A machine learning-assisted differential evolution method based on taboo space delineation

Abstract

The present invention discloses a machine learning-assisted differential evolution method based on taboo space delineation, which is applicable to the service time maximization problem of the drive shaft of an azimuth propeller, including: considering the structural parameters and alternating loads of the drive shaft to initialize the population and parameters, and construct a determined taboo space; designing a machine learning-driven potential space determination strategy to construct a potential space and a machine learning model; constructing a greedy co-evolution operation based on the greedy information of the potential space and the reverse position information of the taboo space; adopting a binomial crossover strategy to obtain a corresponding candidate offspring individual pool; constructing a characteristic-derived candidate offspring individual uncertainty assessment method based on machine learning; updating the population and reconstructing the taboo space and potential space, and repeatedly judging whether the convergence conditions are met until the optimized solution is output. The present invention can construct a real-time taboo space and potential space for the service time maximization problem of the drive shaft of an azimuth propeller, accelerate the optimization efficiency of the optimal structural parameters, and shorten the design cycle.

Description

Technical Field

[0001] The present invention relates to the general technical field of artificial intelligence in electronic information, and more specifically, to a machine learning-assisted differential evolution method based on taboo space delineation. Background Art

[0002] The service time of the drive shaft of the full-revolving propeller requires the use of time-consuming fatigue simulation to obtain the service time of the drive shaft under alternating stress. The fatigue simulation evaluation time for the full-revolving propeller drive shaft is between half an hour and one hour, which makes the traditional evolutionary method that is not time-consuming to obtain the target response and requires a large number of target response functions. It is impossible to give satisfactory drive shaft structural design parameters within a reasonable design cycle and cannot meet the service time requirements. Therefore, how to design an effective intelligent optimization method for the service time problem of the full-revolving propeller drive shaft is one of the key technical difficulties currently faced.

[0003] In recent years, design optimization methods based on machine learning have emerged. These methods use the autonomous learning ability of machine learning to approximate and replace time-consuming target simulations, thereby greatly reducing the number of unnecessary and time-consuming simulation calls, thereby significantly improving the optimization efficiency of the method and shortening the design cycle.

[0004] In the process of implementing the technical method of the embodiment of the present invention, the inventor of this patent found at least the following technical problems in the prior art:

[0005] At present, researchers have only used experimental design methods to maximize the service life of the azimuth thruster drive shaft, resulting in poor performance. In addition, existing design optimization techniques based on machine learning focus on how to explore high-potential search directions for optimization problems, but ignore how to avoid or utilize location information of historical low-potential areas during the optimization process. This also makes the method easy to fall into historical local optimality and has low optimization efficiency.

[0006] In summary, how to combine the characteristics of the simulation model of the service time of the azimuth propeller drive shaft with the characteristics of the key structural parameters to delineate the low-potential areas in the machine learning optimization process and avoid the misleading of the low-potential key structural parameters, and how to use the reverse guidance information of the taboo space and the forward guidance information of the high-potential area to carry out collaborative evolution, so as to improve the optimization efficiency of the algorithm for the problem of maximizing the service time of the azimuth propeller drive shaft. Summary of the invention

[0007] In view of the fact that the service time maximization problem of the drive shaft of the full-rotation propeller in the prior art involves time-consuming simulation and the prior art ignores the characteristics of low-potential areas, a machine learning-assisted differential evolution method based on taboo space delineation is studied and designed. The method effectively delineates the taboo space and high-potential space in the optimization process, and designs a co-evolution operation based on the difference in position information between the two to generate high-potential candidate offspring individuals, thereby greatly reducing the screening pressure of the machine learning model and significantly improving the optimization efficiency of the method.

[0008] To achieve the above objectives, an embodiment of the present invention provides a machine learning-assisted differential evolution method based on taboo space delineation, the method being applicable to the service time maximization problem of a full-rotation propeller drive shaft, the method comprising:

[0009] Step 1: The service time maximization problem of the azimuth propeller drive shaft is transformed into the service time negative value minimization problem. The population and key iterative parameters are initialized based on the number and value range of the key structural parameters of the drive shaft. The azimuth propeller geometric model and fatigue life simulation model are constructed according to the alternating load spectrum and force analysis, and the fatigue simulation response value of the initial population is obtained. A machine learning model is constructed and the machine learning-driven optimization strategy is executed to determine the taboo space of the key structural parameter values, and the convergence conditions of the method are given.

[0010] Step 2: Design a machine learning-driven potential space determination strategy to construct the potential space, and update the machine learning model based on the potential individuals, in order to build an accurate machine learning prediction model for the fatigue life simulation model corresponding to the service time of the azimuth propeller drive shaft;

[0011] Step 3: Based on the greedy information in the potential space and the reverse position information in the taboo space, a greedy co-evolution operation that can inherit the greedy information and stay away from the reverse position information is constructed to generate a pool of candidate mutant offspring individuals for each potential individual in the hope of obtaining a set of key structural parameter vectors with excellent service time performance;

[0012] Step 4: For each candidate mutant offspring individual, use the binomial crossover strategy to obtain the corresponding candidate offspring individual pool;

[0013] Step 5: Based on machine learning, a method for evaluating the uncertainty of candidate offspring individuals is constructed to derive characteristics, and individuals with high potential and high uncertainty are selected from the candidate offspring individual pool as real offspring individuals;

[0014] Step 6: Evaluate the real offspring individuals, update the population and key iteration parameter information, reconstruct the taboo space and potential space based on machine learning-driven optimization, and determine whether the optimized structure meets the service time index. If so, output the optimal key structure parameter vector obtained by the method, otherwise go to step 2 until the optimized structure obtained by the method meets the service time index.

[0015] Furthermore, the step 1 of constructing a machine learning model and executing a machine learning driven optimization strategy to determine the taboo space of key structural parameter values ​​specifically includes the following sub-steps:

[0016] The initial population individuals are used as the modeling sample input set, and the target true response values ​​of the initial population individuals are used as the modeling sample output set to construct a radial basis function machine learning model. The specific expression is as follows:

[0017] ;

[0018] In the above formula, Evaluate the radial basis function machine learning model expression, Indicates The weight coefficients of the basis functions are: represents the number of modeling samples, Indicates Modeling sample points, represents the basis function, x Indicates new sample points that need to be predicted using radial basis functions;

[0019] The particle swarm optimization method convergence condition is set as the population has not been updated for ten consecutive generations. The classical particle swarm optimization method is used to search for the optimization problem constructed by the radial basis function machine learning model. The specific expression is as follows:

[0020] ;

[0021] In the above formula, and They represent the lower and upper bounds of the design space of the optimization problem respectively;

[0022] Set the radius of the taboo space to , then multiple taboo spaces can be determined by the final population converged by the particle swarm method. If the final population size is , then we get Taboo space, that is, the first Individual The corresponding taboo space is .

[0023] Furthermore, the step 2 specifically includes the following sub-steps:

[0024] The convergence condition of the differential evolution method is set as the population has not been updated for ten consecutive generations. The classical differential evolution method is used to search for the optimization problem constructed by the radial basis function machine learning model. The specific expression is as follows:

[0025] ;

[0026] If the final population obtained by differential evolution is , then the potential space consists of the final population The enclosed hyperrectangular area is composed of ;in, xxx 1 represents the first individual of the population obtained by differential evolution, xxx NDE represents the NDEth individual in the population obtained by differential evolution, and NDE represents the number of individuals in the population obtained by differential evolution.

[0027] Furthermore, the step three specifically includes the following sub-steps:

[0028] For each individual in the potential space, a greedy coevolution operation is constructed to generate 1000 candidate mutation offspring individuals, thereby forming a candidate mutation offspring individual pool. The expression of the greedy coevolution operation is as follows:

[0029] ;

[0030] In the above formula, Indicates Candidate mutant offspring individuals, represents each individual in the potential space, represents an individual randomly selected from the potential space, where for xxx r1 The corresponding index, represents an individual randomly selected from the taboo space, where for xx r2 The corresponding index, Represents a random number between 0 and 1.

[0031] Furthermore, the crossover formula in step 4 is as follows:

[0032] ;

[0033] In the above formula, Indicates Candidate offspring individuals, represents the crossover probability, represents a random index, Indicates the current dimension index on which to perform binomial crossover.

[0034] Furthermore, the uncertainty assessment function formula of the uncertainty assessment method in step 5 is as follows:

[0035] ;

[0036] In the above formula, Indicates that for Candidate offspring individuals The uncertainty information value of represents the first of the 100 evaluated real individuals close to Individuals, Represents the radial basis function machine learning model for The predicted value of y(tt i ) express The corresponding true objective function value.

[0037] Furthermore, the step 5 of selecting high potential and high uncertainty individuals from the candidate offspring individual pool as real offspring individuals specifically includes the following sub-steps:

[0038] Selecting an individual with the smallest predicted value of the radial basis function machine learning model from the candidate offspring individual pool as a high-potential offspring individual;

[0039] The individual with the largest uncertainty information value is selected from the candidate offspring individual pool as the high uncertainty individual.

[0040] Furthermore, the specific steps of step six are as follows:

[0041] The fatigue life simulation model is used to evaluate the true target value of high potential and high uncertainty individuals corresponding to each individual in the potential space;

[0042] According to the size of the true target value, find the optimal individual with the minimum true target value among each individual in the potential space and its corresponding high potential and high uncertainty individuals;

[0043] Merge the best individuals corresponding to all individuals in the potential space to form an updated population;

[0044] For the current population, the taboo space and potential space are reconstructed using machine learning-driven optimization strategies;

[0045] Update other key iteration information to prepare for the next generation.

[0046] In a second aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a machine learning-assisted differential evolution method based on taboo space delineation in the aforementioned embodiment are implemented.

[0047] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a machine learning-assisted differential evolution method based on taboo space delineation in the aforementioned embodiment.

[0048] In summary, compared with the prior art, the above technical solution conceived by the present invention provides a machine learning-assisted differential evolution method based on taboo space delineation, which mainly has the following beneficial effects:

[0049] 1. The particle swarm method is used to determine the taboo space corresponding to the low-potential key structural parameter values, and the machine learning model is used to accurately predict the fatigue life simulation model. The global exploration ability of the particle swarm method and the precise prediction ability of the machine learning model are effectively utilized, providing an accurate and reliable low-potential taboo space for the iterative optimization in the main loop.

[0050] 2. The differential evolution method is used to determine the potential space mapped by the machine learning model, which provides stable positive guidance for continuous iteration. At the same time, it inherits all the high-potential offspring individual information that may appear in the historical iteration process and makes full use of all available effective guidance in the search process.

[0051] 3. A collaborative evolutionary operation of forward guidance in potential space and reverse promotion in taboo space was designed, which significantly improved the potential of candidate offspring individuals and significantly reduced the screening pressure of subsequent machine learning models.

[0052] 4. During the iteration process, both high-potential individuals and high-uncertainty individuals are considered to drive optimization. The purpose is to continuously improve the prediction accuracy of the machine learning model as the optimization iteration proceeds, significantly improving the positive guidance role of the machine learning model.

[0053] 5. The present invention solves the service time maximization problem of the azimuth propeller drive shaft involving time-consuming simulation, improves the accuracy of the optimization solution, and can be widely used in the design optimization of similar structural optimization problems involving expensive simulation, which is conducive to promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flowchart of a machine learning-assisted differential evolution method based on taboo space delineation provided by the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below 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 and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0056] See also Figure 1 The present invention provides a machine learning-assisted differential evolution method based on taboo space delineation, which is applicable to the service time maximization problem of the drive shaft of an azimuth propeller. Specifically, the method mainly includes the following steps:

[0057] Step one, transform the service time maximization problem of the azimuth propeller drive shaft into the negative service time minimization problem, initialize the population and key iterative parameters based on the number and value range of key structural parameters of the drive shaft, construct the azimuth propeller geometric model and fatigue life simulation model according to the alternating load spectrum and force analysis, and obtain the fatigue simulation response value of the initial population, construct a machine learning model and execute the machine learning driven optimization strategy to determine the taboo space of key structural parameter values, and give the convergence conditions of the method.

[0058] Among them, constructing a machine learning model and executing a machine learning-driven optimization strategy to determine the taboo space specifically includes the following sub-steps:

[0059] (11) Taking the initial population individuals as the modeling sample input set and the target true response values ​​of the initial population individuals as the modeling sample output set, a radial basis function machine learning model is constructed. The specific expression is as follows:

[0060] ;

[0061] In the above formula, Evaluate the radial basis function machine learning model expression, Indicates The weight coefficients of the basis functions are: represents the number of modeling samples, Indicates Modeling sample points, represents the basis function, x Indicates new sample points that need to be predicted using radial basis functions;

[0062] (12) The convergence condition of the particle swarm optimization method is set as the population has not been updated for ten consecutive generations. The classical particle swarm optimization method is used to search for the optimization problem constructed by the radial basis function machine learning model. The specific expression is as follows:

[0063] ;

[0064] In the above formula, and They represent the lower and upper bounds of the design space of the optimization problem respectively;

[0065] (13) Set the radius of the taboo space to , then multiple taboo spaces can be determined by the final population converged by the particle swarm method. If the final population size is , then we get Taboo space, that is, the first Individual The corresponding taboo space is .

[0066] Step 2: Design a machine learning-driven potential space determination strategy to construct the potential space, and update the machine learning model based on the potential individuals, in order to build an accurate machine learning prediction model for the fatigue life simulation model and complex geometric model corresponding to the service time of the azimuth thruster drive shaft.

[0067] Step 2: The specific steps are as follows:

[0068] (21) The convergence condition of the differential evolution method is set as the population has not been updated for ten consecutive generations. The classical differential evolution method is used to search for the optimization problem constructed by the radial basis function machine learning model. The specific expression is as follows:

[0069] ;

[0070] (22) If the final population obtained by differential evolution is , then the potential space consists of the final population The enclosed hyperrectangular area is composed of ;in, xxx 1 represents the first individual of the population obtained by differential evolution, xxx NDE represents the NDEth individual in the population obtained by differential evolution, and NDE represents the number of individuals in the population obtained by differential evolution.

[0071] Step three, based on the greedy information in the potential space and the reverse position information in the taboo space, a greedy coevolution operation is constructed that can inherit the greedy information and stay away from the reverse position information, and a pool of candidate mutant offspring individuals is generated for each potential individual, in the hope of obtaining a set of key structural parameter vectors with excellent service time performance.

[0072] Step 3 specifically includes the following sub-steps:

[0073] (31) For each individual in the potential space, a greedy coevolution operation is constructed to generate 1000 candidate mutant offspring individuals, thereby forming a candidate mutant offspring individual pool. The expression of the greedy coevolution operation is as follows:

[0074] ;

[0075] In the above formula, Indicates Candidate mutant offspring individuals, represents each individual in the potential space, represents an individual randomly selected from the potential space, where for xxx r1 The corresponding index, represents an individual randomly selected from the taboo space, where for xx r2 The corresponding index, Represents a random number between 0 and 1.

[0076] Step 4: For each candidate mutant offspring individual, a binomial crossover strategy is used to obtain the corresponding candidate offspring individual pool.

[0077] The crossover formula is as follows:

[0078] ;

[0079] In the above formula, Indicates Candidate offspring individuals, represents the crossover probability, represents a random index, Indicates the current dimension index on which to perform binomial crossover.

[0080] Step five, construct a method for evaluating the uncertainty of candidate offspring individuals based on machine learning characteristics, and select individuals with high potential and high uncertainty from the candidate offspring individual pool as real offspring individuals.

[0081] Among them, the uncertainty assessment function formula is as follows:

[0082] ;

[0083] In the above formula, Indicates that for Candidate offspring individuals The uncertainty information value of represents the first of the 100 evaluated real individuals close to Individuals, Represents the radial basis function machine learning model for The predicted value of y(tt i ) express The corresponding true objective function value.

[0084] Individuals with high potential and high uncertainty are selected from the candidate offspring individual pool as real offspring individuals, which specifically includes the following sub-steps:

[0085] (51) Selecting an individual with the smallest radial basis function machine learning model prediction value from the candidate offspring individual pool as a high-potential offspring individual;

[0086] (52) Select the individual with the largest uncertainty information value from the candidate offspring individual pool as the high uncertainty individual.

[0087] Step six: evaluate the real offspring individuals, update the population and key iteration parameter information, reconstruct the taboo space and potential space based on machine learning-driven optimization, and determine whether the optimized structure meets the service time index. If so, output the optimal key structure parameter vector obtained by the method, otherwise go to step two until the optimized structure obtained by the method meets the service time index.

[0088] Step six specifically includes the following sub-steps:

[0089] (61) The fatigue life simulation model is used to evaluate the true target values ​​of high potential and high uncertainty individuals corresponding to each individual in the potential space;

[0090] (62) According to the size of the true target value, find the optimal individual with the smallest true target value among each individual in the potential space and its corresponding high potential and high uncertainty individuals;

[0091] (63) Merge the optimal individuals corresponding to all individuals in the potential space to form an updated population;

[0092] (64) For the current population, the taboo space and potential space are reconstructed using machine learning-driven optimization strategies;

[0093] (65) Update other key iteration information to prepare for the next generation.

[0094] Example

[0095] This embodiment uses a classic test problem to illustrate a machine learning-assisted differential evolution method based on taboo space delineation provided in this embodiment. The expression of this embodiment is as follows:

[0096] ;

[0097] In the formula, For the design variables, To optimize the objective function, To find the optimal solution to the classic testing problem, To minimize the classic test problem, are the constraints that need to be satisfied in the classic test problem.

[0098] In view of the above problems, a machine learning-assisted differential evolution method based on taboo space delineation in an embodiment of the present invention is used for processing. In order to further illustrate this embodiment, the maximum number of calls to the real simulation model is set to 1000, and the experimental results are shown in the table below. According to the results in the table below, the optimal solution obtained by the method of this embodiment is infinitely close to the real optimal solution, which reflects the effectiveness of this embodiment in improving the optimization accuracy. It can be considered that the method of this embodiment has good performance in the optimization problem of maximizing the service life of the full-rotation propeller drive shaft.

[0099] Comparison table of optimization results of different methods

[0100] method The optimal solution obtained by this method The real optimal solution Target value 0.013 0

[0101] The present invention provides a machine learning-assisted differential evolution method based on taboo space delineation for the service time maximization problem of the azimuth propeller drive shaft. An accurate machine learning model is constructed for the fatigue life simulation model, and real-time taboo space and potential space are constructed to prevent individuals from falling into low-potential areas on the historical iteration trajectory. At the same time, individuals are guided to move toward high-potential areas of historical iterations, which effectively accelerates the convergence speed and obtains an optimization solution with better accuracy, providing a systematic solution to the service time maximization problem of the azimuth propeller drive shaft.

[0102] Another embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a machine learning-assisted differential evolution method based on taboo space delineation in the aforementioned embodiment are implemented.

[0103] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of a machine learning-assisted differential evolution method based on taboo space delineation in the aforementioned embodiment are implemented.

[0104] It will be easily understood by those skilled in the art that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A machine learning-assisted differential evolution method based on taboo space delineation, characterized in that: The method is applicable to the service time maximization problem of the azimuth propeller drive shaft, and the method comprises: Step 1: The service time maximization problem of the azimuth propeller drive shaft is transformed into the service time negative value minimization problem. The population and key iterative parameters are initialized based on the number and value range of the key structural parameters of the drive shaft. The azimuth propeller geometric model and fatigue life simulation model are constructed according to the alternating load spectrum and force analysis, and the fatigue simulation response value of the initial population is obtained. A machine learning model is constructed and the machine learning-driven optimization strategy is executed to determine the taboo space of the key structural parameter values, and the convergence conditions of the method are given. Step 2: Design a machine learning-driven potential space determination strategy to construct the potential space, and update the machine learning model based on the potential individuals, in order to build an accurate machine learning prediction model for the fatigue life simulation model corresponding to the service time of the azimuth propeller drive shaft; Step 3: Based on the greedy information in the potential space and the reverse position information in the taboo space, a greedy co-evolution operation that can inherit the greedy information and stay away from the reverse position information is constructed to generate a pool of candidate mutant offspring individuals for each potential individual in the hope of obtaining a set of key structural parameter vectors with excellent service time performance; Step 4: For each candidate mutant offspring individual, use the binomial crossover strategy to obtain the corresponding candidate offspring individual pool; Step 5: Based on machine learning, a method for evaluating the uncertainty of candidate offspring individuals is constructed to derive characteristics, and individuals with high potential and high uncertainty are selected from the candidate offspring individual pool as real offspring individuals; Step 6: Evaluate the real offspring individuals, update the population and key iteration parameter information, reconstruct the taboo space and potential space based on machine learning-driven optimization, and determine whether the optimized structure meets the service time index. If so, output the optimal key structure parameter vector obtained by the method, otherwise go to step 2 until the optimized structure obtained by the method meets the service time index; The step 1 of constructing a machine learning model and executing a machine learning driven optimization strategy to determine the taboo space of key structural parameter values ​​specifically includes the following sub-steps: The initial population individuals are used as the modeling sample input set, and the target true response values ​​of the initial population individuals are used as the modeling sample output set to construct a radial basis function machine learning model. The specific expression is as follows: ； In the above formula, Evaluate the radial basis function machine learning model expression, Indicates The weight coefficients of the basis functions are: represents the number of modeling samples, Indicates Modeling sample points, represents the basis function, x Indicates new sample points that need to be predicted using radial basis functions; The particle swarm optimization method convergence condition is set as the population has not been updated for ten consecutive generations. The classical particle swarm optimization method is used to search for the optimization problem constructed by the radial basis function machine learning model. The specific expression is as follows: ； In the above formula, and They represent the lower and upper bounds of the design space of the optimization problem respectively; Set the radius of the taboo space to , then multiple taboo spaces can be determined by the final population converged by the particle swarm method. If the final population size is , then we get Taboo space, that is, the first Individual The corresponding taboo space is ; The step 2 specifically includes the following sub-steps: The convergence condition of the differential evolution method is set as the population has not been updated for ten consecutive generations. The classical differential evolution method is used to search for the optimization problem constructed by the radial basis function machine learning model. The specific expression is as follows: ； If the final population obtained by differential evolution is , then the potential space consists of the final population The enclosed hyperrectangular area is composed of ;in, xxx 1 represents the first individual of the population obtained by differential evolution, xxx NDE represents the NDEth individual of the population obtained by differential evolution, and NDE represents the number of individuals in the population obtained by differential evolution; The step three specifically includes the following sub-steps: For each individual in the potential space, a greedy coevolution operation is constructed to generate 1000 candidate mutation offspring individuals, thereby forming a candidate mutation offspring individual pool. The expression of the greedy coevolution operation is as follows: ； In the above formula, Indicates Candidate mutant offspring individuals, represents each individual in the potential space, represents an individual randomly selected from the potential space, where for xxx r1 The corresponding index, represents an individual randomly selected from the taboo space, where for xx r2 The corresponding index, Represents a random number between 0 and 1.

2. The method according to claim 1, characterized in that The crossover formula in step 4 is as follows: ； In the above formula, Indicates Candidate offspring individuals, represents the crossover probability, represents a random index, Indicates the current dimension index on which to perform binomial crossover.

3. The method according to claim 1, characterized in that The uncertainty assessment function formula of the uncertainty assessment method in step 5 is as follows: ； In the above formula, Indicates that for Candidate offspring individuals The uncertainty information value of represents the first of the 100 evaluated real individuals close to Individuals, Represents the radial basis function machine learning model for The predicted value of y(tt i ) express The corresponding true objective function value.

4. The method according to claim 1, characterized in that The step 5 of selecting high potential and high uncertainty individuals from the candidate offspring individual pool as real offspring individuals specifically includes the following sub-steps: Selecting an individual with the smallest predicted value of the radial basis function machine learning model from the candidate offspring individual pool as a high-potential offspring individual; The individual with the largest uncertainty information value is selected from the candidate offspring individual pool as the high uncertainty individual.

5. The method according to claim 1, characterized in that The specific steps of step six are as follows: The fatigue life simulation model is used to evaluate the true target value of high potential and high uncertainty individuals corresponding to each individual in the potential space; According to the size of the true target value, find the optimal individual with the minimum true target value among each individual in the potential space and its corresponding high potential and high uncertainty individuals; Merge the best individuals corresponding to all individuals in the potential space to form an updated population; For the current population, the taboo space and potential space are reconstructed using machine learning-driven optimization strategies; Update other key iteration information to prepare for the next generation.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images