A multi-objective fuel cell cooling channel optimization design method

Abstract

The present invention relates to a multi-objective fuel cell cooling channel optimization design method, comprising: utilizing control equation coupling to construct a two-dimensional, two-phase, non-isothermal fuel cell mathematical model, which, after model validation, serves as a data-driven source; extracting cooling channel structural optimization parameters, optimization objectives, optimization parameter variation ranges, and constraints; substituting the varied optimization parameters into the mathematical model based on the optimization parameter variation ranges and constraints to output an original data set; applying a machine learning algorithm based on the original data set to construct a data-driven, multi-objective proxy model; and optimizing and solving the multi-objective proxy model using a genetic algorithm to obtain optimal cooling channel structural parameters. Compared with existing technologies, the present invention can comprehensively optimize cooling channel structural parameters from multiple angles, quickly, and accurately, providing guidance for practical fuel cell structural design.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fuel cells, in particular to a multi-objective fuel cell cooling channel optimization design method. BACKGROUND

[0002] Hydrogen-oxygen proton exchange membrane fuel cells have high energy density, no pollution, no catalyst poisoning and other characteristics, and have been the preferred direction for fuel cell industrialization. However, the high heat load brought by high energy density has become an important factor hindering its development, mainly because a large amount of reaction heat will affect the temperature distribution of the cell, causing the entire or local cell temperature to be too high. As a low-temperature fuel cell, the typical working temperature of the proton exchange membrane fuel cell needs to be controlled at 60-85℃, and the temperature difference with the environment is very small. It is difficult to maintain the working temperature by only the cell itself.

[0003] To deal with the above-mentioned cell heating problem, the current mainstream method is to use a bipolar plate containing a cooling water channel to cool the fuel cell using cooling water. A good cooling water channel structure design and arrangement can greatly help the fuel cell thermal management, which can control the cell temperature, improve the internal temperature uniformity, and further improve the cell performance.

[0004] In the related optimization work of proton exchange membrane fuel cells, the traditional method is to directly compare multiple experiments to design an optimized cooling channel structure, but this will inevitably consume a lot of manpower and resources. Therefore, simulation modeling has become an important means of current design optimization. Using simulation modeling can calculate the internal temperature distribution and material transport of the cell that is difficult to access in experiments, providing valuable information for solving practical problems and greatly reducing the cost of cell optimization design. However, the traditional fuel cell model optimization often uses the control variable method, and the optimization target is single. Due to the limitations of the control variable method, the application scenarios are severely limited, and the advantages of modeling simulation cannot be fully utilized. Simply using an optimization algorithm combined with a mathematical model can solve this problem to some extent, but the optimization process requires thousands of calculations on the mathematical model, which will greatly increase the time and labor costs. SUMMARY

[0005] The purpose of the present application is to overcome the defects of the prior art and provide a multi-objective fuel cell cooling channel optimization design method. The channel parameters are optimized from multiple angles to quickly and accurately optimize the cooling channel structure of the fuel cell.

[0006] The purpose of the present application can be achieved by the following technical solution: a multi-objective fuel cell cooling channel optimization design method, comprising the following steps:

[0007] S1, build a mathematical model: a two-dimensional two-phase non-isothermal fuel cell model is built by coupling control equations, and the model is discretized and calculated by using the finite element analysis method through grid division;

[0008] S2, build a multi-objective agent model: determine the cooling channel structure optimization parameters, optimization objectives, optimization parameter variation range and constraint conditions;

[0009] According to the optimization parameter variation range and the constraint condition, the changed optimization parameters are substituted into the mathematical model to obtain the original data set;

[0010] Based on the original data set, a multi-objective agent model is constructed by machine learning training;

[0011] S3, genetic algorithm optimization: genetic algorithm is used to optimize and solve the multi-objective agent model to obtain the optimal cooling channel structure parameters.

[0012] Further, the control equation in step S1 includes mass conservation equation, momentum conservation equation, energy conservation equation, material conservation equation, charge conservation equation, concentration-dependent Butler-Volmer equation and oxygen transport process equation, wherein the mass conservation equation includes gas phase and liquid phase mass conservation equation, and the momentum conservation equation includes gas phase and liquid phase momentum conservation equation.

[0013] Further, the step S2 specifically includes the following steps:

[0014] S21, according to the structure of the cooling channel, the following structure parameters are selected as optimization parameters:

[0015] X=(Bias,W CH,w ,H CH,w ,H BP,a ,H BP,c ) T

[0016] Wherein, X is the cooling channel structure parameter set, Bias is the fuel channel-cooling water channel center axis deviation, W CH,w is the cooling water channel width, H CH,w is the half cooling water channel height, H BP,a is the anode fuel-cooling water plate thickness, H BP,c is the cathode fuel-cooling water plate thickness;

[0017] The optimization objective is determined as the rated power density of the battery and the cross-section temperature difference;

[0018] The optimization parameter variation range is determined as:

[0019] {Bias|(0,0.6)}

[0020] {W CH,w |(0.4,1)}

[0021] {H CH,w |(0.1,0.5)}

[0022] {H BP,a |(0.2,1)}

[0023] {H BP,c |(0.2,1)}

[0024] {H total |(1.2,1.6)}

[0025] Meanwhile, the thickness of the bipolar plate needs to meet the following constraint condition:

[0026] H total = H BP,a + H BP,c + 2H CH,w

[0027] {H total |(1.2,1.6)}

[0028] wherein H total is the thickness of the bipolar plate;

[0029] S22, a plurality of sets of optimization parameters are randomly generated according to the variation range of each optimization parameter as input data sets, the input data sets are substituted into the mathematical model to obtain output data sets, and the original data sets are obtained by corresponding combination of the input data sets and the output data sets;

[0030] S23, a multi-objective proxy model is constructed based on the original data sets through machine learning training.

[0031] Further, the step S23 specifically comprises the following steps:

[0032] S231, the input and output of the original data sets are normalized and preprocessed, and the parameter normalization mapping rule is recorded, and the preprocessed original data sets are divided into a training set and a test set according to a set proportion;

[0033] S232, support vector machines are used as a machine learning training algorithm, the training set is trained by using the support vector machines, the proxy model accuracy is verified by using the test set, a multi-objective proxy model about different optimization targets is obtained, and the multi-objective proxy model includes a battery rated point power density proxy model and a cross-section temperature difference proxy model.

[0034] Further, the configuration of the support vector machine is specifically: selecting a radial basis function as a kernel function, and obtaining a model-determining highest regularization parameter c and a Gaussian kernel width g through a grid search method.

[0035] Further, the multi-objective agent model is specifically:

[0036]

[0037]

[0038]

[0039] wherein, is a battery rated power density agent model, is a cross-sectional temperature difference agent model.

[0040] Further, the step S3 specifically includes the following steps:

[0041] S31, configuring a genetic algorithm and initializing a population: configuring a population number, an elite number, a maximum number of generations and a stagnation number, and using the same optimization parameter variation range and constraint conditions as in the original data set to generate an initialized population;

[0042] S32, normalizing and preprocessing the initialized population, and then encoding the normalized population;

[0043] S33, substituting the population individuals into the multi-objective agent model to obtain individual fitness through calculation;

[0044] S34, performing genetic operation according to the individual fitness, calculating a new population, and using a set stop condition to judge convergence to obtain an optimal parameter set;

[0045] S35, decoding and anti-normalizing the optimal parameter set to obtain optimal cooling flow passage structure parameters.

[0046] Further, the step S33 specifically includes the following steps:

[0047] S331, substituting the population individuals into the multi-objective agent model to obtain corresponding output results;

[0048] S332, performing weighted summation calculation on the output results obtained in step S331 to obtain individual fitness:

[0049] Fit=W T Y

[0050] W T =(w1,w2)

[0051] wherein Fit is the individual fitness, W is the parameter weight set, Y is the output summary data set of the multi-objective surrogate model, w1 is the weight value corresponding to the battery rated point power density surrogate model in the multi-objective surrogate model, and w2 is the weight value corresponding to the cross-section temperature difference surrogate model in the multi-objective surrogate model. T wherein Fit is the individual fitness, W is the parameter weight set, Y is the output summary data set of the multi-objective surrogate model, w1 is the weight value corresponding to the battery rated point power density surrogate model in the multi-objective surrogate model, and w2 is the weight value corresponding to the cross-section temperature difference surrogate model in the multi-objective surrogate model.

[0052] Further, the genetic operation in the step S34 includes a selection function, a crossover function and a mutation function, wherein the selection function adopts a random uniform distribution selection method, the crossover function adopts a discrete recombination method, and the mutation function adopts a Gaussian mutation method.

[0053] Further, the stop condition set in the step S34 is specifically a plurality of stop conditions, and the plurality of stop conditions include a first stop condition and a second stop condition, the first stop condition is that the population generation number reaches a maximum generation number, and the second stop condition is that, within a stagnation generation number, the change of the weighted average value of the population fitness is less than 10 -6 ;

[0054] When the first stop condition or the second stop condition is met, it is judged that convergence is reached.

[0055] Compared with the prior art, the present application aims at the optimization design of the fuel cell cooling water flow channel structure, builds a mathematical model as a data set source, extracts required optimization parameters to generate an original data set, adopts a multi-objective surrogate model optimization mode, substitutes the data set into machine learning to generate a surrogate model, calculates the optimal parameters based on the surrogate model using a genetic algorithm, and realizes the optimization of the cooling water flow channel structure.

[0056] The present application uses a mathematical model as an original data source, greatly reduces the battery optimization design cost, provides information such as the internal temperature distribution of the battery which is difficult to obtain in experiments, and provides the possibility of generating a multi-objective surrogate model; the multi-objective surrogate model optimization mode is adopted, the original data is trained using machine learning to obtain a multi-objective surrogate model instead of a mathematical model for optimization, which can greatly reduce the optimization time; compared with traditional modes such as the control variable method, the multi-objective surrogate model optimization mode can effectively shorten the optimization time; and the fuel cell can be comprehensively optimized from multiple angles, and more scientific and reasonable results can be obtained compared with single-objective optimization; the genetic algorithm is used for optimization calculation, which is more rapid and accurate than traditional qualitative optimization methods, has a wider parameter coverage, and can obtain specific optimal solutions that are difficult to obtain by traditional methods.

[0057] Therefore, the method of the present application can comprehensively optimize the flow channel parameters from multiple angles, quickly and accurately, and provide guidance for actual fuel cell structure design. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1A process flow diagram of the method of the present application;

[0059] Figure 2 A process diagram of the method of the present application applied in the example;

[0060] Figure 3 A diagram of the calculation domain of the two-dimensional two-phase non-isothermal fuel cell model in the example;

[0061] Figure 4 A diagram of the selected optimization parameters in the example;

[0062] Figure 5 A comparison diagram of the mathematical model and the actual stack polarization curve in the example;

[0063] Figure 6a A comparison diagram of the rated power density of the training set proxy model and the mathematical model in the example;

[0064] Figure 6b A comparison diagram of the rated power density of the test set proxy model and the mathematical model in the example;

[0065] Figure 6c A comparison diagram of the maximum temperature difference of the training set proxy model and the mathematical model in the example;

[0066] Figure 6d A comparison diagram of the maximum temperature difference of the test set proxy model and the mathematical model in the example;

[0067] Marked in the figure: 1, anode gas diffusion layer, 2, anode microporous layer, 3, anode catalyst layer, 4, proton exchange membrane, 5, cathode catalyst layer, 6, cathode microporous layer, 7, cathode gas diffusion layer, 8, anode cooling water flow channel, 9, anode polar plate, 10, anode fuel flow channel, 11, cathode fuel flow channel, 12, cathode polar plate, 13, cathode cooling water flow channel. DETAILED DESCRIPTION

[0068] The present application will be described in detail below in conjunction with the accompanying drawings and specific examples.

[0069] Example

[0070] As Figure 1 shown, a multi-objective fuel cell cooling channel optimization design method includes the following steps:

[0071] S1, build a mathematical model: a two-dimensional two-phase non-isothermal fuel cell model is built by coupling control equations, discrete calculation is performed by using finite element analysis method through grid division, and the mathematical model is built, wherein the control equations include mass conservation equation (including gas phase and liquid phase mass conservation equation), momentum conservation equation (gas phase and liquid phase momentum conservation equation), energy conservation equation, material conservation equation, charge conservation equation, concentration-dependent Butler-Volmer equation and oxygen transport process equation;

[0072] S2, build a multi-objective agent model: determine the cooling flow channel structure optimization parameters, optimization targets, optimization parameter variation range and constraint conditions;

[0073] According to the optimization parameter variation range and the constraint conditions, the changed optimization parameters are substituted into the mathematical model to output the original data set;

[0074] Based on the original data set, a multi-objective agent model is built by machine learning training;

[0075] Specifically:

[0076] First, according to the structure of the cooling flow channel, the following structure parameters are selected as optimization parameters:

[0077] X=(Bias, W CH,w , H CH,w , H BP,a , H BP,c ) T

[0078] Wherein, X is the cooling flow channel structure parameter set, Bias is the fuel flow channel-cooling water flow channel center axis deviation, W CH,w is the cooling water flow channel width, H CH,w is the half cooling water flow channel height, H BP,a is the anode fuel-cooling water polar plate thickness, H BP,c is the cathode fuel-cooling water polar plate thickness;

[0079] The optimization target is determined to be the rated power density of the battery and the cross-section temperature difference;

[0080] The optimization parameter variation range is determined to be:

[0081] {Bias|(0,0.6)}

[0082] {W CH,w |(0.4,1)}

[0083] {H CH,w |(0.1,0.5)}

[0084] {H BP,a |(0.2,1)}

[0085] {H BP,c |(0.2,1)}

[0086] {H total |(1.2,1.6)}

[0087] At the same time, the thickness of the bipolar plate needs to meet the following constraint conditions:

[0088] H total =H BP,a +H BP,c +2H CH,w

[0089] {H total |(1.2,1.6)}

[0090] wherein H total is the thickness of the bipolar plate;

[0091] Then, according to the variation range of each optimization parameter, a plurality of sets of optimization parameters are randomly generated as an input data set, the input data set is substituted into the mathematical model to obtain an output data set, and the input data set and the output data set are correspondingly combined to obtain an original data set;

[0092] Based on the original data set, the input and output of the original data set are normalized and pretreated, and the normalization mapping rules of each parameter are recorded, and the pretreated original data set is divided into a training set and a test set according to a set proportion;

[0093] Support vector machine (selecting radial basis function as kernel function and obtaining the highest regularization parameter c and Gaussian kernel width g of model determination coefficient through grid search method) is used as a machine learning training algorithm, the support vector machine is used to train the proxy model of the training set, and the test set is used to verify the accuracy of the proxy model, and a multi-objective proxy model about different optimization objectives is obtained, the multi-objective proxy model includes a battery rated point power density proxy model and a cross-section temperature difference proxy model:

[0094]

[0095]

[0096]

[0097] wherein, is the battery rated power density proxy model, is the cross-section temperature difference proxy model;

[0098] S3, genetic algorithm optimization: genetic algorithm is used to optimize and solve the multi-objective proxy model to obtain the optimal cooling flow channel structure parameter, and specifically:

[0099] S31, configure a genetic algorithm and initialize a population: configure the population size, elite number, maximum number of generations, and stasis number, and use the same optimization parameter variation range and constraint conditions as in the original data set to generate an initialized population;

[0100] S32, normalize the initialized population for preprocessing, and then encode the normalized population;

[0101] S33, substitute the population individuals into the multi-objective agent model to obtain the corresponding output results;

[0102] The output results of the multi-objective agent model are further weighted and summed to obtain the individual fitness:

[0103] Fit=W T Y

[0104] W T =(w1,w2)

[0105] Wherein, Fit is the individual fitness, W T is the parameter weight set, Y is the output summary data set of the multi-objective agent model, w1 is the weight value corresponding to the battery rated point power density agent model in the multi-objective agent model, and w2 is the weight value corresponding to the cross-section temperature difference agent model in the multi-objective agent model. The individual fitness is calculated;

[0106] S34, genetic operation is performed according to the individual fitness (including selection function, crossover function and mutation function, wherein the selection function adopts random uniform distribution selection method, the crossover function adopts discrete recombination method, and the mutation function adopts Gaussian mutation method), a new population is calculated, and convergence judgment is performed using a set of stop conditions (the stop conditions include a first stop condition and a second stop condition, the first stop condition is that the population generation reaches the maximum generation, and the second stop condition is that the change of the weighted average value of the population fitness is less than 10 -6 % within the stasis number); When the first stop condition or the second stop condition is met, it is judged that the convergence is reached, and the optimal parameter set is obtained;

[0107] S35, decoding and reverse normalization of the optimal parameter set are performed to obtain the optimal cooling flow channel structure parameters.

[0108] As can be seen from the above, the present application is aimed at optimizing the design of the fuel cell cooling water flow channel, builds a mathematical model as a data set source, extracts the required optimization parameters to generate an original data set, uses a multi-objective agent model optimization mode, substitutes the data set into machine learning to generate an agent model, and calculates the optimal parameters based on the agent model using a genetic algorithm to realize the optimization of the cooling water flow channel structure.

[0109] The above method is applied in this embodiment, and the main process is as followsFigure 2 The method comprises building a mathematical model, generating an agent model, genetic algorithm optimization, and reliability verification of the optimization results.

[0110] 1) Building a mathematical model: according to the optimization requirements of the model, a two-dimensional two-phase non-isothermal fuel cell model is built under the condition of considering the accuracy and computational economy of the model. The specific calculation domain of the model is shown in Figure 3 , including an anode gas diffusion layer 1, an anode microporous layer 2, an anode catalyst layer 3, a proton exchange membrane 4, a cathode catalyst layer 5, a cathode microporous layer 6, a cathode gas diffusion layer 7, an anode cooling water flow channel 8, an anode polar plate 9, an anode fuel flow channel 10, a cathode fuel flow channel 11, a cathode polar plate 12, and a cathode cooling water flow channel 13. The basic control equations of the model are as follows:

[0111] Gas phase and liquid phase mass conservation equations:

[0112] ▽·(ρ g u g )=m g

[0113] ▽·(ρ l u l )=m l

[0114] Wherein, ρ g , ρ l are the gas phase and liquid phase density, u g , u l are the gas phase and liquid phase velocity, m g , m l are the gas phase and liquid phase mass source term;

[0115] Gas phase and liquid phase momentum conservation equations:

[0116]

[0117]

[0118] Wherein, K is the absolute permeability in porous media, k r,g , k r,l are the gas phase and liquid phase relative permeability, p g , p l are the gas phase and liquid phase pressure;

[0119] The capillary pressure p c is used to couple the gas-liquid two-phase motion:

[0120]

[0121] Wherein, σ is the surface tension of liquid water, θ cis the liquid water contact angle, ε is the porosity, K is the absolute permeability, and s is the liquid water volume fraction;

[0122] Energy conservation equation:

[0123]

[0124] where ε is the porosity, C p is the component specific heat capacity, T is the component temperature, λ i is the component thermal conductivity, S T is the temperature source term;

[0125] Mass conservation equation:

[0126]

[0127] where C i is the component concentration, is the component effective diffusivity, is the component thermal diffusivity, S i is the component molar production rate;

[0128] Charge conservation equation:

[0129]

[0130]

[0131] where, are the effective protonic and electronic conductivities, φ l and φ s are the electrolyte and electrode potentials, respectively; Q l and Q s are the protonic and electronic production rates, respectively;

[0132] The hydrogen oxidation reaction and oxygen reduction reaction occurring at the anode and cathode, respectively, are described using the concentration-dependent Butler-Volmer equation:

[0133]

[0134]

[0135] where j 0,a and j 0,c are the anode and cathode exchange current densities, respectively, are the hydrogen and oxygen reference concentrations, and α a and α c are the anode and cathode reaction charge transfer coefficients, respectively, F is the Faraday constant, and R is the gas constant;

[0136] Aa 、A c are the active specific surface areas of the anode catalyst layer 3 and the cathode catalyst layer 5, respectively, a ,η c is the anode and cathode overpotential, which can be calculated by the following formula:

[0137]

[0138]

[0139] η=φ s -φ l -E eq

[0140] Among them, E a 、E c are the electrochemical surface areas of the anode catalyst layer 3 and the cathode catalyst layer 5, m pt,a 、m pt,c are the platinum loadings of the anode catalyst layer 3 and the cathode catalyst layer 5, respectively. ACL 、H CCL are the thicknesses of the anode catalyst layer 3 and the cathode catalyst layer 5 respectively;

[0141] is the concentration of hydrogen and oxygen at the activation site, The one-dimensional oxygen transport process equation is used for calculation:

[0142]

[0143]

[0144] Among them, δ w , δ n are the thickness of water film and ionomer film on the surface of catalyst agglomerates, are the diffusion coefficients of oxygen in liquid water and ionomer respectively, C1 is the equilibrium concentration of oxygen on the water film surface, is the oxygen concentration in the cathode catalyst layer 5, H n is the Henry coefficient of the ionomer surface, k n 、k pt are the adsorption coefficients of oxygen on the ionomer and platinum surfaces, respectively;

[0145] The net flux of liquid water transmission between the anode and cathode on both sides of the proton exchange membrane 4 is:

[0146]

[0147] Among them, n d is the electrical permeability coefficient, I is the reaction current density, p l,c-a is the pressure difference between the anode and cathode liquid water on both sides of the proton exchange membrane 4, Molar mass of liquid water, μ l Dynamic viscosity of liquid water, H MEM Thickness of the proton exchange membrane 4;

[0148] Based on the above mathematical equation coupling, the grid is divided, and the finite element analysis method is used for discrete calculation, and the mathematical model is built. As shown in Figure 5 The actual fuel cell stack test polarization curve is used to verify the effectiveness of the mathematical model, and the curve fitting determines the coefficient R 2 = 0.9893, which proves that the model is effective, and the verified mathematical model is used as a data set source;

[0149] 2) Generate a proxy model:

[0150] 2.1) Determine the optimization parameters and optimization targets, determine the optimization parameter variation range and constraint conditions: refer to Figure 4 , select five cooling water flow channel structure parameters as shown in the figure as optimization parameters, respectively: fuel flow channel-cooling water flow channel center axis deviation Bias, cooling water flow channel width W CH,w , half cooling water flow channel height H CH,w , anode fuel-cooling water polar plate thickness H BP,a , cathode fuel-cooling water polar plate thickness H BP,c , which is summarized as: X=(Bias,W CH,w ,H CH,w ,H BP,a ,H BP,c ) T In this embodiment, the optimization parameter value of the original version is: X=(0, 0.8, 0.6, 0.4, 0.45) T ;

[0151] Select the battery rated point power density P rated and the cross-sectional temperature difference ΔT rated as the optimization target, to comprehensively optimize from the battery performance and temperature uniformity, in order to ensure the rationality of the battery structure and the battery volume power density without excessive change, the optimization parameter, the bipolar plate thickness H total =H BP,a +H BP,c +2H CH,w The variation range is within the specified range, and the optimization parameter value range is respectively: {Bias|(0, 0.6)}mm, {W CH,w |(0.4, 1)}mm, {H CH,w |(0.1, 0.5)}mm, {H BP,a |(0.2, 1)}mm, {H BP,c |(0.2, 1)}mm, {H total |(1.2, 1.6)}.

[0152] 2.2) Change optimization parameters into mathematical model to generate original data set: Based on the completed and verified mathematical model, a script is written to randomly generate 150 groups of optimization parameters within the specified range as input data set, which is substituted into the mathematical model to guide the calculation and solve, and the required output data set P is recorded rated , ΔT rated , to obtain the original data set;

[0153] 2.3) Normalization preprocessing and data set division of original data set: In order to improve the model training accuracy and eliminate the influence caused by different data magnitudes, the input and output of the original data set are normalized and the normalization mapping rules of various parameters are recorded, and the preprocessed data set is divided into training set and test set according to the proportion of 0.75:0.25;

[0154] 2.4) Substitute machine learning training to generate multi-objective agent model: Use machine learning algorithm support vector machine (Support Vector Machine, SVM) to train data set to generate agent model, write script to configure algorithm and train model, wherein the specific configuration of support vector machine is: select radial basis function (Radial Basis Function, RBF) as kernel function, and obtain the highest regularization parameter c and Gaussian kernel width g of the model through simple grid search method;

[0155] Use the training set to train the model, and use the test set to verify the accuracy of the model, to obtain the agent model about battery performance P rated , cross-sectional temperature difference ΔT rated :

[0156]

[0157]

[0158] Among them, are normalized results, which can be summarized as

[0159] As shown in Figures 6a to 6d , the training accuracy of the trained agent model is:

[0160] determination coefficient

[0161]

[0162] The training accuracy meets the training requirements, and the agent model is used for subsequent optimization;

[0163] 3) Genetic algorithm optimization:

[0164] 3.1) Initialization of population and configuration of genetic algorithm: after the multi-objective agent model is established, the genetic algorithm is further used for model optimization. First, the genetic algorithm is configured. In this embodiment, the population size is 60, the elite number is 2, the maximum number of generations is 300, the stall generation is 50, and the same parameter range and limit condition as in the original data set is used to generate the initial population;

[0165] 3.2) Population data normalization and encoding: the initial population is normalized and preprocessed by using the parameter normalization mapping rule of the original data, and the normalized population is encoded for subsequent genetic operation;

[0166] 3.3) Application of multi-objective agent model to calculate fitness: the population individuals are substituted into the multi-objective agent model to obtain the corresponding output results. Since the output results of the multi-objective agent model are normalized results, there is no magnitude deviation problem, and the weighted sum of the output results of each agent model can be directly determined as the individual fitness Fit = W T Y, where Y is the output summary data set, W T is the parameter weight set, W T = (w1, w2). In this embodiment, the purpose is to obtain lower battery temperature difference, and at the same time, better battery performance is also expected. The specific calculation method is

[0167] 3.4) Genetic operation to determine new population. After the individual fitness is calculated, genetic operation is performed according to the fitness. The genetic operation includes selection function, crossover function and mutation function. In this embodiment, the selection function adopts stochastic uniform distribution selection method (Stochastic uniform), the crossover function adopts scattered recombination method (Scattered), the crossover probability is 0.8, the mutation function adopts Gaussian mutation (Gaussian), and the mutation probability is 0.2. Through genetic operation, the statistical results obtain a new population;

[0168] 3.5) Application of stopping condition to determine whether to converge: in order to ensure the stability of the algorithm, multiple stopping conditions are set for stopping judgment. When any of the following conditions is met, the loop is exited: the population generation reaches the maximum generation or the weighted average value of the population fitness changes less than 10 -6 , the above stopping conditions are met, the loop is exited, and the parameter set optimization is completed;

[0169] 3.6) Parameter set decoding and inverse normalization to get the optimal solution: parameter decoding is performed on the optimization result, inverse normalization is performed according to the parameter normalization mapping rule, and the optimization result is converted into the actual product required parameters, and in this embodiment, the parameter values obtained by optimization are: X opti =(0.6, 0.65, 0.8, 0.2, 0.2);

[0170] 4) Reliability verification of the optimization result: the parameter result obtained by optimization is substituted into the mathematical model to calculate the rated point power density P rated and the cross-sectional temperature difference ΔT rated , compared with the original version of the results, the reliability of the optimization result is verified, and the optimization design is completed.

[0171] By using the method provided by the application, the structure parameters of the cooling flow channel can be comprehensively optimized in multiple angles, quickly and accurately, and the method can provide guidance for actual fuel cell structure design; compared with traditional modes such as the control variable method, the multi-objective agent model optimization mode can greatly shorten the optimization time.

Claims

1. A multi-objective fuel cell cooling channel optimization design method, characterized in that: The following steps are involved: S1. Build a mathematical model: Build a two-dimensional two-phase non-isothermal fuel cell model using the control equation coupling. Complete the mathematical model building by meshing and performing discrete calculations using the finite element analysis method. S2. Construct a multi-objective agent model: determine the cooling channel structure optimization parameters, optimization objectives, optimization parameter variation range and constraint conditions; According to the optimization parameter variation range and constraint conditions, the changed optimization parameters are substituted into the mathematical model to output the original data set; Based on the original data set, a multi-objective proxy model is constructed through machine learning training; S3. Genetic algorithm optimization: Genetic algorithm is used to optimize and solve the multi-objective agent model to obtain the optimal cooling channel structure parameters; Step S2 specifically includes the following steps: S21. According to the structure of the cooling channel, the following structural parameters are selected as optimization parameters: X=(Bias,W CH,w ,H CH,w ,H BP,a ,H BP,c ) T Among them, X is the cooling channel structure parameter set, Bias is the central axis deviation between the fuel channel and the cooling water channel, and W CH,w is the width of the cooling water channel, H CH,w is the height of the semi-cooling water channel, H BP,a is the thickness of the anode fuel-cooling water plate, H BP,c is the thickness of the cathode fuel-cooling water plate; The optimization targets are determined to be the rated power density of the battery and the cross-sectional temperature difference; The specific range of optimization parameter variation is determined as follows: {Bias|(0,0.6)} {W CH,w |(0.4,1)} {H CH,w |(0.1,0.5)} {H BP,a |(0.2,1)} {H BP,c |(0.2,1)} {H total |(1.2,1.6)} At the same time, the thickness of the bipolar plate must meet the following constraints: H total =H BP,a +H BP,c +2H CH,w {H total |(1.2,1.6)} Among them, H total is the bipolar plate thickness; S22. Randomly generate multiple sets of optimization parameters according to the variation range of each optimization parameter, use them as input data sets, substitute the input data sets into the mathematical model to obtain output data sets, and combine the input data sets and the output data sets to obtain the original data sets; S23. Based on the original data set, a multi-objective agent model is constructed through machine learning training.

2. A multi-objective fuel cell cooling channel optimization design method according to claim 1, characterized in that: The control equations in step S1 include the mass conservation equation, the momentum conservation equation, the energy conservation equation, the material conservation equation, the charge conservation equation, the concentration-dependent Butler-Volmer equation, and the oxygen transfer process equation, wherein the mass conservation equation includes the gas phase and liquid phase mass conservation equations, and the momentum conservation equation includes the gas phase and liquid phase momentum conservation equations.

3. The multi-objective fuel cell cooling channel optimization design method according to claim 1, characterized in that: The step S23 specifically includes the following steps: S231, performing normalization preprocessing on the input and output of the original data set, and recording the normalization mapping rules of each parameter, and dividing the preprocessed original data set into a training set and a test set according to a set ratio; S232. Using a support vector machine as a machine learning training algorithm, the support vector machine is used to train a proxy model on the training set, and the test set is used to verify the accuracy of the proxy model to obtain a multi-objective proxy model for different optimization objectives, wherein the multi-objective proxy model includes a battery rated point power density proxy model and a cross-sectional temperature difference proxy model.

4. A multi-objective fuel cell cooling channel optimization design method according to claim 3, characterized in that: The configuration of the support vector machine is specifically as follows: a radial basis function is selected as a kernel function, and a regularization parameter c and a Gaussian kernel width g with the highest model determination coefficient are obtained through a grid search method.

5. The multi-objective fuel cell cooling channel optimization design method according to claim 3, characterized in that: The multi-objective agent model is specifically: in, is the battery rated power density proxy model, It is the cross-section temperature difference proxy model.

6. A multi-objective fuel cell cooling channel optimization design method according to claim 5, characterized in that: The step S3 specifically includes the following steps: S31. Configure the genetic algorithm and initialize the population: configure the population size, the number of elites, the maximum number of generations, and the number of stagnant generations, and use the same optimization parameter variation range and constraints as in the original data set to generate the initialized population; S32, performing normalization preprocessing on the initialized population, and then encoding the normalized population; S33, substituting the individuals in the population into the multi-objective agent model, and obtaining the individual fitness by calculation; S34, performing genetic operations based on individual fitness, calculating a new population, and using the set cutoff conditions to perform convergence judgment to obtain the optimal parameter set; S35. Decode and denormalize the optimal parameter set to obtain optimal cooling channel structural parameters.

7. The multi-objective fuel cell cooling channel optimization design method according to claim 6, characterized in that: The step S33 specifically includes the following steps: S331, substituting the population individuals into the multi-objective agent model to obtain the corresponding output results; S332: Perform weighted sum calculation on the output results obtained in step S331 to obtain the individual fitness: Fit=W T Y’ IN T =(w1,w2) Among them, Fit is the individual fitness, W T is the parameter weight set, Y' is the output summary data set of the multi-objective surrogate model, w1 is the weight value corresponding to the battery rated point power density surrogate model in the multi-objective surrogate model, and w2 is the weight value corresponding to the cross-sectional temperature difference surrogate model in the multi-objective surrogate model.

8. The multi-objective fuel cell cooling channel optimization design method according to claim 6, characterized in that: The genetic operation in step S34 includes a selection function, a crossover function and a mutation function, wherein the selection function adopts a random uniform distribution selection method, the crossover function adopts a discrete recombination method, and the mutation function adopts a Gaussian mutation method.

9. The multi-objective fuel cell cooling channel optimization design method according to claim 6, characterized in that: The cut-off condition set in step S34 is specifically a multiple cut-off condition, which includes a first cut-off condition and a second cut-off condition. The first cut-off condition is that the population generation reaches the maximum generation, and the second cut-off condition is that the change in the weighted average of the population fitness is less than 10 within the stagnant generation. -6 ; When the first cutoff condition or the second cutoff condition is met, it is determined that convergence is achieved.

Images