An optimization method for integrated micro-modules based on complex power distribution

Abstract

The application discloses an optimization method of integrated micro-modules based on complex power distribution, and relates to the technical field of integrated circuit thermal management, comprising the following steps: constructing a power distribution characteristic matrix by multi-physical field coupling analysis based on a structured data set, and generating micro-channel geometric parameters by using a gradient descent algorithm; constructing a structure thermal conduction efficiency mapping model, combining the micro-channel geometric parameters to obtain flow channel cross-section thermal resistance characteristic change data, and outputting a thermal conduction efficiency optimization matrix; extracting thermal resistance distribution characteristics in the thermal conduction efficiency optimization matrix, combining the flow channel cross-section thermal resistance characteristic change data to generate a collaborative optimization parameter instruction; obtaining thermal power distribution characteristics by executing the collaborative optimization parameter instruction, calculating an equivalent thermal resistance gradient, and obtaining a heat dissipation efficiency regulation and control instruction through dynamic mapping. By constructing the structure thermal conduction efficiency mapping model, real-time prediction of the micro-channel geometric parameters to thermal performance is realized, the efficiency is improved, and the global performance optimization is ensured.

Description

Technical Field

[0001] This invention relates to the field of integrated circuit thermal management technology, and in particular to an optimization method for integrated micromodules based on complex power distribution. Background Technology

[0002] With the rapid development of 3D integrated circuits and high-power-density electronic devices, thermal management has become a key technology restricting performance and reliability. Microfluidic cooling technology, due to its efficient heat dissipation capabilities, is widely used in integrated circuits, power modules, and advanced packaging. Traditional microfluidic design methods mainly rely on parametric scanning and empirical rules. In recent years, multiphysics coupling simulation technology has been introduced into the microfluidic design process, achieving coordinated analysis of temperature, flow, and stress fields by coupling physical fields such as fluid mechanics, heat conduction, and structural mechanics. In addition, intelligent optimization algorithms are also applied in microfluidic optimization to handle multi-objective, high-dimensional design space search problems.

[0003] However, most existing technologies are still limited to single-physics field or static power assumptions, failing to adequately address the dynamic and non-uniform complex power distribution scenarios in actual operation. Existing technologies suffer from two main bottlenecks. First, most optimization methods assume uniform power distribution or are based on idealized heat source models, failing to effectively couple real-time power data with geometric design. This leads to localized overheating in high-power gradient regions and wasted flow channel space in low-power regions. Second, existing optimization frameworks have limitations in the integration of multi-physics coupling and intelligent decision-making. Although multi-physics simulation tools can achieve electro-thermal-mechanical coupling analysis, the simulation process is usually separated from the optimization algorithm, requiring multiple iterations and incurring high computational costs. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an optimization method based on an integrated micro-module with complex power allocation to solve the problems of local overheating in high power gradient regions and low efficiency of intelligent optimization decision-making.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an optimization method for integrated micromodules based on complex power distribution, comprising:

[0008] The system collects and monitors operational data and preprocesses it to obtain a structured dataset. Based on the structured dataset, it constructs a power distribution feature matrix through multiphysics coupling analysis and generates microchannel geometric parameters using a gradient descent algorithm. It then constructs a structural heat conduction efficiency mapping model, combines the microchannel geometric parameters to obtain data on the change in thermal resistance characteristics of the channel cross-section, and outputs a heat conduction efficiency optimization matrix. It extracts the thermal resistance distribution features from the heat conduction efficiency optimization matrix and generates collaborative optimization parameter instructions based on the change in thermal resistance characteristics of the channel cross-section. Finally, it executes the collaborative optimization parameter instructions to obtain the thermal power distribution features and calculates the equivalent thermal resistance gradient, obtaining heat dissipation efficiency control instructions through dynamic mapping.

[0009] As a preferred embodiment of the optimization method for integrated micro-modules based on complex power distribution described in this invention, the monitoring operation data includes equipment performance indicators, environmental parameters, and equipment operating status.

[0010] The preprocessing includes data cleaning, normalization, and spatiotemporal coding.

[0011] As a preferred embodiment of the optimization method for integrated micro-modules based on complex power allocation described in this invention, the structured dataset includes ambient temperature values, power values, spatiotemporal feature tensors, and device operating parameters.

[0012] The power distribution feature matrix is ​​constructed based on the structured dataset through multiphysics coupling analysis. The specific steps are as follows:

[0013] The structured dataset is input into the multiphysics coupling equation for transient solution, and the global physical field distribution data is output.

[0014] Based on global physical field distribution data, the energy flux density vector is calculated, and the power distribution feature matrix is ​​constructed through spatial discretization.

[0015] As a preferred embodiment of the optimization method for integrated micro-modules based on complex power distribution described in this invention, the microchannel geometric parameters refer to the statistical characteristics extracted from the power distribution feature matrix and the power distribution uniformity calculated, which are generated through topology optimization.

[0016] As a preferred embodiment of the optimization method for integrated micro-modules based on complex power allocation described in this invention, the specific steps for constructing the structural heat conduction efficiency mapping model are as follows:

[0017] The geometric parameters of the microchannel are extracted using tensor decomposition to generate geometric feature tensors;

[0018] Based on geometric feature tensors, fluid temperature field and structural thermal stress field are generated through two-way thermal coupling analysis, and global energy flux density distribution field is generated through spatial gradient algorithm.

[0019] The thermal resistance characteristic distribution field is generated by calculating the grid equivalent thermal resistance based on the global energy flux density distribution field, and statistical feature vectors are generated by multi-order moment statistics.

[0020] Based on the thermal resistance characteristic distribution field and statistical eigenvectors, a thermal resistance geometric performance correlation network is constructed to obtain a structural thermal conduction performance mapping model.

[0021] As a preferred embodiment of the optimization method for integrated micro-modules based on complex power distribution described in this invention, the step of obtaining the thermal resistance characteristic change data of the flow channel cross-section by combining the micro-channel geometric parameters and outputting the thermal conduction efficiency optimization matrix refers to inputting the micro-channel geometric parameters into the structural thermal conduction efficiency mapping model to obtain the thermal resistance characteristic change data of the flow channel cross-section, and using a multi-objective particle swarm optimization algorithm to output the thermal conduction efficiency optimization matrix.

[0022] As a preferred embodiment of the optimization method for integrated micro-modules based on complex power distribution described in this invention, the extraction of thermal resistance distribution features in the thermal conduction efficiency optimization matrix refers to the use of an asymmetric feature extraction algorithm to extract the thermal resistance distribution features in the thermal conduction efficiency optimization matrix.

[0023] As a preferred embodiment of the optimization method for integrated micro-modules based on complex power distribution described in this invention, the specific steps for generating collaborative optimization parameter instructions by combining flow channel cross-sectional thermal resistance characteristic variation data are as follows:

[0024] By combining the thermal resistance distribution characteristics with the thermal resistance characteristic variation data of the flow channel cross section, a thermal resistance synergistic feature map is constructed.

[0025] Extract the spatiotemporal coupled thermal resistance distribution from the thermal resistance cooperative feature map, and generate cooperative optimization parameter instructions by locating the key heat conduction path.

[0026] As a preferred embodiment of the optimization method for integrated micro-modules based on complex power allocation described in this invention, the step of executing collaborative optimization parameter instructions to obtain thermal power distribution characteristics refers to executing collaborative optimization parameter instructions to generate temperature field distribution data, using recursive least squares method to calculate spatial temperature gradient, and obtaining thermal power distribution characteristics.

[0027] As a preferred embodiment of the optimization method for integrated micro-modules based on complex power allocation described in this invention, the specific steps for calculating the equivalent thermal resistance gradient and obtaining heat dissipation performance control commands through dynamic mapping are as follows:

[0028] Based on the characteristics of thermal power distribution, the equivalent thermal resistance gradient is calculated by combining temperature field distribution data.

[0029] The thermal power distribution characteristics and the equivalent thermal resistance gradient are fused into a joint thermal feature vector, and the heat dissipation performance control command is obtained through an optimization learning algorithm.

[0030] The beneficial effects of this invention are as follows: by constructing a structural heat conduction efficiency mapping model, real-time prediction and multi-objective collaborative optimization of microchannel geometric parameters to thermal performance are realized, improving efficiency and ensuring optimal global performance; by forming a closed-loop control through dynamic mapping and recursive least squares method, the overall thermal management scheme optimizes the flow channel volume ratio while reducing peak temperature, providing an efficient thermal management solution for high power density electronic devices. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of an optimization method for integrated micromodules based on complex power allocation.

[0033] Figure 2 A flowchart for generating microchannel geometric parameters.

[0034] Figure 3 A flowchart for generating the heat conduction efficiency optimization matrix.

[0035] Figure 4 A flowchart for generating instructions to regulate heat dissipation performance. Detailed Implementation

[0036] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0037] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0038] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0039] Reference Figures 1-4This is one embodiment of the present invention, which provides an optimization method for an integrated micromodule based on complex power allocation, comprising the following steps:

[0040] S1: Collect monitoring and operation data and obtain a structured dataset through preprocessing;

[0041] S1.1: Monitoring and operational data includes equipment performance indicators, environmental parameters, and equipment operating status;

[0042] It should be noted that device performance metrics are primarily obtained through monitoring agents deployed on the server or by actively querying the device's voltage and current via SNMP and JMX protocols. The core function of device performance metrics is to quantitatively assess the utilization of hardware resources and computing power, helping operations and maintenance personnel identify performance bottlenecks, predict capacity risks (such as insufficient disk space), and provide a basis for resource scheduling and elastic scaling decisions.

[0043] Environmental parameters (such as temperature, humidity, air pressure, and dust concentration) are collected by temperature sensors, vibration sensors, or environmental monitoring sensors deployed inside the equipment or in the equipment room. The main purpose of environmental parameters is to monitor the physical environment in which the equipment operates and to prevent equipment failure or performance degradation caused by environmental factors.

[0044] Equipment operating status data is obtained through applications and log files, network traffic analysis, and API calls. This data includes equipment uptime, baseline temperature, and baseline power. The core value of this data lies in directly reflecting the real-time health and availability of business applications and software. It is used to track user experience, locate the root cause of faults, discover security threats, and provide key evidence for ensuring service continuity.

[0045] S1.2: Preprocessing includes data cleaning, normalization, and spatiotemporal coding.

[0046] It should be noted that data cleaning mainly addresses outliers, missing values, and noise in the monitoring and operational data. This involves identifying and removing outlier data points that significantly deviate from the normal range; for missing values, statistical values ​​(mean, median) based on the monitoring and operational data can be used to fill in the gaps; and simultaneously, filtering algorithms (such as median filtering or wavelet denoising) are used to obtain cleaned environmental temperature values, power values, and equipment operating parameters to suppress high-frequency noise interference.

[0047] The core function of data cleaning is to improve the accuracy and consistency of monitoring and operational data, avoid distortion that could lead to false alarms or biased analysis conclusions, and provide reliable input for subsequent modeling.

[0048] Normalization aims to eliminate the influence of dimensional differences and numerical ranges among different characteristic parameters. It linearly transforms the data to the [0,1] interval. For example, after normalization, equipment temperature values ​​(unit: °C) and voltage values ​​(unit: V) can be directly compared and weighted for calculation.

[0049] Normalization enhances the comparability of temperature and voltage values, prevents certain features from dominating the model training process due to excessively large numerical ranges, and improves the algorithm's convergence efficiency and stability.

[0050] Spatiotemporal coding is used for monitoring and operational data that possesses both temporal and spatial attributes (such as sequences collected over time by multiple sensors at different locations). It uses timestamps for unified alignment (correcting timing discrepancies caused by transmission delays or different sampling frequencies), spatial location identification (assigning a unique location code to each sensor), and reconstructs the data into a spatiotemporal feature tensor.

[0051] Spatiotemporal coding explicitly characterizes the correlation and evolution of data in the spatiotemporal dimension, providing a structured data foundation for fault location (such as vibration signal propagation path analysis) and trend prediction (such as the spatiotemporal evolution of temperature field).

[0052] S2: Based on the structured dataset, a power distribution feature matrix is ​​constructed through multiphysics coupling analysis, and the gradient descent algorithm is used to generate the microchannel geometric parameters;

[0053] S2.1: The structured dataset includes ambient temperature values, power values, spatiotemporal feature tensors, and equipment operating parameters;

[0054] Specifically, the mean and standard deviation of environmental parameters are cleaned to remove outliers; and normalization is performed to convert the read millidegree Celsius values ​​to degrees Celsius, thus obtaining accurate temperature values ​​with clear physical meaning.

[0055] The power values ​​include active power and reactive power. The collected voltage and current are normalized and converted into voltage and current values ​​with actual physical units. Active power is obtained by calculating instantaneous power (the product of voltage and current); reactive power is obtained by extracting the fundamental component of instantaneous power using discrete Fourier transform.

[0056] This involves integrating and structuring the monitoring and operational data of sensors at different spatial locations that change over time. This typically includes aligning the timestamps of multiple sensor nodes to obtain the sensor's time-series data, thus eliminating time-series misalignments; and extracting temporal and frequency domain features from the sensor's time-series data, integrating these features, and generating a spatiotemporal feature tensor through spatiotemporal coding.

[0057] The system acquires equipment operating status that reflects the working state of the equipment, performs data cleaning to handle missing values ​​and obvious outliers, and eliminates dimensional differences between different parameters through data integration and normalization to form equipment operating parameters that can be used for analysis.

[0058] S2.2: Input the structured dataset into the multiphysics coupling equation for transient solution and output global physical field distribution data;

[0059] It should be noted that the computational domain of the multiphysics coupling equations is meshed to obtain spatial mesh nodes, and all generated spatial mesh nodes are mapped to a coordinate system to form a spatial mesh node coordinate set. The device operating time is divided into a series of time steps with fixed step sizes, each time step representing a specific moment.

[0060] Within each time step, the governing equations for each physical field (such as fluid Navier-Stokes equations, structural mechanics equations, heat transfer equations, etc.) are established and the space is discretized (using finite element method, finite volume method, or finite difference method) to transform the partial differential equations into a system of algebraic equations. Then, the time is discretized, and a time step is introduced to transform the transient problem into a series of steady-state problems over time steps. Within each time step, the velocity and pressure values ​​of all grid nodes are obtained by discretizing the multiphysics coupling equations and gradually advancing the time step.

[0061] Iterate through all grid nodes, integrate temperature, velocity, and pressure values ​​as corresponding physical quantity values ​​for each grid node, and output global physical field distribution data including spatial grid node coordinates, time step sequence, and corresponding physical quantity values.

[0062] A superior approach is to use mathematical language to precisely describe the interactions and transformations between fluid mass, momentum, and energy. The mass conservation equation ensures a balance of mass within the computational domain; the Navier-Stokes equations solve for the flow and pressure fields and reveal the relationship between velocity changes and forces; and the energy conservation equation predicts the temperature field distribution. Only by solving these equations simultaneously can accurate simulations of complex coupled phenomena such as convection-thermal-solid processes be achieved.

[0063] S2.3: Based on the global physical field distribution data, calculate the energy flux density vector and construct the power distribution feature matrix through spatial discretization.

[0064] It should be noted that the velocity values ​​of the spatial grid nodes are extracted from the global physical field distribution data as the spatial grid node velocity vectors, and the pressure values ​​of the spatial grid nodes are extracted as the pressure vectors.

[0065] The product of the velocity vector and the pressure vector is used as the stress tensor; the total number of all spatial grid nodes is counted, and the ratio of the stress tensor to the total number of all spatial grid nodes is used as the energy flux density vector.

[0066] The energy flux density vector is mapped to a regular grid according to spatial coordinates; the power flux density is obtained by binding the energy flux density vector in each regular grid with a timestamp; and the power flux density of all regular grids is integrated to construct a power distribution feature matrix.

[0067] S2.4: Extract statistical characteristics from the power distribution feature matrix, calculate the power distribution uniformity, and generate microchannel geometric parameters through topology optimization.

[0068] It should be noted that the power values ​​are extracted from the power distribution feature matrix, and the total number of grids and the total power value in the power distribution feature matrix are counted. The ratio of the total power value to the total number of grids is taken as the power average. The power distribution uniformity is calculated based on the power average of the power distribution feature matrix, and the expression is:

[0069] ;

[0070] in, For power distribution uniformity, This represents the total number of rows in the power distribution feature matrix. This represents the total number of columns in the power distribution characteristic matrix. The power distribution characteristic matrix is ​​the first... Line number Power values ​​at grid points. This represents the index value of the row in the power distribution feature matrix. This refers to the index value of a column in the power distribution characteristic matrix. This represents the average power.

[0071] Using power distribution uniformity as the optimization objective, the microchannel geometry parameters are directly adjusted based on the power values. For regions where the power value is higher than the average total power, the microchannel width or depth is increased accordingly; for regions where the power value is lower than the average total power, the microchannel width or depth is decreased accordingly. Simultaneously, the microchannel branching direction is adjusted according to the power distribution characteristic matrix to make the channels more densely cover the high-power regions. The microchannel geometry parameters (width, depth, and branching angle) are generated.

[0072] S3: Construct a structural heat conduction efficiency mapping model, combine microchannel geometric parameters to obtain data on the change of thermal resistance characteristics of the channel cross section, and output the heat conduction efficiency optimization matrix;

[0073] S3.1: The geometric parameters of the microchannel are extracted using tensor decomposition to generate geometric feature tensors;

[0074] It should be noted that the width, depth, and branch angle parameters of the microchannel are arranged in spatial coordinates to construct a three-dimensional array of microchannel geometric parameters; the three-dimensional array of microchannel geometric parameters is decomposed into the sum of multiple tensors, and the width factor matrix, depth factor matrix, and branch angle factor matrix are obtained after decomposition. Each factor matrix contains several columns of factor vectors; factor vectors that characterize the geometric features of the microchannel are extracted from the factor matrix, and the factor vectors are recombined into a geometric feature tensor.

[0075] S3.2: Based on the geometric feature tensor, the fluid temperature field and the structural thermal stress field are generated through two-way thermal coupling analysis, and the global energy flux density distribution field is generated through the spatial gradient algorithm.

[0076] It should be noted that the spatial distribution ranges of the fluid and solid regions are determined using geometric characteristic tensors. In the fluid region, the fluid undergoes energy changes due to flow and heat conduction; in the solid region, the solid undergoes mechanical responses due to thermal expansion and contraction, resulting in structural changes. The fluid temperature field affects the temperature of the solid region, and the deformation of the solid region caused by the structural thermal stress field affects the shape of the fluid region. The fluid temperature field is formed by separating the temperature values ​​of the spatial grid nodes in the fluid domain from the global physical field distribution data; the structural thermal stress field is formed by separating the temperature values ​​of the spatial grid nodes in the solid domain from the global physical field distribution data.

[0077] The fluid temperature field and the structural thermal stress field are mapped to the same coordinate system, gridded, and the grid spacing is obtained. The temperature difference between the adjacent left and right nodes of each spatial grid node is obtained, and the ratio of the temperature difference to the grid spacing is used as the temperature gradient. The global energy flux density distribution field is generated by combining the temperature gradient, the fluid temperature field, and the structural thermal stress field.

[0078] Superiorly, the spatial gradient algorithm accurately quantifies and analyzes the continuous variation patterns and directional characteristics of physical quantities in space, thus overcoming the limitations of traditional discrete or homogenized analysis methods. By capturing subtle spatial heterogeneity, the spatial gradient algorithm can also reveal hidden physical mechanisms and provide a reliable basis for multi-physics coupling by quantifying spatial dependencies.

[0079] S3.3: Calculate the equivalent thermal resistance of the grid based on the global energy flux density distribution field to generate the thermal resistance characteristic distribution field, and generate the statistical characteristic vector through multi-order moment statistics;

[0080] It should be noted that the heat flux density value of each grid is extracted based on the global energy flux density distribution field, and the average thermal resistance value is calculated. The average thermal resistance values ​​of all grids are then arranged according to spatial coordinates to generate a thermal resistance characteristic distribution field.

[0081] Multi-moment statistics are performed on all thermal resistance values ​​in the thermal resistance characteristic distribution field. The first moment (mean thermal resistance value) and the second moment (variance thermal resistance value) are calculated to generate a statistical feature vector describing the central tendency, dispersion, asymmetry, and sharpness of the distribution pattern of the thermal resistance. The expression for calculating the mean thermal resistance value is:

[0082] ;

[0083] in, This is the average thermal resistance value. This represents the total thermal resistance value. This is the index value of the total number of thermal resistance values. This is the thermal resistance value.

[0084] The expression for calculating the variance of thermal resistance is:

[0085] ;

[0086] in, The variance of thermal resistance. This represents the average thermal resistance value.

[0087] Superiorly, multi-moment statistics, by analyzing higher-order characteristics of data distribution, can more comprehensively characterize the shape, tail behavior, and extreme value features of data; thus improving the accurate perception and quantification of uncertainty and risk. It avoids model errors caused by traditional methods that rely on specific distribution assumptions (such as normal distribution) or fitted curves, enhancing the robustness and applicability of statistical inference.

[0088] S3.4: Based on the thermal resistance characteristic distribution field and statistical eigenvectors, a thermal resistance geometric performance correlation network is constructed to obtain a structural thermal conduction performance mapping model;

[0089] It should be noted that the thermal resistance value of each grid is extracted based on the thermal resistance characteristic distribution field, and the thermal resistance value of each grid is used as a network node; the spatial adjacency relationship between the thermal resistance characteristic distribution field and the statistical feature vector is constructed, where the first moment (mean) represents the central tendency of the distribution and the second moment (variance) represents the degree of dispersion; a thermal resistance geometric performance correlation network is constructed with thermal resistance value as node attribute and spatial adjacency relationship as edge;

[0090] By using a thermal resistance geometric efficiency correlation network, complex three-dimensional structures are discretized into thermal resistance node networks. Heat flow is input into the thermal resistance node networks to obtain the transmission efficiency and resistance characteristics in the path. The distribution of thermal resistance values ​​is correlated with the thermal conductivity to generate a structural thermal conductivity efficiency mapping model.

[0091] S3.5: Input the microchannel geometric parameters into the structural thermal conduction efficiency mapping model to obtain the thermal resistance characteristic change data of the channel cross section, and use the multi-objective particle swarm optimization algorithm to output the thermal conduction efficiency optimization matrix.

[0092] It should be noted that the microchannel geometric parameters are input into the structural thermal conduction efficiency mapping model. The structural thermal conduction efficiency mapping model generates thermal resistance values ​​and their distribution variations of the channel cross-section under different geometric configurations based on the heat flow path and spatial adjacency relationship in the thermal resistance geometric efficiency correlation network.

[0093] Based on thermal resistance variation data, an optimization problem is established with the objective functions of minimizing average thermal resistance and maximizing thermal uniformity (e.g., minimizing thermal resistance variance). A multi-objective particle swarm optimization algorithm is used to initialize the particle swarm, where each particle position represents a combination of microchannel geometric parameters. By updating the particle velocity and position, the microchannel geometric parameter combination with the smallest particle position offset and the largest particle velocity is integrated and output as a thermal conduction efficiency optimization matrix.

[0094] A superior approach is to use structural thermal conductivity mapping models to establish quantitative relationships between material microstructure, geometric parameters, and macroscopic thermal conductivity, thereby enabling accurate prediction and optimized design of the thermal behavior of complex structures. Structural thermal conductivity mapping models improve the prediction accuracy and design efficiency of thermal management, and reduce development costs by simulating temperature distribution and heat flow paths under different operating conditions in a virtual environment.

[0095] S4: Extract the thermal resistance distribution characteristics in the thermal conduction efficiency optimization matrix, and generate collaborative optimization parameter instructions by combining the thermal resistance characteristic change data of the flow channel section.

[0096] S4.1: An asymmetric feature extraction algorithm is used to extract the thermal resistance distribution features in the thermal conduction efficiency optimization matrix;

[0097] It should be noted that, to determine the spatial coordinates of the central grid, four positively adjacent grids and four diagonally adjacent grids are selected to form the asymmetric neighborhood of the central grid. The positively adjacent grids share a complete boundary with the central grid, while the diagonally adjacent grids are connected only by vertices.

[0098] Traverse each grid in the thermal conductivity optimization matrix, calculate the total thermal resistance of all grids in the asymmetric neighborhood and the total number of grids, and use the ratio of the total thermal resistance to the total number of grids as the neighborhood average. The neighborhood average represents the average thermal resistance level of the area surrounding the central grid.

[0099] The thermal resistance distribution characteristics can be extracted by comparing the thermal resistance value of the central grid with the average value of the neighborhood. If the central value is higher than the average value, it is marked as a high-resistance feature point; if the central value is lower than the average value, it is marked as a low-resistance feature point.

[0100] S4.2: Combine the thermal resistance distribution characteristics with the thermal resistance characteristic variation data of the flow channel cross section to construct a thermal resistance cooperative feature map;

[0101] It should be noted that the thermal resistance distribution feature map and the thermal resistance characteristic variation data of the flow channel cross-section are mapped one-to-one on the spatial grid. The thermal resistance distribution feature and the thermal resistance characteristic variation data of the flow channel cross-section are decomposed into a base layer and a detail layer, respectively; the base layer contains low-frequency global information of thermal resistance distribution, and the detail layer contains high-frequency local information of thermal resistance distribution and gradient variation information of the flow channel cross-section.

[0102] By linearly superimposing the base layer and the detail layer, a comprehensive feature value for each spatial location is generated. The comprehensive feature values ​​of all locations are arranged according to the spatial structure to form a thermal resistance synergistic feature map.

[0103] S4.3: Extract the spatiotemporal coupled thermal resistance distribution from the thermal resistance collaborative feature map, and generate collaborative optimization parameter instructions by locating the key heat conduction path.

[0104] It should be noted that wavelet transform is performed on the thermal resistance co-conductivity feature map to separate spatial and temporal features; the spatial features reflect the distribution differences of thermal resistance values ​​between grids, while the temporal features reflect the dynamic changes of thermal resistance values ​​as the heat conduction process progresses.

[0105] Based on spatial dimension features, the heat flow direction is continuously tracked, and the path with the highest heat flow density and the smallest change in thermal resistance is identified as the critical heat conduction path. The critical heat conduction path consists of a series of spatial coordinate points, each containing a thermal resistance value and a heat flow direction vector. The spatiotemporal coupled thermal resistance distribution is extracted point by point along the critical heat conduction path to obtain the average thermal resistance and the variance of thermal resistance fluctuation. Based on temporal dimension features, the thermal resistance value is bound to a timestamp of a unit time, and the difference in thermal resistance value between adjacent timestamps is obtained. The ratio of the thermal resistance difference to the unit time is used as the thermal resistance change rate, generating parameter adjustment instructions to reduce the mean thermal resistance and the variance of thermal resistance fluctuation along the path.

[0106] S5: Execute collaborative optimization parameter instructions to obtain thermal power distribution characteristics, calculate the equivalent thermal resistance gradient, and obtain heat dissipation performance control instructions through dynamic mapping.

[0107] S5.1: Execute the collaborative optimization parameter command to generate temperature field distribution data, use the recursive least squares method to calculate the spatial temperature gradient, and obtain the thermal power distribution characteristics;

[0108] It should be noted that the analytical co-optimization parameter instructions clearly define the adjustment requirements for the mean and fluctuation of path thermal resistance. Based on the co-optimization parameter instructions, the thermal management strategy is adjusted, and the relationship between temperature and time and space is described using a discretization method. The heat source term is set according to the real-time power load, and the boundary conditions are treated as adiabatic conditions (i.e., the boundary heat flux is zero), thus obtaining temperature field distribution data in the spatiotemporal dimensions.

[0109] After acquiring the temperature field distribution data, the ratio of the temperature difference to the grid spacing is used as the spatial temperature gradient. The lowest convergent state is marked when the change in the spatial temperature gradient reaches zero. The product of the lowest convergent state spatial temperature gradient and the material's thermal conductivity is used as the thermal power distribution feature to generate a thermal power distribution feature map.

[0110] It should be noted that the thermal conductivity (also known as thermal conductivity coefficient) of a material is mainly determined experimentally to quantify its thermal conductivity; for example, the thermal conductivity coefficient of a gas ranges from [value missing]. The range of values ​​for the thermal conductivity of liquids is: The range of values ​​for the thermal conductivity of metals is: .

[0111] S5.2: Calculate the equivalent thermal resistance gradient based on the characteristics of thermal power distribution and temperature field distribution data;

[0112] It should be noted that, based on the characteristics of heat power distribution, the average local thermal resistance value is extracted from the temperature values ​​in the temperature field distribution data as the local heat flux density; the ratio of the local heat flux density to the thermal conductivity of the material is used as the local temperature gradient; isotherms are identified along the heat flow direction in the temperature field distribution data, and the normal distance between adjacent isotherms is obtained. The quotient of the temperature difference and the heat flux density is used as the equivalent thermal resistance, and finally, the equivalent thermal resistance gradient is obtained.

[0113] S5.3: The thermal power distribution characteristics and the equivalent thermal resistance gradient are fused into a joint thermal feature vector, and the heat dissipation performance control instructions are obtained through an optimization learning algorithm.

[0114] It should be noted that the local heat flux density extracted from the thermal power distribution characteristics and the spatial thermal resistance change rate obtained from the equivalent thermal resistance gradient are spliced ​​together in a multi-dimensional feature to form a joint thermal feature vector.

[0115] Based on the joint thermal feature vector, a mapping relationship from features to the optimal control strategy is established, and control commands, such as adjusting fan speed, are attempted. The effect of this control command is monitored, such as whether the temperature is more stable and the energy consumption. Based on the effect, the system learns and updates itself. Control commands with good results are strengthened and remembered, while commands with poor results are weakened and avoided. Through a "trial-observation-learning" loop, an optimal heat dissipation control command is output.

[0116] In addition, optimization learning algorithms improve training efficiency by refining training strategies and parameter update mechanisms; they accelerate the convergence process and avoid getting trapped in local optima by adjusting the learning rate or introducing momentum mechanisms, thereby achieving higher prediction accuracy while reducing computation time and resource consumption; furthermore, optimization learning algorithms support complex scenarios and application requirements, providing key guarantees for practical deployment.

[0117] This embodiment also provides a computer device applicable to the optimization method of integrated micro-modules based on complex power allocation, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the optimization method of integrated micro-modules based on complex power allocation as proposed in the above embodiment.

[0118] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0119] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the optimization method for implementing integrated micromodules based on complex power allocation as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0120] In summary, this invention achieves real-time prediction and multi-objective collaborative optimization of microchannel geometric parameters to thermal performance by constructing a structural heat conduction efficiency mapping model, thereby improving design efficiency and ensuring optimal global performance. By forming a closed-loop control through dynamic mapping and recursive least squares method, the overall thermal management scheme optimizes the flow channel volume ratio while reducing peak temperature, providing an efficient thermal management solution for high power density electronic devices.

[0121] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An optimization method for integrated micromodules based on complex power distribution, characterized in that: include, Collect monitoring and operational data and obtain structured datasets through preprocessing; The monitoring and operation data includes equipment performance indicators, environmental parameters, and equipment operating status; The device performance indicators are mainly obtained by actively querying the device's voltage and current through a monitoring agent deployed on the server or through SNMP and JMX protocols. The environmental parameters are collected by temperature sensors, vibration sensors, or environmental monitoring sensors deployed inside the equipment or in the computer room. The device operating status is obtained by using applications and log files, network traffic analysis, and API calls to obtain device operating time, reference temperature value, and reference power. The structured dataset includes ambient temperature values, power values, spatiotemporal feature tensors, and equipment operating parameters; The ambient temperature value is obtained by cleaning the mean and standard deviation of the environmental parameters to remove outliers, and then scaling the data through normalization. The power values ​​include active power and reactive power. The collected voltage and current are normalized and converted into voltage and current values ​​with actual physical units. The spatiotemporal feature tensor refers to the integrated and structured representation of monitoring and operational data from sensors at different spatial locations that change over time. The equipment operating parameters are obtained by cleaning the equipment operating status data to remove missing values ​​and obvious anomalies, and by data integration and normalization. The power distribution feature matrix of the structured dataset is constructed through multiphysics coupling analysis, and the microchannel geometric parameters are generated using the gradient descent algorithm. A structural heat conduction efficiency mapping model is constructed. Microchannel geometric parameters are input to obtain data on the variation of thermal resistance characteristics of the channel cross-section, and a heat conduction efficiency optimization matrix is ​​output. The specific steps are as follows: The geometric parameters of the microchannel are extracted using tensor decomposition to generate geometric feature tensors; Based on geometric feature tensors, fluid temperature field and structural thermal stress field are generated through two-way thermal coupling analysis, and global energy flux density distribution field is generated through spatial gradient algorithm. The thermal resistance characteristic distribution field is generated by calculating the grid equivalent thermal resistance based on the global energy flux density distribution field, and statistical feature vectors are generated by multi-order moment statistics. Based on the thermal resistance characteristic distribution field and statistical feature vector, a thermal resistance geometric performance correlation network is constructed to obtain a structural thermal conduction performance mapping model; The microchannel geometric parameters are input into the structural thermal conduction efficiency mapping model to obtain the thermal resistance characteristic variation data of the channel cross section, and the multi-objective particle swarm optimization algorithm is used to perform non-dominated sorting and output the thermal conduction efficiency optimization matrix. Extract the thermal resistance distribution characteristics from the thermal conduction efficiency optimization matrix, and generate collaborative optimization parameter instructions by combining the thermal resistance characteristic change data of the flow channel section. The process involves executing collaborative optimization parameter commands to obtain thermal power distribution characteristics and calculating the equivalent thermal resistance gradient. Then, dynamic mapping is used to obtain heat dissipation performance control commands. The specific steps are as follows: The system executes collaborative optimization parameter instructions to generate temperature field distribution data, and uses the recursive least squares method to calculate the spatial temperature gradient and obtain the thermal power distribution characteristics. Based on the characteristics of thermal power distribution, the equivalent thermal resistance gradient is calculated by combining temperature field distribution data. The thermal power distribution characteristics and the equivalent thermal resistance gradient are fused into a joint thermal feature vector, and the heat dissipation performance control command is obtained through an optimization learning algorithm.

2. The optimization method for integrated micromodules based on complex power distribution as described in claim 1, characterized in that: The monitoring and operation data includes equipment performance indicators, environmental parameters, and equipment operating status; The preprocessing includes data cleaning, normalization, and spatiotemporal coding.

3. The optimization method for integrated micromodules based on complex power distribution as described in claim 2, characterized in that: The structured dataset includes ambient temperature values, power values, spatiotemporal feature tensors, and equipment operating parameters; The specific steps for constructing the power distribution feature matrix from the structured dataset through multiphysics coupling analysis are as follows: The structured dataset is input into the multiphysics coupling equation for transient solution, and the global physical field distribution data is output. Based on global physical field distribution data, the energy flux density vector is calculated, and the power distribution feature matrix is ​​constructed through spatial discretization.

4. The optimization method for integrated micromodules based on complex power distribution as described in claim 3, characterized in that: The microchannel geometric parameters are generated through topology optimization by extracting statistical characteristics from the power distribution feature matrix and calculating the power distribution uniformity.

5. The optimization method for integrated micromodules based on complex power distribution as described in claim 4, characterized in that: The extraction of thermal resistance distribution features from the thermal conduction efficiency optimization matrix refers to using an asymmetric feature extraction algorithm to analyze the gradient change law of thermal resistance distribution features and extract the thermal resistance distribution features from the thermal conduction efficiency optimization matrix.

6. The optimization method for integrated micromodules based on complex power distribution as described in claim 5, characterized in that: The process of combining the thermal resistance variation data of the flow channel cross-section with the thermal resistance distribution characteristics to generate collaborative optimization parameter instructions involves the following specific steps. By combining the thermal resistance distribution characteristics with the thermal resistance characteristic variation data of the flow channel cross section, a thermal resistance synergistic feature map is constructed. Extract the spatiotemporal coupled thermal resistance distribution from the thermal resistance cooperative feature map, and generate cooperative optimization parameter instructions by locating the key heat conduction path.

Images