Multi-objective-mean clustering fin optimization method and double-layer micro-channel radiator

Abstract

The invention belongs to the technical field of micro-channel heat dissipation, and discloses a multi-objective-mean clustering fin optimization method and a double-layer micro-channel radiator. Wherein the double-layer micro-channel radiator comprises a heat sink middle portion, the heat sink middle portion comprises fins located at the two ends and two fins located in the middle, a distance exists between the two fins, and the optimal value # imgabs0 # of the length of the two fins is obtained through the multi-objective-mean value clustering fin optimization method; according to the invention, the double-layer micro-channel radiator not only can reduce the voltage drop, but also can ensure a relatively low peak temperature value, and meanwhile, can also avoid too large voltage drop and too much energy dissipation.

Description

Technical Field

[0001] The present invention relates to the technical field of microchannel heat dissipation, and in particular to a multi-objective-mean clustering fin optimization method and a double-layer microchannel radiator. Background Art

[0002] A microchannel heat sink is a device that achieves efficient heat dissipation through tiny channels (usually with a channel size between 10-1000 μm). Due to its excellent heat dissipation performance, it is widely used in electronic devices (such as chips), new energy vehicles, aerospace and other fields.

[0003] In microchannel cooling system design, the fin parameter settings directly impact the balance between flow resistance and thermal performance. However, traditional methods rely on manual experience to set key parameters such as fin height, spacing, and thickness, which has significant optimization limitations. Manual trial-and-error methods make it difficult to find the optimal values ​​for the coordinated optimization of flow and thermal resistance, making it difficult to accurately capture the optimal balance between the two.

[0004] The above limitations directly lead to core problems: the system peak temperature cannot be reduced to the theoretical minimum value, especially in high power density scenarios, where local overheating significantly affects the stability of equipment operation; in pursuit of heat dissipation effects, the system is often forced to accept excessive pressure drops, resulting in a surge in pump power loss; in addition, the manual optimization process requires repeated experimental verification, which not only prolongs the R&D cycle, but also makes it difficult to adapt to the development trend of miniaturization and integration of microchannel structures, restricting the application and expansion of heat dissipation systems in high-end fields such as new energy vehicles and aerospace. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-objective-mean clustering fin optimization method and a double-layer microchannel radiator to overcome the problems existing in the prior art. The double-layer microchannel radiator of the present invention can not only reduce the pressure drop but also ensure a relatively low peak temperature, while also avoiding excessive pressure drop and excessive energy dissipation.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a multi-objective-mean clustering fin optimization method, comprising the following steps: Step 1: Acquire experimental data, normalize the experimental data, and obtain a thermal performance model of the maximum temperature with respect to the fin length and a thermal performance model of the maximum pressure drop with respect to the fin length by fitting the normalized experimental data; Step 2, initializing the multi-objective optimization algorithm, generating an initialization population through the initialized multi-objective optimization algorithm, the initialization population including several individuals of fin length values; Step 3, iteratively optimize the initialized population through the initialized multi-objective optimization algorithm, and retain several generations of Pareto frontiers formed during the iterative process; Step 4, calculating the maximum temperature and maximum pressure drop of several generations of Pareto fronts respectively through the thermal performance model of maximum temperature with respect to fin length and the thermal performance model of maximum pressure drop with respect to fin length, and constructing a decision matrix through the several generations of Pareto fronts and the maximum temperature and maximum pressure drop of several generations of Pareto fronts; Step 5: normalize the decision matrix, define the positive ideal solution and the negative ideal solution according to the normalized decision matrix, calculate the Euclidean distance between the positive ideal solution and the negative ideal solution, calculate the relative proximity according to the Euclidean distance between the positive ideal solution and the negative ideal solution, and obtain the optimal value of the fin length according to the relative proximity; Furthermore, the formula of the thermal performance model of the maximum temperature with respect to the fin length specifically includes:

[0007] Where, Indicates the fin length; Indicates the maximum temperature along the fin length; The formula of the thermal performance model of the maximum pressure drop with respect to the fin length specifically includes:

[0008] Where, Indicates the fin length; Indicates the maximum pressure drop across the fin length; Furthermore, the step 2 specifically includes: Initialization parameters of the multi-objective optimization algorithm are set, where the initialization parameters include population size, maximum number of iterations, crossover probability, and mutation probability. Then, the fin length range and step size of the multi-objective optimization algorithm are defined. According to the initialization parameters, fin length range, and step size, several fin length value individuals are generated within the fin length range. The number of fin length value individuals is equal to the population size, and the several fin length value individuals constitute the initialization population. Furthermore, the step 3 specifically includes: Step 31: Perform a first non-dominated sort on the initialized population, then perform a first crowding distance calculation on the initialized population after the first non-dominated sort. Select a number of parent generations based on the results of the first non-dominated sort and the first crowding distance calculation, and generate children from the parent generations until the number of parents and children reaches the population size specified in the initialization parameter. Then, merge the parents and children to obtain a merged population. Step 32: Perform a second non-dominated sort on the merged population, where the highest non-dominated front is the Pareto front. Perform a second crowding distance calculation on the merged population after the second non-dominated sort. Select the next generation population based on the results of the second non-dominated sort and the second crowding distance calculation, completing an iterative optimization. The number of the next generation population is the population size in the initialization parameter. Step 33, executing step 31 again on the next generation population, repeating the iteration until the maximum number of iterations in the initialization parameter is reached, completing the iterative optimization, wherein the Pareto front of each generation is retained, and K-means clustering is performed on the Pareto front retained every five generations; Furthermore, the step 31 specifically includes: Assign several fin length value individuals of the initialized population to non-dominated frontiers of different levels, calculate the crowding distance of fin length value individuals of the same level non-dominated frontier, randomly select two fin length value individuals each time, and select the one with a higher non-dominated frontier level as the parent among the two selected fin length value individuals. If the non-dominated frontier levels of the two selected fin length value individuals are the same, select the one with a higher crowding distance as the parent, and then use each two parents as parents, and pass the parents and random coefficients to obtain the parent. Linear combination crossover is used to generate two offspring, and Gaussian perturbation mutation is performed on each of the two offspring to obtain Gaussian perturbation mutated offspring until the number of parents and offspring reaches the population size in the initialization parameter. The parents and offspring are then merged to obtain a merged population. The step 32 specifically includes: The fin length value individuals in the merged population are redistributed to non-dominated fronts of different levels, the crowding distance of the fin length value individuals on the non-dominated front of the same level is calculated, and the fin length value individuals are selected as the next generation population according to the non-dominated front level from high to low and the crowding distance from large to small. The number of the next generation population is the population size in the initialization parameter. Furthermore, the step 4 specifically includes: The maximum temperature and maximum pressure drop of several generations of Pareto fronts are calculated using the thermal performance model of maximum temperature with respect to fin length and the thermal performance model of maximum pressure drop with respect to fin length, and the decision matrix is ​​constructed by taking the several generations of Pareto fronts as rows and the several generations of Pareto fronts as columns. Furthermore, in step 5, the decision matrix is ​​standardized, and the formula specifically includes:

[0009] Where, Indicates the The Pareto frontier is Standardized value under each thermal performance index; Indicates the The Pareto frontier is The original value of each thermal performance index; Indicates the The first of all Pareto fronts of thermal performance indicators The maximum value of a thermal performance index; Indicates the maximum value; Indicates the Thermal performance indicators; Represents the row of the decision matrix, representing the a Pareto frontier; Represents the columns of the decision matrix, representing the Thermal performance indicators; The Euclidean distances of the positive ideal solution and the negative ideal solution are calculated in step 5, and the formulas specifically include:

[0010]

[0011] Where, represents the Euclidean distance of the positive ideal solution; represents the Euclidean distance to the negative ideal solution; Indicates summation; Indicates the The weight of each thermal performance indicator; Indicates the The Pareto frontier is Standardized value under each thermal performance index; Indicates the Positive ideal solution of a thermal performance index; Indicates the Negative ideal solution of a thermal performance index; In step 5, the relative proximity is calculated based on the Euclidean distance between the positive ideal solution and the negative ideal solution, and the formula specifically includes:

[0012] Where, Indicates the the relative proximity of the two Pareto frontiers; represents the Euclidean distance to the negative ideal solution; represents the Euclidean distance of the positive ideal solution; In step 5, the optimal value of the fin length is obtained according to the relative proximity, and the formula specifically includes:

[0013] Where, Indicates the optimal value of fin length Indicates the The maximum relative proximity of the Pareto frontier; Indicates the maximum value; Indicates the The relative proximity of two Pareto fronts.

[0014] In a second aspect, the present invention provides a double-layer microchannel heat sink, based on the above-mentioned multi-objective-mean clustering fin optimization method, comprising a heat sink base, a heat sink middle portion and a heat sink top portion, wherein the heat sink base and the heat sink middle portion are connected by a plurality of first baffles, and a plurality of lower microchannels are formed between two adjacent first baffles, and the heat sink middle portion and the heat sink top portion are connected by a plurality of second baffles, and a plurality of upper microchannels are formed between two adjacent second baffles, and each upper microchannel and lower microchannel is provided with an inlet and outlet unit; The middle part of the heat sink includes fins at both ends and two fins in the middle. There is a distance between the two fins. The optimal value of the length of the two fins is It is obtained by using a multi-objective-mean clustering fin optimization method mentioned above; Furthermore, the inlet and outlet unit includes a top inlet, an upper layer inlet, a lower layer inlet, an upper layer outlet and a lower layer outlet; The top inlet is located in the middle of the top of the heat sink, the upper inlet and the upper outlet are respectively arranged at the two ends of the upper microchannel, the lower inlet and the lower outlet are respectively arranged at the two ends of the lower microchannel, the upper inlet and the lower inlet are located on the same side of the upper microchannel and the lower microchannel, and the upper outlet and the lower outlet are located on the same side of the upper microchannel and the lower microchannel; The two fins are arranged in parallel, and the distance between the two fins is equal to the length of the top inlet; Furthermore, the angle between the fins is β 0° ~ 20°.

[0015] The above technical solution has the following advantages or beneficial effects: In the first aspect, the present invention provides a multi-objective-mean clustering fin optimization method, which effectively balances the contradiction between heat dissipation performance and fluid resistance by establishing a dual-objective thermal performance model of maximum temperature and maximum pressure drop, and overcomes the limitation of traditional single-objective optimization that it is difficult to take into account multiple performance indicators; based on the normalization processing and model fitting of experimental data, the accuracy of the thermal performance model is significantly improved, and the uncertainty of pure numerical simulation is avoided; a non-dominated sorting algorithm is used to generate the Pareto front to ensure diversity and convergence, and provide a rich set of non-inferior solutions for decision-making; by constructing a decision matrix and combining a multi-criteria decision-making method, the Euclidean distance and relative proximity of the positive and negative ideal solutions are used to quantify the pros and cons, and finally determine the optimal fin length, so that the double-layer microchannel radiator can not only reduce the pressure drop but also ensure a relatively low peak temperature, while also avoiding excessive pressure drop and excessive energy dissipation.

[0016] Furthermore, by constructing a thermal performance model of maximum temperature and maximum pressure drop with respect to fin length, it is possible not only to accurately quantify the impact of changes in fin length on these two key thermal parameters, but also to clearly grasp the changing trends of maximum temperature and maximum pressure drop under different fin lengths during the optimization process, thereby improving the heat exchange efficiency of the fins.

[0017] Furthermore, by reasonably setting the initialization parameters of the multi-objective optimization algorithm, such as population size and maximum number of iterations, a solid framework is built for the optimization process to ensure that the algorithm can run and converge effectively.

[0018] Furthermore, the parent generation is screened through the first non-dominated sort and crowding distance calculation to generate offspring, and the offspring are merged to form a merged population, which not only ensures the diversity of the population, but also allows the excellent individuals to be retained and reproduced; by performing secondary processing on the merged population, the Pareto front is determined and the next generation population is selected. By repeating the iteration to the maximum number of iterations, the search space can be fully explored to avoid premature convergence. The Pareto front of each generation is retained and clustered every five generations. The optimization process and result distribution can be clearly presented, which helps to analyze the changing trend of the optimal solution, provide a comprehensive, accurate and diversified optimal solution for fin design, and significantly improve the optimization effect.

[0019] Furthermore, selecting parents according to the non-dominated frontier level and crowding distance can ensure that excellent and diverse individuals participate in reproduction. The linear combination crossover and Gaussian perturbation mutation of parents not only inherits excellent characteristics but also introduces new changes, enriches the population genes, and avoids premature convergence. Through redistribution and calculation, the next generation of population is selected according to level and distance, and the optimization is gradually guided. The above-mentioned fine operations can comprehensively search the solution space, balance multiple optimization objectives, and find individuals with better wing length values.

[0020] Furthermore, by utilizing the established thermal performance model of maximum temperature and maximum pressure drop with respect to fin length, the values ​​of these two key indicators corresponding to several generations of Pareto frontiers are accurately calculated, providing a solid data foundation for subsequent analysis; by constructing a decision matrix, the different generations of Pareto frontiers and the corresponding maximum temperature and maximum pressure drop are orderly integrated, so that the complex multi-objective optimization results are presented in a clear and intuitive matrix form, which helps decision makers to comprehensively and systematically compare the performance of different schemes on multiple objectives, quickly identify the optimal fin design under different demand preferences, improve decision-making efficiency and accuracy, and effectively promote fin optimization.

[0021] Furthermore, by precisely applying the established thermal performance model to calculate the key indicator values ​​of several generations of Pareto frontiers, the accuracy and reliability of the data are guaranteed; by constructing a decision matrix, the complex multi-objective optimization results of multiple generations can be orderly integrated, simplifying the complex and clearly showing the performance differences of different schemes in multiple objectives such as maximum temperature and maximum pressure drop in an intuitive matrix form.

[0022] Secondly, the present invention provides a double-layer microchannel radiator. By adopting the above-mentioned specific fin optimization method, the complex relationship between multiple performance indicators is comprehensively considered through multi-objective non-dominated sorting, and the optimal fin length value is accurately found. At this length, the double-layer microchannel radiator of the present invention achieves an optimized balance of performance, which can not only effectively reduce pressure drop, reduce fluid flow resistance, reduce energy consumption, and extend the service life of the equipment, but also ensure that the temperature is at a relatively low level, improve heat dissipation efficiency, ensure stable operation of the equipment, and avoid performance degradation or damage caused by excessive temperature; the upper and lower microchannels are formed by connecting the base, middle and top of the heat sink through baffles, and an inlet and outlet unit is set, which increases the heat dissipation area and enhances the heat dissipation effect; the reasonable layout of the fins at both ends and the middle fins further optimizes the heat transfer path; the double-layer microchannel radiator of the present invention can reduce energy consumption while ensuring efficient heat dissipation, provide a reliable heat dissipation solution for electronic equipment, etc., and improve the overall performance and reliability of the equipment.

[0023] Furthermore, the inlet and outlet units of the present invention are reasonably arranged, with the top inlet in the center, and the inlets and outlets of the upper and lower microchannels located at both ends and corresponding to the same side, so that the cooling fluid can flow into and out of each microchannel in an orderly manner, forming an efficient and stable fluid circulation path, enhancing the uniformity and smoothness of heat dissipation, and effectively improving the heat dissipation efficiency; the two fins are parallel and the spacing is equal to the length of the top inlet, which not only ensures that there is enough space between the fins for the fluid to fully flow and heat exchange, but also reasonably utilizes the internal space of the radiator and optimizes the heat dissipation structure.

[0024] Furthermore, a smaller angle can make the fluid flow smoother, reduce local eddy currents and energy loss, and lower pressure drop. Properly increasing the angle can increase the contact area and impact force between the fluid and the heat sink, enhance the heat exchange efficiency, and better ensure that the temperature is at a relatively low value. Through the adjustable angle design, it can be optimized and adjusted according to different heat dissipation requirements and operating conditions, so that the radiator can perform excellent heat dissipation performance in a variety of scenarios, thereby improving the adaptability and versatility of the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Schematic diagram of a multi-objective-mean clustering fin optimization method of the present invention; Figure 2 This is a schematic structural diagram of a double-layer microchannel radiator with multi-objective-mean clustering fin optimization according to the present invention; Figure 3 Schematic diagram of the flow path of the double-layer microchannel heat sink of Example 2 of the present invention; Figure 4 Schematic diagram of the flow path of the double-layer microchannel heat sink of Example 3 of the present invention; Figure 5 Schematic diagram of the flow path of the double-layer microchannel heat sink of Example 4 of the present invention; Figure 6 Schematic diagram of the flow path of the double-layer microchannel heat sink of Example 5 of the present invention; Figure 7 The difference between the maximum pressure drop and the minimum pressure drop, the Reynolds number and the angle β 3D schematic diagram of Figure 8 2. It is a two-dimensional schematic diagram of the contact thermal resistance and the Reynolds number of the present invention; Figure 9 is a two-dimensional schematic diagram of the maximum temperature and Reynolds number of the present invention; Figure 10 It is a two-dimensional schematic diagram of the comprehensive heat transfer coefficient and Reynolds number of the present invention; Figure 11 A schematic diagram of the curve drawn by the fitting equation of the present invention and the temperature value points of the experimental data; Figure 12 The residual graph of the curve drawn by the fitting equation of the present invention and the temperature value points of the experimental data; Figure 13 Schematic diagram of the curve drawn by the fitting equation of the present invention and the pressure drop value points of experimental data; Figure 14 The residual graph of the curve drawn by the fitting equation of the present invention and the pressure drop value points of the experimental data; Figure 15 Schematic diagram of the Pareto solution and random solution of the present invention; Figure 16Schematic diagram of generation distance comparison between the original algorithm and the improved algorithm of the present invention; Figure 17 Schematic diagram of the hypervolume comparison between the original algorithm and the improved algorithm of the present invention; Figure 18 Schematic diagram for comparing the spacing metrics of the original algorithm and the improved algorithm of the present invention; Figure 19 Schematic diagram showing the comparison of the Pareto front sizes between the original algorithm and the improved algorithm of the present invention; Figure 20 Schematic diagram showing the comparison of the running time of the original algorithm and the improved algorithm of the present invention; In the figure, 1, heat sink base; 2, fin; 3, heat sink top; 4, top inlet; 5, fin; 11, first baffle; 12, second baffle; 41, upper layer inlet; 42, lower layer inlet; 43, upper layer outlet; 44, lower layer outlet. DETAILED DESCRIPTION

[0026] The present invention will be further described in detail below with reference to specific embodiments, which are intended to explain the present invention rather than to limit it. In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention. It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] Example 1: See also Figure 1 The present invention provides a multi-objective-mean clustering fin optimization method, comprising the following steps: Step 1: Get the experimental data, normalize the experimental data, and get the highest temperature by fitting the normalized experimental data. Thermal performance model and maximum pressure drop with respect to fin length Thermal performance model with respect to fin length; Specifically, the normalized experimental data were input into the curve fitter of MATLAB (Matrix Laboratory) to obtain the maximum temperature. The equations for the thermodynamic performance model with respect to fin length include:

[0028] Where, Indicates the fin length; Indicates the maximum temperature along the fin length; Maximum pressure drop The equations for the thermodynamic performance model with respect to fin length include:

[0029] Where, Indicates the fin length; Indicates the maximum pressure drop across the fin length; Specifically, the experimental data and the normalized experimental data are shown in the following table: Table 1 Experimental data and normalized experimental data

[0030] In the table, T represents the maximum temperature after normalization, P represents the maximum pressure drop after normalization; Specifically, the experimental steps include: assembling the double-layer microchannel radiator and the water tank, reinforcing the assembly parts and the inlet and outlet units of the double-layer microchannel radiator with glass glue or resin glue to prevent water leakage; fixing the thin film resistor on the heat source surface of the double-layer microchannel radiator with thermal conductive silicone, and spraying three layers of anti-rust paint to prevent the thin film resistor from dissipating heat; connecting the experimental equipment in sequence according to the circuit diagram, and checking whether the pipe joints are leaking; turning on the constant temperature water bath, adding an appropriate amount of water, setting its temperature slightly higher than 20°C, and waiting for heating to the set temperature; turning on the peristaltic pump water tank to fill the entire experimental device with water, and checking whether the experimental device is leaking. If it is leaking, use glass fiber to reinforce the waterproofing; turning on the DC power supply, heating the double-layer microchannel radiator to simulate the heat source, and checking the temperature of the double-layer microchannel radiator itself. Check whether the pressure rises and wait for the temperature of the double-layer microchannel radiator to stabilize; set the peristaltic pump flow mode to continuous operation mode, then select the system flow, turn on the peristaltic pump, and the coolant quickly enters the entire channel; after the coolant flows through the entire double-layer microchannel radiator, observe the pressure, temperature and solid wall temperature of the double-layer microchannel radiator inlet and outlet units on the intelligent pressure measuring instrument and temperature measuring instrument. When the values ​​remain basically unchanged (that is, the upper and lower fluctuations do not exceed 5%), record the experimental data; according to the flow range, set part of the flow to the flow of the peristaltic pump, and record it after the data stabilizes; wait for all experimental data to be recorded, turn off the DC power supply, and stop heating; set the rotation mode of the peristaltic pump to reverse, and discharge all the coolant in the double-layer microchannel radiator.

[0031] See also Figures 11 to 14 The prediction curve drawn according to the fitting equation is highly consistent with the experimental data points. The vast majority of data points are closely consistent with the prediction curve or are distributed near the prediction curve, which clearly and strongly proves the excellent fitting performance of the model. The residuals of all data points are less than 0.15. In addition, the maximum temperature The coefficient of determination is 0.8641, indicating that the model can explain The fitting quality is good for 86.41% of the changes in the data; the sum of squared errors (SSE) is 0.2289 and the root mean square error (RMSE) is 0.1235, which are both relatively low and accurately quantify the prediction error; similarly, The coefficient of determination is 0.9643, indicating that the model can explain The fitting quality is excellent for 96.43% of the changes in the data, close to perfect fitting; the sum of squared errors (SSE) is 0.05397 and the root mean square error (RMSE) is 0.05998, both at extremely low levels, reflecting the high accuracy of the model in predicting pressure drop.

[0032] Step 2, initializing the multi-objective optimization algorithm, generating an initialization population through the initialized multi-objective optimization algorithm, the initialization population including several individuals of fin length values; Specifically, the initialization parameters of the multi-objective optimization algorithm are set, wherein the initialization parameters include population size, maximum number of iterations, crossover probability and mutation probability, and then the fin length range and step size of the multi-objective optimization algorithm are defined. According to the initialization parameters, fin length range and step size, several fin length value individuals are generated within the fin length range. The number of fin length value individuals is equal to the value of the population size. Several fin length value individuals constitute the initialization population, which are respectively selected through the maximum temperature Thermal performance model and maximum pressure drop with respect to fin length Thermodynamic performance model of fin length is used to calculate the maximum temperature of several fin length values ​​in the initial population. and maximum pressure drop ; Preferably, the multi-objective optimization algorithm may be NSGA-II (Non-dominated Sorting Genetic Algorithm II), which is suitable for solving engineering design problems with two or more objectives; Preferably, the population size is preferably set to 300, the maximum number of iterations is preferably set to 50, the crossover probability is preferably set to 0.8, the mutation probability is preferably set to 0.05, the fin length range is preferably set to 3.55~7.55 mm, the step size is preferably set to 0.001, and the number of fin length value individuals is 300.

[0033] Step 3, iteratively optimize the initialized population through the initialized multi-objective optimization algorithm, and retain several generations of Pareto frontiers formed during the iterative process; Step 31: Perform a first non-dominated sort on the initialized population, then perform a first crowding distance calculation on the initialized population after the first non-dominated sort. Select a number of parent generations based on the results of the first non-dominated sort and the first crowding distance calculation, and generate children from the parent generations until the number of parents and children reaches the population size specified in the initialization parameter. Then, merge the parents and children to obtain a merged population. Specifically, several fin length value individuals of the initialized population are assigned to non-dominated frontiers of different levels, and the crowding distance of the fin length value individuals of the same level non-dominated frontier is calculated. Two fin length value individuals are randomly selected each time, and the one with the higher non-dominated frontier level is selected as the parent among the two selected fin length value individuals. If the non-dominated frontier level of the two selected fin length value individuals is the same, the one with the higher crowding distance is selected as the parent, and then every two parents are taken as parents, and the parents are combined with the random coefficient to obtain the parent. Linear combination crossover is used to generate two offspring, and Gaussian perturbation mutation is performed on each of the two offspring to obtain Gaussian perturbation mutated offspring until the number of parents and offspring reaches the population size in the initialization parameter. The parents and offspring are then merged to obtain a merged population. Preferably, the crowding distance of individual fin length values ​​of the same level of non-dominated front is calculated, and the formula includes:

[0034] Where, Indicates the The fin length value of the individual in the The crowding distance on the thermal performance model of the maximum temperature on the Pareto front and the thermal performance model of the maximum pressure drop on the fin length; Indicates the sorted The fin length value of the individual in the The values ​​of the thermal performance model of the maximum temperature of the Pareto front and the thermal performance model of the maximum pressure drop with respect to the fin length; Indicates the position of the individual fin length values; Indicates the position of the Pareto front; Represents the number of fin length value individuals in the non-dominated front; the larger the crowding distance, the sparser the distribution of solutions around the fin length value individual, the better the diversity, and the greater the possibility of being retained; Indicates the sorted The fin length value of the individual in the non-dominated front is The values ​​of the thermal performance model of the maximum temperature of the Pareto front and the thermal performance model of the maximum pressure drop with respect to the fin length; Preferably, by parents and random coefficients Linear combination crossover, the formula includes:

[0035]

[0036] Where, Represents one of the offspring; Indicates another offspring; Indicates the random coefficient (0~1 random number); Represents one of the parents; Indicates another parent;

[0037] Where, Indicates the dynamic adjustment of crossover probability; Indicates the current iteration number; Indicates the maximum number of iterations, the maximum number of iterations is 50; represents population diversity, calculated as the standard deviation of the Euclidean distance between individuals of fin length values; at first, Smaller, (High crossover rate, strengthening global search, expanding search space), then, Iterate to , (crossover rate is reduced, which is conducive to local fine search), when When >0.6, the second item increases dynamically , to prevent the population from converging prematurely and maintain the ability to explore the solution set;

[0038]

[0039] Where, Indicates the dynamic adjustment of mutation probability; represents the maximum population diversity; when (threshold), the probability of mutation will increase, thereby increasing the disturbance of genotype and getting rid of the local optimal state; if When (threshold), a low baseline variation rate (0.05) is maintained to ensure the stability of local convergence in the later stage; Preferably, Gaussian perturbation mutation is performed on the two offspring respectively to obtain the offspring after Gaussian perturbation mutation, and the formula includes:

[0040]

[0041]

[0042] Where, represents one of the offspring after Gaussian perturbation mutation; represents another offspring after Gaussian perturbation mutation; represents the random disturbance amount; Indicates the maximum value; Indicates the minimum value; Represents a random number generation function, which is used to generate random numbers that conform to the standard normal distribution (mean 0, standard deviation 1); Preferably, the number of parents is 300, the number of offspring is 300, and the number of the combined population is 600; Step 32: Perform a second non-dominated sort on the merged population, where the highest non-dominated front is the Pareto front. Perform a second crowding distance calculation on the merged population after the second non-dominated sort. Select the next generation population based on the results of the second non-dominated sort and the second crowding distance calculation, completing an iterative optimization. The number of the next generation population is the population size in the initialization parameter. Specifically, several individuals with the same fin length value in the merged population are redistributed to non-dominated fronts of different levels, and the crowding distances of the individuals with the same level of non-dominated fronts are calculated. Several individuals with the same fin length value are selected as the next generation population according to the non-dominated front level from high to low and the crowding distance from large to small, which can ensure the continued retention of the optimal solution of the Pareto front and the diversity of the population. The number of the next generation population is the population size in the initialization parameter. Preferably, the number of next generation population is 300; Step 33: execute step 31 again on the next generation population, and repeat the iteration to the maximum number of iterations in the initialization parameter to complete the iterative optimization, wherein the Pareto front of each generation is retained. To ensure a reasonable distance metric, the Pareto front of each generation is first normalized (classified into the range of 0 to 1), and K-means clustering is performed on the Pareto front retained every five generations to obtain several clusters. The crowding degree of each cluster is calculated, and the average crowding degree and crowding standard deviation of each cluster are calculated based on the crowding degree of each cluster. A dynamic threshold is set based on the average crowding degree and the crowding standard deviation, and the Pareto front of each cluster whose crowding degree is greater than or equal to the dynamic threshold is used as the sparse representative solution of each cluster. Preferably, the number of K-means clusters K=3 is set (which can be adjusted according to the scale of the problem), and the cluster centers are initialized. For example, 3 fin length value individuals are randomly selected, and the fin length value individuals are assigned to the cluster center of the nearest cluster according to the minimum Euclidean distance. The center point of each cluster is updated until the distribution within the cluster no longer changes, indicating that the clustering has converged; then the crowding degree of several fin length value individuals in each cluster is calculated, that is, within the same cluster, they are re-sorted according to the target value and recalculated using the standard NSGA-II crowding degree formula. ; Preferably, a dynamic threshold is set according to the average congestion level and the standard deviation of the congestion level, and the formula includes:

[0043] Where, represents the dynamic threshold; represents the average degree of crowding; represents the standard deviation of crowding; See also Figure 15, shows the Pareto frontier points obtained after the algorithm is improved (optimized) by the present invention. It can be seen that the highest temperature of the Pareto frontier point is higher than that of 10,000 random points. and maximum pressure drop The values ​​are all small, which proves the effectiveness of the Pareto frontier; the cluster centers and cluster regions illustrate how to use K-means clustering to automatically divide the frontier into three performance regions. In each region, the most representative solutions with the highest density will be selected and retained to the next generation, effectively avoiding the potential optimal solutions in the sparse areas from being squeezed out by the extreme solutions on the boundary, ensuring the diversity and spatial uniformity of the solution set.

[0044] Step 4, pass the maximum temperature Thermal performance model and maximum pressure drop with respect to fin length Calculation of the maximum temperature of several generations of Pareto fronts based on the thermodynamic performance model of fin length and maximum pressure drop , a decision matrix is ​​constructed through several generations of Pareto fronts, the highest temperatures and the maximum pressure drops of several generations of Pareto fronts; Specifically, the maximum temperature Thermal performance model and maximum pressure drop with respect to fin length Calculation of the maximum temperature of several generations of Pareto fronts based on the thermodynamic performance model of fin length and maximum pressure drop , take several generations of Pareto frontiers as rows of the decision matrix, and take the highest temperature of several generations of Pareto frontiers as and maximum pressure drop As the columns of the decision matrix, the construction of the decision matrix is ​​completed.

[0045] Step 5: Standardize the decision matrix, define the positive ideal solution and the negative ideal solution according to the standardized decision matrix, calculate the Euclidean distance of the positive ideal solution and the negative ideal solution respectively, calculate the relative proximity according to the Euclidean distance of the positive ideal solution and the negative ideal solution, and obtain the optimal value of the fin length according to the relative proximity. =5.477 mm; Preferably, the decision matrix is ​​standardized, and the formula includes:

[0046] Where, Indicates the The Pareto frontier is Standardized value under each thermal performance index; Indicates the The Pareto frontier is The original value of each thermal performance index; Indicates the The first of all Pareto fronts of thermal performance indicators The maximum value of a thermal performance index; Indicates the maximum value; Indicates the Thermal performance indicators; Represents the row of the decision matrix, representing the a Pareto frontier; Represents the columns of the decision matrix, representing the Thermal performance indicators; Preferably, the Euclidean distances of the positive ideal solution and the negative ideal solution are calculated separately, and the formula includes:

[0047]

[0048] Where, represents the Euclidean distance of the positive ideal solution; represents the Euclidean distance to the negative ideal solution; Indicates summation; Indicates the The weight of each thermal performance indicator; Indicates the The Pareto frontier is Standardized value under each thermal performance index; Indicates the Positive ideal solution of a thermal performance index; Indicates the Negative ideal solution of a thermal performance index; Preferably, set the maximum temperature Weight , maximum pressure drop Weight ; Preferably, the relative proximity is calculated based on the Euclidean distance between the positive ideal solution and the negative ideal solution, and the formula includes:

[0049] Where, Indicates the the relative proximity of the two Pareto frontiers; represents the Euclidean distance to the negative ideal solution; represents the Euclidean distance of the positive ideal solution; Preferably, the optimal value of the fin length is obtained according to the relative proximity, and the formula specifically includes:

[0050] Where, Indicates the optimal value of fin length; Indicates the The maximum relative proximity of the Pareto frontier; Indicates the maximum value; Indicates the The relative proximity of two Pareto fronts.

[0051] The present invention compares a multi-objective-mean clustering fin optimization method with the original NSGA-II, wherein the present invention is an improved algorithm and the original NSGA-II is the original algorithm. In order to fully verify the effectiveness of the improved algorithm proposed in the present invention in multi-objective optimization, the present invention compares and analyzes the performance differences between the original algorithm and the improved algorithm from multiple dimensions such as convergence, solution quality, distribution uniformity, computational efficiency and robustness. Through multiple rounds of experiments and statistical tests, the significance and reliability of the improved algorithm are confirmed. The specific indicators are as follows: Hypervolume (HV): Hypervolume is one of the most commonly used indicators to measure the quality and distribution of solution sets. It is used to describe the volume (or area) covered by the current Pareto front solution set in the target space. Simply put, the larger the hypervolume, the closer the non-dominated solution set is to the ideal frontier and the more comprehensive the coverage area. In this paper, the hypervolume uses the ideal reference point (i.e., the maximum temperature) to represent the optimal solution set. and maximum pressure drop The theoretically maximized point) is used as the reference boundary of the hypervolume. Each Pareto front corresponds to a target value. The target value point and the ideal reference point form a rectangle or polygon, and then the sum of the coverage areas is calculated.

[0052] Generational distance (GD): The generational distance is used to measure the average Euclidean distance between the current solution set and the true (theoretical) Pareto frontier, reflecting the convergence of the solution: the smaller the generational distance, the closer the current solution set is to the theoretical optimal frontier; in this invention, the theoretical Pareto frontier is obtained by converting the known maximum temperature and maximum pressure drop The fitting curve is discretized into 10,000 high-precision reference points. For each Pareto front currently generated, the Euclidean distance to the nearest point on the reference front is calculated, and then the average distance of all solutions is calculated to obtain the generation distance value.

[0053] Spacing Metric (SP): The spacing metric is used to measure the uniformity of the distribution of solutions on the Pareto front in the target space. The smaller the spacing metric, the more uniform the distribution of solutions, without local clustering or blank areas.

[0054] The number of solutions: refers to the number of non-dominated solutions retained by the current Pareto front. The greater the number of solutions, the wider the area covered by the Pareto front and the finer the distribution.

[0055] Running time (t): The running time is used to measure the computational efficiency of the entire multi-objective optimization process. The unit is seconds (s). The shorter the time, the smaller the computational burden of the improved parameter dynamic adjustment and clustering mechanism, and the higher the efficiency.

[0056] The original algorithm and the improved algorithm were run 30 times independently. The above indicators were recorded for each generation of each experiment. The average values ​​of the last generation of the 30 runs are shown in Table 2: Table 2 Comparison of the average values ​​of the last generation in 30 runs

[0057] From the results, it can be seen that the improved algorithm is significantly better than the original algorithm in all indicators. The dynamic parameter mechanism allows the crossover probability and mutation probability Adaptive adjustments based on population diversity and iteration progress effectively balance search and exploration, increasing the hypervolume by 10.62%, indicating a wider coverage of the Pareto front and faster convergence. The adaptive crowding threshold and cluster retention strategy ensure that potential high-quality solutions are not ignored, reducing the distance (generational distance) between the solution set and the theoretical optimal frontier by 99.99%, demonstrating a significant improvement in the proximity of the solutions. K-means clustering divides the local area and, combined with the crowding threshold, retains sparse representative solutions from each cluster, significantly reducing over-clustering within the frontier. The spacing metric is reduced by 86.37%, and the distribution of the solution set becomes more even. The improved algorithm retains more boundary solutions and local representative solutions, and the number of solutions increases by 79.12% compared to the original algorithm, further demonstrating a stronger exploration capability. Finally, despite the introduction of dynamic adjustment in cluster selection, the overall runtime is reduced by 26.61%, indicating that the proposed innovative mechanism improves performance while maintaining high computational feasibility. The present invention performs runtime tests on all key indicators. P The (pressure drop) values ​​are all significantly less than 0.05, which proves that the improvement effect is statistically significant and not caused by random fluctuations.

[0058] See also Figure 16The initial generation distance of the original algorithm is very small, and it seems to converge very well. However, this often indicates insufficient population diversity, premature convergence to local optimality, and missed global optimality. In contrast, the initial generation distance of the improved algorithm is larger, reflecting that the dynamic crossover mutation and clustering mechanism maintains a higher search span for the population, avoiding falling into local optimality. As the iteration proceeds, the generation distance steadily decreases and is always smaller than the original algorithm in the later stage, which indicates that the final frontier is closer to the theoretical optimal curve. The improved algorithm effectively avoids premature convergence and ensures population diversity through dynamic parameter adjustment and local clustering selection. Ultimately, the smaller generation distance and the frontier closer to the theoretical optimal greatly improve the global convergence of multi-objective optimization.

[0059] See also Figure 17 , the hypervolume of the original algorithm improves slowly and the plateau value is low, indicating that the solution set fails to fully cover the potential feasible solution area; the improved algorithm enhances the global search capability of the solution set by introducing dynamic crossover and mutation rate, and avoids the localization of the solution set through K-means clustering, so the hypervolume value rises rapidly and remains at a high level for a long time; the improvement of the hypervolume index shows that the improved algorithm significantly enhances the spatial coverage capability of the Pareto front and the diversity of the solution set, which fully verifies that the proposed mechanism can effectively identify more potential high-quality compromise solutions, thereby improving the comprehensiveness of solution selection.

[0060] See also Figure 18 , the spacing metric index of the original algorithm is relatively high, and the initial fluctuation is large, which indicates that there are blank areas on the frontier and the solutions are too concentrated. In addition, the solutions within some boundaries may be scarce or missing, resulting in the loss of compromise solutions. In contrast, the spacing metric index of the improved algorithm has dropped significantly, and the curve is smoother and more stable, which shows that cluster screening effectively ensures the representativeness of local areas. At the same time, the dynamic crowding threshold suppresses the excessive aggregation of dense areas. The significant reduction in the spacing metric shows that the improved algorithm not only generates more non-dominated solutions, but also the solution set is T (Temperature) Space and P The distribution of the (pressure drop) space is more uniform, greatly improving the diversity of understanding and global coverage capabilities.

[0061] See also Figure 19The solution set size of the original algorithm grows slowly and remains at around 166 after convergence, which indicates that the exclusion calculation and elite retention mechanism of the original algorithm are insufficient to maintain a sufficient number of non-dominated solutions. Some potential compromise solutions are squeezed out, resulting in a "sparse" Pareto front. In the improved algorithm, the solution set size expands rapidly in the first few generations and stabilizes at around 299 (close to the upper limit of the solution set size of 300). This is because the global search is enhanced by the dynamic crossover / mutation rate in the early stage, thereby quickly identifying more effective solutions; at the same time, the clustering screening mechanism forcibly retains representative solutions in different local areas, preventing local clustering from causing sparse solutions in certain areas; the significant increase in the number of solutions directly indicates that the improved algorithm can generate a richer non-dominated solution set with the same number of iterations, providing a larger design space for subsequent multi-attribute decision-making.

[0062] See also Figure 20 The original algorithm does not contain dynamic adjustment or clustering operations, so the running time of each generation remains almost unchanged. In contrast, the improved algorithm introduces dynamic crossover rate, mutation rate and K-means clustering screening in the early stage, which leads to slightly higher initial computational overhead. As the number of iterations increases, the adaptive mechanism of the dynamic parameters automatically reduces the crossover probability and mutation probability. Clustering screening can also eliminate redundant individuals, thereby gradually shortening the running time. Ultimately, compared with the original algorithm, the running time of each iteration of the improved algorithm in the later stages is significantly shortened; this shows that although the improved algorithm introduces additional mechanisms, it achieves "more exploration in the early stages and rapid convergence in the later stages" through adaptive adjustment. Compared with the original algorithm, the running time of each iteration is significantly shortened, which greatly improves the computational efficiency, which proves the engineering feasibility and practicality of the algorithm of the present invention in large-scale optimization.

[0063] Example 2: See also Figure 2 In one embodiment of the present invention, a double-layer microchannel heat sink is provided, comprising a heat sink base 1, a heat sink middle portion, a heat sink top portion 3, a first baffle 11, a second baffle 12, and an inlet and outlet unit; the heat sink middle portion comprises fins 2 and fins 5; the inlet and outlet unit comprises a top inlet 4, an upper layer inlet 41, a lower layer inlet 42, an upper layer outlet 43, and a lower layer outlet 44; The heat sink base 1 and the middle of the heat sink are connected by a number of first baffles 11, and a number of lower microchannels are formed between two adjacent first baffles 11. The middle of the heat sink and the top 3 of the heat sink are connected by a number of second baffles 12, and a number of upper microchannels are formed between two adjacent second baffles 12. Each upper microchannel and lower microchannel is provided with an inlet and outlet unit. The top inlet 4 of the inlet and outlet unit is located in the middle of the top 3 of the heat sink, and the upper inlet 41 and the upper outlet 43 are respectively provided in the upper microchannel. The lower inlet 42 and the lower outlet 44 are respectively arranged at the two ends of the lower microchannel, the upper inlet 41 and the lower inlet 42 are located on the same side of the upper microchannel and the lower microchannel, and the upper outlet 43 and the lower outlet 44 are located on the same side of the upper microchannel and the lower microchannel; the middle part of the heat sink includes fins 2 located at both ends and two fins 5 located in the middle, the two fins 5 are arranged in parallel, there is a distance between the two fins 5, the distance is equal to the length of the top inlet 4, and the optimal value of the length of the two fins 5 It is obtained by using a multi-objective-mean clustering fin optimization method mentioned above; Preferably, the optimal value of the length of the two fins 5 is 5.477 mm; Preferably, see Figure 3 , the angle between fin 2 and fin 5 β is 20°; Specifically, the flow path of the double-layer microchannel radiator of this embodiment is as follows: coolant is introduced into the top inlet 4, the upper layer inlet 41 and the lower layer inlet 42 at the same time; under the action of the fins 5, a part of the coolant passing through the top inlet 4 enters the lower layer microchannel through the gap between the two fins 5, and another part of the coolant passing through the top inlet 4 enters the upper layer microchannel; under the action of the fins 2, a part of the coolant passing through the lower layer inlet 42 enters the upper layer microchannel through the fins 2, and another part of the coolant passing through the lower layer inlet 42 continues to pass through the lower layer microchannel and flows out from the lower layer outlet 44; under the action of the fins 5, a part of the coolant passing through the upper layer inlet 41 enters the lower layer microchannel through the gap between the two fins 5, and another part of the coolant passing through the upper layer inlet 41 continues to remain in the upper layer microchannel; under the action of the fins 2, a part of the other part of the coolant that continues to remain in the upper microchannel enters the lower layer microchannel through the fins 2, and the remaining part that does not enter the lower microchannel through the fins 2 flows out through the upper layer outlet 43.

[0064] See also Figure 7 , Reynolds number ( Re ) reflects the laminar or turbulent flow, that is, the nature of the fluid flow. The pressure drop increases with the increase of Reynolds number. This is because the fluid velocity increases, causing the fluid to show a turbulent state, which increases the additional dissipation of energy. As can be seen from the figure, the disturbance of the fluid by fin 2 increases with the increase of the angle with the horizontal. The angle between fin 2 and fin 5 isβ The pressure drop increases fastest when the angle is 20°, and the peak value can reach 8900 Pa. β The 4500 Pa at 15° is nearly doubled. Therefore, in the pursuit of high-disturbance heat exchange scenarios, Example 2 chooses the angle β Double-layer microchannel radiator at 20°.

[0065] See also Figure 8 , contact thermal resistance ( R ) reflects the heat transfer capacity between media. The contact thermal resistance decreases with the increase of Reynolds number. The smaller the contact thermal resistance, the faster the heat transfer rate. As the inlet flow rate increases, the contact thermal resistance gradually decreases, indicating a faster heat transfer rate and better heat transfer effect. Among them, the angle between fin 2 and fin 5 is β The lowest contact thermal resistance value is 0.051 K / W at 20°, which is nearly half of the 0.097 K / W of the reference model, indicating that fin 2 significantly disturbs the fluid and enhances the heat transfer effect.

[0066] See also Figure 9 The bottom peak temperature decreases as the Reynolds number increases. The lower the bottom peak temperature, the better the cooling effect of the radiator. The angle between fin 2 and fin 5 is β When the temperature is 20°, the lowest peak temperature at the bottom of the model is about 313 K, which is about 19 K lower than the maximum temperature of the reference model of about 332 K, indicating that the cooling effect is significant under the same heat flux density.

[0067] See also Figure 10 , the comprehensive heat transfer coefficient formula is ,when When it is >1, it indicates that the overall heat dissipation effect of the double-layer microchannel radiator is better. represents the Nusselt number of the designed model, represents the Nusselt number of the reference model, represents the pressure drop of the design model, Represents the pressure drop of the reference model. The figure shows that the comprehensive heat transfer coefficient values ​​of the four design models are all greater than 1, indicating that they have better heat dissipation effects than the reference model. The angle between fin 2 and fin 5 is β When the angle is 20°, after multi-objective optimization of pressure drop and temperature is achieved by using a multi-objective-mean clustering fin optimization method in Example 1 of the present invention, the comprehensive heat transfer coefficient is further improved, the heat dissipation effect of the double-layer microchannel radiator is improved, and the coordinated optimization of thermal resistance and flow resistance is achieved.

[0068] Example 3: See also Figure 4 , which is different from Example 1 and Example 2 in that the angle between fin 2 and fin 5 is βis 15°.

[0069] Example 4: See also Figure 5 , which is different from Example 1, Example 2 and Example 3 in that the angle between fin 2 and fin 5 is β is 10°.

[0070] Example 5: See also Figure 6 , which is different from Example 1, Example 2, Example 3 and Example 4 in that the angle between fin 2 and fin 5 is β is 0°; Specifically, the flow path of the double-layer microchannel radiator of this embodiment is as follows: coolant is introduced into the top inlet 4, the upper inlet 41 and the lower inlet 42 at the same time. Under the action of the fins 5, a part of the coolant passing through the top inlet 4 enters the lower microchannel through the gap between the two fins 5, and another part of the coolant passing through the top inlet 4 enters the upper microchannel; the coolant passing through the lower inlet 42 flows out from the lower outlet 44 through the lower microchannel; a part of the coolant passing through the upper inlet 41 enters the lower microchannel through the gap between the two fins 5, and another part of the coolant passing through the upper inlet 41 continues to stay in the upper microchannel and flows out through the upper outlet 43.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-objective-mean clustering fin optimization method, characterized in that: The following steps are involved: S1, acquiring experimental data, normalizing the experimental data, and fitting the normalized experimental data to obtain a thermal performance model of the maximum temperature with respect to the fin length and a thermal performance model of the maximum pressure drop with respect to the fin length; S2, initializing the multi-objective optimization algorithm, generating an initialization population through the initialized multi-objective optimization algorithm, the initialization population including several individuals of fin length values; S3, iteratively optimize the initialized population through the initialized multi-objective optimization algorithm, and retain several generations of Pareto frontiers formed in the iterative process; S4, calculating the maximum temperature and maximum pressure drop of several generations of Pareto fronts respectively through the thermal performance model of maximum temperature with respect to fin length and the thermal performance model of maximum pressure drop with respect to fin length, and constructing a decision matrix through the several generations of Pareto fronts and the maximum temperature and maximum pressure drop of several generations of Pareto fronts; S5, standardize the decision matrix, define the positive ideal solution and the negative ideal solution according to the standardized decision matrix, calculate the Euclidean distance of the positive ideal solution and the negative ideal solution respectively, calculate the relative proximity according to the Euclidean distance of the positive ideal solution and the negative ideal solution, and obtain the optimal value of the fin length according to the relative proximity.

2. A multi-objective-mean clustering fin optimization method according to claim 1, characterized in that: The formula of the thermal performance model of the maximum temperature with respect to the fin length specifically includes: Where, Indicates the fin length; Indicates the maximum temperature along the fin length; The formula of the thermal performance model of the maximum pressure drop with respect to the fin length specifically includes: Where, Indicates the fin length; Indicates the maximum pressure drop over the fin length.

3. A multi-objective-mean clustering fin optimization method according to claim 1, characterized in that: Said S2 specifically includes: Set the initialization parameters of the multi-objective optimization algorithm, where the initialization parameters include population size, maximum number of iterations, crossover probability, and mutation probability. Then define the fin length range and step size of the multi-objective optimization algorithm. According to the initialization parameters, fin length range, and step size, several fin length value individuals are generated within the fin length range. The number of fin length value individuals is equal to the value of the population size. Several fin length value individuals constitute the initialization population.

4. A multi-objective-mean clustering fin optimization method according to claim 3, characterized in that: Said S3 specifically includes: S31, performing a first non-dominated sort on the initialized population, then performing a first crowding distance calculation on the initialized population after the first non-dominated sort, selecting a number of parent generations based on the results of the first non-dominated sort and the first crowding distance calculation, generating children from the parent generations until the number of parents and children reaches the population size specified in the initialization parameter, and merging the parents and children to obtain a merged population; S32, performing a second non-dominated sort on the merged population, wherein the highest level non-dominated front is the Pareto front, performing a second crowding distance calculation on the merged population after the second non-dominated sort, selecting the next generation population based on the results of the second non-dominated sort and the second crowding distance calculation, and completing an iterative optimization, wherein the number of the next generation population is the population size in the initialization parameter; S33, execute S31 again on the next generation population, repeat the iteration to the maximum number of iterations in the initialization parameter, and complete the iterative optimization, wherein the Pareto front of each generation is retained, and K-means clustering is performed on the Pareto front retained every five generations.

5. A multi-objective-mean clustering fin optimization method according to claim 4, characterized in that: Said S31 specifically includes: Assign several fin length value individuals of the initialized population to non-dominated frontiers of different levels, calculate the crowding distance of fin length value individuals of the same level non-dominated frontier, randomly select two fin length value individuals each time, and select the one with a higher non-dominated frontier level among the two selected fin length value individuals as the parent. If the non-dominated frontier levels of the two selected fin length value individuals are the same, select the one with a higher crowding distance as the parent, and then use each two parents as parents, and pass the parents and random coefficients to obtain the parent. Linear combination crossover is used to generate two offspring, and Gaussian perturbation mutation is performed on each of the two offspring to obtain Gaussian perturbation mutated offspring until the number of parents and offspring reaches the population size in the initialization parameter. The parents and offspring are then merged to obtain a merged population. The S32 specifically includes: Several individuals with fin length values ​​in the merged population are redistributed to non-dominated fronts of different levels, and the crowding distances of the individuals with fin length values ​​on the non-dominated fronts of the same level are calculated. Several individuals with fin length values ​​are selected as the next generation population according to the non-dominated front level from high to low and the crowding distance from large to small. The number of the next generation population is the population size in the initialization parameter.

6. A multi-objective-mean clustering fin optimization method according to claim 1, characterized in that: Said S4 specifically includes: The maximum temperature and maximum pressure drop of several generations of Pareto fronts are calculated respectively through the thermal performance model of maximum temperature with respect to fin length and the thermal performance model of maximum pressure drop with respect to fin length. The several generations of Pareto fronts are used as the rows of the decision matrix, and the maximum temperature and maximum pressure drop of several generations of Pareto fronts are used as the columns of the decision matrix to complete the construction of the decision matrix.

7. The multi-objective-mean clustering fin optimization method according to claim 1, characterized in that: The decision matrix is ​​standardized in S5, and the formula specifically includes: Where, Indicates the The Pareto frontier is Standardized value under each thermal performance index; Indicates the The Pareto frontier is The original value of each thermal performance index; Indicates the The first of all Pareto fronts of thermal performance indicators The maximum value of a thermal performance index; Indicates the maximum value; Indicates the Thermal performance indicators; Represents the row of the decision matrix, representing the a Pareto frontier; Represents the columns of the decision matrix, representing the Thermal performance indicators; The Euclidean distances of the positive ideal solution and the negative ideal solution are calculated in S5, and the formulas specifically include: Where, represents the Euclidean distance to the positive ideal solution; represents the Euclidean distance to the negative ideal solution; Indicates summation; Indicates the The weight of each thermal performance indicator; Indicates the The Pareto frontier is Standardized value under each thermal performance index; Indicates the Positive ideal solution of a thermal performance index; Indicates the Negative ideal solution of a thermal performance index; The relative proximity is calculated according to the Euclidean distance between the positive ideal solution and the negative ideal solution in S5, and the formula specifically includes: Where, Indicates the the relative proximity of the two Pareto frontiers; represents the Euclidean distance to the negative ideal solution; represents the Euclidean distance to the positive ideal solution; The optimal value of the fin length is obtained according to the relative proximity in S5, and the formula specifically includes: Where, Indicates the optimal value of fin length; Indicates the The maximum relative proximity of the Pareto frontier; Indicates the maximum value; Indicates the The relative proximity of two Pareto fronts.

8. A double-layer microchannel radiator, characterized in that: The heat sink comprises a heat sink base (1), a heat sink middle portion and a heat sink top portion (3), wherein the heat sink base (1) and the heat sink middle portion are connected via a plurality of first baffles (11), and a plurality of lower microchannels are formed between two adjacent first baffles (11); the heat sink middle portion and the heat sink top portion (3) are connected via a plurality of second baffles (12), and a plurality of upper microchannels are formed between two adjacent second baffles (12), and each upper microchannel and lower microchannel is provided with an inlet and outlet unit; The middle portion of the heat sink includes fins (2) located at both ends and two fins (5) located in the middle portion. There is a spacing between the two fins (5). The optimal value of the length of the two fins (5) is It is obtained by using a multi-objective-mean clustering fin optimization method as described in any one of claims 1-7.

9. The double-layer microchannel radiator according to claim 8, characterized in that: The inlet and outlet unit includes a top inlet (4), an upper layer inlet (41), a lower layer inlet (42), an upper layer outlet (43) and a lower layer outlet (44); The top inlet (4) is located in the middle of the top of the heat sink (3), the upper inlet (41) and the upper outlet (43) are respectively arranged at the two ends of the upper microchannel, the lower inlet (42) and the lower outlet (44) are respectively arranged at the two ends of the lower microchannel, the upper inlet (41) and the lower inlet (42) are located on the same side of the upper microchannel and the lower microchannel, and the upper outlet (43) and the lower outlet (44) are located on the same side of the upper microchannel and the lower microchannel; The two fins (5) are arranged in parallel, and the distance between the two fins (5) is equal to the length of the top inlet (4).

10. The double-layer microchannel heat sink according to claim 8, characterized in that: The angle between the fin (2) and the fin (5) β 0° ~ 20°.

Images