CN118262844A - Integrated learning-based high-hardness high-entropy alloy component design method - Google Patents

Abstract

The invention relates to the technical field of high entropy alloy hardness prediction, and more specifically to a high hardness high entropy alloy composition design method based on integrated learning. The design method comprises the following steps: obtaining a data set for predicting the hardness of the high entropy alloy based on a high entropy alloy system, then self-encoding the composition data in the initial data set, obtaining the hardness distribution latent space of the high entropy alloy, sampling the latent space using a Gaussian mixture distribution model, obtaining sample points involved in the prediction, then modeling different feature combinations of multiple machine learning, constructing multiple models, training and evaluating the models, screening out qualified models and then forming an integrated model, combining multiple machine learning models to form an integrated model by using an integrated learning method, predicting the participating sample points, then preparing and testing the alloy, and finally realizing the design of the high hardness high entropy alloy composition.

Description

A method for designing high-hardness and high-entropy alloy composition based on ensemble learning

Technical Field

The invention relates to the technical field of high entropy alloy hardness prediction, and more specifically to a high-hardness high entropy alloy composition design method based on integrated learning.

Background technique

High entropy alloys are a class of alloys composed of five or more main elements in similar atomic percentages. These alloys have attracted widespread attention due to their unique microstructures and excellent physical, chemical and mechanical properties. Compared with traditional alloys, high entropy alloys exhibit properties such as high hardness, excellent corrosion resistance and good high temperature performance, making them potentially valuable for application in aerospace, military, automotive manufacturing and energy fields.

Despite the many advantages of high-entropy alloys, there are still some challenges in alloy design. Traditional alloy design methods are often based on trial and error, which is not only time-consuming and costly, but also difficult to systematically explore and optimize complex composition spaces.

The role of data-driven methods represented by machine learning in alloy research and development is becoming increasingly prominent. By establishing a complex relationship between input features and material targets, machine learning methods can achieve rapid prediction of material properties, and thus play an important role in guiding the design of materials including high entropy alloys.

The Chinese patent announcement number is CN116092604B, which proposes a data-driven high-strength, high-toughness refractory high-entropy alloy and preparation method. The method can mine data set information from the high-entropy alloy data set and predict the composition points that may have the target properties, which greatly accelerates the design and optimization of high-entropy alloy composition. However, the current machine learning methods in materials science mainly use a single model to model and predict material properties or chemical structures. Although these methods have achieved certain prediction effects, the prediction results are not accurate enough, and the above methods can only make simple predictions, which is still difficult in designing the specific composition of high-entropy alloys. Although machine learning algorithms have excellent learning and modeling prediction capabilities, these models can only guarantee the prediction accuracy within the composition distribution range when the training data set is less than 300, and cannot explore unknown space. Ensemble learning methods can integrate multiple machine learning models to obtain better prediction results than a single machine learning machine.

In summary, the present invention develops a high-hardness and high-entropy alloy composition design method based on ensemble learning, which has important theoretical significance and application value, and plays a vital role in promoting the development of the field of materials science and realizing the rapid iteration of new high-hardness and high-entropy alloys.

Summary of the invention

The present invention provides a high-hardness high-entropy alloy composition design method based on ensemble learning, which obtains the latent space of the hardness distribution of the high-entropy alloy through a differential autoencoder algorithm, and uses a Gaussian mixture model combined with a Markov chain Monte Carlo method to sample the latent space to select sample points participating in candidate screening; then, an ensemble learning method is used to combine multiple machine learning models, and the participating sample points are predicted, and finally high-hardness high-entropy alloy composition points are selected through an efficiency function, so as to solve the problems of low prediction accuracy of a single machine learning model, difficulty in finding high-hardness high-entropy alloy composition points, and difficulty in exploring unknown high-entropy alloy composition space.

The specific technical solutions of the present invention are as follows:

A high-hardness high-entropy alloy composition design method based on ensemble learning, the design method steps are as follows:

S1: Based on the high entropy alloy system, a data set for predicting the hardness of high entropy alloys is obtained. The high entropy alloy system is the Al-Co-Cr-Cu-Fe-Ni system. Based on this system, multiple components and their corresponding hardness data are collected, and combined with the physical characteristics of the high entropy alloy, a data set is formed for further analysis. The number of data sets is 100-300;

Twenty physical characteristics of high entropy alloys are used to describe the basic properties of the alloys and the factors affecting their performance. These characteristics are atomic radius difference (δr), electronegativity difference (Δχ), valence electron concentration (VEC), mixing enthalpy (ΔH), configuration entropy (ΔS), Ω parameter (Ω), Λ parameter (Λ), γ parameter (γ), local electronegativity mismatch (D.χ), number of mobile electrons (e/a), cohesive energy (Ec), modulus mismatch (η), local size mismatch (D.r), energy term (A), Navarro coefficient (F), work function (W), shear modulus (G), shear modulus difference (δG), local modulus mismatch (D.G), and lattice distortion energy (μ). The calculation of these features relies on the element physical property values provided in the existing literature (Hardness prediction of Al CoCrCuFeNi series high entropy alloy based on machine learning, author: Zou Rui), and is calculated according to its formula; and integrated with the composition data to form an initial data set for further analysis;

S2: Then, the composition data in the data set is autoencoded to obtain the latent space of hardness distribution of high entropy alloy;

S3: Use the Gaussian mixture distribution model to sample the latent space, and select 1000-3000 samples to obtain sample points involved in prediction;

S4: Select different feature combinations for multiple machine learning models, build multiple models, train and evaluate the models, screen out qualified models and then form an integrated model;

S5: predict the participating sample points according to the integrated model to obtain the hardness prediction results;

S6: Use the efficiency function to sort and select experimental points to prepare samples, and then test the hardness of the sample points, where the hardness value must reach above 800HV; if the tested hardness value is lower than the 800HV standard, repeat the previous S2 to S5, and use the updated data set to re-train, verify and screen the model to optimize the alloy composition design and ultimately achieve the design of high-hardness high-entropy alloy composition; otherwise, if the hardness value is higher than the 800HV standard, further performance evaluation and application exploration will be carried out.

As a technical solution of the present invention, in S1, based on the Al-Co-Cr-Cu-Fe-Ni system, 278 components and their corresponding hardness data are collected, combined with the physical characteristics of high entropy alloys, there are probably several data under 20 physical characteristics, and the combined data is 278 data. Each piece of data includes three parts, the first part is 6 components, the second part is 20 characteristics, and the third part is hardness, which is the target feature. Therefore, a total of 278 rows x 27 columns are configured to form the initial data set; and the initial data set is divided into a training set and a test set, wherein the training set and the test set are randomly allocated in a ratio of 4:1, with a total of 222 training sets and 56 test sets.

As a technical solution of the present invention, in S2, the potential representation of the Al-Co-Cr-Cu-Fe-Ni high entropy alloy data set is learned by differential autoencoding. The encoder and decoder parts of the differential autoencoder are both represented by neural networks. The function of the encoder is to accept input data and convert it into a lower-dimensional latent space, and reconstruct these latent representations back to the original high-dimensional data through the decoder; the input data passes through the encoder of the differential autoencoder, and the component data is mapped to a two-dimensional latent space, and then the two-dimensional latent space data is mapped back to the original component data through the decoder, and the training parameters of the differential autoencoder are adjusted by comparing the loss.

As a technical solution of the present invention, in S2, the latent space is divided into a high hardness area greater than 600 HV and other areas, and a neural network classifier is trained by a neural network method, and the area greater than 600 HV is set as a high hardness area, and the area less than 600 HV is set as a low hardness area.

As a technical solution of the present invention, the grid training parameters of the neural network classifier are set as follows: the initial learning rate is set to 0.0001, the training batch size is set to 32, and the number of training rounds is set to 400.

As a technical solution of the present invention, in S3, an initial sample is randomly selected from the Gaussian mixture model and 10,000 iterations are performed. In each iteration, a suggested next sample is first generated based on the current sample using the multivariate normal distribution method, and the current sample and the suggested sample are merged and passed to the classifier trained in S2. The classifier outputs the classification probability of the two samples. If the acceptance probability of the suggested sample is higher than that of the current sample, the suggested sample is retained as the sample for the next iteration, otherwise it is discarded. Through this method, 1,352 sample points participating in subsequent screening are obtained.

As a technical solution of the present invention, in S4, the models involved in the integration include but are not limited to SVR, CatBoost, LightGBM, Back Propagation Neural Networks, Random Forest, XGBoost, and AdaBoost models.

As a technical solution of the present invention, in S4, a method of iteratively adding features is used to train the model under different training set distributions. The iterative addition of features starts from an initial subset, which only contains the components of the elements, and then gradually adds new features, only one feature is added each time. In each iteration, the final feature set of the model is selected by the change of the RMSE value of the model; the RMSE calculation process is as follows:

Where _yi represents the true hardness, represents the prediction hardness of the model, and n represents the number of samples.

As a technical solution of the present invention, in S5, multiple machine learning models trained in S4 are used to predict the hardness values of the high entropy alloy component points involved in the screening, and the average of the predicted hardness values provided by the integrated model and the standard deviation of the predicted hardness values are calculated.

As a technical solution of the present invention, in S6, the performance function is an upper confidence limit (UCB) function, and the calculation formula is:

UCB(x)＝ μ(x)+ κσ(x) (3)

Where μ(x) is the predicted mean of the component point, and σ(x) is the predicted standard deviation of the component point. In order to take into account both the utilization of the model and the development of unknown points, k is set to 0.2.

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

1. The present invention obtains the latent space of the hardness distribution of high entropy alloys by adopting a differential autoencoder algorithm, and uses a Gaussian mixture model combined with a Markov chain Monte Carlo method to sample the latent space to select sample points participating in candidate screening. After that, an integrated learning method is used to combine multiple machine learning models to form an integrated model, and the participating sample points are predicted, and then the alloy is prepared and tested, and finally the design of high-hardness high-entropy alloy composition is realized.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of high entropy alloy composition design of the present invention;

FIG2 is a two-dimensional latent space diagram of a high entropy alloy composition data set of the present invention;

FIG3 is a graph of Spearman rank correlation coefficients between different features of the present invention;

FIG4 is a graph showing the RMSE results of multiple machine learning models of the present invention under different training set ratios.

Detailed ways

The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

Embodiment 1:

Step 1: As shown in Figures 1-4, the high entropy alloy system in the present invention is an Al-Co-Cr-Cu-Fe-Ni system, and 278 components and their corresponding hardness data are collected based on this system.

At the same time, 20 physical features were identified to describe the basic properties of the alloy and the factors affecting its performance, which are atomic radius difference (δr), electronegativity difference (Δχ), valence electron concentration (VEC), mixing enthalpy (ΔH), configuration entropy (ΔS), Ω parameter (Ω), Λ parameter (Λ), γ parameter (γ), local electronegativity mismatch (D.χ), number of mobile electrons (e/a), cohesive energy (Ec), modulus mismatch (η), local size mismatch (D.r), energy term (A), Nabarro coefficient (F), work function (W), shear modulus (G), shear modulus difference (δG), local modulus mismatch (D.G), lattice distortion energy (μ). The calculation of these features relies on the element physical property values provided in the literature (Hardness prediction of Al CoCrCuFeNi series high entropy alloy based on machine learning, author: Zou Rui), and is calculated according to its formula; and integrated with the composition data to form an initial data set for further analysis.

Step 2: Autoencode the composition data in the data set to obtain the latent space of hardness distribution of high entropy alloy. The latent space is divided into high hardness areas greater than 600 HV and other areas.

It should be noted that the component dataset proposed in this article refers to the component part of 278 data, and the corresponding latent space refers to the two-dimensional space mapped from the high-dimensional space by the encoder of the differential autoencoder.

In the present invention, differential autoencoders are used to learn the potential representation of Al-Co-Cr-Cu-Fe-Ni high entropy alloy data sets. The role of the encoder is to accept input data and convert it into a lower dimensional potential space. In this process, the encoder attempts to retain the important features of the data, which are necessary for reconstructing the input data. The neural network decoder receives the encoding in the potential space and attempts to reconstruct the input data. The goal of the decoder is to generate an output as close as possible to the original input, so as to learn the effective representation of the data, and then reconstruct these potential representations back to the original high-dimensional data through the decoder. The encoder and decoder parts of the differential autoencoder are both represented by neural networks. Then, the optimal encoder and decoder network parameters can be obtained by optimizing the loss function, and then the variational parameters are used to sample and reconstruct the signal; in a specific embodiment, the training parameters of the network are set: the initial learning rate is set to 0.001, the weight decay value is set to 0.0001, the training batch size is set to 64, and the number of training rounds is set to 600.

Loss = Loss _{{reconstruction loss}} + Loss _{{DL divergence}} (1)

The input data passes through the encoder of the differential autoencoder to map the component data to a two-dimensional latent space, and then the two-dimensional latent space data is mapped back to the original component data through the decoder. The training parameters of the differential autoencoder are adjusted by comparing the final loss.

Finally, the two-dimensional latent space diagram of the component data distribution in this specific embodiment is shown in Figure 2. The component points of different systems produce relatively obvious boundaries in the two-dimensional latent space.

After that, the present invention trained a neural network classifier through a neural network method, set the area greater than 600HV as a high hardness area, and set the area less than 600HV as a low hardness area. We randomly allocated the training set and the test set in a ratio of 4:1, with a total of 222 items in the training set and 56 items in the test set. The grid training parameters of the classifier are set as follows: the initial learning rate is set to 0.0001, the training batch size is set to 32, the number of training rounds is set to 400, and five-fold cross validation is performed. The accuracy of the neural network classifier on the training set is 0.89, and the accuracy on the test set is 0.85. In the existing materials disciplines, it is assumed that 0.6 or above has a good accuracy.

Step 3: Use the Gaussian mixture distribution model to sample the latent space, and select 1000-3000 samples to obtain sample points involved in prediction;

The present invention uses the component data set of the system to train the Gaussian mixture model; in order to determine the most suitable number of Gaussian components in the Gaussian mixture model, this embodiment uses the elbow algorithm, which evaluates the impact of different numbers of components on the goodness of fit of the model and finds an inflection point, that is, adding more Gaussian components no longer significantly improves the goodness of fit of the model. According to the results of the elbow algorithm, the optimal number of Gaussian components is determined to be 6, and the average negative log likelihood is -1.16.

After determining the optimal number of Gaussian components, the present invention further uses the Markov Chain Monte Carlo method to sample the Gaussian mixture model (the Gaussian mixture model when the number of Gaussian components is 6).

This embodiment first randomly extracts an initial sample from the Gaussian mixture model and performs 10,000 iterations. In each iteration, a suggested next sample is first generated based on the current sample using the multivariate normal distribution method, and the current sample and the suggested sample are merged and passed to the classifier trained in step 2. The classifier outputs the classification probability of the two samples. If the acceptance probability of the suggested sample is higher than that of the current sample, the suggested sample is retained as the sample for the next iteration, otherwise it is discarded. Through this method, we obtained 1,352 sample points for subsequent screening, which are considered to be high-entropy alloy component points that may have high hardness values.

It should be noted that 10,000 iterations refer to an initial sample. A recommended sample is generated after one iteration. If it is retained, the recommended sample is used as the initial sample for the next time. The next recommended sample is generated after another iteration, for a total of 10,000 times. If it is still retained after 10,000 iterations, it is used as a sample point. Through measurement, a total of 1,352 sample points are retained.

Step 4: Select different feature combinations for multiple machine learning models, build multiple models, train and evaluate the models, screen out qualified models and then form an integrated model;

The present invention uses machine learning methods such as SVR, CatBoost, LightGBM, Back Propagation Neural Networks, Random Forest, XGBoost (decision tree classifier), AdaBoost, etc.

First, the Spearman rank correlation coefficient (Spearman) is used to screen out features with a correlation greater than 0.95. The Spearman rank correlation coefficient is used to evaluate the correlation between two variables, that is, their degree of correlation and the direction in which the variable values change in the same way. Its value ranges from -1 to +1. The closer the absolute value is to 1, the greater the correlation. Because there are highly correlated features in the feature data, one of them can be used to replace the other; the Spearman rank correlation coefficient diagram between different features is shown in Figure 3.

It should be noted that the "two variables" mentioned above refer to two variables in the concept, and what is actually involved is the comparison of characteristics other than composition and hardness.

Then, the iterative feature addition method was used to train the model under different training set distributions. The iterative feature addition started from the initial subset, which only contained the components of the elements, and then gradually added new features, one feature at a time. In each iteration, the final feature set of the model was selected by the change of the model RMSE value. For each machine learning method, we trained 100 machine models and selected the top 20 models with the lowest RMSE values as the integrated model. For each machine learning algorithm, we built and trained 100 different model instances. These models include SVR, CatBoost, LightGBM, Back Propagation Neural Networks, RandomForest, and XGBoost, each of which was applied in different combinations of feature sets and training distributions. After training, we used the root mean square error (RMSE) as a performance indicator to evaluate these models. RMSE is a commonly used metric that measures the difference between the model's predicted value and the actual observed value. A lower RMSE value usually means that the model has better prediction accuracy. The final RMSE result graph of each model at different scales is shown in Figure 4.

Where _yi represents the true hardness, represents the prediction hardness of the model, and n represents the number of samples.

After training and evaluating all models, we selected the top 20 models with the lowest RMSE values from the 100 models built by each machine learning method. These models were considered the best performing models and were selected to build the final ensemble model. The ensemble model improves the accuracy and robustness of the prediction by combining the prediction results of multiple models. In this way, we hope that the ensemble model can combine the advantages of different single models, reduce possible overfitting, and ultimately provide more reliable predictions when facing unknown data.

Step 5: Select different feature combinations for multiple machine learning models, build multiple models, train and evaluate the models, select qualified models and then form an integrated model;

According to an embodiment of the present invention, the present invention uses multiple machine learning models trained in step 4 to predict the hardness values of the high entropy alloy component points involved in the screening. In order to predict the hardness value of each specific high entropy alloy component point, we use an integrated learning method.

The core advantage of ensemble learning technology lies in its multi-model fusion strategy. In the present invention, multiple different machine learning models are organically integrated, and each model independently learns and captures different characteristics and inherent laws of the data during the training process. In the prediction stage, by comprehensively considering the output results of each model, not only the advantageous information in a single model is extracted, but also the possible limitations of any model are avoided, thereby optimizing the overall prediction performance.

Specifically, for each component point, we not only calculated the mean of the predicted hardness values provided by the ensemble model, but also calculated the standard deviation of these predicted values. The predicted mean gives the model's collective prediction of the hardness of the component point, while the predicted standard deviation provides a quantitative indicator of the uncertainty of the prediction. For each machine learning method, there are 20 machine learning models, so for this example, a total of 180 machine learning models are involved in the prediction of the hardness of the component point.

Step 6: Select experimental points for preparation according to the efficiency function ranking and test the hardness of the sample points;

When the hardness obtained in step six does not meet the preset requirement, the characteristic data of the alloy composition is added to the data set, and steps four to five are repeated until the hardness obtained in step six meets the preset requirement.

According to an embodiment of the present invention, the performance function in the present invention is an upper confidence bound (UCB) function.

UCB(x)＝ μ(x)+ κσ(x) (3)

μ(x) is the predicted mean of the component point, and σ(x) is the predicted standard deviation of the component point. In order to take into account both the utilization of the model and the development of unknown points, k is set to 0.2. UCB only represents a ranking that combines the predicted mean and standard deviation.

Finally, we selected Al45Co23Cr18Cu1Fe7Ni5, Al44Co15Cr22Cu5Fe8Ni6 and Al44Co16Cr15Fe13Ni 12 as the final composition points.

To ensure the high quality of the alloy, we strictly select metal materials with a purity of more than 99.9% as the raw materials of the alloy. The preparation process starts with grinding the surface of the raw materials to carefully remove the surface oxide layer that may affect the performance of the alloy. Subsequently, it is deeply cleaned by ultrasonic cleaning technology to remove any remaining impurities and contaminants, and then the cleaned materials are placed in an oven for thorough drying to eliminate potential problems caused by moisture.

After preparing the raw materials, we accurately calculate the mass of each metal raw material required based on the molar ratio of each alloy component. Then, the various metal raw materials are prepared after ultrasonic cleaning and accurate weighing for the next alloy melting process. We use induction melting to prepare the alloy, which can provide a uniform and controlled heating environment to promote the full fusion of the metal. In order to ensure the uniformity of the alloy composition, we repeatedly melt each alloy sample at least six times, and wait for the alloy to cool and turn it over after each melting to ensure that the components are evenly distributed in the alloy.

After the alloy is prepared, we conduct hardness tests on the obtained high entropy alloy samples. The purpose of the test is to ensure that the hardness value of the alloy meets our strict requirements for high-performance materials, that is, the hardness value must reach 800HV or above. If the measured hardness value meets this standard, then we can consider that the alloy preparation is successful and further performance evaluation and application exploration can be carried out.

If the hardness value does not reach 800HV, the actual measured hardness values of the three component points will be recorded and fed back to our machine learning data set. Such a data feedback mechanism can help us continuously improve and adjust the algorithm model and further improve the accuracy of alloy design through iterative learning.

Next, we will repeat steps 2 to 5 and use the updated data set to retrain, validate, and screen the model to optimize the alloy composition design. Through this dynamic iterative method, we gradually approach the ideal alloy ratio and ultimately achieve the design of high-hardness high-entropy alloy composition.

The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein.

Claims (10)

1. A high-hardness high-entropy alloy composition design method based on ensemble learning, characterized in that the design method steps are as follows: S1: Based on the high entropy alloy system, a data set for predicting the hardness of high entropy alloys is obtained. The high entropy alloy system is the Al-Co-Cr-Cu-Fe-Ni system. Based on this system, multiple components and their corresponding hardness data are collected, and combined with the physical characteristics of the high entropy alloy to form a data set for further analysis. The number of data sets is 100-300; S2: Then, the composition data in the data set is autoencoded to obtain the latent space of hardness distribution of high entropy alloy; S3: Use the Gaussian mixture distribution model to sample the latent space, and select 1000-3000 samples to obtain sample points involved in prediction; S4: Select different feature combinations for multiple machine learning models, build multiple models, train and evaluate the models, screen out qualified models and then form an integrated model; S5: predict the participating sample points according to the integrated model to obtain the hardness prediction results; S6: Use the efficiency function to sort and select experimental points to prepare samples, and then test the hardness of the sample points, where the hardness value must reach above 800HV; if the tested hardness value is lower than the 800HV standard, repeat the previous S2 to S5, and use the updated data set to re-train, verify and screen the model to optimize the alloy composition design and ultimately achieve the design of high-hardness high-entropy alloy composition; otherwise, if the hardness value is higher than the 800HV standard, further performance evaluation and application exploration will be carried out. 2. According to a high-hardness high-entropy alloy composition design method based on ensemble learning in claim 1, it is characterized in that: in S1, based on the Al-Co-Cr-Cu-Fe-Ni system, 278 components and their corresponding hardness data are collected, combined with the physical characteristics of the high-entropy alloy to form an initial data set; and the initial data set is divided into a training set and a test set, wherein the training set and the test set are randomly allocated in a ratio of 4:1, with a total of 222 items in the training set and 56 items in the test set. 3. According to a high-hardness high-entropy alloy composition design method based on ensemble learning as described in claim 1, it is characterized in that: in S2, the potential representation of the Al-Co-Cr-Cu-Fe-Ni high-entropy alloy data set is learned by differential autoencoding, and the encoder and decoder parts of the differential autoencoder are both represented by neural networks. The encoder is used to accept input data and convert it into a lower-dimensional latent space, and these latent representations are reconstructed back to the original high-dimensional data through the decoder; the input data passes through the encoder of the differential autoencoder, and the component data is mapped to a two-dimensional latent space, and then the two-dimensional latent space data is mapped back to the original component data through the decoder, and the training parameters of the differential autoencoder are adjusted by comparing the loss. 4. According to the high-hardness and high-entropy alloy composition design method based on ensemble learning in claim 3, it is characterized in that: in S2, the latent space is divided into a high hardness area greater than 600 HV and other areas, and a neural network classifier is trained by a neural network method, and the area greater than 600 HV is set as a high hardness area, and the area less than 600 HV is set as a low hardness area. 5. A high-hardness and high-entropy alloy composition design method based on ensemble learning according to claim 4, characterized in that: the grid training parameters of the neural network classifier are set as follows: the initial learning rate is set to 0.0001, the training batch size is set to 32, and the number of training rounds is set to 400. 6. According to the high-hardness and high-entropy alloy composition design method based on ensemble learning described in claim 5, it is characterized in that: in S3, an initial sample is randomly extracted from the Gaussian mixture model and 10,000 iterations are performed. In each iteration, a suggested next sample is first generated based on the current sample using the multivariate normal distribution method, and the current sample and the suggested sample are merged and passed to the classifier trained in S2. The classifier outputs the classification probability of the two samples. If the acceptance probability of the suggested sample is higher than that of the current sample, the suggested sample is retained as the sample for the next iteration, otherwise it is discarded. Through this method, 1,352 sample points participating in subsequent screening are obtained. 7. A high-hardness and high-entropy alloy composition design method based on ensemble learning according to claim 6, characterized in that: in S4, the models involved in the integration include but are not limited to SVR, CatBoost, LightGBM, Back Propagation Neural Networks, Random Forest, XGBoost, and AdaBoost models. 8. A high-hardness high-entropy alloy composition design method based on ensemble learning according to claim 7, characterized in that: in S4, the model is trained under different training set distributions using an iterative feature addition method, the iterative feature addition starts from an initial subset, the initial subset only contains the composition of the element, and then new features are added step by step, only one feature is added each time, and in each iteration, the final feature set of the model is selected by the change of the model RMSE value; the RMSE calculation process is as follows: Where _yi represents the true hardness, represents the prediction hardness of the model, and n represents the number of samples. 9. According to the high-hardness high-entropy alloy composition design method based on integrated learning in claim 8, it is characterized in that: in S5, multiple machine learning models trained in S4 are used to predict the hardness values of the high-entropy alloy component points involved in the screening, and the average of the predicted hardness values provided by the integrated model and the standard deviation of the predicted hardness values are calculated. 10. A high-hardness high-entropy alloy composition design method based on ensemble learning according to claim 9, characterized in that: in S6, the performance function is an upper confidence limit (UCB) function, and the calculation formula is: UCB(x)μ(+Kσ(x)(3) Where μ(x) is the predicted mean of the component point, and σ(x) is the predicted standard deviation of the component point. In order to take into account both the utilization of the model and the development of unknown points, k is set to 0.2.