A reverse design method for high-performance magnesium alloys based on machine learning

Abstract

The present invention discloses a reverse design method for high-performance magnesium alloys based on machine learning. By collecting data on the composition, processing technology and mechanical properties of magnesium alloys, the implicit structure-activity relationship of the composition, processing technology and mechanical properties of magnesium alloys is analyzed by machine learning algorithms, thereby realizing efficient design of new high-performance magnesium alloys based on mechanical performance requirements. The present invention combines an optimization algorithm with a forward model to optimize parameters such as alloy composition, effectively improve the accuracy of prediction (the calculation accuracy can reach 99%), reduce deviations (the error is only 0.5%), and can achieve good fitting effects even with a small amount of alloy data and complex process combinations. The calculation method is simple and easy to implement. The present invention improves the design efficiency of magnesium alloys, reduces experimental costs, and is a reverse design guided by actual performance requirements. It not only solves the problem of small amount of magnesium alloy data, but also provides a new idea for the efficient development and design of high-performance magnesium alloys based on machine learning methods.

Description

Technical Field

[0001] The present invention relates to the technical field of magnesium alloy material design, and in particular to a reverse design method for high-performance magnesium alloys based on machine learning. Background Art

[0002] With the advancement of automotive and aerospace technologies, the material performance requirements for large components such as auto bodies and aircraft engine casings are becoming increasingly stringent. Therefore, developing new, high-performance materials is a key approach to meeting these demands. Magnesium alloys, due to their low density, high specific strength and stiffness, and excellent electromagnetic shielding properties, hold broad application prospects in the automotive and aerospace sectors. However, their poor room-temperature strength and ductility, coupled with poor formability, severely limit their application in large structural components. Alloying is an effective method for manipulating the mechanical properties of magnesium alloys. However, the influence of alloy composition, heat treatment, and deformation processes on their mechanical properties and microstructure is complex, and the mechanisms of influence remain unclear. Furthermore, as magnesium alloy composition increases and process performance continues to be optimized, the combinations of various alloy parameters become increasingly complex, and models that quantitatively describe the relationships between alloy composition, structure, processing, and performance are lacking. Rapidly clarifying the complex relationships between alloy mechanical properties, composition, and processing, and further improving the mechanical properties of magnesium alloys, is a key to developing high-performance magnesium alloys.

[0003] Currently, alloy development still relies on traditional trial-and-error methods, which have long development cycles and rely on extensive resource consumption, resulting in high costs and low development efficiency. Machine learning (ML) methods, on the other hand, offer low computational costs and short development cycles. They can bypass complex equation solving and utilize data analysis to identify implicit relationships between data. Therefore, they are one of the most effective alternatives to repeated laboratory experiments. However, research on the application of machine learning methods in magnesium alloy design is limited, primarily focusing on forward prediction models from composition to performance. For example, patent CN114898821A discloses a method for predicting alloy properties based on the fusion of machine learning models. By mining historical data on alloy composition, processing technology, and performance, the method integrates boosting and stacking fusion in machine learning. Boosting fusion can improve model accuracy and reduce bias, while stacking fusion can make the model more robust. By synergizing these two models, different models can complement each other, and the relationship between composition, processing technology, and performance can be constructed through fusion of the training process and the training results. However, this method cannot achieve fast and efficient search for the global optimal solution, and the accuracy and generalization performance of different machine learning algorithms are also lacking in-depth research. At the same time, the main purpose of magnesium alloy research is to meet the performance requirements of application structures. The performance-oriented reverse design method that has emerged in recent years is conducive to achieving "on-demand design" of alloys, but there is currently little research on reverse design methods for magnesium alloys. For example, invention patent CN110010210A discloses a multi-component alloy composition design method based on machine learning and oriented to performance requirements. By mining a large amount of existing data on alloy composition and performance, machine learning technology is used to unlock the implicit and complex relationship between "composition-performance" to achieve the purpose of quickly and accurately designing alloy composition according to performance requirements. The method includes: S1, establishing a data set based on historical data; S2, establishing and training C2P and P2C models; S3, inputting the target performance as input data into P2C to obtain an initial design composition; S4, inputting the initial design composition as input data into C2P to obtain predicted performance; S5, judging whether the error of the predicted performance relative to the target performance is within an acceptable range. If so, the model is not re-established; if so, the design is completed. However, this method directly uses the neural network for reverse training, and then optimizes the model structure based on the forward model and the reverse neural network in series, so its design accuracy is low, the calculation is complex, the rate is low, and the cost is high. More importantly, the above method does not take into account the subsequent processing technology and parameters, and the influence of processing technology and parameters on the performance of the alloy is also crucial. Furthermore, for the small amount of alloy data, the combination is complex, but there are few mechanical performance indicators, and a simple model reverse training cannot achieve an acceptable fitting effect. Because the algorithm is based on data, if the amount of data is small, the input end latitude is low and the output end latitude is high, a good fitting effect is not easy to achieve with the current common machine learning methods. Summary of the Invention

[0004] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is: how to provide a reverse design method for high-performance magnesium alloys based on machine learning to solve the problems of existing machine learning methods such as complex calculations, low efficiency, low accuracy, weak generalization ability, and poor model fitting effect for small amounts of data.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solution: a reverse design method of high-performance magnesium alloy based on machine learning, comprising the following steps:

[0006] S1: Data preparation: Collect historical data related to magnesium alloys, build a magnesium alloy original database, and then perform data preprocessing and feature selection to obtain a magnesium alloy dataset;

[0007] S2: Establishment of a forward prediction model: Divide the magnesium alloy dataset obtained in step S1 into a training set and a test set, standardize the training set, and construct a machine learning forward prediction model of alloy composition, processing technology → mechanical properties using alloy composition and processing technology as input and mechanical properties as output. The accuracy of the forward prediction model is evaluated using the test set.

[0008] S3: Selection and establishment of sampling optimization model: Particle swarm optimization (PSO) and genetic algorithm (GA) are selected as sampling optimization algorithms respectively, and the model is established by self-programming to perform parameter optimization. The appropriate algorithm is selected by comparing the optimization effects of the two. In this way, the sampling model mainly looks at the optimization effects of the two based on examples and reverse design ideas, and then selects the optimization algorithm with better optimization effect, and finally establishes the magnesium alloy reverse design model.

[0009] S4: Construction of the reverse design model: Determine a parameter space, randomly sample within this space using the sampling optimization algorithm selected in step S3, and then input the sampling results into the forward prediction model after the same standardization process as the forward model training set to obtain the corresponding performance prediction value, and construct a reverse design model of mechanical properties → alloy composition and processing technology;

[0010] S5: Alloy design: The root mean square error of the performance prediction value and the target performance value is used as the fitness function. Through the iteration of the sampling optimization algorithm, some values ​​are updated according to whether the requirements are met. This cycle is repeated until the optimal solution that minimizes the fitness function is obtained. The optimal solution is the combination of alloy composition and processing parameters that matches the target performance.

[0011] Preferably, the data in step S1 include alloy element content, heat treatment process parameters, deformation process parameters and mechanical properties.

[0012] Preferably, the data processing in step S1 includes missing value processing, duplicate value deletion or outlier processing; and the feature selection includes range scaling processing of the solution temperature and the extrusion temperature.

[0013] Preferably, the solution temperature is scaled in intervals of 20°C, and the extrusion temperature is scaled in intervals of 30°C.

[0014] Preferably, the ratio of the training set to the test set in step S2 is 8:2; the training set is standardized by formula (1)

[0015]

[0016] Where x represents each feature sample, μ represents the mean of each feature sample set, and σ represents the standard deviation of each feature sample set.

[0017] Preferably, the method constructed in step S2 includes extreme gradient boosting tree (XGBoost) and random forest (RF).

[0018] Preferably, step S2 further comprises adjusting the hyperparameters of the forward prediction model using a greedy algorithm, that is, each time a parameter is adjusted, a parameter having the greatest impact on the model performance is selected for adjustment until optimal performance is achieved.

[0019] Preferably, the specific implementation of the PSO algorithm is as follows:

[0020] 1) Setting initial parameters, including particle swarm size, particle dimension, and number of iterations; particle swarm size refers to a range of magnesium alloy composition and processing parameters; particle dimension refers to the number of dimensions of the selected initial particles, and the dimension setting is related to the parameters of the alloy combination to be designed; the number of iterations refers to the set condition, that is, how many times the particle is updated before the update ends. After the update ends, the particle with the best fitness among these several iterations is selected.

[0021] 2) Randomly initialize the particle position and velocity, and calculate the fitness function value of the initialized particle;

[0022] 3) Update the particle's velocity and position according to equations (3) and (4), calculate the fitness value of each particle, update the historical optimal position and optimal fitness of each individual, and update the historical optimal position and fitness value of each group; the particle's position refers to a solution to the problem (such as the specific values ​​of the alloy composition and processing parameters); the particle's velocity is a position vector, which refers to the distance and direction of each particle update.

[0023]

[0024]

[0025] In the formula is the historical optimal position of particle i in the d-th dimension in the k-th iteration; is the historical optimal position of the d-th dimension of the group in the k-th iteration, ω is the inertia factor, r1 and r2 are random numbers between (0,1), and c1 and c2 are learning factors;

[0026] 4) Determine whether the maximum number of iterations has been reached or whether the minimum value of the fitness difference between the two generations of particles has been reached. If so, the iteration ends and the optimal position of the individual is output; if not, repeat step 3) until the stopping condition is met and the optimal solution is output.

[0027] Preferably, the specific implementation of the GA algorithm is: importing the geatpy genetic algorithm library and setting initial parameters, which include population size, number of iterations and individual dimensions; using the fitness function to calculate the fitness value of the individual; sampling and optimizing the individual's better fitness value through the differential evolution algorithm template in the library until the iteration termination condition is met; screening the chromosome with the highest fitness according to the size of the fitness and performing a decoding operation to output the optimal solution.

[0028] Preferably, the fitness function is expressed as follows:

[0029]

[0030] In the formula is the predicted value of the forward model, y i is the target performance value.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. The present invention collects data on the composition, processing technology and mechanical properties of magnesium alloys, and uses machine learning algorithms to analyze the implicit structure-activity relationship between the composition, processing technology and mechanical properties of magnesium alloys, thereby realizing the efficient design of new high-performance magnesium alloys based on mechanical performance requirements. The present invention optimizes the input value of the forward model through a sampling optimization algorithm, with the goal of reducing the error between the predicted value of the forward model and the target performance. Then, it continuously searches until it finds an input value that makes the error reach the current optimal value. Through the combination of the two, the accuracy of the model can be effectively improved (the calculation accuracy can reach 99%) and the deviation can be reduced (the error is only 0.5%), which is conducive to the efficient optimization design of new high-performance magnesium alloys and matching process systems. It also provides guidance for the efficient development and design of high-performance magnesium alloys based on machine learning methods, which is of great significance to the application and development of magnesium alloys.

[0033] 2. The present invention provides a method for inverse design of alloy composition and process, guided by actual mechanical property requirements. This method not only considers alloy composition but also subsequent processing techniques and parameters. This method is more comprehensive and reliable, with a simple and easy-to-implement calculation method. It enables "on-demand design" of high-performance magnesium alloys. Parameters such as alloy composition are optimized, not model structure. This significantly reduces modeling time, improves alloy development efficiency, and reduces experimental costs. The present invention combines an optimization algorithm with a forward model to optimize overall model performance, achieving good fitting results even with small amounts of alloy data and complex process combinations. This design method addresses the computational complexity, low efficiency, low accuracy, and weak generalization capabilities of simple model inverse training. Compared to simple model inverse training methods, it offers superior model performance, offers significant advantages for the development of high-performance magnesium alloys, and is an effective strategy to assist in magnesium alloy research and development. Overall, the present invention improves magnesium alloy design efficiency and provides inverse design guided by actual performance requirements. This not only addresses the issue of limited magnesium alloy data but also addresses the currently unseen inverse design problem of mechanical properties to alloy composition in the magnesium alloy field. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Schematic diagram of the process of the reverse design method of the high performance magnesium alloy of the present invention.

[0035] Figure 2 These are the evaluation indicators of XGBoost and RF in the Mg-Gd-Y magnesium alloy test set in the embodiment of the present invention; A~C are the prediction results of XGBoost for YTS, UTS, and EL, respectively, and D~F are the prediction results of EF for YTS, UTS, and EL, respectively.

[0036] Figure 3 The actual mechanical properties diagram of the magnesium alloy predicted by the embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present invention will be further described in detail below with reference to the 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.

[0038] Example

[0039] In this example, Mg-Gd-Y magnesium alloy is selected as a candidate alloy, and the target performance is UTS = 400MPa; EL = 10%. The alloy composition and processing technology are designed through the reverse design model. Figure 1 As shown: According to the composition of the reverse design model, the modeling of the following parts is mainly carried out: data preparation, establishment of the forward prediction model, selection and establishment of the sampling optimization model, and construction of the reverse design model. The specific implementation steps are as follows:

[0040] 1. Data Collection

[0041] 1) Collect historical data related to Mg-Gd-Y magnesium alloys and build an original database of magnesium alloys, where the database content includes parameters such as alloy element content, heat treatment process parameters, deformation process parameters and mechanical properties.

[0042] 2) Analyze the data distribution and data quality based on the original database, and then perform data preprocessing, including missing value processing, duplicate value deletion, and outlier processing. Specifically, delete outliers and missing values ​​in alloy performance reports, and for performance data of the same composition reported in different literature reports, determine the relative error of each reported value, average the duplicate values ​​whose relative error is less than the error threshold, and delete the duplicate values ​​whose relative error is greater than the error threshold.

[0043] 3) Feature selection was performed by comprehensively evaluating the database using various methods, including Pearson correlation analysis and feature importance analysis based on a tree-structured model. To uniformly scale the data for each feature in the dataset and mitigate the impact of large data dispersion on model fitting, the solution and extrusion temperatures were scaled by 20°C intervals for the solution temperature and 30°C intervals for the extrusion temperature. This ultimately resulted in a magnesium alloy dataset of relatively high quality.

[0044] 2. Establishment of Forward Prediction Model

[0045] 1) Before training the machine learning model, the obtained magnesium alloy dataset was divided into a training set and a test set, with the ratio of the training set to the test set being 8:2.

[0046] 2) Since each feature in the dataset has obvious differences, in order to compare and weight indicators of different units or magnitudes during the model fitting process, the training set is standardized to scale its features. The specific implementation method is as follows: Where x represents each feature sample, μ represents the mean of each feature sample set, and σ represents the standard deviation of each feature sample set.

[0047] 3) Based on the training set, using alloy composition and processing technology as input and performance as output, a machine learning forward prediction model from alloy composition and processing technology to mechanical properties was constructed using Extreme Gradient Boosting (XGBoost) and Random Forest (RF). The performance of the forward prediction model on the training set was evaluated using cross-validation.

[0048] 4) The machine learning algorithm includes some hyperparameters that affect the model fitting effect. In order to improve the model fitting effect, a greedy algorithm is used to adjust the hyperparameters of the forward prediction model. That is, each time the parameters are adjusted, the parameter with the greatest impact on the performance of the forward prediction model is selected for adjustment until the optimal performance is achieved.

[0049] 5) In order to test the decision-making ability of the machine learning model fitted on the training set on data that has not been seen before - generalization performance, its generalization performance is evaluated on the test set, and its evaluation models are XGBoost and RF.

[0050] The generalization performance results of the XGBoost model and RF model on the test set are as follows Figure 2 As shown in the figure, according to the evaluation indicators R2 and MAE of the XGBoost model and the RF model, the two models have similar performance in yield strength and tensile strength. However, the XGBoost model has poor generalization performance for EL, with an R2 of only about 0.77, while the R2 of the RF model is about 0.87. Therefore, the present invention selects RF as the main model.

[0051] 3. Selection of Sampling Optimization Model

[0052] Particle swarm optimization (PSO) and genetic algorithm (GA) are selected as sampling optimization algorithms to perform parameter optimization.

[0053] The specific implementation of the PSO algorithm is as follows:

[0054] 1) Setting initial parameters, including particle swarm size, particle dimension, and number of iterations; particle swarm size refers to a range of magnesium alloy composition and processing parameters; particle dimension refers to the number of dimensions of the selected initial particles, and the dimension setting is related to the parameters of the alloy combination to be designed; the number of iterations refers to the set condition, that is, how many times the particle is updated before the update ends. After the update ends, the particle with the best fitness among these several iterations is selected.

[0055] 2) Randomly initialize the position and velocity of the particle and calculate the fitness function value of the initialized particle;

[0056] 3) Update the particle's velocity and position according to equations (3) and (4), calculate the fitness value of each particle, update the historical optimal position and optimal fitness of each individual, and update the historical optimal position and fitness value of each group; the particle's position refers to a solution to the problem (such as the specific values ​​of the alloy composition and processing parameters); the particle's velocity is a position vector, which refers to the distance and direction of each particle update.

[0057]

[0058]

[0059] In the formula is the historical optimal position of particle i in the d-th dimension in the k-th iteration; is the historical optimal position of the d-th dimension of the group in the k-th iteration, ω is the inertia factor, r1 and r2 are random numbers between (0,1), and c1 and c2 are learning factors;

[0060] 4) Determine whether the maximum number of iterations has been reached or whether the minimum value of the fitness difference between the two generations of particles has been reached. If so, the iteration ends and the optimal position of the individual is output; if not, repeat step 3) until the stopping condition is met and the optimal solution is output.

[0061] The specific implementation method of the GA algorithm is as follows: import the geatpy genetic algorithm library and set the initial parameters, which include population size, number of iterations and individual dimensions; use the fitness function to calculate the fitness value of the individual; sample and optimize the individual's better fitness value through the differential evolution algorithm template in the library until the iteration termination condition is met; select the chromosome with the highest fitness according to the size of the fitness and perform decoding operation to output the optimal solution.

[0062] 4. Construction of reverse design model

[0063] Determine a parameter space, randomly sample in this space through the selected sampling optimization algorithm, and then input the sampling results into the forward prediction model after the same standardization processing as the forward model training set to obtain the corresponding performance prediction value, and construct a reverse design model of mechanical properties → alloy composition and processing technology. According to the optimization algorithm category, the reverse design model is divided into PSO-RF and GA-RF.

[0064] 5. Determination of alloy design

[0065] Using the root mean square error (RMSE) between the predicted performance and the target performance as the fitness function, the PSO-RF and GA-RF algorithms were used to determine the optimal solution that minimized the fitness function. This optimal solution was the combination of alloy composition and processing parameters that matched the target performance. Since similar performance values ​​can lead to different design results, to ensure the accuracy of the model, the design was repeated five times. The average value was taken as the model's accuracy, and the design with the lowest fitness was selected as the final result. Based on the design results, the PSO-RF and GA-RF models achieved an accuracy of 99% when yield strength was not constrained. The final design results are shown in Table 1. The alloy composition and processing parameters were also introduced into the forward prediction model, resulting in performance predictions, as shown in Table 2.

[0066] Table 1

[0067]

[0068] Table 2

[0069]

[0070] According to the model design results, since the fitness of the design result of GA-RF is smaller than that of PSO-RF, the design result of GA-RF is finally selected.

[0071] 6. Verification

[0072] Combined with the actual experimental conditions, the alloy composition and processing parameters were as close as possible to the model design results to prepare the Mg-Gd-Y-Zn-Zr alloy. Table 3 shows the actual composition of the experimentally prepared alloy and the related processing parameters.

[0073] Table 3

[0074]

[0075] According to the above composition and process parameters, the prepared alloy tensile specimens were obtained and their room temperature tensile mechanical properties were tested. The results are as follows: Figure 3 and as shown in Table 4.

[0076] Table 4

[0077]

[0078] It can be seen from the results that the error between the tensile strength of the Mg-Gd-Y-Zn-Zr alloy designed by the GA-RF of the present invention and the target strength is only 0.5%.

[0079] The forward model RF was used to predict the mechanical properties of the actual alloy, and the results are shown in Table 5. The errors between the predicted values ​​of the tensile yield strength, tensile strength, and elongation of the experimental alloy and the actual measured values ​​were 18 MPa, 4 MPa, and 5.5%, respectively. Therefore, the forward prediction part of the present invention has a good predictive effect on the mechanical properties of the alloy.

[0080] Table 5

[0081]

[0082] In summary, the performance-oriented magnesium alloy reverse design model of the present invention provides a fast and effective method for achieving efficient optimization design of new high-performance magnesium alloys and matching process systems.

[0083] The above description is only a preferred embodiment of the present invention and does not 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 scope of protection of the present invention.

Claims

1. A reverse design method for high-performance magnesium alloys based on machine learning, characterized in that: The following steps are involved: S1: Data preparation: Collect historical data related to magnesium alloys, build a magnesium alloy original database, and then perform data preprocessing and feature selection to obtain a magnesium alloy dataset; S2: Establishment of a forward prediction model: Divide the magnesium alloy dataset obtained in step S1 into a training set and a test set, and standardize the training set. Using alloy composition and processing technology as input and mechanical properties as output, a machine learning forward prediction model of alloy composition, processing technology → mechanical properties is constructed, and the accuracy of the forward prediction model is evaluated using the test set. S3: Establishment and selection of sampling optimization model: Select particle swarm optimization and genetic algorithm as sampling optimization algorithms, establish the model by self-programming, perform parameter optimization, and select the appropriate optimization algorithm by comparing the optimization effects of the two; S4: Construction of the reverse design model: Determine a parameter space, and randomly sample within this space using the sampling optimization algorithm selected in step S3. The sampling results are normalized in the same way as the forward model training set and then input into the forward prediction model to obtain the corresponding performance prediction values. A reverse design model of mechanical properties → alloy composition and processing technology is constructed. S5: Alloy design: The root mean square error (RMS) between the predicted performance value and the target performance value is used as the fitness function. The optimal solution that minimizes the fitness function is obtained through iterative solution using a sampling optimization algorithm. This optimal solution is the combination of alloy composition and processing parameters that matches the target performance.

2. The reverse design method for high-performance magnesium alloys based on machine learning according to claim 1, characterized in that: The data in step S1 include alloy element content, heat treatment process parameters, deformation process parameters and mechanical properties.

3. The reverse design method for high-performance magnesium alloys based on machine learning according to claim 1, characterized in that: The data processing in step S1 includes missing value processing, duplicate value deletion or outlier processing; the feature selection includes range scaling processing of the solution temperature and the extrusion temperature.

4. The reverse design method for high-performance magnesium alloys based on machine learning according to claim 3, characterized in that: The solution temperature is scaled in intervals of 20°C, and the extrusion temperature is scaled in intervals of 30°C.

5. The reverse design method of high-performance magnesium alloy based on machine learning according to claim 1, characterized in that: The ratio of the training set to the test set in step S2 is 8:2; the training set is standardized by formula (1) Where x represents each feature sample, μ represents the mean of each feature sample set, and σ represents the standard deviation of each feature sample set.

6. The reverse design method for high-performance magnesium alloy based on machine learning according to claim 1, characterized in that: The methods constructed in step S2 include extreme gradient boosting trees and random forests.

7. The reverse design method for high-performance magnesium alloy based on machine learning according to claim 1, characterized in that: Step S2 also includes using a greedy algorithm to adjust the hyperparameters of the forward prediction model, that is, each time a parameter is adjusted, a parameter that has the greatest impact on the model performance is selected for adjustment until the optimal performance is achieved.

8. The reverse design method for high-performance magnesium alloys based on machine learning according to claim 1, characterized in that: The specific implementation of the PSO algorithm is as follows: 1) Setting initial parameters, including particle swarm size, particle dimension, and number of iterations; 2) Randomly initialize the position and velocity of the particle and calculate the fitness function value of the initialized particle; 3) Update the particle speed and position according to equations (3) and (4), calculate the fitness value of each particle, update the historical optimal position and optimal fitness of each individual, and update the historical optimal position and fitness value of each group; In the formula is the historical optimal position of particle i in the d-th dimension in the k-th iteration, is the historical optimal position of the d-th dimension of the group in the k-th iteration, ω is the inertia factor, r1 and r2 are random numbers between (0,1), and c1 and c2 are learning factors; 4) Determine whether the maximum number of iterations has been reached or whether the minimum value of the fitness difference between the two generations of particles has been reached. If so, the iteration ends and the optimal position of the individual is output; if not, repeat step 3) until the stopping condition is met, and then output the optimal solution.

9. The reverse design method for high-performance magnesium alloys based on machine learning according to claim 1, characterized in that: The specific implementation of the GA algorithm is as follows: import the geatpy genetic algorithm library and set initial parameters, including population size, number of iterations and individual dimensions; use the fitness function to calculate the fitness value of the individual; sample and optimize the individual's better fitness value through the differential evolution algorithm template in the library until the iteration termination condition is met; screen the chromosome with the highest fitness according to the size of the fitness and perform decoding operation to output the optimal solution.

10. The reverse design method of high-performance magnesium alloy based on machine learning according to claim 1, characterized in that: The fitness function is expressed as follows: In the formula is the predicted value of the forward model, y i is the target performance value.

Images