exploration of the population. A two-point crossover operation is performed on the gene fragment group to be crossed. Two-point crossover is a gene recombination method. By selecting two crossover points, the parent gene fragments are exchanged between the two points to generate a new daughter gene sequence. The recombined daughter gene sequence is parameter decoded, and the discretized gene sequence is mapped to the actual microchannel structure parameter value, including the channel cross-sectional area, depth, bending angle and other geometric features. After decoding, the SIMPLE algorithm is used to perform fluid dynamics calculations on these actual parameter values. The SIMPLE algorithm is an efficient algorithm for solving the fluid control equations and is suitable for numerical simulation of incompressible fluids. In the iterative solution process of the SIMPLE algorithm, the geometric parameters of the microchannel structure are substituted into the continuity equation and momentum equation of the fluid, and the numerical solutions of the pressure field and velocity field are gradually updated. The pressure correction and velocity field correction are performed alternately until convergence, and a stable solution that satisfies the control equations is obtained. After the numerical solutions of the pressure field and velocity field are completed, the heat transfer process in the microchannel is calculated. The heat transfer calculation adopts a combination of the finite volume method and the central difference format. The finite volume method divides the entire calculation domain into several control volumes. By discretizing the energy conservation equations on the control volumes, the heat transfer process in the microchannel can be effectively calculated. The central difference format is used to discretize the spatial derivative terms with high precision to ensure the accuracy of the temperature field calculation. Through this calculation process, the temperature field distribution results at each position in the channel are obtained, thereby supplementing the solution of the fluid control equation. The numerical solutions of the pressure field and velocity field and the temperature field distribution results are used to analyze the fluid characteristics. By extracting key characteristics of the fluid such as local velocity gradient, pressure change rate and temperature distribution uniformity, the dynamic behavior of the fluid in the microchannel and its coupling relationship with the channel geometry are revealed. By comprehensively evaluating the flow field distribution characteristics of each individual, the offspring individuals with excellent performance are identified.

[0033] The flow field distribution characteristics are analyzed, and the core information describing the flow field characteristics is extracted by calculating the velocity distribution function, pressure distribution function and temperature distribution function of the fluid in the channel. These functions reflect the motion state, pressure change law and temperature distribution characteristics of the fluid in the channel respectively. By analyzing these data, a set of flow field characteristic functions is generated. Based on the set of flow field characteristic functions, a local optimization objective function is constructed to form an optimization model. The local optimization objective function consists of multiple performance indicators, such as minimizing flow resistance, maximizing heat transfer efficiency and optimizing the uniformity of temperature distribution. These performance indicators need to be optimized within the parameter constraints of the local structure. Parameter constraints are set for the optimization model, and the optimization variable space is constructed by limiting the value range of the diversion ratio R∈[0.4,0.6] and the confluence angle θ∈[30°,60°]. The optimization variable space is input into the gradient calculation module to calculate the partial derivative of the objective function with respect to the optimization variable and obtain the parameter gradient vector. Each dimension of the gradient vector corresponds to the change direction and change rate of an optimization variable, which is the core information in the optimization process. Based on these gradient vectors, a parameter update equation is constructed to define the iterative update rules in the optimization process. The parameter update equation adopts the standard gradient descent form, in which the update amplitude of each step is controlled by the learning rate. In order to accelerate convergence and avoid falling into the local optimum, the optimization strategies such as momentum term or adaptive learning rate are combined to enhance the robustness of gradient descent. When applying the iterative update rule, the convergence of the optimization process is strictly analyzed. The key to the convergence analysis is to calculate the change of the optimization parameters between adjacent iterations. If the change gradually decreases to a certain acceptable range, it means that the optimization process has converged to the optimal parameters of the local structure. By comparing the parameter change with the preset convergence threshold, it is judged whether the optimization process reaches the stop condition. After meeting the convergence criterion, the optimal split ratio and the optimal confluence angle are obtained. These two parameters directly determine the performance of the splitter and confluence structure. Based on the optimal split ratio and the optimal confluence angle, the splitter and confluence structure of the microchannel is reconstructed. The spacing of the splitter plates and the angle of the confluence are adjusted to match the geometric parameters required by the optimization results. The adjustment of the splitter plate spacing affects the velocity distribution of the fluid in the splitter section, while the optimization of the confluence angle directly affects the mixing effect and flow resistance of the fluid. By fine-tuning these geometric parameters, the optimal design of the local structure is generated.

[0034] In one example, the heat conduction coefficient and convection heat transfer coefficient of the microchannel heat exchange module are input into the Rayleigh-Ritz coupling model for dynamic analysis to obtain channel structure parameter data, including:

[0035] Perform characteristic decomposition operation on the microchannel heat exchange module to obtain the boundary data of the functional area including the inlet section, the heat exchange section, and the outlet section;

[0036] The fluid phase characteristics of the functional area boundary data are analyzed to obtain the fluid physical property data set, and the fluid interface characteristics are extracted from the fluid physical property data set and the microchannel diameter range parameters. The interface heat transfer characteristic vector is obtained by calculating the convection heat transfer coefficient and the thermal conductivity coefficient.

[0037] The interface heat transfer eigenvector is input into the Rayleigh-Ritz coupling model for orthogonalization to obtain the coupled dynamic equations, and the fluid characteristics are analyzed based on the coupled dynamic equations to obtain multi-dimensional fluid distribution data;

[0038] The cross-sectional area is optimized based on the multi-dimensional fluid distribution data to obtain the channel cross-sectional area parameters, and the channel cross-sectional area parameters are deeply optimized to obtain the channel depth value;

[0039] The bending angle field equation and the diversion structure field equation are numerically solved based on the channel depth to obtain the channel structure parameter data.

[0040] In this example, the microchannel heat exchange module is subjected to feature decomposition to obtain the boundary data of the functional area including the inlet section, heat exchange section, and outlet section. Assume that the total length of the microchannel is , the fluid velocity distribution is , the pressure distribution is , the temperature gradient is , through feature recognition, determine the regional demarcation point and , the following conditions are met:

[0041] ;

[0042] in The formula indicates that the inlet section is mainly dominated by the pressure gradient, the heat exchange section is characterized by a significant temperature gradient, and the outlet section is characterized by a redistribution of pressure and velocity. The fluid phase characteristics are analyzed on the functional area boundary data to obtain the fluid property data set. The fluid property data is composed of density , Viscosity Specific heat capacity and thermal conductivity Characterization, the calculation of which relies on the fluid state equation:

[0043] ;

[0044] in For pressure, is the gas constant, is the temperature. Combining the pressure and temperature distribution of the region, the distribution function of the fluid physical parameters at different locations is calculated. By combining the fluid physical property data and the microchannel diameter range , extract the fluid interface characteristics. Convective heat transfer coefficient and thermal conductivity The calculation formulas are:

[0045] ;

[0046] in is the Nusselt number, is the Prandtl number, satisfying By calculating the and , and obtain the interface heat transfer eigenvector The interface heat transfer eigenvector is input into the Rayleigh-Ritz coupling model for orthogonalization to obtain the coupled dynamic equations. Assume that the temperature field is , the heat transfer potential energy function is:

[0047] ;

[0048] Solving by calculus of variations , the temperature field Decompose into a linear combination of a set of orthogonal basis functions:

[0049] ;

[0050] in are orthogonal basis functions, is the corresponding weight coefficient, generating a coupled dynamic equation set to describe the multidimensional fluid distribution. Based on the coupled dynamic equation set, the channel cross-sectional area Optimize. The distribution of cross-sectional area must satisfy the fluid continuity equation:

[0051] ;

[0052] By adjusting To optimize heat transfer performance. Calculated by the following formula:

[0053] ;

[0054] in is the temperature difference. Optimization To make Maximize. Deeply optimize the channel cross-sectional area parameters and calculate the channel depth . Deep optimization is based on the energy conservation equation:

[0055] ;

[0056] in It represents the heat flow carried by the fluid per unit time, by adjusting To achieve the best heat transfer effect. Solve the bending angle field equation and the diversion structure field equation to obtain the channel structure parameter data. Assume that the bending angle field is , the split ratio is , satisfying the condition of minimum fluid resistance:

[0057] ;

[0058] The optimization constraint of the split ratio is:

[0059] ;

[0060] in and is the flow rate of the diversion channel. Through the above optimization process, the three-dimensional channel structure parameters are obtained to achieve an efficient design scheme for the microchannel heat exchange module.

[0061] In one example, the temperature field and pressure field of the channel structure parameter data are collected, and the temporal features are extracted by the time encoder of the dual spatiotemporal autoencoder, and the spatial features are extracted by the spatial encoder of the dual spatiotemporal autoencoder. After the temporal features and the spatial features are processed by the co-attention mechanism, the fluid dynamic feature data matrix is ​​obtained, including:

[0062] According to the channel structure parameter data, the sensor network composed of temperature sensors and pressure sensors performs multi-channel parallel sampling to obtain the original data of temperature field and pressure field;

[0063] The original data of the temperature field and the original data of the pressure field are standardized to obtain a normalized time series data set;

[0064] The normalized time series dataset is input into the time encoder of the dual spatiotemporal autoencoder to extract time series features and obtain a time feature map. The time encoder contains three one-dimensional convolutional layers, and each one-dimensional convolutional layer uses a ReLU activation function.

[0065] The temporal feature map is input into the spatial encoder of the dual spatiotemporal autoencoder for spatial feature extraction to obtain the spatial feature map. The spatial encoder contains four two-dimensional convolutional layers, each of which uses Batch Normalization and LeakyReLU activation functions.

[0066] The spatial feature map is input into the middle layer of the dual spatiotemporal autoencoder for nonlinear transformation to obtain the potential feature vector. The middle layer contains two fully connected layers.

[0067] Perform multi-head attention calculation on the potential feature vector to obtain the attention weight matrix, and perform linear projection and splicing operations on the attention weight matrix to obtain the fused feature tensor;

[0068] The fused feature tensor is input into the feature reconstruction network for upsampling and reconstruction to obtain the fluid dynamic feature data matrix. The feature reconstruction network contains 3 transposed convolutional layers, each of which uses the ReLU activation function and residual connection.

[0069] In this example, based on the channel structure parameter data, the sensor network composed of temperature sensors and pressure sensors performs multi-channel parallel sampling. By arranging the sensor array, the temperature of the fluid is collected in real time in key areas of the channel (such as the inlet section, heat exchange section, and outlet section). and pressure Distribution, get the original data of temperature field changing with time And the original data of pressure field . is the spatial position, is a time variable. The sampling rate of temperature and pressure data follows ,satisfy To ensure signal integrity, aliasing is avoided according to the Nyquist sampling theorem. The raw data is normalized to eliminate the difference in data dimensions and form a normalized time series data set. The normalization process is completed by the following formula:

[0070] ;

[0071] in and are the mean values ​​of temperature and pressure, respectively. and The standardized data and Composition of normalized time series dataset , providing a unified input for subsequent feature extraction. The normalized time series dataset is input into the dual spatiotemporal autoencoder for processing. The time series features are extracted through the time encoder. The time encoder contains three one-dimensional convolutional layers, with convolution kernels Extract the change pattern in the time dimension. For the input data , the output of each convolutional layer is calculated as:

[0072] ;

[0073] in For the The convolution kernel weight matrix of the layer, is the bias term, * represents the convolution operation, is the activation function. After 3 layers of convolution, we get the temporal feature map , this map contains the dynamic pattern of temperature and pressure data changing over time. The temporal feature map is input into the spatial encoder for spatial feature extraction. The spatial encoder contains four 2D convolutional layers. The convolution operation is performed through a 2D convolution kernel Extract the spatial distribution characteristics, and the calculation formula is:

[0074] ;

[0075] Where LeakyReLU is the activation function, and BatchNorm represents the batch normalization operation, which is used to accelerate training and reduce overfitting. After 4 layers of convolution, the spatial feature map is obtained. , reflecting the spatial distribution of temperature and pressure fields. The spatial feature map is input into the middle layer of the dual spatiotemporal autoencoder for nonlinear transformation to generate the potential feature vector z. The middle layer consists of two fully connected layers, and the calculation formula is:

[0076] ;

[0077] in and is the weight matrix, and is the bias term, is the activation function. After obtaining the potential feature vector, the attention weight matrix is ​​calculated through the multi-head attention mechanism , used to capture the correlation between different features. The weight calculation formula for each head is:

[0078] ;

[0079] in are query, key, and value matrices respectively, is the feature dimension. Through linear projection and concatenation, the results of each head are integrated into a fused feature tensor The fused feature tensor is input into the feature reconstruction network for upsampling and reconstruction to generate the fluid dynamic feature data matrix. The feature reconstruction network consists of three transposed convolutional layers, and the output calculation formula of each layer is:

[0080] ;

[0081] in Represents the transposed convolution operation. The initial input and output are added through the residual connection to ensure that the feature information is not lost. The final fluid dynamic feature data matrix is The spatiotemporal dynamic distribution of temperature and pressure fields is accurately characterized.

[0082] In one example, the fluid dynamic characteristic data matrix is ​​input into the recursive window Bayesian network, and the channel optimization objective function is obtained by calculating the heat transfer efficiency weight coefficient, the fluid resistance weight coefficient, and the heat transfer uniformity weight coefficient, including:

[0083] The fluid dynamic characteristic data matrix is ​​input into the Bayesian state estimator in the recursive window Bayesian network to construct the state space equation including the fluid velocity field, temperature field and pressure field, and the Kalman filter recursive calculation is performed in combination with the measurement equation to obtain the state probability distribution;

[0084] The multi-parameter weight calculation is performed on the state probability distribution, the heat transfer efficiency objective function is constructed, and the maximum a posteriori probability MAP criterion is used for parameter estimation to obtain the initial weight vector of heat transfer efficiency.

[0085] The initial weight vector of heat transfer efficiency is input into the weight optimization network for optimization iteration to obtain the weight coefficient of heat transfer efficiency;

[0086] Perform fluid resistance analysis on the state probability distribution, build a Bayesian probability model including Reynolds number and friction coefficient, and calculate the posterior probability density function to obtain the fluid resistance weight coefficient;

[0087] Based on the state probability distribution, the heat transfer uniformity analysis is carried out, the conditional probability distribution of the temperature field variance and skewness coefficient is calculated, and the heat transfer uniformity weight coefficient is obtained by combining the heat flux density distribution function;

[0088] Based on the heat transfer efficiency weight coefficient, fluid resistance weight coefficient and heat transfer uniformity weight coefficient, a penalty function including the heat transfer efficiency term, fluid resistance loss term and heat transfer uniformity term is constructed to obtain the channel optimization objective function.

[0089] In this example, the fluid dynamics data matrix The input is sent to the Bayesian state estimator in the recursive window Bayesian network. By constructing the state space equation containing the fluid velocity field, temperature field and pressure field, and combining it with the measurement equation for Kalman filter recursive calculation, the state probability distribution is obtained. The basic form of the state space equation includes the state transfer equation and the measurement equation. The state transfer equation describes the dynamic evolution of the system in the form of:

[0090] ;

[0091] in Indicates The state vector at time t, including the velocity field , Temperature field and pressure field is the state transfer matrix, which defines the dynamic laws of the system; is process noise, which follows a zero-mean normal distribution ,in is the covariance matrix of the process noise. The measurement equation describes the relationship between the observations and the state in the form of:

[0092] ;

[0093] in is the measurement vector, containing sensor data of temperature and pressure; is the measurement matrix; is the measurement noise, which follows a normal distribution ,in is the covariance matrix of the measurement noise. The estimated value of the state is recursively calculated through the Kalman filter algorithm, which is divided into a prediction step and an update step. The prediction step is:

[0094] ;

[0095] The update steps are:

[0096] ;

[0097] ;

[0098] in is the estimated value of the state vector, is the state covariance matrix, is the Kalman gain matrix. Based on the state probability distribution, the multi-parameter weights of the fluid are calculated to construct the heat transfer efficiency objective function. The heat transfer efficiency objective function is based on the local convection heat transfer coefficient As the core, it is defined as:

[0099] ;

[0100] in is the local temperature difference, is the area of ​​the heat exchange region. The objective function is estimated by the maximum a posteriori probability criterion to find the optimal parameters satisfy:

[0101] ;

[0102] in is the likelihood function, is the prior distribution. The initial weight vector of heat transfer efficiency is obtained through the optimization process The initial weight vector of heat transfer efficiency is input into the weight optimization network for optimization iteration. The optimization goal is to minimize the following loss function:

[0103] ;

[0104] in is the target heat transfer efficiency, is the model prediction value. It is adjusted iteratively through back-propagation and gradient descent , and obtain the final heat transfer efficiency weight coefficient. At the same time, the fluid resistance analysis is performed on the state probability distribution to construct a model including the Reynolds number and friction coefficient The Bayesian probability model, Reynolds number is defined as:

[0105] ;

[0106] in is the density, is the average speed, is the channel diameter, is the dynamic viscosity. Friction coefficient Calculated using the Darcy-Weisbach equation:

[0107] ;

[0108] in is the voltage drop, is the channel length. The Bayesian model is based on the posterior probability density function Estimate the fluid resistance weight coefficient. Perform heat transfer uniformity analysis based on state probability distribution and define the variance of the temperature field. and the skewness coefficient :

[0109] ;

[0110] in is the average temperature. Combined with the heat flux distribution function , and obtain the heat transfer uniformity weight coefficient. The heat transfer efficiency weight coefficient, fluid resistance weight coefficient and heat transfer uniformity weight coefficient are combined to construct the penalty function F:

[0111] ;

[0112] in is the weight factor, and the channel optimization objective function is obtained and used for structural optimization design.

[0113] In one example, according to the channel optimization objective function, a genetic algorithm is used to perform global iterative calculations on the channel main structure, and numerical simulation is used to perform local optimization on the flow-dividing and converging structure, to obtain a three-dimensional channel structure optimization solution, including:

[0114] A genetic optimization model of the channel main structure is established according to the channel optimization objective function, and the value range of the channel cross-sectional area, the value range of the channel depth and the value range of the bending angle are determined to generate an optimization parameter coding sequence;

[0115] Generate an initial population based on the optimized parameter coding sequence, build a population fitness evaluation system based on the initial population, and perform a selection operation on the initial population according to the population fitness evaluation system to obtain a parent population;

[0116] Perform gene crossover operation on the parent population to obtain the offspring population, and interpret the offspring population into channel structure parameters. Use SIMPLE algorithm to numerically solve the fluid control equations to obtain the flow field distribution characteristics of each individual.

[0117] Based on the flow field distribution characteristics, a local optimization model of the flow diversion and confluence structure is constructed, and the optimal parameters of the local structure are obtained by gradient descent iteration through parameter constraints of the diversion ratio and confluence angle.

[0118] Multi-level optimization is performed on the local structural optimal parameters and the globally optimized individual parameters to obtain the three-dimensional channel structure optimization solution.

[0119] In this example, a genetic optimization model of the channel main structure is established according to the channel optimization objective function, and the value range of key geometric parameters is set, including the channel cross-sectional area. , channel depth and bending angle These parameters directly affect the heat transfer performance, fluid resistance and heat transfer uniformity. In the optimization process, it is necessary to establish a reasonable parameter encoding method to discretize the continuous geometric parameter space into an optimized parameter encoding sequence. Parameter encoding uses binary or floating point encoding. For example, the cross-sectional area The value range is set to , channel depth The value range is , bending angle The value range is Taking binary coding as an example, assuming that the coding length of the cross-sectional area is , then the discretization expression of the cross-sectional area is:

[0120] ;

[0121] in is the binary code corresponding to the cross-sectional area, int Represents the operation of converting binary to decimal. Similar encoding methods are applied to channel depth and bending angle to generate a complete encoding sequence of optimization parameters. Based on these encoding sequences, the initial population is randomly generated. , where each individual is a combination of a set of parameter encodings, representing a possible channel design solution. Determined according to the complexity of the problem and computing resources. A fitness evaluation system is constructed for each population individual, and the fitness function is based on the optimization objective function Definition, including heat transfer efficiency , fluid resistance and heat transfer uniformity Three parts:

[0122] ;

[0123] in is the weight coefficient used to balance the priorities among various objectives. The calculation formula is:

[0124] ;

[0125] Fluid resistance Calculated using the Darcy-Weisbach equation:

[0126] ;

[0127] in is the pressure drop, is the channel diameter, is the channel length, is the fluid velocity. Heat transfer uniformity Represents the variance of the temperature distribution:

[0128] ;

[0129] in is the average temperature of the region. Based on the fitness value, the initial population is selected using methods including roulette selection and tournament selection, and the parent population is formed by preferentially retaining individuals with higher fitness. Based on the parent population, a gene crossover operation is performed, and the offspring population is generated using the two-point crossover method. For two parent individuals, their binary codes exchange gene fragments at two randomly selected intersections to generate new offspring individuals. After the offspring population is interpreted as channel structure parameters, the SIMPLE algorithm is used to numerically solve the fluid control equations to obtain the flow field distribution characteristics corresponding to each individual. The fluid control equations include the continuity equation and the momentum equation:

[0130] ;

[0131] ;

[0132] in is the velocity vector field, For pressure, is the dynamic viscosity. Through numerical iterative calculation, the velocity field and pressure field distribution of each individual under the specified channel structure parameters are obtained. Based on the flow field distribution characteristics, a local optimization model of the flow diversion and confluence structure is constructed. The local optimization parameters include the diversion ratio and confluence angle , the optimization goal is to minimize fluid resistance and improve mixing uniformity. The constraints are , The optimization process uses the gradient descent method, and the objective function is:

[0133] ;

[0134] By calculating the partial derivative of the objective function with respect to the optimization variable, the gradient update formula is obtained:

[0135] ;

[0136] in is the learning rate, Represents the number of iterations. Through gradient update iteration, it finally converges to the optimal parameters of the local structure. The optimal parameters of the local structure are integrated with the global optimization individual parameters to perform multi-level optimization. Global optimization focuses on the adjustment of the overall channel geometry, while local optimization focuses on the detailed design of key parts. Combining the optimization results of the two parts, a complete three-dimensional channel structure optimization solution is generated to achieve a comprehensive improvement in heat exchange efficiency, fluid resistance and heat transfer uniformity.

[0137] In one example, a gene crossover operation is performed on the parent population to obtain a child population, which is then interpreted as channel structure parameters. The fluid control equations are numerically solved using the SIMPLE algorithm to obtain the flow field distribution characteristics of each individual, including:

[0138] Select coding sites based on the parent population to obtain a gene crossover site matrix, and pair the parent individuals according to the gene crossover site matrix to obtain a gene segment group to be crossed;

[0139] Performing a double-point crossover operation on the gene fragment group to be crossed to obtain a recombined offspring gene sequence, and performing parameter decoding on the offspring gene sequence to obtain the actual parameter value of the microchannel structure;

[0140] Through SIMPLE algorithm, the actual parameter values ​​of the microchannel structure are input into the fluid control equations for iterative solution to obtain the numerical solutions of the pressure field and velocity field.

[0141] The heat transfer calculation is performed by the finite volume method and the central difference format to obtain the temperature field distribution results;

[0142] The numerical solutions of the pressure field and velocity field and the temperature field distribution results are used to analyze the fluid characteristics and obtain the flow field distribution characteristics of each individual.

[0143] In this example, coding sites are selected from the gene sequences of the parent population to generate a gene crossover site matrix. Assume that the parent population contains individuals, and the length of the gene sequence of each individual is , the crossover sites are determined by random selection or heuristic methods. Gene crossover site matrix is a A Boolean matrix, where Indicates The parent individual The gene loci are crossover points. Based on the crossover site matrix, the parent individuals are paired. The pairing is done randomly or in a roulette wheel matching according to the fitness ranking. and , extract its gene fragment group, where the gene fragment between the first intersection point and the second intersection point is the part to be crossed. Assume that the gene fragment group is , generate the recombined offspring gene sequence through double-point crossover operation The two-point crossover formula is:

[0144] ;

[0145] ;

[0146] in and are the indices of the two intersection points, Represents the fragment splicing operation. After the crossover operation is completed, the recombined offspring gene sequence is parameter decoded to obtain the actual parameter value of the microchannel structure. Assume that the encoding involves three main parameters: channel cross-sectional area , channel depth and bending angle Taking the channel cross-sectional area as an example, the decoding formula of the gene sequence is:

[0147] ;

[0148] in is the binary encoding of the cross-sectional area, is the encoding length. A similar decoding process can be used for channel depth and bending angle. The actual parameter values ​​of the microchannel structure are input into the control equations of fluid dynamics and solved iteratively using the SIMPLE algorithm. The SIMPLE algorithm is suitable for numerical calculations of incompressible fluids. The control equations include the continuity equation and the momentum equation:

[0149] ;

[0150] ;

[0151] in is the velocity field, is the pressure field, is the fluid density, is the dynamic viscosity. The SIMPLE algorithm iteratively updates the solutions of the velocity field and pressure field through the pressure correction step. Assume that The predicted pressure field of the iteration is , by solving the pressure correction equation:

[0152] ;

[0153] Update Pressure Field and velocity field ,in and is the correction amount. After obtaining the pressure field and velocity field, the heat transfer equation is calculated by the finite volume method and the central difference format to obtain the distribution of the temperature field. The heat transfer equation is:

[0154] ;

[0155] in is the temperature field, is the specific heat capacity, is the thermal conductivity, is the heat source term. By discretizing, the equation is integrated in the control volume to obtain the discrete form:

[0156] ;

[0157] in and are the temperature values ​​at the center and on the surface of the control volume, respectively. and is the coefficient of dispersion, The central difference format is used to calculate the spatial derivative term, and a stable temperature field distribution is obtained through iterative solution. The numerical solution of the pressure field, velocity field and temperature field distribution results are used to analyze the fluid characteristics and extract key performance indicators, including the local velocity gradient. , Pressure Drop and temperature distribution uniformity For example, the temperature distribution uniformity is calculated by the following formula:

[0158] ;

[0159] in is the average temperature. By analyzing these characteristic indicators, the performance of each offspring individual is evaluated, providing a basis for subsequent population evolution.

[0160] In one example, a local optimization model of the flow diversion and confluence structure is constructed based on the flow field distribution characteristics, and gradient descent iteration is performed through parameter constraints of the diversion ratio and confluence angle to obtain the optimal parameters of the local structure, including:

[0161] Conduct characteristic analysis on the flow field distribution characteristics, and obtain the flow field characteristic function set by calculating the velocity distribution function, pressure distribution function and temperature distribution function of the fluid in the channel;

[0162] Based on the flow field characteristic function set, a local optimization objective function is constructed to obtain a local optimization model, and parameter constraints are set for the local optimization model. By limiting the value range of the diversion ratio R∈[0.4,0.6] and the confluence angle θ∈[30°,60°], the optimization variable space is obtained.

[0163] The optimization variable space is input into the gradient calculation module, and the parameter gradient vector is obtained by calculating the partial derivative of the objective function with respect to the optimization variable. The parameter update equation is constructed based on the parameter gradient vector to obtain the iterative update rule.

[0164] The convergence analysis of the iterative update rule is carried out. By calculating the parameter changes of adjacent iteration steps, the convergence criterion is obtained. The convergence criterion is compared with the preset threshold to obtain the optimal diversion ratio and the optimal confluence angle.

[0165] The microchannel flow diversion and confluence structure is reconstructed based on the optimal diversion ratio and the optimal confluence angle, and the optimal parameters of the local structure are obtained by updating the distance between the diversion plates and the confluence angle.

[0166] In this example, the flow field distribution characteristics are analyzed by calculating the velocity distribution function, pressure distribution function, and temperature distribution function of the fluid in the channel, extracting the key relationship between fluid dynamics and heat transfer characteristics, and constructing a flow field characteristic function set. Assume that the velocity field is , the pressure field is , the temperature field is , which are obtained through numerical simulation or experimental measurement. Velocity distribution function The characteristics of the local velocity gradient and the divergence of the velocity vector field Description, where:

[0167] ;

[0168] in is the velocity vector, is the spatial coordinate. Similarly, the pressure distribution function is given by the pressure gradient The temperature distribution function is expressed by the mean value of the temperature field and variance To characterize:

[0169] ;

[0170] in is the total volume in the microchannel. Through the above characteristic analysis, the flow field characteristic function set containing the velocity field, pressure field and temperature field characteristics is obtained. Based on the flow field characteristic function set, a local optimization objective function is constructed to evaluate the performance of the flow diversion and confluence structure. The objective function comprehensively considers the heat transfer efficiency, fluid resistance and heat transfer uniformity, and the form is:

[0171] ;

[0172] in is the heat transfer efficiency, is the fluid resistance, is the uniformity index of the temperature field, is a weighting coefficient used to balance the priorities of different objectives. represents the heat transfer capacity of the channel and is defined as:

[0173] ;

[0174] in is the local heat transfer coefficient, is the local temperature difference, is the heat transfer surface area. Fluid resistance Calculation based on the Darcy-Weisbach equation:

[0175] ;

[0176] in is the pressure drop, is the channel diameter, is the channel length, is the fluid density, is the flow rate. The key to optimizing the objective function is to constrain the parameter space and ensure the split ratio and confluence angle Within the physically reasonable range. The value range is [0.4,0.6], the confluence angle The value range is , The optimization variable space is input into the gradient calculation module, and the objective function Take partial derivatives of the optimization variables and get the parameter gradient vector . Use the gradient vector to construct the parameter update equation and define the update rule for each iteration as:

[0177] ;

[0178] in is the learning rate, is the number of iterations. The goal of parameter updating is to find the optimal and During the update process, a convergence analysis is performed by calculating the parameter changes between adjacent iteration steps. and , to determine whether the convergence criterion is met:

[0179] ;

[0180] in is the preset convergence threshold. When the changes of all parameters are less than the threshold, the iteration process is terminated and the optimal split ratio is output. and the optimal confluence angle Based on the optimal split ratio and confluence angle, the splitter and confluence structure of the microchannel is reconstructed. The splitter structure is optimized by adjusting the spacing between the splitter plates. and the geometry of the diversion area, while the confluence structure is achieved by optimizing the confluence angle The optimization of these geometric parameters combined with fluid mechanics simulation ensures uniform distribution of the fluid in the diversion section and mixing effect in the confluence section.

[0181] Reference Figure 2 This embodiment provides a three-dimensional channel structure design device for a microchannel heat exchange module, comprising:

[0182] Analysis module 1, used for inputting the heat conduction coefficient and convection heat transfer coefficient of the microchannel heat exchange module into the Rayleigh-Ritz coupling model for dynamic analysis to obtain channel structure parameter data;

[0183] Extraction module 2 is used to collect the temperature field and pressure field of the channel structure parameter data, and extract the time series features through the time encoder of the dual-time-space autoencoder, and extract the space features through the space encoder of the dual-time-space autoencoder, and obtain the fluid dynamic feature data matrix after the time series features and space features are processed by the co-attention mechanism;

[0184] The calculation module 3 is used to input the fluid dynamic characteristic data matrix into the recursive window Bayesian network, and obtain the channel optimization objective function by calculating the heat transfer efficiency weight coefficient, the fluid resistance weight coefficient, and the heat transfer uniformity weight coefficient;

[0185] The optimization module 4 is used to perform global iterative calculation on the main structure of the channel using a genetic algorithm according to the channel optimization objective function, and to perform local optimization on the flow-dividing and converging structure using numerical simulation to obtain a three-dimensional channel structure optimization solution.

[0186] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0187] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0188] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0189] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0190] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0191] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0192] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A three-dimensional channel structure design method for a microchannel heat exchange module, characterized in that: The following steps are involved: The heat conduction coefficient and convection heat transfer coefficient of the microchannel heat exchange module are input into the Rayleigh-Ritz coupling model for dynamic analysis to obtain the channel structure parameter data; The temperature field and pressure field of the channel structure parameter data are collected, and the time encoder of the dual-time-space autoencoder is used to extract the time series features, and the space encoder of the dual-time-space autoencoder is used to extract the space features, and the time series features and the space features are processed by the co-attention mechanism to obtain the fluid dynamic feature data matrix; Specifically, the method comprises: performing multi-channel parallel sampling on a sensor network composed of a temperature sensor and a pressure sensor according to the channel structure parameter data to obtain raw temperature field data and raw pressure field data; performing standardization processing on the raw temperature field data and the raw pressure field data to obtain a normalized time series data set; inputting the normalized time series data set into a time encoder of a dual spatiotemporal autoencoder to extract time series features to obtain a time feature map, wherein the time encoder comprises three one-dimensional convolutional layers, and each one-dimensional convolutional layer uses a ReLU activation function; inputting the time feature map into a spatial encoder of a dual spatiotemporal autoencoder to extract spatial features to obtain a spatial feature map, wherein the spatial encoder comprises four two-dimensional convolutional layers, and each two-dimensional convolutional layer uses a Batch Normalization and LeakyReLU activation functions; inputting the spatial feature map into the middle layer of the dual spatiotemporal autoencoder for nonlinear transformation to obtain a potential feature vector, wherein the middle layer includes two fully connected layers; performing multi-head attention calculation on the potential feature vector to obtain an attention weight matrix, and performing linear projection and concatenation operations on the attention weight matrix to obtain a fused feature tensor; inputting the fused feature tensor into a feature reconstruction network for upsampling and reconstruction to obtain a fluid dynamic feature data matrix, wherein the feature reconstruction network includes three transposed convolutional layers, each of which uses a ReLU activation function and a residual connection; The fluid dynamic characteristic data matrix is ​​input into the recursive window Bayesian network, and the channel optimization objective function is obtained by calculating the heat transfer efficiency weight coefficient, the fluid resistance weight coefficient, and the heat transfer uniformity weight coefficient; specifically comprising: inputting the fluid dynamic characteristic data matrix into the Bayesian state estimator in the recursive window Bayesian network, constructing a state space equation including the fluid velocity field, the temperature field, and the pressure field, and performing Kalman filter recursive calculation in combination with the measurement equation to obtain a state probability distribution; performing multi-parameter weight calculation on the state probability distribution, constructing a heat transfer efficiency objective function, and using the maximum a posteriori probability MAP criterion for parameter estimation to obtain an initial heat transfer efficiency weight vector; the initial heat transfer efficiency The weight vector is input into the weight optimization network for optimization iteration to obtain the heat transfer efficiency weight coefficient; the fluid resistance analysis is performed on the state probability distribution, a Bayesian probability model including the Reynolds number and the friction coefficient is constructed, and the posterior probability density function is calculated to obtain the fluid resistance weight coefficient; based on the state probability distribution, the heat transfer uniformity analysis is performed, the conditional probability distribution of the temperature field variance and the skewness coefficient is calculated, and the heat transfer uniformity weight coefficient is obtained in combination with the heat flux density distribution function; based on the heat transfer efficiency weight coefficient, the fluid resistance weight coefficient and the heat transfer uniformity weight coefficient, a penalty function including the heat transfer efficiency term, the fluid resistance loss term and the heat transfer uniformity term is constructed to obtain the channel optimization objective function; According to the channel optimization objective function, a genetic algorithm is used to perform global iterative calculations on the channel main structure, and numerical simulation is used to perform local optimization on the flow-dividing and converging structure to obtain a three-dimensional channel structure optimization solution.

2. The three-dimensional channel structure design method of the microchannel heat exchange module according to claim 1, characterized in that: The heat conduction coefficient and convection heat transfer coefficient of the microchannel heat exchange module are input into the Rayleigh-Ritz coupling model for dynamic analysis to obtain channel structure parameter data, including: Perform characteristic decomposition operation on the microchannel heat exchange module to obtain the boundary data of the functional area including the inlet section, the heat exchange section, and the outlet section; Performing fluid phase characteristic analysis on the functional area boundary data to obtain a fluid physical property data set, and extracting fluid interface characteristics on the fluid physical property data set and microchannel diameter range parameters, and obtaining an interface heat transfer characteristic vector by calculating a convection heat transfer coefficient and a heat conduction coefficient; The interface heat transfer characteristic vector is input into the Rayleigh-Ritz coupling model for orthogonalization to obtain a coupled dynamic equation group, and a fluid characteristic analysis is performed based on the coupled dynamic equation group to obtain multi-dimensional fluid distribution data; Optimizing the cross-sectional area based on the multi-dimensional fluid distribution data to obtain a channel cross-sectional area parameter, and performing a depth optimization operation on the channel cross-sectional area parameter to obtain a channel depth value; The bending angle field equation and the diversion structure field equation are numerically solved based on the channel depth to obtain channel structure parameter data.

3. The three-dimensional channel structure design method of the microchannel heat exchange module according to claim 1, characterized in that: According to the channel optimization objective function, a genetic algorithm is used to perform global iterative calculation on the channel main structure, and numerical simulation is used to perform local optimization on the flow-dividing and converging structure to obtain a three-dimensional channel structure optimization scheme, including: A genetic optimization model of the channel main structure is established according to the channel optimization objective function, and an optimization parameter coding sequence is generated according to the value range of the channel cross-sectional area, the value range of the channel depth and the value range of the bending angle; Generate an initial population based on the optimization parameter coding sequence, construct a population fitness evaluation system based on the initial population, and perform a selection operation on the initial population according to the population fitness evaluation system to obtain a parent population; Performing a gene crossover operation on the parent population to obtain a child population, and interpreting the child population into channel structure parameters, numerically solving the fluid control equations by SIMPLE algorithm to obtain the flow field distribution characteristics of each individual; A local optimization model of the flow diversion and confluence structure is constructed based on the flow field distribution characteristics, and gradient descent iteration is performed through parameter constraints of the diversion ratio and the confluence angle to obtain the optimal parameters of the local structure; The optimal parameters of the local structure and the globally optimized individual parameters are optimized at multiple levels to obtain a three-dimensional channel structure optimization solution.

4. The three-dimensional channel structure design method of the microchannel heat exchange module according to claim 3, characterized in that: The gene crossover operation is performed on the parent population to obtain a child population, and the child population is interpreted as channel structure parameters, and the fluid control equations are numerically solved by the SIMPLE algorithm to obtain the flow field distribution characteristics of each individual, including: Selecting coding sites based on the parent population to obtain a gene crossover site matrix, and pairing parent individuals according to the gene crossover site matrix to obtain a gene segment group to be crossed; Performing a double-point crossover operation on the gene segment group to be crossed to obtain a recombined offspring gene sequence, and performing parameter decoding on the offspring gene sequence to obtain an actual parameter value of the microchannel structure; The actual parameter values ​​of the microchannel structure are input into the fluid control equations for iterative solution through SIMPLE algorithm to obtain numerical solutions of pressure field and velocity field; The heat transfer calculation is performed by the finite volume method and the central difference format to obtain the temperature field distribution results; Fluid characteristic analysis is performed on the numerical solutions of the pressure field and velocity field and the temperature field distribution results to obtain the flow field distribution characteristics of each individual.

5. The three-dimensional channel structure design method of the microchannel heat exchange module according to claim 4, characterized in that: The local optimization model of the flow diversion and confluence structure is constructed based on the flow field distribution characteristics, and gradient descent iteration is performed through parameter constraints of the diversion ratio and the confluence angle to obtain the optimal parameters of the local structure, including: Performing characteristic analysis on the flow field distribution characteristics, and obtaining a flow field characteristic function set by calculating a velocity distribution function, a pressure distribution function, and a temperature distribution function of the fluid in the channel; A local optimization objective function is constructed based on the flow field characteristic function set to obtain a local optimization model, and parameter constraints are set for the local optimization model to obtain an optimization variable space by limiting the value range of the diversion ratio R∈[0.4,0.6] and the confluence angle θ∈[30°,60°]; Input the optimization variable space into the gradient calculation module, obtain the parameter gradient vector by calculating the partial derivative of the objective function with respect to the optimization variable, and construct the parameter update equation according to the parameter gradient vector to obtain the iterative update rule; Performing convergence analysis on the iterative update rule, obtaining a convergence criterion by calculating the parameter changes of adjacent iteration steps, and comparing the convergence criterion with a preset threshold to obtain an optimal split ratio and an optimal confluence angle; The microchannel flow diversion and confluence structure is reconstructed based on the optimal diversion ratio and the optimal confluence angle, and the optimal parameters of the local structure are obtained by updating the distance between the diversion plates and the confluence angle.

6. A three-dimensional channel structure design device for a microchannel heat exchange module, characterized in that: For implementing the steps of the method according to any one of claims 1 to 5, the three-dimensional channel structure design device of the microchannel heat exchange module comprises: The analysis module is used to input the heat conduction coefficient and convection heat transfer coefficient of the microchannel heat exchange module into the Rayleigh-Ritz coupling model for dynamic analysis to obtain channel structure parameter data; The extraction module is used to collect the temperature field and pressure field of the channel structure parameter data, and extract the time series features through the time encoder of the dual space-time autoencoder, and extract the spatial features through the space encoder of the dual space-time autoencoder, and obtain the fluid dynamic feature data matrix after the time series features and spatial features are processed by the co-attention mechanism; specifically including: according to the channel structure parameter data, the sensor network composed of temperature sensors and pressure sensors performs multi-channel parallel sampling to obtain the original temperature field data and the original pressure field data; the original temperature field data and the original pressure field data are standardized to obtain a normalized time series data set; the normalized time series data set is input into the time encoder of the dual space-time autoencoder to extract the time series features to obtain the time feature map, the time encoder includes 3 one-dimensional convolution layers, each one-dimensional convolution layer uses the ReLU activation function; the time feature map is input into the space encoder of the dual space-time autoencoder to extract the spatial features to obtain the spatial feature map, the space encoder includes 4 two-dimensional convolution layers, each two-dimensional convolution layer uses the Batch Normalization and LeakyReLU activation functions; inputting the spatial feature map into the middle layer of the dual spatiotemporal autoencoder for nonlinear transformation to obtain a potential feature vector, wherein the middle layer includes two fully connected layers; performing multi-head attention calculation on the potential feature vector to obtain an attention weight matrix, and performing linear projection and concatenation operations on the attention weight matrix to obtain a fused feature tensor; inputting the fused feature tensor into a feature reconstruction network for upsampling and reconstruction to obtain a fluid dynamic feature data matrix, wherein the feature reconstruction network includes three transposed convolutional layers, each of which uses a ReLU activation function and a residual connection; The calculation module is used to input the fluid dynamic characteristic data matrix into the recursive window Bayesian network, and obtain the channel optimization objective function by calculating the heat exchange efficiency weight coefficient, the fluid resistance weight coefficient, and the heat transfer uniformity weight coefficient; specifically comprising: inputting the fluid dynamic characteristic data matrix into the Bayesian state estimator in the recursive window Bayesian network, constructing a state space equation including the fluid velocity field, the temperature field, and the pressure field, and performing Kalman filter recursive calculation in combination with the measurement equation to obtain the state probability distribution; performing multi-parameter weight calculation on the state probability distribution, constructing the heat exchange efficiency objective function and using the maximum a posteriori probability MAP criterion to perform parameter estimation to obtain the heat exchange efficiency initial weight vector; The initial efficiency weight vector is input into the weight optimization network for optimization iteration to obtain the heat exchange efficiency weight coefficient; the state probability distribution is subjected to fluid resistance analysis, a Bayesian probability model including the Reynolds number and the friction coefficient is constructed, and the posterior probability density function is calculated to obtain the fluid resistance weight coefficient; based on the state probability distribution, a heat transfer uniformity analysis is performed, the conditional probability distribution of the temperature field variance and the skewness coefficient is calculated, and the heat transfer uniformity weight coefficient is obtained in combination with the heat flux density distribution function; based on the heat exchange efficiency weight coefficient, the fluid resistance weight coefficient and the heat transfer uniformity weight coefficient, a penalty function including the heat exchange efficiency term, the fluid resistance loss term and the heat transfer uniformity term is constructed to obtain the channel optimization objective function; The optimization module is used to perform global iterative calculation on the main structure of the channel using a genetic algorithm according to the channel optimization objective function, and to perform local optimization on the flow-dividing and converging structure using numerical simulation to obtain a three-dimensional channel structure optimization solution.

7. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images