Active learning method for energy equipment packaging stress uniformity optimization
By adopting active learning methods under the Bayesian optimization framework, the shape and size of enhanced partitions of energy equipment are optimized, and the problem of optimizing stress uniformity of energy equipment packaging is solved, achieving efficient and safe design and operation.
Patent Information
- Application Number
- CN202411654207.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-19
- Publication Date
- 2025-04-04
AI Technical Summary
The prior art is difficult to effectively optimize packaging stress uniformity in energy equipment design, resulting in high calculation costs, long design time, and difficult to meet the constraints of user needs.
The active learning method is adopted under the Bayesian optimization framework, and the shape and size of the partition are optimized and enhanced through parameterized modeling and finite element simulation to reduce the deformation non-uniformity of the top electrode plate and realize the on-demand design of energy equipment.
Through active learning methods and Bayesian optimization algorithms, the dependence on finite element calculations is significantly reduced, design efficiency is improved, the efficient and safe operation of energy equipment is achieved, and the constraints of user needs are met.
Smart Images
Figure CN119761093A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of optimization design, and in particular to an active learning method for optimizing the uniformity of energy equipment packaging stress. Background Art
[0002] Energy equipment is rooted in the continued growth of global energy demand and the pursuit of sustainable development goals. While traditional energy resources such as oil, natural gas and coal are becoming increasingly depleted and energy security issues are intensifying, the challenges of environmental pollution and climate change have prompted countries to accelerate the transformation of their energy structures and promote the development and application of new energy. As the core technology carrier for energy production, conversion, transportation and storage, energy equipment plays a vital role in this process. Whether it is the use of renewable energy such as wind energy and solar energy, or the efficient mining and clean use of fossil energy, it is inseparable from the technical support of advanced energy equipment. In addition, the rapid development of nuclear energy, energy storage technology, smart grid and other fields has also put forward higher requirements for energy equipment.
[0003] The significance of energy equipment is reflected in many aspects. First, it is an important guarantee for achieving energy security and stable energy supply, and can ensure efficient and reliable transmission and distribution of energy. Secondly, energy equipment promotes the improvement of energy efficiency, reduces energy waste through technological progress, and reduces losses in the production and transmission process. At the same time, it promotes the green transformation of the energy industry and reduces the negative impact on the environment. For example, through the promotion of renewable energy equipment, greenhouse gas emissions can be greatly reduced, helping the world to cope with climate change. In addition, the energy equipment industry is also a catalyst for technological innovation, promoting the coordinated development of multiple fields such as materials science, automatic control, and artificial intelligence, and improving the technical level of the entire industrial system. Overall, energy equipment is not only a key supporting force for the energy industry, but also a cornerstone for achieving a sustainable energy future.
[0004] Structural optimization technology has a wide range of applications in the field of energy equipment design. It can significantly improve the internal balance of energy equipment and achieve the optimization design goals while meeting the constraints. Size optimization, as a common optimization technology, usually optimizes the structural dimensions or material parameters by combining optimization algorithms and numerical simulation technologies to search for the optimal microstructure parameters that meet the constraints. However, since the numerical simulation process is time-consuming and the entire optimization process requires multiple simulations, the computational cost is often unacceptable in actual engineering product development. To solve this problem, a proxy model is usually used instead of finite element calculation as the objective function of the optimization problem to speed up the solution process. However, building a proxy model also requires a large amount of data, and if this data comes from true calculations, it will still take a lot of time.
[0005] In order to further improve the optimization efficiency, the Bayesian optimization method is introduced. This method uses the posterior information provided by the initial proxy model to design the corresponding point addition criteria, and maximizes or minimizes the criteria through an external optimization algorithm to obtain the most potential new sample points, which are added to the initial sample set, and then the proxy model is continuously reconstructed. Through this cyclic iteration, the global optimal solution can be gradually approached, the dependence on finite element calculations can be reduced, and the design time can be shortened.
[0006] Therefore, establishing an active learning method for optimizing the uniformity of energy equipment packaging stress can effectively improve the performance of energy equipment and ensure its efficient and safe operation in practical applications. Summary of the invention
[0007] In view of the above problems, the present invention provides an active learning method for optimizing the uniformity of energy equipment packaging stress.
[0008] The present invention adopts the following technical solution: an active learning method for optimizing the uniformity of energy equipment packaging stress, comprising the following steps:
[0009] S1: Establish a finite element model of energy equipment and use python script to perform parametric modeling;
[0010] S2: Determine the mathematical model of the optimization problem, simulate and solve the energy equipment, extract relevant results and obtain the top electrode plate deformation non-uniformity index through data processing, which is used as the objective function value in the optimization design of energy equipment;
[0011] S3: Using active learning methods, under the Bayesian optimization framework, the enhanced partitions in the parametric modeling of energy equipment are used as design variables, the unevenness calculation results are used as objective functions, and user needs are used as constraints to calculate the modeling parameters corresponding to the minimum unevenness, thereby realizing on-demand design of energy equipment.
[0012] The step S1 comprises:
[0013] S101: Establish the geometric model of energy equipment, assemble it, and realize parameterization;
[0014] The energy equipment is composed of end plates, reinforced separators, electrode plates, electrolyte layers, seals and bolts, among which two electrode plates, seals and electrolyte layers constitute a single cell. After the single cells are connected in series, they form a stacking structure of the energy equipment together with the lower end plate, reinforced separator and upper end plate.
[0015] The energy equipment as a whole can be equivalent to a symmetrical structure. In order to simplify the calculation, the following processing is adopted. One quarter of the structure of the overall model is selected for study; the bolts are modeled using beam units.
[0016] S102: Establish material properties and assign properties to each component, and implement parameterization;
[0017] S103: Establish the corresponding analysis steps required for energy equipment simulation and implement them by parameterization;
[0018] In the simulation calculation of the microfluidic reactor, the force on the microfluidic reactor is static, so the static analysis step is adopted, and the total time is set to 1, the initial step size is 1e-03, the minimum step size is 1e-05, and the maximum step size is 1.
[0019] S104: Establish the contact between the components of the energy equipment and implement it by parameterization;
[0020] In the simulation calculation of energy equipment, it is necessary to set the contact properties between components. In the present invention, binding constraints are used between the various components, and MPC constraints are used between the bolts and the upper and lower end plates.
[0021] Each component interacts with each other by binding, the bolt preload is 8000N, and considering the influence of gravity, symmetric boundary conditions are adopted for the x and y symmetric planes in the model, and the displacement constraint condition of 0 is adopted for the center point of the bottom surface of the model. The static general solver is used for simulation and solution.
[0022] S105: Establish loads and boundary conditions and implement them parameterized;
[0023] S106: Grid settings are performed for each component of the energy equipment and implemented by parameterization;
[0024] Discretize each component of the energy equipment into hexahedral solid units and specify the mesh size of each component.
[0025] S107: Create a job to perform simulation calculation and implement it with parameterization.
[0026] The step S2 is as follows:
[0027] S201: Determine design variables and establish an optimization mathematical model;
[0028] For energy equipment, after packaging, the packaging load is transmitted to the inside of the battery through the end plate. The structure of the end plate will have a great impact on the uniformity of the force inside the energy equipment. However, since the processing cost of modifying the structure of the end plate is relatively high, a reinforced partition is set between the upper end plate and the top electrode plate in the energy equipment structure, and the structure of the reinforced partition is optimized to improve the internal pressure distribution of the fuel cell.
[0029] The optimization model of the shape and size of the enhanced partition can be expressed as:
[0030]
[0031] stKU=F
[0032] 1≤x1≤50
[0033] 0.01≤x2≤0.03
[0034] 100≤x3≤160
[0035] 100≤x4≤180
[0036] S202: calling the simulation software solver to read the parameterized modeling script written in step S1 to perform simulation calculation;
[0037] The statics solver is used to calculate the model, and the node displacement matrix U is obtained according to the finite element equation, where K is the overall stiffness matrix and F is the node load matrix.
[0038] KU=F
[0039] S203: extract the vertical displacement of the lower surface of the middle plate, output a report file, and obtain the non-uniformity of the top electrode plate deformation by reading the report file and performing data processing, which is used as an indicator for subsequent optimization design.
[0040] When the design of energy equipment is unreasonable, the deformation of the top electrode plate will be too large. Excessive deformation of the top electrode plate will affect the internal reaction and performance, so the deformation of the top electrode plate needs to be as uniform as possible. Therefore, the lower surface area of the top electrode plate is selected, and the vertical displacement field in the area in the simulation results is extracted. The relative standard deviation of the displacement is calculated according to the following formula as the goal of the energy equipment structure optimization design.
[0041]
[0042] In the formula, u i The node displacement of the selected area of the top electrode plate, n is the number of nodes.
[0043] The step S3 comprises:
[0044] S301: Initialize data through Latin hypercube sampling according to the requirements of variables and design domain;
[0045] S302: using the initialization data as modeling parameters for parametric modeling in step S1, using the modulus prediction program in step S2 to calculate its modulus as a data set, and training the proxy model;
[0046] The proxy model uses a Gaussian regression proxy model:
[0047] y(x)~N(μ(x),k(x,x′))
[0048] Choose radial basis as kernel function, that is:
[0049]
[0050] in l is a hyperparameter.
[0051] The expression of the prediction point is determined according to the posterior probability, namely:
[0052]
[0053] have to:
[0054] y * |y~N(K * K -1 y,Cov(x))
[0055] Where K is the covariance matrix, K* is the covariance matrix of the sample point x and the prediction point x*, and K** is the covariance matrix of the prediction point x*. The expression is as follows:
[0056]
[0057] K * =[k[x1,x * ],k[x2,x * ],k[x3,x * ],...,k[x n ,x * ]]
[0058] K ** =[k[x * ,x * ]]
[0059] S303: Optimizing the proxy model using the grey wolf algorithm with introduced constraints, and storing the obtained optimal value;
[0060] The hunting behavior of the gray wolf in the gray wolf algorithm includes: encirclement stage, pursuit stage and attack stage.
[0061] During the encirclement phase, the gray wolf group updates its position according to the following formula.
[0062]
[0063] Where D is the distance between the current individual and the prey, t is the number of iterations, is the coefficient vector, X p is the position of the prey, X w is the current position of the gray wolf. Vector and The update formula is as follows:
[0064]
[0065] The value of a decreases linearly from 2 to 0, and r1 and r2 are random numbers uniformly distributed on [0,1].
[0066] In the hunting stage, the α, β, and δ wolves in the gray wolf optimization algorithm lead the rest of the gray wolf individuals to move to the area close to the prey in the search space, so the selection of α, β, and δ wolves determines whether the algorithm can find the extreme value. When optimizing the solution, each gray wolf corresponds to a target value, so in the optimization iteration, the optimal wolf, the second-best wolf, and the third-best wolf of the current gray wolf population are selected as the α, β, and δ wolves of the current iteration step.
[0067] In the hunting stage, the position of wolf δ is updated according to the positions of wolves α, β, and δ. The position update formula of the gray wolf in the stage of hunting prey is as follows:
[0068] D α =|C1·X α -X|
[0069] D β =|C2·X β -X|
[0070] D δ =|C3·X δ -X|
[0071]
[0072] Among them, C1, C2, C3 are the three component vectors of C, and X is the position of the current solution. α is the optimal solution of the population α, the position of the wolf, X β is the position of the population's suboptimal solution β wolf, X δ It is the third best solution of the population, δ, the position of the wolf.
[0073] In the attack phase, the gray wolf group completes the process of attacking prey according to the change of the value of a. The initial value of a is 2. In the process of a gradually decreasing to 0, according to the formula Its corresponding The value of also changes in the interval [-a, a]. visible, The value range of a gradually narrows. When a changes from 2 to 1, the search range of the gray wolf individual is larger, and the algorithm mainly performs global search. When a changes from 1 to 0, the search range of the gray wolf individual gradually narrows, and the algorithm's ability to perform local search is enhanced.
[0074] S304: Determine whether the stored optimal value converges, and if so, terminate the optimization process;
[0075] S305: A grey wolf algorithm with introduced constraints is used to optimize the double-addition point criterion, and the optimal point obtained is added to the sample set as a new point.
[0076] In the invention, a double-point addition rule is adopted, and the expression of the addition rule is as follows:
[0077]
[0078] Where κ is the control coefficient. When optimizing the maximum value, κ takes 1, and when optimizing the minimum value, κ takes -1.
[0079] Through this point addition criterion, each round of iteration in the optimization process will generate two points, one point represents the point closest to the extreme point under the current proxy model, and the other point represents the point with the highest uncertainty under the current proxy model. In this way, the optimization algorithm can balance development and search capabilities. However, in the optimization calculation process, different iterations may get the same point through the optimization point addition criterion. Calculating the same sample point not only increases the calculation cost, but also does not improve the accuracy of the proxy model. Therefore, in the Bayesian algorithm under the double-point point addition criterion, such duplicate points need to be removed to avoid repeated calculation and addition to the data set.
[0080] S306: Repeat steps S302 to S305 until the iteration converges.
[0081] The beneficial effects of the present invention are:
[0082] 1. The present invention uses Python script to perform parameterized modeling on the energy equipment provided by the user.
[0083] 2. Determine the mathematical model of the optimization problem, simulate and solve the energy equipment, extract relevant results and obtain the top electrode plate deformation non-uniformity index through data processing as the objective function value in the optimization design of energy equipment.
[0084] 3. Using active learning methods, under the Bayesian optimization framework, the enhanced partitions in the parametric modeling of energy equipment are used as design variables, the unevenness calculation results are used as objective functions, and user needs are used as constraints to calculate the modeling parameters corresponding to the minimum unevenness, thereby realizing on-demand design of energy equipment.
[0085] 4. The present invention adopts the Bayesian optimization method in active learning, reduces the number of calculations of the objective function, improves the solution efficiency, and adopts the double-addition criterion to meet the optimization algorithm's consideration of both development capability and search capability. BRIEF DESCRIPTION OF THE DRAWINGS
[0086] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments will be briefly introduced below. Obviously, the drawings in the following description only relate to some embodiments of the present invention, but are not intended to limit the present invention.
[0087] Figure 1 It is a flowchart of active learning under the double-point criterion in the present invention;
[0088] Figure 2 It is a geometrical schematic diagram of the energy device in the present invention;
[0089] Figure 3 A schematic diagram of some codes for parametric modeling and assembly of energy equipment in the present invention;
[0090] Figure 4 A schematic diagram of some codes assigned to the parametric modeling attributes of energy equipment in the present invention;
[0091] Figure 5 A schematic diagram of some codes set for the energy equipment parameter modeling analysis step in the present invention;
[0092] Figure 6 A schematic diagram of part of the code for the contact setting of the parametric modeling of the energy equipment in the present invention;
[0093] Figure 7 A schematic diagram of part of the code for setting boundary conditions for parameterized modeling of energy equipment in the present invention;
[0094] Figure 8 It is a schematic diagram of some codes for the sub-network setting of the energy equipment parameter modeling in the present invention;
[0095] Fig. 9 It is a partial code schematic diagram of the energy equipment parameter modeling operation in the present invention;
[0096] Fig.10 It is the Latin hypercube sampling flow chart of the present invention;
[0097] Fig.11 It is the flow chart of the grey wolf algorithm in the present invention;
[0098] Fig.12 It is a flow chart of the double-pointing criterion in the present invention;
[0099] Fig.13 It is a schematic diagram of the enhanced partition in the present invention;
[0100] Fig.14 It is a schematic diagram of the iterative process in the present invention;
[0101] Fig.15 It is a schematic diagram of the optimization results in the present invention;
[0102] Fig.16 A schematic diagram showing the deformation comparison before and after optimization in the present invention;
[0103] Fig.17 Schematic diagram of the shape change of the enhanced partition before and after optimization in the present invention. DETAILED DESCRIPTION
[0104] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0105] Unless otherwise defined, the technical terms or scientific terms used in the present disclosure shall have the usual meanings understood by persons with ordinary skills in the field to which the present disclosure belongs. The words "include" or "comprise" and the like used in the present disclosure mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0106] The present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0107] Example:
[0108] like Figure 1 As shown, an active learning method for optimizing the uniformity of energy equipment packaging stress includes the following steps:
[0109] S1: Establish a finite element model of energy equipment and use python script to perform parametric modeling;
[0110] S2: Determine the mathematical model of the optimization problem, simulate and solve the energy equipment, extract relevant results and obtain the top electrode plate deformation non-uniformity index through data processing, which is used as the objective function value in the optimization design of energy equipment;
[0111] S3: Using active learning methods, under the Bayesian optimization framework, the enhanced partitions in the parametric modeling of energy equipment are used as design variables, the unevenness calculation results are used as objective functions, and user needs are used as constraints to calculate the modeling parameters corresponding to the minimum unevenness, thereby realizing on-demand design of energy equipment.
[0112] In this embodiment, the above step S1 specifically includes:
[0113] S101: Establish the geometric model of energy equipment, assemble it, and realize parameterization. The model is as follows: Figure 2 As shown, some of the procedures are as follows Figure 3 As shown;
[0114] The energy equipment is composed of end plates, reinforced separators, electrode plates, electrolyte layers, seals and bolts, among which two electrode plates, seals and electrolyte layers constitute a single cell. After the single cells are connected in series, they form a stacking structure of the energy equipment together with the lower end plate, reinforced separator and upper end plate.
[0115] The energy equipment as a whole can be equivalent to a symmetrical structure. In order to simplify the calculation, the following processing is adopted. One quarter of the structure of the overall model is selected for study; the bolts are modeled using beam units.
[0116] S102: Establish material properties and assign properties to each component, and implement parameterization. Some procedures are as follows Figure 4 As shown;
[0117] S103: Establish the corresponding analysis steps required for energy equipment simulation and implement them with parameters. Some procedures are as follows Figure 5 As shown;
[0118] In the simulation calculation of the microfluidic reactor, the force on the microfluidic reactor is static, so the static analysis step is adopted, and the total time is set to 1, the initial step size is 1e-03, the minimum step size is 1e-05, and the maximum step size is 1.
[0119] S104: Establish the contact between the components of the energy equipment and implement it with parameterization. Some procedures are as follows Figure 6 As shown;
[0120] In the simulation calculation of energy equipment, it is necessary to set the contact properties between components. In the present invention, binding constraints are used between the various components, and MPC constraints are used between the bolts and the upper and lower end plates.
[0121] Each component interacts with each other by binding, the bolt preload is 8000N, and considering the influence of gravity, symmetric boundary conditions are adopted for the x and y symmetric planes in the model, and the displacement constraint condition of 0 is adopted for the center point of the bottom surface of the model. The static general solver is used for simulation and solution.
[0122] S105: Establish loads and boundary conditions and implement them parameterized. Some procedures are as follows Figure 7 As shown;
[0123] S106: Set up grids for each component of energy equipment and implement them parameterized. Some programs include Figure 8 As shown;
[0124] Discretize each component of the energy equipment into hexahedral solid units and specify the mesh size of each component.
[0125] S107: Create a job for simulation calculation and implement it with parameterization. Some programs are as follows Fig. 9 shown.
[0126] In this embodiment, the above step S2 specifically includes:
[0127] S201: Determine design variables and establish an optimization mathematical model;
[0128] For energy equipment, after packaging, the packaging load is transmitted to the inside of the battery through the end plate. The structure of the end plate will have a great impact on the uniformity of the force inside the energy equipment. However, since the processing cost of modifying the structure of the end plate is relatively high, a reinforced partition is set between the upper end plate and the top electrode plate in the energy equipment structure, and the structure of the reinforced partition is optimized to improve the internal pressure distribution of the fuel cell.
[0129] The optimization model of the shape and size of the enhanced partition can be expressed as:
[0130]
[0131] stKU=F
[0132] 1≤x1≤50
[0133] 0.01≤x2≤0.03
[0134] 100≤x3≤160
[0135] 100≤x4≤180
[0136] S202: calling the simulation software solver to read the parameterized modeling script written in step S1 to perform simulation calculation;
[0137] The statics solver is used to calculate the model, and the node displacement matrix U is obtained according to the finite element equation, where K is the overall stiffness matrix and F is the node load matrix.
[0138] KU=F
[0139] S203: extract the vertical displacement of the lower surface of the middle plate, output a report file, and obtain the non-uniformity of the top electrode plate deformation by reading the report file and performing data processing, which is used as an indicator for subsequent optimization design.
[0140] When the design of energy equipment is unreasonable, the deformation of the top electrode plate will be too large. Excessive deformation of the top electrode plate will affect the internal reaction and performance, so the deformation of the top electrode plate needs to be as uniform as possible. Therefore, the lower surface area of the top electrode plate is selected, and the vertical displacement field in the area in the simulation results is extracted. The relative standard deviation of the displacement is calculated according to the following formula as the goal of the energy equipment structure optimization design.
[0141]
[0142] In the formula, u i The node displacements of the selected area of the top electrode plate, n is the number of nodes.
[0143] In this embodiment, the above step S3 specifically includes:
[0144] S301: Initialize data through Latin hypercube sampling according to the requirements of variables and design domain. The Latin hypercube sampling process is as follows: Fig.10 ;
[0145] S302: using the initialization data as modeling parameters for parametric modeling in step S1, using the modulus prediction program in step S2 to calculate its modulus as a data set, and training the proxy model;
[0146] The proxy model uses a Gaussian regression proxy model:
[0147] y(x)~N(μ(x),k(x,x′))
[0148] Choose radial basis as kernel function, that is:
[0149]
[0150] in l is a hyperparameter.
[0151] The expression of the prediction point is determined according to the posterior probability, namely:
[0152]
[0153] have to:
[0154] y * |y~N(K * K -1 y,Cov(x))
[0155] Where K is the covariance matrix, K* is the covariance matrix of the sample point x and the prediction point x*, and K** is the covariance matrix of the prediction point x*. The expression is as follows:
[0156]
[0157] K* =[k[x1,x * ],k[x2,x * ],k[x3,x * ],...,k[x n ,x * ]]
[0158] K ** =[k[x * ,x * ]]
[0159] S303: Use the gray wolf algorithm with introduced constraints to optimize the proxy model and store the optimal value. The principle of the gray wolf algorithm is as follows: Fig.11 ;
[0160] The hunting behavior of the gray wolf in the gray wolf algorithm includes: encirclement stage, pursuit stage and attack stage.
[0161] During the encirclement phase, the gray wolf group updates its position according to the following formula.
[0162]
[0163] Where D is the distance between the current individual and the prey, t is the number of iterations, is the coefficient vector, X p is the position of the prey, X w is the current position of the gray wolf. Vector and The update formula is as follows:
[0164]
[0165] The value of a decreases linearly from 2 to 0, and r1 and r2 are random numbers uniformly distributed on [0,1].
[0166] In the hunting stage, the α, β, and δ wolves in the gray wolf optimization algorithm lead the rest of the gray wolf individuals to move to the area close to the prey in the search space, so the selection of α, β, and δ wolves determines whether the algorithm can find the extreme value. When optimizing the solution, each gray wolf corresponds to a target value, so in the optimization iteration, the optimal wolf, the second-best wolf, and the third-best wolf of the current gray wolf population are selected as the α, β, and δ wolves of the current iteration step.
[0167] In the hunting stage, the position of wolf δ is updated according to the positions of α, β, and δ. The position update formula of the gray wolf in the stage of hunting prey is as follows:
[0168] D α =|C1·X α -X|
[0169] Dβ =|C2·X β -X|
[0170] D δ =|C3·X δ -X|
[0171]
[0172] Among them, C1, C2, C3 are the three component vectors of C, and X is the position of the current solution. α is the optimal solution of the population α, the position of the wolf, X β is the position of the population's suboptimal solution β wolf, X δ It is the third best solution of the population, δ, the position of the wolf.
[0173] In the attack phase, the gray wolf group completes the process of attacking prey according to the change of the value of a. The initial value of a is 2. In the process of a gradually decreasing to 0, according to the formula Its corresponding The value of also changes in the interval [-a, a]. visible, The value range of a gradually narrows. When a changes from 2 to 1, the search range of the gray wolf individual is larger, and the algorithm mainly performs global search. When a changes from 1 to 0, the search range of the gray wolf individual gradually narrows, and the algorithm's ability to perform local search is enhanced.
[0174] S304: Determine whether the stored optimal value converges, and if so, terminate the optimization process;
[0175] S305: Use the grey wolf algorithm with introduced constraints to optimize the double-addition criterion, and the optimal point is added to the sample set as a new point, such as Fig.12 As shown;
[0176] In the invention, a double-point addition rule is adopted, and the expression of the addition rule is as follows:
[0177]
[0178] Where κ is the control coefficient. When optimizing the maximum value, κ takes 1, and when optimizing the minimum value, κ takes -1.
[0179] Through this point addition criterion, each round of iteration in the optimization process will generate two points, one point represents the point closest to the extreme point under the current proxy model, and the other point represents the point with the highest uncertainty under the current proxy model. In this way, the optimization algorithm can balance development and search capabilities. However, in the optimization calculation process, different iterations may get the same point through the optimization point addition criterion. Calculating the same sample point not only increases the calculation cost, but also does not improve the accuracy of the proxy model. Therefore, in the Bayesian algorithm under the double-point point addition criterion, such duplicate points need to be removed to avoid repeated calculation and addition to the data set.
[0180] S306: Repeat steps S302 to S305 until the iteration converges.
[0181] Figure 13-Figure 17 In order to design reinforced separators of different shapes on demand, the deformation distribution non-uniformity of the specified area of the top electrode plate is taken as the objective function, the shape parameters R, m, n of the reinforced separator and the thickness H of the reinforced separator are taken as design variables, and the size and shape of the reinforced separator are optimized.
[0182] The shape of the enhanced partition is determined by the following formula:
[0183]
[0184] Where n is the chamfer parameter. When n is larger, the chamfer radius of the enhanced partition is smaller and closer to a right angle. m is the semi-axis length of the enhanced partition in the x direction, and n is the semi-axis length of the enhanced partition in the y direction. Fig.13 Schematic diagram of the enhanced partition shape.
[0185] The optimization model of the shape and size of the enhanced partition can be expressed as:
[0186]
[0187] stKU=F
[0188] 1≤x1≤50
[0189] 0.01≤x2≤0.03
[0190] 100≤x3≤160
[0191] 100≤x4≤180
[0192] Among them, the objective function RSD is the relative standard deviation of the displacement of the lower surface of the top electrode plate; U and F are the overall displacement vector and force vector of the design space (i.e., the enhanced partition), respectively; K is the overall stiffness matrix; is the vector of design variables; KU=F is the control equation under static conditions.
[0193] The Bayesian optimization algorithm is used to optimize the size and parameters of the fuel cell enhanced partition structure. First, 10 initial sample points are generated by Latin hypercube sampling, and the corresponding objective function values of the sample points are obtained by finite element simulation and the initial proxy model is established. The number of Bayesian iterations is set to 100.
[0194] The gray wolf algorithm is used to optimize the point addition criterion. The population size of the gray wolf algorithm is set to 100 and the number of iterations is set to 500.
[0195] Fig.14 The historical convergence curve of Bayesian optimization is given. The horizontal axis represents the number of iterations, and the vertical axis represents the current optimal objective function value. It can be seen from the figure that the optimization algorithm reaches convergence after 10 iterations.
[0196] The non-uniformity of the top electrode plate in the final optimized solution is 0.73%, and the deformation non-uniformity is reduced by 40.65% compared with the initial structure. Fig.15 shown. Fig.16 To compare the deformation cloud images of the top electrode plate before and after optimization, the reduction of the red area in the image indicates that the deformation uniformity of this area has been improved to a certain extent. Fig.17 For optimization of the front and rear reinforcement baffles.
[0197] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the technical solution of the present invention.
Claims
1. An active learning method for optimizing the uniformity of energy equipment packaging stress, characterized in that: The following steps are involved: S1: Establish a finite element model of energy equipment and use python script to perform parametric modeling; S2: Determine the mathematical model of the optimization problem, simulate and solve the energy equipment, extract relevant results and obtain the top electrode plate deformation non-uniformity index through data processing, which is used as the objective function value in the optimization design of energy equipment; S3: Using active learning methods, under the Bayesian optimization framework, the enhanced partitions in the parametric modeling of energy equipment are used as design variables, the unevenness calculation results are used as objective functions, and user needs are used as constraints to calculate the modeling parameters corresponding to the minimum unevenness, thereby realizing on-demand design of energy equipment.
2. The active learning method for optimizing energy equipment packaging stress uniformity according to claim 1, characterized in that: The step S1 comprises: S101: Establish geometric model of energy equipment and assemble it; S102: Establish material properties and assign properties to each component; S103: Establishing the corresponding analysis steps required for energy equipment simulation; S104: Establishing contacts between components of energy equipment; S105: Establish loads and boundary conditions; S106: Grid setting for each component of the energy equipment; S107: Create a job to perform simulation calculation.
3. An active method for optimizing stress uniformity of energy equipment packaging according to claim 1 A learning method, characterized in that The step S2 is as follows: S201: Determine design variables and establish an optimization mathematical model; S202: calling the simulation software solver to read the parameterized modeling script written in step S1 to perform simulation calculation; S203: extract the vertical displacement of the lower surface of the middle plate, output a report file, and obtain the non-uniformity of the top electrode plate deformation by reading the report file and performing data processing, which is used as an indicator for subsequent optimization design.
4. The active learning method for optimizing energy equipment packaging stress uniformity according to claim 1, characterized in that: The step S3 comprises: S301: Initialize data through Latin hypercube sampling according to the requirements of variables and design domain; S302: using the initialization data as modeling parameters for parametric modeling in step S1, using the modulus prediction program in step S2 to calculate its modulus as a data set, and training the proxy model; S303: Optimizing the proxy model using the grey wolf algorithm with introduced constraints, and storing the obtained optimal value; S304: Determine whether the stored optimal value converges, and if so, terminate the optimization process; S305: A grey wolf algorithm with introduced constraints is used to optimize the double-addition point criterion, and the optimal point obtained is added to the sample set as a new point. S306: Repeat steps S302 to S305 until the iteration converges.