Optimization Design Method of Cabinet Radiator Based on GWO-SVM Model

Abstract

The present invention discloses an optimization design method for a chassis radiator based on a GWO-SVM model, which includes the following steps: S1. Construct a geometric model of the radiator, and obtain a data set after processing the geometric model through finite element analysis simulation software; S2. Use the GWO algorithm to optimize the penalty factor c and kernel function g of the SVM model to construct a GWO-SVM model; S3. Train the constructed GWO-SVM model through the data set, and perform reliability analysis on the GWO-SVM model; if the GWO-SVM model is reliable, execute S4; if the GWO-SVM model is unreliable, then add an update factor to retrain the GWO-SVM model; S4. Obtain the simulation prediction result, and calculate the GPU simulation temperature value through finite element analysis simulation software; S5. Screen the structural parameters in the data set whose temperature values are below the Tn value; obtain the optimal solution for adjusting the structural parameters of the radiator corresponding to the GPU temperature. Training the SVM model according to the gray wolf algorithm can quickly obtain the predicted temperature value and reduce the trial production time.

Description

Technical Field

[0001] The present invention relates to the technical field of radiator optimization design. Specifically, it relates to an optimization design method for the cage radiator based on the GWO-SVM model. Background Art

[0002] The existing radiator design is carried out by engineers based on experience. After the prototype is completed, experienced engineers will modify the radiator according to the experimental situation. There are also methods of finite element analysis to perform mesh division on the radiators designed by engineers, simulate the working conditions of the radiators and apply boundary conditions for simulation analysis to replace real experimental tests.

[0003] The existing technology usually requires a relatively long practical test. Even if the finite element analysis method is adopted, when heat dissipation problems are found, engineers need to make continuous attempts. And for each design change, mesh division needs to be carried out again, and boundary conditions need to be reset again. In order to be closer to the real results, the number of meshes is often too large, resulting in too long simulation analysis case time, so that the optimization design cannot be achieved within the specified time and the design process cannot be accelerated. As a result, the finite element analysis cannot really meet the requirements of fast design. Summary of the Invention

[0004] In view of the above technical problems, the present invention proposes an optimization design method for the cage radiator based on the GWO-SVM model, which adopts a method of combining machine learning with finite element analysis. According to a small amount of finite element analysis results, it replaces manual experience to find a better design scheme; according to the model trained by SVM, the predicted temperature value can be quickly obtained; according to the predicted better scheme, finite element analysis is carried out again to obtain the accurate simulation temperature, so as to judge whether the requirements are met; the optimal solution is calculated through the gray wolf algorithm, and processing is carried out according to the accurate solution of the finite element analysis to reduce the trial production time.

[0005] To achieve the above technical objectives, a technical solution provided by the present invention is an optimization design method for the cage radiator based on the GWO-SVM model, including the following steps:

[0006] S1. Construct a geometric model of the radiator, and obtain a data set after processing the geometric model through finite element analysis simulation software;

[0007] S2. Optimize the penalty factor c and kernel function g of the SVM model by using the GWO algorithm to construct a GWO-SVM model;

[0008] S3. Train the constructed GWO-SVM model through the data set and perform reliability analysis on the GWO-SVM model; if the GWO-SVM model is reliable, execute S4; if the GWO-SVM model is unreliable, add an update factor to retrain the GWO-SVM model;

[0009] S4. Obtain the simulation prediction results, and calculate the GPU simulation temperature value through the finite element analysis simulation software.

[0010] S5. Screen the structural parameters in the dataset whose temperature values are below the Tn value; combine the temperature distribution map of the radiator and the spatial maximum envelope to obtain the optimal solution for adjusting the structural parameters of the radiator corresponding to the GPU temperature.

[0011] In this solution, the finite element analysis simulation software includes one of flotherm, Fluent, ABAQUS, domestic simetherm, and ANSYS ICPAK, which is not limited here; to reduce the computer simulation calculation time and improve the simulation efficiency, preprocess and postprocess the geometric model of the radiator. After obtaining a batch of simulation results, input them into the support vector machine model as samples, and at the same time use the grey wolf optimization algorithm to adjust the SVM model parameters, so that a large number of parameters can be modified and accurate simulation results can be obtained, saving labor costs and time.

[0012] Preferably, a dataset is obtained after the geometric model is processed by the finite element analysis simulation software, including de-featuring the geometric model, setting boundary conditions, setting material properties, and setting mesh parameters. The dataset is labeled according to the range interval of the temperature value, and then the original label set is obtained.

[0013] Preferably, the dataset is divided into a training set, a calibration set, and a test set according to a certain ratio by the KS algorithm. For example, the division ratio of the training set, the calibration set, and the test set is 3:1:1; and the dataset is normalized; the training set is used as the input data when training the GWO-SVM model, the calibration set is used as the input data when determining the reliability of the GWO-SVM model, and the test set is used as the input data when verifying the reliability of the GWO-SVM model.

[0014] In this solution, in order to maintain the consistency of the data distribution, the original dataset is divided into three parts according to the ratio of 3:1:1 by the KS algorithm: the training set, the calibration set, and the test set. The training set is a batch of data initially input into the model for training. The model parameters are corrected by the prediction accuracy of the calibration set. Finally, the trained model is evaluated by the test set.

[0015] Due to the inconsistency of the radiator parameter values, the large difference in the order of magnitude between the data will affect the prediction results of the SVM algorithm, and the SVM algorithm is sensitive to the data between [0, 1]. To improve the accuracy, the sample set is normalized so that its range is between [0, 1].

[0016] Preferably, in S2, the steps of optimizing the penalty factor and kernel function of the SVM model using the GWO algorithm are as follows: S21. Initialize the GWO algorithm, including determining the gray wolf population N; determining the parameters to be optimized, namely the penalty factor c and the kernel function g; the maximum number of iterations t, initializing the positions of the α wolf, β wolf, and γ wolf; setting the boundary values of the wolf pack positions, and initializing the fitness values of the individual wolves in the wolf pack;

[0017] S22. Randomly update the positions of the α wolf, β wolf, and γ wolf, and adjust the positions of the wolf pack;

[0018] S23. Traverse each wolf, calculate the fitness values of the individual wolves in the wolf pack. If the fitness value of the new individual is better than that of the old individual, then replace the position of the old individual with the position of the new individual, and retain the fitness value of the new individual as the target value;

[0019] S24. Continuously iterate the GWO algorithm. When the maximum number of iterations t is reached, output the penalty factor c and the kernel function g. Otherwise, continuously execute S22 - S24.

[0020] Preferably, adjusting the positions of the wolf pack includes:

[0021] If the position of the wolf pack is within the boundary values, then determine the current position of the wolf pack; if the position of the wolf pack is outside the boundary values, it is processed in two cases;

[0022] Case 1: If it is greater than the upper limit of the boundary value, then adjust the current position of the wolf pack to the upper limit of the boundary value;

[0023] Case 2: If it is less than the lower limit of the boundary value, then adjust the current position of the wolf pack to the lower limit of the boundary value.

[0024] Preferably, in S22, the formulas for randomly updating the positions of the α wolf, β wolf, and γ wolf are as follows:

[0025]

[0026] In the formula, D is the distance from the prey, t represents the number of iterations, X p (t) represents the current position of the prey, X(t) represents the current position of the wolf pack, and X(t + 1) represents the updated position of the wolf pack. A and C represent random vectors.

[0027] Preferably, the random vectors A and C are represented by the following formulas:

[0028]

[0029] Among them, a is a value that linearly decreases with the number of iterations within the range of [0, 2], r 1 and r 2 are random numbers within the range of [0, 1].

[0030] Preferably, the reliability analysis of the GWO-SVM model includes the following steps:

[0031] Determine the predicted label value according to the temperature value corresponding to the simulation prediction result, and then determine the predicted label set. Compare the predicted label value with the original label value. If the error between the predicted label value and the original label set is less than h, it is determined that the GWO-SVM model is reliable. Optimize the parameters of SVM through the GWO algorithm to improve the accuracy of sample classification. The input of SVM is the main structural parameters of the radiator and the GWO initialization parameters, and the output is the optimal SVM parameters and the simulation prediction result.

[0032] Preferably, label the GPU temperature value according to the temperature value range of the GPU temperature. The temperature value range corresponding to the first label value is 70°C to 75°C; the temperature range of the second label value is: 75°C to 80°C; the temperature range of the third label value is: above 80°C.

[0033] Preferably, the radiator structure parameters include the installation position in the X direction, width position, total length, width length, base height, total height, number of fins, thickness, and GPU temperature. By analyzing the temperature characteristic curve of the heat sink, the temperature effect of the heat sink is positively correlated with the number of fins, fin height, material thermal conductivity, base height, thickness, and length-width; it is negatively correlated with the thickness; based on the temperature characteristics, guide the optimization direction of the structural parameters. By adjusting one structural parameter and arranging and combining the remaining parameters, several test sets are obtained, and the test sets are screened by the temperature value label.

[0034] Advantages of the present invention: The present invention proposes an optimization design method for the cage radiator based on the GWO-SVM model, which adopts a method combining machine learning and finite element analysis. According to a small amount of finite element analysis results, it replaces manual experience to find a better design scheme; according to the model trained by SVM, the predicted temperature value can be quickly obtained; according to the predicted better scheme, finite element analysis is carried out again to obtain the accurate simulation temperature, so as to judge whether the requirements are met; the optimal solution is calculated through the gray wolf algorithm, and processing is carried out according to the accurate solution of the finite element analysis, reducing the trial production time. Brief Description of the Drawings

[0035] Figure 1 It is a flowchart of the optimization design method for the cage radiator based on the GWO-SVM model of the present invention.

[0036] Figure 2 It is an algorithm flowchart of the GWO-SVM model of the present invention.

[0037] Figure 3 It is a schematic diagram of the fitness curve corresponding to the GWO-SVM model of the present invention.

[0038] Figure 4 This is a schematic diagram of the accuracy of the calibration set of the GWO-SVM model of the present invention.

[0039] Figure 5 This is a schematic diagram of the accuracy of the test set of the GWO-SVM model of the present invention. Specific embodiments

[0040] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only the best embodiments of the present invention, which are only used to explain the present invention and do not limit the protection scope of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0041] Embodiment: As Figure 1 shown, an optimization design method for a chassis radiator based on a GWO-SVM model includes the following steps:

[0042] S1. Construct a geometric model of the radiator, and obtain a data set after processing the geometric model through finite element analysis simulation software;

[0043] S2. Use the GWO algorithm to optimize the penalty factor c and kernel function g of the SVM model to construct a GWO-SVM model;

[0044] S3. Train the constructed GWO-SVM model through the data set and perform reliability analysis on the GWO-SVM model; if the GWO-SVM model is reliable, execute S4; if the GWO-SVM model is unreliable, add an update factor to retrain the GWO-SVM model;

[0045] S4. Obtain the simulation prediction result, and calculate the GPU simulation temperature value through finite element analysis simulation software;

[0046] S5. Screen the structural parameters in the data set whose temperature values are below the Tn value; combine the temperature distribution map and the spatial maximum envelope of the radiator to obtain the optimal solution for adjusting the structural parameters of the radiator corresponding to the GPU temperature.

[0047] In this embodiment, the finite element analysis simulation software includes one of flotherm, Fluent, ABAQUS, domestic simetherm, and ANSYS ICPAK, which is not limited herein; to reduce the computer simulation calculation time and improve the simulation efficiency, preprocess and postprocess the geometric model of the radiator. After obtaining a batch of simulation results, input them into the support vector machine model as samples, and at the same time use the grey wolf optimization algorithm to adjust the SVM model parameters, so as to modify the parameters in large quantities and obtain accurate simulation results, saving labor costs and time.

[0048] The GWO-SVM model proposed in this embodiment. As Figure 2 shown, the GWO-SVM model is an efficient model with a support vector machine as the basic framework structure and an intelligent swarm optimization algorithm update strategy. In this model, a parameter combination of SVM is represented by the position information of particles. Through grey wolf optimization, the information of the optimal particle is obtained, and then the optimal classifier can be established according to the optimal particle to find the optimal classification hyperplane. This model optimizes the parameters of SVM through the GWO algorithm to improve the accuracy of sample classification; the input of SVM is the main structural parameters of the radiator and the GWO initialization parameters, and the output is the optimal SVM parameters and the simulation prediction results.

[0049] The geometric model is processed by finite element analysis simulation software to obtain a data set, including de-featuring the geometric model, setting boundary conditions, setting material properties, and setting mesh parameters. The data set is labeled according to the range interval of temperature values to obtain the original label set.

[0050] The data set is divided into a training set, a calibration set, and a test set according to a certain ratio by the KS algorithm. For example, the division ratio of the training set, the calibration set, and the test set is 3:1:1; and the data set is normalized; the training set is used as the input data when training the GWO-SVM model, the calibration set is used as the input data when determining the reliability of the GWO-SVM model, and the test set is used as the input data when verifying the reliability of the GWO-SVM model.

[0051] In this embodiment, in order to maintain the consistency of data distribution, the original data set is divided into three parts according to the ratio of 3:1:1 by the KS algorithm: a training set, a calibration set, and a test set. The training set is a batch of data for initial input into the model for training. The model parameters are corrected by the prediction accuracy of the calibration set. Finally, the trained model is evaluated by the test set.

[0052] Due to the inconsistency of the radiator parameter values, the large difference in the order of magnitude between data will affect the prediction results of the SVM algorithm, and the SVM algorithm is more sensitive to data between [0, 1]. In order to improve the accuracy, the sample set is normalized so that its range is between [0, 1].

[0053] In S2, optimizing the penalty factor and kernel function of the SVM model by the GWO algorithm includes the following steps:

[0054] S21. Initialize the GWO algorithm, including determining the gray wolf population N; determining the parameters to be optimized, namely the penalty factor c and the kernel function g; the maximum number of iterations t, initializing the positions of the α-wolf, β-wolf, and γ-wolf; setting the boundary values of the wolf pack positions, and initializing the fitness values of the individual wolves in the wolf pack;

[0055] S22. Randomly update the positions of the α-wolf, β-wolf, and γ-wolf, and adjust the positions of the wolf pack;

[0056] S23. Traverse each wolf, calculate the fitness values of the individual wolves in the wolf pack. If the fitness value of the new individual is better than that of the old individual, then replace the position of the old individual with the position of the new individual, and retain the fitness value of the new individual as the target value;

[0057] S24. Continuously iterate the GWO algorithm. When the maximum number of iterations t is reached, output the penalty factor c and the kernel function g, otherwise continue to execute S22 - S24.

[0058] Adjust the positions of the wolf pack, including:

[0059] If the position of the wolf pack is within the boundary value range, then determine the current position of the wolf pack; if the position of the wolf pack is outside the boundary value range, it is processed in two cases;

[0060] Case 1: If it is greater than the upper limit of the boundary value, then adjust the current position of the wolf pack to the upper limit of the boundary value;

[0061] Case 2: If it is less than the lower limit of the boundary value, then adjust the current position of the wolf pack to the lower limit of the boundary value.

[0062] In S22, the formulas for randomly updating the positions of the α-wolf, β-wolf, and γ-wolf are as follows:

[0063]

[0064] In the formula, D is the distance from the prey, t represents the number of iterations, X p (t) represents the current position of the prey, X(t) represents the current position of the wolf pack, and X(t + 1) represents the updated position of the wolf pack, and A and C represent random vectors.

[0065] The formulas for the random vectors A and C are expressed as follows:

[0066]

[0067] Among them, a is a value that linearly decreases with the number of iterations within the range of [0, 2], r 1 and r 2 are random numbers within the range of [0, 1].

[0068] The reliability analysis of the GWO - SVM model includes the following steps:

[0069] Determine the predicted label value according to the temperature value corresponding to the simulation prediction result, and then determine the predicted label set. Compare the predicted label value with the original label value. If the error between the predicted label value and the original label set is less than h, it is determined that the GWO-SVM model is reliable.

[0070] Label the GPU temperature values according to the temperature value range of the GPU temperature. The temperature value range corresponding to the first label value is 70°C to 75°C; the temperature range of the second label value is: 75°C to 80°C; the temperature range of the third label value is: above 80°C.

[0071] The radiator structure parameters include the installation position in the X direction, width position, total length, width length, base height, total height, number of fins, thickness, and GPU temperature. By analyzing the temperature characteristic curve of the heat sink, the temperature effect of the heat sink is positively correlated with the number of fins, fin height, material thermal conductivity, base height, thickness, and length width; it is negatively correlated with the thickness; based on the temperature characteristics, guide the optimization direction of the structure parameters. By adjusting one structure parameter and arranging and combining the remaining parameters, several test sets are obtained, and the test sets are screened through the temperature value labels.

[0072] The specific simulation experiment is as follows:

[0073] By changing the input of the radiator structure parameters, perform simulation calculations in the ANSYS Icepak simulation software to obtain 40 groups of simulation data as shown in Table 1. The GPU temperature between 70°C and 75°C is label category 1, 75°C to 80°C is label category 2, and above 80°C is label category 3. Divide these 40 initial data into a training set, a calibration set, and a test set according to a ratio of 3:1:1 and input them into the GWO-SVM model. Input the 40 groups of simulation data as a sample set into the GWO-SVM model to obtain the best values of the penalty factor and the kernel function; as Figure 3 、 Figure 4 、 Figure 5 shown, when c takes 440.05 and g takes 0.01 when achieving the highest accuracy in the calibration set, the accuracy of the test set reaches 100%, which can meet the prediction of the radiator simulation results.

[0074] The base thickness, fin spacing, and thickness are all factors affecting the heat dissipation effect. Increasing the base thickness and decreasing the fin spacing and thickness can all increase the heat dissipation area. However, excessive dimensions will bring a series of negative effects. The radiator dimensions should be reasonably selected under the overall design requirements of the product and the actual process conditions.

[0075] As can be seen from Table 1, based on the maximum spatial envelope and temperature field distribution of the heat sink, it is concluded that within the operating temperature range, the temperature effect of the heat sink is positively correlated with the number of fins, fin height, and material thermal conductivity, and negatively correlated with the thickness; it is positively correlated with the base height, thickness, and width. Therefore, when optimizing the structural parameters of the next version, the parameters can be modified according to the corresponding relationships.

[0076] Table 1. Simulation data sample table.

[0077] Serial number Xe Xs Total length L Width length Base height H1 Total height H Number of sheets n Thickness t GPU Temperature range 1 -62.63 -126.63 64 82 3 11.8 13 1.5 110.4 3 2 -62.63 -126.63 64 100 3 11.8 13 1.5 110.4 3 3 -41.43 -131.63 90.2 100 6 28 33 0.6 93.093 3 4 -42.63 -136.63 94 100 6 17.8 25 1 84.5 3 5 -42.63 -136.63 94 100 6 19.8 25 1 82 3 6 -42.63 -136.63 94 100 7 22.8 20 1 81.6 3 7 -42.63 -136.63 94 100 7 21.8 25 1 80.4 3 8 -42.63 -136.63 94 100 6 21.8 25 1 80.1 3 9 -42.63 -136.63 94 100 6 21.8 28 1 79.3 2 10 -42.63 -136.63 94 100 7 22.8 25 1 79.3 2 11 -42.63 -136.63 94 100 5 23.8 25 1 78.8 2 12 -42.63 -136.63 94 100 7 23.8 25 1 78.7 2 13 -42.63 -136.63 94 100 5 24.8 25 1 78.1 2 14 -42.63 -136.63 94 100 8 23.8 28 1 78 2 15 -42.63 -136.63 94 100 6 24.8 25 1 77.8 2 16 -42.63 -136.63 94 100 6 25.8 25 1 77.2 2 17 -42.63 -136.63 94 100 6 26.8 25 1 76.7 2 18 -42.63 -136.63 94 100 7 26.8 25 1 76.5 2 19 -42.63 -136.63 94 100 6 26.8 26 0.8 76.1 2 20 -38.03 -128.63 90.6 82 6 26.8 26 0.6 75.6 2 21 -46.03 -136.63 90.6 82 6 26.8 26 0.6 75.6 2 22 -38.03 -128.63 90.6 82 6 26.8 31 0.5 75.2 2 23 -38.03 -128.63 90.6 82 6 26.8 31 0.5 75.19 2 24 -46.63 -136.63 90 100 6 26.8 30 0.5 75 1 25 -46.03 -136.63 90.6 82 6 26.8 31 0.6 74.9 1 26 -43.33 -136.63 93.3 100 6 26.8 30 0.5 74.9 1 27 -38.23 -128.63 90.4 100 6 28 32 0.5 73.8 1 28 -41.03 -131.63 90.6 100 6 28 26 0.6 73.5394 1 29 -38.03 -128.63 90.6 100 6 28 31 0.5 73.46 1 30 -30.23 -131.63 10.4 100 6 28 29 0.6 73.4185 1 31 -38.03 -128.63 90.6 100 6 28 31 0.6 73.38 1 32 -41.43 -131.63 90.2 100 6 28 29 0.6 73.3774 1 33 -34.63 -136.63 102 100 6 28 36 0.5 72.46 1 34 37.03 -136.63 99.6 100 7 28 34 0.6 72.4 1 35 -37.03 -136.63 99.6 100 7 28 34 0.6 72.38 1 36 -37.03 -136.63 99.6 100 6 28 40 0.5 72.33 1 37 -36.13 -136.63 100.5 100 6 28 41 0.5 72.33 1 38 37.03 -136.63 99.6 100 6 28 40 0.5 72.3 1 39 36.13 -136.63 100.5 100 6 28 41 0.5 72.3 1 40 -36.13 -136.63 100.5 100 7 28 41 0.5 72.25 1

[0078] The specific process is as follows:

[0079] Step1: By modifying the values of each parameter and arranging and combining the remaining parameters, a list of structural parameters to be optimized is obtained, totaling 98 groups;

[0080] Step2: Input them into the GWO - SVM model to obtain the simulation prediction results;

[0081] Step3: Screen out each group of structural parameters with a predicted result below 75°C, obtaining columns 2 - 9 in Table 2, a total of 20 groups;

[0082] Step4: Conduct simulation calculations according to each group of structural parameters in Table 2 to obtain the specific simulation temperature values, as shown in the last column of Table 2.

[0083] Table 2. Improved radiator parameter table.

[0084] Serial number Xe Xs Total length L Width length Base height H1 Total height H Number of sheets n Thickness t GPU Temperature range Original -62.63 -126.63 64 82 3 11.8 13 1.5 110.4 3 1 -38.03 -128.63 90.6 90 6 26.8 31 0.6 74.46 1 2 -37.03 -128.63 91.6 100 6 26.8 36 0.6 73.93 1 3 -36.13 -128.63 92.5 100 6 28 41 0.5 73.2 1 4 -38.03 -128.63 90.6 100 6 28 31 0.5 73.46 1 5 -38.03 -128.63 90.6 100 7 28 31 0.5 73.1 1 6 37.03 -136.63 99.6 100 6 26.8 34 0.6 73.04 1 7 -38.13 -128.63 90.5 100 7 28 37 0.5 72.2 1 8 -38.23 -128.63 90.4 100 7 28 32 0.5 73.54 1 9 -38.23 -128.63 90.4 100 6 28 32 0.5 73.8 1 10 37.03 -136.63 99.6 100 6 28 39 0.6 72.6 1 11 -37.03 -136.63 99.6 100 6 26.8 34 0.6 73.05 1 12 37.03 -136.63 99.6 100 6 28 34 0.5 72.5 1 13 -37.03 -136.63 99.6 100 6 28 34 0.6 72.68 1 14 37.03 -136.63 99.6 100 6 28 40 0.5 72.3 1 15 36.13 -136.63 100.5 100 6 28 41 0.5 72.3 1 16 36.13 -136.63 100.5 100 7 28 41 0.5 72.1 1 17 -37.03 -136.63 99.6 100 6 28 34 0.5 72.63 1 18 -36.13 -136.63 100.5 100 7 28 41 0.5 72.25 1 19 -34.63 -136.63 102 100 6 28 36 0.5 72.46 1 20 -34.63 -136.63 102 100 7 28 36 0.5 72.18 1

[0085] According to the GPU temperature under different parameters in the table, combined with the temperature distribution map and the maximum spatial envelope of the heat sink, a better solution to reduce the GPU temperature is selected. The total length changes from the original 64 mm to 102 mm. The total width changes from the original 82 mm to 100 mm. The total height changes from the original 11.8 mm to 28 mm, where the base height increases by 4 mm to 6 mm, and the fin height increases by 14.8 mm to 22 mm. The number of fins changes from the original 13 to 36, and the thickness changes from the original 1.5 to 0.5 mm.

[0086] The above - described specific implementation manner is the preferred implementation manner of the optimization design method of the cabinet heat sink based on the GWO - SVM model of the present invention, and does not limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to this specific implementation manner. Any equivalent changes made according to the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. Optimization design method of cage radiator based on GWO-SVM model, Characterized in that, It includes the following steps: S1. Construct a geometric model of the radiator, and obtain a data set after processing the geometric model through finite element analysis simulation software; S2. Use the GWO algorithm to optimize the penalty factor c and kernel function g of the SVM model to construct a GWO-SVM model; S3. Train the constructed GWO-SVM model through the data set, and perform reliability analysis on the GWO-SVM model; if the GWO-SVM model is reliable, execute S4; if the GWO-SVM model is unreliable, then add an update factor to retrain the GWO-SVM model; S4. Obtain the simulation prediction result, and calculate the GPU simulation temperature value through finite element analysis simulation software; S5. Screen the structural parameters in the data set whose temperature values are below the Tn value; combine the temperature distribution map and the spatial maximum envelope of the radiator to obtain the optimal solution for adjusting the structural parameters of the radiator corresponding to the GPU temperature; In S2, using the GWO algorithm to optimize the penalty factor and kernel function of the SVM model includes the following steps: S21. Initialize the GWO algorithm, including determining the gray wolf population N; determining the parameters to be optimized, namely the penalty factor c and the kernel function g; the maximum number of iterations t, and initializing wolves, wolves, the positions of wolves; setting the boundary values of the positions of the wolf pack, and initializing the fitness values of the individuals in the wolf pack; S22. Randomized update Wolf, Wolf, the position of the wolves, and adjust the position where the wolf pack is located; S23. Traverse each wolf, calculate the fitness value of the wolf pack individuals. If the fitness value of the new individual is better than that of the old individual, then the position of the new individual replaces the position of the old individual, and retain the fitness value of the new individual as the target value; S24. Continuously iterate the GWO algorithm. When the maximum iteration number t is reached, output the penalty factor c and kernel function g, otherwise continuously execute S22-S24.

2. The optimization design method of the cage radiator based on the GWO-SVM model according to claim 1, Characterized in that, The data set obtained after processing the geometric model through finite element analysis simulation software includes: de-featuring the geometric model, setting boundary conditions, setting material properties, and setting mesh parameters, and label the data set according to the range interval of temperature values, and then obtain the original label set.

3. The optimization design method of the cage radiator based on the GWO-SVM model according to claim 1 or 2, Characterized in that, The data set is divided into a training set, a calibration set and a test set according to a certain proportion by the KS algorithm, and the data set is normalized; the training set is used as the input data when training the GWO-SVM model, the calibration set is used as the input data when determining the reliability of the GWO-SVM model, and the test set is used as the input data when verifying the reliability of the GWO-SVM model.

4. The optimization design method of the cage radiator based on the GWO-SVM model according to claim 1, Characterized in that, Adjusting the position of the wolf pack includes: If the position of the wolf pack is within the boundary value range, then determine the current position of the wolf pack; if the position of the wolf pack is outside the boundary value range, it is processed in two cases; Case 1: If it is greater than the upper limit of the boundary value, then adjust the current position of the wolf pack to the upper limit of the boundary value; Case 2: If it is less than the lower limit of the boundary value, then adjust the current position of the wolf pack to the lower limit of the boundary value.

5. The optimization design method of the cage radiator based on the GWO-SVM model according to claim 1, Characterized in that, In S22, randomized update wolf, wolf, The position formula of the wolf is as follows: ； Wherein, is the distance to the prey, represents the number of iterations, represents the position of the current prey, represents the position of the current wolf pack, represents the position of the updated wolf pack, and represent random vectors.

6. The optimization design method of the cage radiator based on the GWO-SVM model according to claim 1, characterized in that, Random vector and are expressed by the following formula: ； Among them, is a value that linearly decreases with the number of iterations within a certain range, and are random numbers within a certain range.

7. The optimization design method of the cage radiator based on the GWO-SVM model according to claim 2, characterized in that, The reliability analysis of the GWO-SVM model includes the following steps: Determine the prediction label value according to the temperature value corresponding to the simulation prediction result, and then determine the prediction label set. Compare the prediction label value with the original label value. If the error between the prediction label value and the original label set is less than h, it is determined that the GWO-SVM model is reliable.

8. The optimization design method of the cage radiator based on the GWO-SVM model according to claim 1, characterized in that, Label the GPU temperature value according to the temperature value range of the GPU temperature. The temperature value range corresponding to the first label value is 70°C to 75°C; the temperature range of the second label value is: 75°C to 80°C; the temperature range of the third label value is: above 80°C.

9. The optimization design method of the cage radiator based on the GWO-SVM model according to claim 8, characterized in that, The radiator structure parameters include the X-direction installation position, width position, total length, width length, base height, total height, number of fins, thickness, and GPU temperature. By analyzing the temperature characteristic curve of the heat sink, the temperature effect of the heat sink is positively correlated with the number of fins, fin height, material thermal conductivity, base height, thickness, and length width; negatively correlated with the thickness; guiding the optimization direction of the structure parameters according to the temperature characteristics, adjusting one of the structure parameters, and arranging and combining the remaining parameters to obtain several test sets, and screening the test sets through the temperature value label.

Images