Co-simulation design method and system for thermoelectric cooler based on neural network acceleration

Abstract

The present invention discloses a collaborative simulation design method and system for thermoelectric coolers based on neural network acceleration, the method comprising the following steps: (1) building an equivalent physical field simulation model of a thermoelectric cooler; (2) simulating or experimentally measuring the equivalent physical field simulation model of the thermoelectric cooler, collecting structural parameters, and drawing an influencing parameter-performance parameter curve according to the corresponding relationship between the length, width, height, and current parameters of the thermoelectric arm and the performance of the thermoelectric cooler to find the parameters affecting the thermoelectric cooler; (3) building a neural network model, collecting influencing parameter-performance parameter data sets, and performing neural network training and testing to obtain a trained neural network model; (4) building a particle swarm optimization algorithm for optimization, and continuously approaching the target performance parameters by controlling the influencing parameters, and finally obtaining the thermoelectric cooler structural parameters that meet the design requirements. The present invention greatly shortens the design cycle of the thermoelectric cooler and improves the design efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of thermoelectric cooler simulation agent tool design, and in particular relates to a thermoelectric cooler collaborative simulation design method and system based on neural network acceleration. Background Art

[0002] As the semiconductor industry's advanced process technology advances, process nodes are gradually approaching physical limits, which means that the semiconductor industry, which relies on scaling technology under the guidance of Moore's Law, has encountered a bottleneck. However, with the rapid development of advanced packaging technology, 2.5D / 3D integration technology can realize the organic integration of multiple chips at different process nodes, so this technology is regarded as one of the most powerful paths to "More than Moore's Law".

[0003] 3D Heterogeneous Integration is an advanced semiconductor manufacturing technology that combines chips or components with different materials, different manufacturing processes and different functions in a vertical stack to form a highly integrated system. However, it is precisely because of the close arrangement or even vertical stacking of multiple chips in the same package that the chip heat dissipation and heat flux density increase significantly; at the same time, the uneven arrangement of multiple chips with different power levels will also cause local hot spots. Therefore, there is an urgent need for a thermal management method that can achieve integrated applications and precise cooling.

[0004] Traditional thermal management methods (such as fin cooling, forced air cooling, water-cooled convection, etc.) cannot effectively deal with thermal problems such as high heat dissipation and local hot spots. The fin structure is difficult to be effectively used in microsystems due to its large size; forced air cooling technology has problems such as high energy consumption, high noise, and high maintenance costs; the reliability of water-cooled convection structure cannot be ignored. Therefore, thermoelectric coolers (TECs), as an active thermal management method based on the thermoelectric effect, are widely used in production and life because of their small size, light weight, high reliability, accurate temperature control, and rapid cooling. They play an important role in the field of microsystem thermal management.

[0005] However, current research on thermoelectric coolers mainly relies on commercial finite element simulation tools such as Ansys, COMSOL, Multiphysics, etc. Although such numerical simulation software is powerful, it is necessary to solve a set of partial differential equations during the simulation of complex multi-physics problems involved in the thermoelectric effect. This process is time-consuming and inefficient, causing inconvenience to technicians in the development, design and application of thermoelectric coolers. Summary of the invention

[0006] The purpose of the present invention is to address the deficiencies of the prior art and provide a thermoelectric cooler collaborative simulation design method and system based on neural network acceleration.

[0007] The objective of the present invention is achieved through the following technical solutions:

[0008] The collaborative simulation design method of thermoelectric cooler based on neural network acceleration specifically includes the following steps:

[0009] (1) Build an equivalent physical field simulation model of a thermoelectric cooler;

[0010] (2) Through simulation or experimental measurement of the equivalent physical field simulation model of the thermoelectric cooler, the corresponding relationship between different structural parameters, current parameters and the performance of the thermoelectric cooler is collected, and the influencing parameter-performance parameter curve is drawn to find the parameters that mainly affect the thermoelectric cooler;

[0011] (3) Build a neural network model, collect influencing parameter-performance parameter data sets, and perform neural network training and testing to obtain a trained neural network model;

[0012] (4) A particle swarm optimization (PSO) algorithm was built to find the optimal solution. By controlling the main influencing parameters, the target performance parameters were continuously approached, and finally the optimal thermoelectric cooler structural parameters that met the design requirements were obtained.

[0013] Furthermore, in step (1), the accuracy of the equivalent physical field simulation model of the thermoelectric cooler is verified by experiments.

[0014] Furthermore, step (2) specifically includes the following sub-steps:

[0015] (2.1) Using the principle of controlled variables, study the influence of the ceramic thickness, conductor thickness, rubber thickness, logarithm of the thermoelectric arm, cross-sectional area of ​​the thermoelectric arm, height of the thermoelectric arm, and current on the performance of the thermoelectric cooler;

[0016] (2.2) Based on the collected data, draw the curve of influencing parameters-performance parameters according to the corresponding relationship between current parameters and thermoelectric cooler performance;

[0017] (2.3) According to the development trend of the curve, find the main influencing parameters of the thermoelectric cooler.

[0018] Furthermore, step (3) specifically includes the following sub-steps:

[0019] (3.1) Building a neural network model, the neural network model has five input values, namely, the logarithm N of the thermoelectric arm of the thermoelectric cooler, the length L of the thermoelectric arm, the width W of the thermoelectric arm, the height H of the thermoelectric arm, and the input current I; three hidden layers are set, and the number of neurons in the hidden layers is 10, 10, and 10 respectively; there are four output values, namely, the cold surface temperature of the thermoelectric cooler , temperature difference between hot and cold surfaces DT, cooling capacity , the power consumed P;

[0020] (3.2) 1250 sets of influencing parameter-performance parameter data were collected, where each set of influencing parameter-performance parameter data included the number of pairs of thermoelectric arms N, the length L of the thermoelectric arms, the width W of the thermoelectric arms, the height H of the thermoelectric arms, the input current I, the cold surface temperature , temperature difference between hot and cold surfaces DT, cooling capacity , the consumed electric energy P; the 1250 sets of data are divided into training set and test set in the ratio of 5:2;

[0021] (3.3) Train the constructed neural network model through the training set and test it through the test set to obtain the trained neural network model.

[0022] Furthermore, in step (3.2), 1250 sets of influencing parameter-performance parameter data are collected through experiments, simulations and hybrid methods.

[0023] Furthermore, in step (3.3), the constructed neural network model is trained with the maximum number of training times as the convergence criterion.

[0024] Furthermore, step (4) specifically includes:

[0025] Build a particle swarm optimization algorithm for target design:

[0026] The design goal is read from the input end and recorded as:

[0027] ;

[0028] in, represents the expected temperature difference, Indicates the expected power consumption, Indicates the expected cooling capacity;

[0029] In a D-dimensional target search space, an initial particle swarm consisting of M particles is created, where the velocity and position of the i-th particle at time t are expressed as:

[0030] ; ∈ ;

[0031] ; ∈ ;

[0032] in, and Represent the minimum and maximum speed respectively. and Represent the lower and upper limits of each particle’s search space, respectively. , ;

[0033] The particles are predicted by the neural network respectively, and the prediction result of the i-th particle at time t is expressed as:

[0034] ;

[0035] in, represents the temperature difference of the thermoelectric cooler predicted by the neural network, represents the power consumption of the thermoelectric cooler predicted by the neural network, represents the cooling capacity of the thermoelectric cooler predicted by the neural network, ;

[0036] The individual fitness of the i-th particle at time t is calculated as:

[0037] ;

[0038] The individual best fitness of the i-th particle at time t and the group best fitness of the g-th iteration are:

[0039] ;

[0040] ;

[0041] in, represents the expected weight of the target temperature difference, represents the expected weight of the target power consumption, represents the expected weight of the target cooling capacity, , ;

[0042] Update the individual optimal position of the i-th particle at time t and the group optimal position completed in the g-th iteration respectively:

[0043] ;

[0044] ;

[0045] in, for The corresponding particle position, , ;

[0046] The speed and position of the dth dimension of the i-th particle at time t+1 are updated as follows:

[0047] ;

[0048] ;

[0049] in, is the inertia weight, and is a random number between (0,1), and represents the learning factor, , , ;

[0050] The above process is repeated, and after G iterations, the thermoelectric cooler structure and current parameters that can meet the design objectives are finally determined.

[0051] Furthermore, the thermoelectric cooler includes cross-distributed P-type thermoelectric arms, N-type thermoelectric arms and metal wires, and the P-type thermoelectric arms and N-type thermoelectric arms are cross-connected in series to form a matrix array structure. The columns of the matrix array structure are staggered end to end, and the metal wires alternately connect the P-type thermoelectric arms and the N-type thermoelectric arms at the head and tail.

[0052] The present invention also discloses a thermoelectric cooler collaborative simulation design system based on neural network acceleration, which is used to execute the above method and includes the following modules:

[0053] Model building module: build an equivalent physical field simulation model of a thermoelectric cooler;

[0054] Influencing parameter acquisition module: By simulating or experimentally measuring the equivalent physical field simulation model of the thermoelectric cooler, different structural parameters are collected, and the influencing parameter-performance parameter curve is drawn according to the length, width, height of the thermoelectric arm, and the corresponding relationship between the input current parameters and the performance of the thermoelectric cooler to find the parameters that affect the thermoelectric cooler;

[0055] Neural network model building and training module: build a neural network model, collect influencing parameter-performance parameter data sets, and conduct neural network training and testing to obtain a trained neural network model;

[0056] Optimal parameter acquisition module: Build a particle swarm optimization algorithm for optimization, continuously approach the target performance parameters by controlling the influencing parameters, and finally obtain the thermoelectric cooler structural parameters that meet the design requirements.

[0057] The beneficial effects of the present invention are:

[0058] 1) Compared with traditional commercial finite element simulation methods such as Ansys and COMSOL Multiphysics, the collaborative simulation design method of thermoelectric coolers proposed in this invention based on neural network acceleration greatly shortens the design cycle of thermoelectric coolers and improves design efficiency while ensuring simulation accuracy;

[0059] 2) The collaborative simulation design method of thermoelectric coolers based on neural network acceleration in the present invention is also combined with a meta-heuristic algorithm to achieve the optimal structural design of thermoelectric coolers that meet user performance requirements and realize intelligent design. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 3. The three views of the thermoelectric cooler, wherein (a) is the front view of the thermoelectric cooler, (b) is the left view of the thermoelectric cooler, and (c) is the top view of the thermoelectric cooler;

[0061] Figure 2 It is a schematic diagram of the dimensions of the thermoelectric arm in the thermoelectric cooler;

[0062] Figure 3 This is the working principle diagram of the thermoelectric cooler;

[0063] Figure 4 Design a flow chart for the co-simulation method of thermoelectric cooler based on neural network acceleration;

[0064] Figure 5 The graphs of the influencing parameters and performance parameters of the thermoelectric cooler are shown in Figure 1, where (a) is the ceramic thickness-temperature difference and power consumption curve, (b) is the conductor thickness-temperature difference and power consumption curve, (c) is the rubber thickness-temperature difference and power consumption curve, (d) is the cross-sectional area of ​​the thermoelectric arm-temperature difference and power consumption curve, (e) is the height of the thermoelectric arm-temperature difference and power consumption curve, and (f) is the input current-temperature difference and power consumption curve.

[0065] Figure 6 To build a neural network model diagram using the MATLAB platform as an example;

[0066] Figure 7 The figure is a statistical histogram of the prediction error of the performance parameters of the thermoelectric cooler based on the neural network (the error is calculated by the difference between the predicted value and the actual value, and the vertical axis is the number of data sets with errors within a certain range), where (a) is a statistical histogram of the cold surface temperature prediction error based on the neural network, (b) is a statistical histogram of the cold and hot surface temperature difference prediction error based on the neural network, (c) is a statistical histogram of the power consumption prediction error based on the neural network, and (d) is a statistical histogram of the cooling capacity prediction error based on the neural network;

[0067] Figure 8 The figure is a comparison diagram of the finite element simulation results of the equivalent physical field model of the thermoelectric cooler and the experimental measurement values, wherein (a) is a comparison diagram of the cold surface temperature value based on the finite element simulation and the experimental measurement value, (b) is a comparison diagram of the temperature difference between the hot and cold surfaces based on the finite element simulation and the experimental measurement value, and (c) is a comparison diagram of the power consumption value based on the finite element simulation and the experimental measurement value;

[0068] Fig. 9 is a flow chart of a particle swarm optimization algorithm of a preferred embodiment;

[0069] Fig.10 This is a block diagram of a thermoelectric cooler collaborative simulation design system based on neural network acceleration according to a preferred embodiment of the present invention.

[0070] In the figure: N-type thermoelectric arm 1, P-type thermoelectric arm 2, and metal wire 3. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical scheme and advantages of the present invention more clear, the present invention is further described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0072] The preferred embodiment of the present invention provides a collaborative simulation design method for thermoelectric coolers based on neural network acceleration, which is implemented on a thermoelectric cooler made of bismuth telluride. The method can achieve rapid evaluation and inverse design of thermoelectric cooler performance by introducing an artificial intelligence-based thermoelectric cooler influencing parameter-performance parameter prediction algorithm and a meta-heuristic intelligent algorithm represented by a particle swarm algorithm. Compared with traditional commercial finite element simulation methods, the present invention significantly shortens the design cycle of thermoelectric coolers while ensuring simulation accuracy, providing a fast and accurate design method.

[0073] This embodiment is a thermoelectric cooler collaborative simulation design method based on neural network acceleration. In order to effectively explain the collaborative simulation design method in detail, the thermoelectric cooler is first briefly introduced. It is particularly noted that the thermoelectric cooler described in this embodiment is only used to illustrate the proposed thermoelectric cooler collaborative simulation design method based on neural network acceleration, and is not the only applicable structure.

[0074] like Figure 1 FIG. 1 is a thermoelectric cooler for predictive design in a preferred embodiment of the present invention. The front view, left view and top view of the thermoelectric cooler are shown in FIG. Figure 1The thermoelectric cooler comprises cross-distributed P-type thermoelectric arms 2, N-type thermoelectric arms 1 and metal wires 3, wherein 127 P-type thermoelectric arms 2 and 127 N-type thermoelectric arms 1 are cross-connected in series to form a 16×16 matrix array structure, wherein the columns of the 16×16 matrix array structure are staggered head to tail, and the metal wires 3 are alternately connected head to tail of the P-type thermoelectric arms 2 and the N-type thermoelectric arms 1.

[0075] The P-type thermoelectric arm 2 and the N-type thermoelectric arm 1 in the thermoelectric cooler are of the same size; Figure 2 As shown, the length of each thermoelectric arm in the thermoelectric cooler is L, the width is W, and the height is H; the distance between adjacent thermoelectric arms in the same plane of the thermoelectric cooler is D.

[0076] For example, the length of each thermoelectric arm in the thermoelectric cooler is L=1.3mm, the width is W=1.36mm, the height is H=0.5mm, and the spacing between adjacent thermoelectric arms in the same plane in the x direction is D=1mm, and the spacing in the y direction is D=1mm.

[0077] A working principle of a thermoelectric cooler for predictive design: Figure 3 As shown, the current flows from bottom to top through the N-type thermoelectric arm 1, passes through the metal wire 3, and finally flows from top to bottom through the P-type thermoelectric arm 2, forming a phenomenon of absorbing heat below the thermocouple and releasing heat above the thermocouple, thereby achieving the purpose of cooling.

[0078] Example 1

[0079] Based on the above thermoelectric cooler, the preferred embodiment of the present invention proposes a thermoelectric cooler collaborative simulation design method based on neural network acceleration. Figure 4 As shown, the method specifically comprises the following steps:

[0080] (1) Use COMSOL Multiphysics to build an equivalent physical field simulation model of a thermoelectric cooler and conduct experiments to verify the accuracy of the model;

[0081] In this embodiment, Figure 8 The figure shows the comparison between the finite element simulation results and the experimental measurement values ​​of the equivalent physical field model of the thermoelectric cooler, where (a) is the comparison between the cold surface temperature value based on the finite element simulation and the experimental measurement value, (b) is the comparison between the hot and cold surface temperature difference value based on the finite element simulation and the experimental measurement value, and (c) is the comparison between the power consumption value based on the finite element simulation and the experimental measurement value. Figure 8 It can be proved that the equivalent physical field simulation model of the thermoelectric cooler built by COMSOL is consistent with the experimental structure. Therefore, the accuracy of the equivalent physical field simulation model of the thermoelectric cooler can be proved.

[0082] (2) By simulating or experimentally measuring the thermoelectric cooler model, different structural parameters are collected, and the main influencing parameters are found by drawing the influencing parameter-performance parameter curve according to the corresponding relationship between the current parameters and the thermoelectric cooler performance;

[0083] The step (2) specifically includes the following sub-steps:

[0084] (2.1) Using the principle of controlled variables, the effects of the ceramic thickness, conductor thickness, rubber thickness, logarithm of the thermoelectric arm, cross-sectional area of ​​the thermoelectric arm, height of the thermoelectric arm, and input current on the performance of the thermoelectric cooler are studied.

[0085] (2.2) Based on the collected data and the corresponding relationship, a curve of the influencing parameter and the performance parameter is drawn; in this embodiment, the curve of the influencing parameter and the performance parameter is as follows: Figure 5 As shown, (a) is the ceramic thickness-temperature difference, power consumption curve, (b) is the conductor thickness-temperature difference, power consumption curve, (c) is the rubber thickness-temperature difference, power consumption curve, (d) is the cross-sectional area of ​​the thermoelectric arm-temperature difference, power consumption curve, (e) is the height of the thermoelectric arm-temperature difference, power consumption curve, (f) is the input current size-temperature difference, power consumption curve.

[0086] (2.3) By observing the curve Figure 5 Based on the development trend of (a)-(f) in the figure, the main influencing parameters of the thermoelectric cooler are found.

[0087] In this embodiment, the main parameters affecting the performance of the thermoelectric cooler are the number of thermoelectric arms, the cross-sectional area of ​​the thermoelectric arms (the length of the thermoelectric arms×the width of the thermoelectric arms), the height of the thermoelectric arms, and the magnitude of the input current.

[0088] (3) Build a neural network model, then collect the influencing parameter-performance parameter data set and perform neural network training and testing to obtain a trained neural network model;

[0089] The step (3) specifically includes the following sub-steps:

[0090] (3.1) Taking MATLAB platform as an example, the neural network model is constructed as shown in the figure Figure 6 As shown in FIG. 1 , a neural network model is constructed. The neural network model has five input values, namely, the logarithm N of the thermoelectric arm of the thermoelectric cooler, the length L of the thermoelectric arm, the width W of the thermoelectric arm, the height H of the thermoelectric arm, and the input current I; three hidden layers are set, and the number of neurons in the hidden layers is 10, 10, and 10, respectively; there are four output values, namely, the cold surface temperature of the thermoelectric cooler , temperature difference between hot and cold surfaces DT, cooling capacity , the consumed electric energy P.

[0091] (3.2) The training data collection for the artificial neural network prediction model in the present invention is not limited to a fixed method, which may include experiments, simulations and mixed methods. In this embodiment, the performance of the thermoelectric cooler is transiently simulated by the finite element method, and 1250 sets of influencing parameter-performance parameter data are collected, wherein each set of influencing parameter-performance parameter data includes the number of logarithms N of the thermoelectric arm of the thermoelectric cooler, the length L of the thermoelectric arm, the width W of the thermoelectric arm, the height H of the thermoelectric arm, the input current I, the cold surface temperature , temperature difference between hot and cold surfaces DT, cooling capacity , the consumed electric energy P; divide the 1250 sets of data into training set and test set in the ratio of 5:2.

[0092] (3.3) Train the constructed neural network model through the training set and test it through the test set to obtain the trained neural network model.

[0093] The training convergence standard is not unique. In this embodiment, the maximum number of training times is used as the convergence standard. Then, the trained neural network is tested to determine the quality of the training effect according to the specific application problem. Figure 7 Statistical histogram of prediction error of thermoelectric cooler performance parameters by neural network (the error is calculated by the difference between the predicted value and the simulated value, and the vertical axis is the number of data sets with errors within a certain range), where (a) is the statistical histogram of cold surface temperature prediction error based on neural network, (b) is the statistical histogram of hot and cold surface temperature difference prediction error based on neural network, (c) is the statistical histogram of power consumption prediction error based on neural network, and (d) is the statistical histogram of cooling capacity prediction error based on neural network. It can be seen from (a) that the predicted temperature difference of the cold surface temperature of the thermoelectric cooler is basically maintained within 0.5°C. The predicted temperature difference of the cold surface temperature of the thermoelectric cooler is ; As can be seen from (b), the predicted temperature difference between the hot and cold surfaces of the thermoelectric cooler is basically kept within 0.5K. The predicted temperature difference of the cold surface temperature of the thermoelectric cooler is ; From (c), it can be seen that the predicted difference in the power consumption of the thermoelectric cooler is kept within 0.6W, and the predicted difference in the power consumption of the thermoelectric cooler is ; From (d), it can be seen that the predicted difference of the cooling capacity of the thermoelectric cooler is kept within 1.2W, and the predicted difference of the cooling capacity of the thermoelectric cooler is .

[0094] in, represents the cold surface temperature predicted by the neural network, represents the simulated cold surface temperature; represents the temperature difference between hot and cold surfaces predicted by the neural network, Indicates the simulated temperature difference between hot and cold surfaces; represents the power consumption predicted by the neural network, Indicates the power consumption of the simulation; represents the cooling capacity predicted by the neural network, Represents the simulated cooling capacity.

[0095] (4) Build a particle swarm optimization (PSO) algorithm for optimization. By controlling the main influencing parameters, the target performance parameters are continuously approached, and finally the optimal thermoelectric cooler structure that meets the design requirements is obtained.

[0096] like Fig. 9 As shown, step (4) specifically includes the following sub-steps:

[0097] Build a particle swarm optimization algorithm for target design:

[0098] The design goal is read from the input end and recorded as:

[0099] ;

[0100] in, represents the expected temperature difference, Indicates the expected power consumption, Indicates the expected cooling capacity;

[0101] In a D-dimensional target search space, an initial particle swarm consisting of M particles is created, where the velocity and position of the i-th particle at time t are expressed as:

[0102] ; ∈ ;

[0103] ; ∈ ;

[0104] in, and Represent the minimum and maximum speed respectively. and Represent the lower and upper limits of each particle’s search space, respectively. , ;

[0105] The particles are predicted by the neural network respectively, and the prediction result of the i-th particle at time t is expressed as:

[0106] ;

[0107] in, represents the temperature difference of the thermoelectric cooler predicted by the neural network, represents the power consumption of the thermoelectric cooler predicted by the neural network, represents the cooling capacity of the thermoelectric cooler predicted by the neural network, ;

[0108] The individual fitness of the i-th particle at time t is calculated as:

[0109] ;

[0110] The individual best fitness of the i-th particle at time t and the group best fitness of the g-th iteration are:

[0111] ;

[0112] ;

[0113] in, represents the expected weight of the target temperature difference, represents the expected weight of the target power consumption, represents the expected weight of the target cooling capacity, , ;

[0114] Update the individual optimal position of the i-th particle at time t and the group optimal position completed in the g-th iteration respectively:

[0115] ;

[0116] ;

[0117] in, for The corresponding particle position, , ;

[0118] The speed and position of the dth dimension of the i-th particle at time t+1 are updated as follows:

[0119] ;

[0120] ;

[0121] in, is the inertia weight, and is a random number between (0,1), and represents the learning factor, , , ;

[0122] The above process is repeated, and after G iterations, the thermoelectric cooler structure and current parameters that can meet the design objectives are finally determined.

[0123] Performance test experiment:

[0124] In order to verify the effectiveness of the present invention, commercial multi-physics field simulation software and commercial mathematical software are used here to compare the design speed of the present invention with the finite element simulation speed to verify the effect of the present invention in shortening the design cycle of the thermoelectric cooler.

[0125] The present invention is a novel design method applied to the field of thermoelectric cooler simulation design. Its specific object is various thermoelectric coolers, and it can realize the rapid design of thermoelectric coolers with existing performance targets.

[0126] Table 1 is a comparison table of the time consumed by the finite element simulation and the proxy model of the present invention. It can be seen from Table 1 that the present invention greatly shortens the design cycle of the thermoelectric cooler:

[0127] Table 1 Comparison of the time consumed by finite element simulation and the proxy model of the present invention

[0128]

[0129] As can be seen from the above, the performance test experimental results fully prove the effectiveness of the thermoelectric cooler collaborative simulation design method based on neural network acceleration proposed in the present invention. Since the thermoelectric cooler collaborative simulation design method based on neural network acceleration proposed in the present invention is only for illustrating the implementation of the simulation method, and is not limited to a thermoelectric cooler of a specific structure, its collaborative simulation design results are generally applicable to thermoelectric coolers.

[0130] Example 2

[0131] like Fig.10 As shown, this embodiment discloses a thermoelectric cooler collaborative simulation design system based on neural network acceleration, which is used to execute the method of the above embodiment, and includes the following modules:

[0132] Model building module: Use COMSOL Multiphysics to build an equivalent physical field simulation model of a thermoelectric cooler;

[0133] Influencing parameter acquisition module: By simulating or experimentally measuring the equivalent physical field simulation model of the thermoelectric cooler, different structural parameters are collected, and the influencing parameter-performance parameter curve is drawn according to the length, width, height of the thermoelectric arm, and the corresponding relationship between the input current parameters and the performance of the thermoelectric cooler to find the parameters that affect the thermoelectric cooler;

[0134] Neural network model building and training module: build a neural network model, collect influencing parameter-performance parameter data sets, and conduct neural network training and testing to obtain a trained neural network model;

[0135] Optimal parameter acquisition module: Build a particle swarm optimization algorithm for optimization, continuously approach the target performance parameters by controlling the influencing parameters, and finally obtain the thermoelectric cooler structural parameters that meet the design requirements.

[0136] For other contents of this embodiment, please refer to the above method embodiment.

[0137] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope of the present invention.

Claims

1. A collaborative simulation design method for thermoelectric coolers based on neural network acceleration, characterized by: The following steps are involved: S1, build an equivalent physical field simulation model of a thermoelectric cooler; S2, by simulating or experimentally measuring the equivalent physical field simulation model of the thermoelectric cooler, collecting structural parameters, and drawing an influencing parameter-performance parameter curve according to the length, width, height of the thermoelectric arm, and the corresponding relationship between the input current parameter and the performance of the thermoelectric cooler, to find the parameters that affect the thermoelectric cooler; S3, building a neural network model, collecting influencing parameter-performance parameter data sets and performing neural network training and testing to obtain a trained neural network model; S4, build a particle swarm optimization algorithm for optimization, and continuously approach the target performance parameters by controlling the influencing parameters, and finally obtain the thermoelectric cooler structural parameters that meet the design requirements; Step S2 specifically includes the following steps: S21, using the principle of controlled variables, respectively study the influence of the ceramic thickness, conductor thickness, rubber thickness, logarithm of the thermoelectric arm, cross-sectional area of ​​the thermoelectric arm, height of the thermoelectric arm, and input current on the performance of the thermoelectric cooler; S22, drawing an influencing parameter-performance parameter curve according to the collected different structural parameters and the corresponding relationship between the current parameter and the performance of the thermoelectric cooler; S23, according to the development trend of the curve, find the parameters that affect the thermoelectric cooler; Step S3 specifically includes the following steps: S31, building a neural network model, the neural network model has 5 input values, namely, the logarithm N of the thermoelectric arm of the thermoelectric cooler, the length L of the thermoelectric arm, the width W of the thermoelectric arm, the height H of the thermoelectric arm, and the input current I; 3 hidden layers are set, and the number of neurons in the hidden layers is 10, 10, and 10 respectively; there are 4 output values, namely, the cold surface temperature T c , temperature difference between hot and cold surfaces DT, cooling capacity Q c , the power consumed P; S32, collecting influencing parameter-performance parameter data, wherein each group of influencing parameter-performance parameter data includes the number of pairs of thermoelectric arms N, the length L of the thermoelectric arms, the width W of the thermoelectric arms, the height H of the thermoelectric arms, the input current I, the cold surface temperature T c , temperature difference between hot and cold surfaces DT, cooling capacity Q c , the consumed electric energy P; the collected data is divided into a training set and a test set in a ratio of 5:2; S33, training the constructed neural network model through the training set, and testing it through the test set to obtain a trained neural network model; In step S4, a particle swarm optimization algorithm is constructed for target design: The design goal is read from the input end and recorded as: goal=(goal DT ,goal P ,goal Qc ); Among them, goal DT Indicates the expected temperature difference, goal P Indicates the expected power consumption, goal Qc Indicates the expected cooling capacity; In a D-dimensional target search space, an initial particle swarm consisting of M particles is created, where the velocity and position of the i-th particle at time t are expressed as: Among them, v min and v max Represent the minimum and maximum values ​​of the speed, respectively, x min and x max Represent the lower and upper limits of each particle’s search space, 1≤i≤M, 1≤d≤D; The particles are predicted by the neural network respectively, and the prediction result of the i-th particle at time t is expressed as: in, represents the temperature difference of the thermoelectric cooler predicted by the neural network, represents the power consumption of the thermoelectric cooler predicted by the neural network, represents the cooling capacity of the thermoelectric cooler predicted by the neural network; The individual fitness of the i-th particle at time t is calculated as: The individual best fitness of the i-th particle at time t and the group best fitness of the g-th iteration are: Among them, ω DT represents the expected weight of the target temperature difference, ω P represents the expected weight of the target power consumption, ω Qc represents the expected weight of the target cooling capacity; Update the individual optimal position of the i-th particle at time t and the group optimal position completed in the g-th iteration respectively: in, for The corresponding particle position; The speed and position of the dth dimension of the i-th particle at time t+1 are updated as follows: Among them, ω is the inertia weight, r1 and r2 are random numbers between (0,1), c1 and c2 represent learning factors, 1≤g≤G, and G represents the maximum number of iterations; Determine whether the target parameters are met. If not, return the fitness value of the particle executing the calculation. If so, determine the thermoelectric cooler structure parameters and current parameters that meet the design goals.

2. The collaborative simulation design method for thermoelectric cooler based on neural network acceleration according to claim 1 is characterized in that: In step S1, the accuracy of the equivalent physical field simulation model of the thermoelectric cooler is verified through experiments.

3. The collaborative simulation design method for thermoelectric cooler based on neural network acceleration according to claim 1 is characterized in that: In step S32, 1250 sets of influencing parameter-performance parameter data are collected through experiments, simulations and hybrid methods.

4. The collaborative simulation design method for thermoelectric cooler based on neural network acceleration according to claim 1 is characterized in that: In step S33, the constructed neural network model is trained with the maximum number of training times as the convergence criterion.

5. The collaborative simulation design method for thermoelectric cooler based on neural network acceleration according to any one of claims 1 to 4, characterized in that: The thermoelectric cooler includes cross-distributed P-type thermoelectric arms, N-type thermoelectric arms and metal wires, which are cross-connected in series to form a matrix array structure. The columns of the matrix array structure are staggered end to end, and the metal wires alternately connect the P-type thermoelectric arms and the N-type thermoelectric arms from head to tail.

6. A thermoelectric cooler collaborative simulation design system based on neural network acceleration, used to execute the method according to any one of claims 1 to 5, characterized in that Includes the following modules: Model building module: build an equivalent physical field simulation model of a thermoelectric cooler; Influencing parameter acquisition module: By simulating or experimentally measuring the equivalent physical field simulation model of the thermoelectric cooler, different structural parameters are collected, and the influencing parameter-performance parameter curve is drawn according to the length, width, height of the thermoelectric arm, and the corresponding relationship between the input current parameters and the performance of the thermoelectric cooler to find the parameters that affect the thermoelectric cooler; Neural network model building and training module: build a neural network model, collect influencing parameter-performance parameter data sets, and conduct neural network training and testing to obtain a trained neural network model; Optimal parameter acquisition module: Build a particle swarm optimization algorithm for optimization, continuously approach the target performance parameters by controlling the influencing parameters, and finally obtain the thermoelectric cooler structural parameters that meet the design requirements.

Images