Machine learning based room temperature high capacity body-centered cubic hydrogen storage alloy design method, device, computer equipment and storage medium

Abstract

This application relates to a machine learning-based method, apparatus, computer equipment, and storage medium for designing high-capacity body-centered cubic hydrogen storage alloys at room temperature. The method includes: extracting compositional features corresponding to candidate alloy composition data of the alloy to be designed; inputting the compositional features into a hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data; selecting candidate alloy composition data whose predicted hydrogen release capacity is greater than a preset hydrogen release capacity from the candidate alloy composition data, and using this as the alloy composition data to be verified; performing experimental verification processing on the alloy composition data to be verified to obtain the experimental verification results; and, if the experimental verification results indicate that the alloy composition to be verified has passed verification, obtaining the target alloy composition data of the alloy to be designed based on the alloy composition data to be verified. Using this method can improve the accuracy of determining the composition of body-centered cubic hydrogen storage alloys.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for designing high-capacity body-centered cubic hydrogen storage alloys at room temperature based on machine learning. Background Technology

[0002] Solid-state hydrogen storage technology has become one of the key technologies supporting the large-scale application of the hydrogen energy industry due to its advantages such as low hydrogen storage pressure, high safety, and high volumetric hydrogen storage density. Among the many solid-state hydrogen storage material systems, body-centered cubic hydrogen storage alloys are regarded as an important candidate system for next-generation room-temperature high-capacity hydrogen storage materials due to their high theoretical hydrogen storage capacity and excellent hydrogen absorption and desorption kinetics near room temperature.

[0003] In traditional technologies, empirical criteria (such as the difference in metal-hydrogen electronegativity and cell parameter thresholds) are generally used to determine the composition of body-centered cubic hydrogen storage alloys. However, this method relies on only limited physical property thresholds, which can easily screen out alloy components that do not meet the room temperature performance requirements, resulting in low accuracy in determining the composition of body-centered cubic hydrogen storage alloys. Summary of the Invention

[0004] Therefore, it is necessary to provide a machine learning-based method, apparatus, computer equipment, computer-readable storage medium, and computer program product for designing high-capacity, room-temperature body-centered cubic hydrogen storage alloys, which can improve the accuracy of determining the composition of body-centered cubic hydrogen storage alloys, in response to the above-mentioned technical problems.

[0005] In a first aspect, this application provides a machine learning-based design method for high-capacity room-temperature body-centered cubic hydrogen storage alloys, including:

[0006] In response to the alloy design instruction of the alloy to be designed, candidate alloy composition data of the alloy to be designed are determined, and the composition features corresponding to the candidate alloy composition data are extracted; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy.

[0007] The compositional features are input into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data.

[0008] From the candidate alloy composition data, the candidate alloy composition data with the predicted hydrogen release capacity greater than the preset hydrogen release capacity are selected as the alloy composition data to be verified for the alloy to be designed.

[0009] The composition data of the alloy to be verified is processed experimentally to obtain the experimental verification results of the composition data of the alloy to be verified.

[0010] If the experimental verification results indicate that the composition of the alloy to be verified has passed verification, the target alloy composition data of the alloy to be designed is obtained based on the composition data of the alloy to be verified.

[0011] In one embodiment, the trained hydrogen release capacity prediction model is obtained by training it in the following manner:

[0012] Obtain the sample alloy composition data of the sample alloy, and extract the sample composition features corresponding to the sample alloy composition data;

[0013] The sample component features are segmented to obtain the training dataset and test dataset for the hydrogen emission capacity prediction model to be trained.

[0014] Based on the training dataset, the hydrogen emission capacity prediction model to be trained is iteratively trained to obtain the trained hydrogen emission capacity prediction model.

[0015] The trained hydrogen emission capacity prediction model was validated based on the test dataset, and the validation results of the trained hydrogen emission capacity prediction model were obtained.

[0016] If the verification results meet the preset verification conditions, the trained hydrogen release capacity prediction model is obtained based on the trained hydrogen release capacity prediction model.

[0017] In one embodiment, extracting the sample composition features corresponding to the sample alloy composition data includes:

[0018] Extract the initial composition features corresponding to the sample alloy composition data;

[0019] The correlation coefficients between the initial component features are determined. Redundant component features with correlation coefficients greater than a preset correlation coefficient are selected from the initial component features and deleted from the initial component features to obtain the first candidate component features corresponding to the sample alloy composition data.

[0020] The importance of the first candidate component features is determined by the random forest algorithm, and a second candidate component feature with an importance greater than a preset importance is selected from each of the first candidate component features.

[0021] The second candidate component features are gradually eliminated using a feature recursive elimination algorithm to obtain the sample component features corresponding to the sample alloy composition data.

[0022] In one embodiment, the step of iteratively training the hydrogen emission capacity prediction model to be trained based on the training dataset to obtain the trained hydrogen emission capacity prediction model includes:

[0023] Based on the training dataset, the hydrogen emission capacity prediction model to be trained is iteratively trained to obtain an initial hydrogen emission capacity prediction model.

[0024] The training dataset is augmented to obtain an augmented training dataset. Based on the augmented training dataset, the initial hydrogen emission capacity prediction model is iteratively trained again to obtain the trained hydrogen emission capacity prediction model.

[0025] In one embodiment, before obtaining the sample alloy composition data of the sample alloy, the method further includes:

[0026] Obtain the elemental data corresponding to the initial alloy, and the frequency of occurrence of the elemental data in the initial alloy;

[0027] From the data of each element, select the target element data whose occurrence frequency is less than the preset occurrence frequency;

[0028] The sample alloy is obtained by deleting the initial alloys containing the target element data from each of the initial alloys.

[0029] In one embodiment, the step of performing experimental verification processing on the alloy composition data to be verified to obtain the experimental verification result of the alloy composition data to be verified includes:

[0030] The alloy composition data to be verified is subjected to alloy preparation verification processing to obtain the first verification result of the alloy composition data to be verified.

[0031] The composition data of the alloy to be verified is subjected to phase characterization verification processing to obtain a second verification result of the composition data of the alloy to be verified.

[0032] The alloy composition data to be verified is subjected to performance verification processing to obtain the third verification result of the alloy composition data to be verified.

[0033] Based on the first verification result, the second verification result, and the third verification result, the experimental verification results of the alloy composition data to be verified are obtained.

[0034] Secondly, this application also provides a machine learning-based room-temperature high-capacity body-centered cubic hydrogen storage alloy design device, comprising:

[0035] The feature extraction module is used to respond to the alloy design instruction of the alloy to be designed, determine the candidate alloy composition data of the alloy to be designed, and extract the composition features corresponding to the candidate alloy composition data; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy.

[0036] The capacity prediction module is used to input the composition features into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data.

[0037] The component screening module is used to screen out the candidate alloy component data whose predicted hydrogen release capacity is greater than the preset hydrogen release capacity from each of the candidate alloy component data, and use them as the alloy component data to be verified for the alloy to be designed.

[0038] The composition verification module is used to perform experimental verification processing on the composition data of the alloy to be verified, and to obtain the experimental verification results of the composition data of the alloy to be verified.

[0039] The composition determination module is used to obtain the target alloy composition data of the alloy to be designed based on the composition data of the alloy to be verified, when the experimental verification results indicate that the composition of the alloy to be verified has been verified.

[0040] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0041] In response to the alloy design instruction of the alloy to be designed, candidate alloy composition data of the alloy to be designed are determined, and the composition features corresponding to the candidate alloy composition data are extracted; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy.

[0042] The compositional features are input into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data.

[0043] From the candidate alloy composition data, the candidate alloy composition data with the predicted hydrogen release capacity greater than the preset hydrogen release capacity are selected as the alloy composition data to be verified for the alloy to be designed.

[0044] The composition data of the alloy to be verified is processed experimentally to obtain the experimental verification results of the composition data of the alloy to be verified.

[0045] If the experimental verification results indicate that the composition of the alloy to be verified has passed verification, the target alloy composition data of the alloy to be designed is obtained based on the composition data of the alloy to be verified.

[0046] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0047] In response to the alloy design instruction of the alloy to be designed, candidate alloy composition data of the alloy to be designed are determined, and the composition features corresponding to the candidate alloy composition data are extracted; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy.

[0048] The compositional features are input into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data.

[0049] From the candidate alloy composition data, the candidate alloy composition data with the predicted hydrogen release capacity greater than the preset hydrogen release capacity are selected as the alloy composition data to be verified for the alloy to be designed.

[0050] The composition data of the alloy to be verified is processed experimentally to obtain the experimental verification results of the composition data of the alloy to be verified.

[0051] If the experimental verification results indicate that the composition of the alloy to be verified has passed verification, the target alloy composition data of the alloy to be designed is obtained based on the composition data of the alloy to be verified.

[0052] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0053] In response to the alloy design instruction of the alloy to be designed, candidate alloy composition data of the alloy to be designed are determined, and the composition features corresponding to the candidate alloy composition data are extracted; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy.

[0054] The compositional features are input into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data.

[0055] From the candidate alloy composition data, the candidate alloy composition data with the predicted hydrogen release capacity greater than the preset hydrogen release capacity are selected as the alloy composition data to be verified for the alloy to be designed.

[0056] The composition data of the alloy to be verified is processed experimentally to obtain the experimental verification results of the composition data of the alloy to be verified.

[0057] If the experimental verification results indicate that the composition of the alloy to be verified has passed verification, the target alloy composition data of the alloy to be designed is obtained based on the composition data of the alloy to be verified.

[0058] The aforementioned machine learning-based design method, apparatus, computer equipment, storage medium, and computer program for high-capacity room-temperature body-centered cubic hydrogen storage alloys first responds to the alloy design instructions of the alloy to be designed, determines the candidate alloy composition data of the alloy to be designed, and extracts the composition features corresponding to the candidate alloy composition data; the alloy to be designed is used to represent the body-centered cubic hydrogen storage alloy; then, the composition features are input into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data; next, from each candidate alloy composition data, candidate alloy composition data with a predicted hydrogen release capacity greater than the preset hydrogen release capacity are selected as the alloy composition data to be verified for the alloy to be designed; then, the alloy composition data to be verified is experimentally verified to obtain the experimental verification results of the alloy composition data to be verified; finally, if the experimental verification results indicate that the alloy composition to be verified has passed the verification, the target alloy composition data of the alloy to be designed is obtained based on the alloy composition data to be verified. In this way, when determining the composition of body-centered cubic hydrogen storage alloys, candidate alloy composition data are first obtained and compositional features are extracted. Then, a trained hydrogen release capacity prediction model is used to predict the hydrogen release capacity of the candidate alloys, pre-screening alloy compositions that meet the requirements for predicted hydrogen release capacity. Experimental verification is then carried out on the screened compositions to determine the target alloy composition. Unlike the traditional method of screening alloys based solely on limited empirical criteria of physical properties, this method no longer relies solely on fixed physical property thresholds for rough evaluation. Instead, a machine learning model is used to explore the intrinsic correlation between composition and room temperature hydrogen release capacity, eliminating a large number of candidate compositions that do not meet the room temperature hydrogen release capacity requirements in advance, reducing invalid experimental screening. This solves the technical problems of traditional empirical criteria easily screening out alloy compositions that do not meet the room temperature performance requirements and the low accuracy of composition determination, thereby improving the accuracy of determining the composition of body-centered cubic hydrogen storage alloys. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a flowchart illustrating a machine learning-based design method for high-capacity, room-temperature body-centered cubic hydrogen storage alloys in one embodiment.

[0061] Figure 2 A schematic diagram showing the ranking of the importance of core features in one embodiment;

[0062] Figure 3This is a schematic diagram illustrating the fitting of the predicted hydrogen release capacity with the actual value of the CatBoost (Category Feature Enhancement Algorithm) model (before data augmentation) in one embodiment.

[0063] Figure 4 This is a schematic diagram illustrating the fitting of the predicted hydrogen release capacity with the actual value using the CatBoost model (after data augmentation) in one embodiment.

[0064] Figure 5 This is a flowchart illustrating a machine learning-based design method for high-capacity, room-temperature body-centered cubic hydrogen storage alloys in another embodiment.

[0065] Figure 6 This is a schematic diagram illustrating the predicted hydrogen release capacity of the alloy composition at 25°C in one embodiment.

[0066] Figure 7 In one embodiment, V 41.35 Ti 21.27 Cr 37.38 A schematic diagram of the XRD (X-ray Diffraction) pattern of the alloy;

[0067] Figure 8 In one embodiment, V 41.35 Ti 21.27 Cr 37.38 A schematic diagram of the hydrogen decomposition PCT (Pressure-Composition-Temperature) curve of the alloy at 25℃;

[0068] Figure 9 This is a structural block diagram of a room-temperature high-capacity body-centered cubic hydrogen storage alloy design device based on machine learning in one embodiment.

[0069] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0071] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0072] In one exemplary embodiment, such as Figure 1 As shown, a machine learning-based design method for high-capacity body-centered cubic hydrogen storage alloys at room temperature is provided. This embodiment illustrates the application of this method to a server; it is understood that this method can also be applied to terminals, and to systems including terminals and servers, and is implemented through interaction between the terminal and the server. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, and tablets; the server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. In this embodiment, the method includes the following steps:

[0073] Step S101: In response to the alloy design instruction of the alloy to be designed, determine the candidate alloy composition data of the alloy to be designed, and extract the composition features corresponding to the candidate alloy composition data; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy.

[0074] Among them, alloy design command refers to the trigger command used to reverse engineer the alloy to be designed.

[0075] Among them, candidate alloy composition data refers to the alloy atomic ratio data of the alloy to be designed, which is generated in batches by an optimization algorithm based on the element space defined by the alloy design instructions.

[0076] Among them, compositional features are used to represent the core data features corresponding to the candidate alloy composition data, including atomic radius, solid solution formation thermodynamic parameters, shear modulus and room temperature (also known as hydrogen expulsion temperature).

[0077] Among them, body-centered cubic hydrogen storage alloy is also known as BCC (Body-Centered Cubic) hydrogen storage alloy. Specifically, it refers to a vanadium-based solid solution hydrogen storage alloy with a body-centered cubic crystal lattice structure that can reversibly absorb and desorb hydrogen at room temperature of 25°C.

[0078] For example, the server receives the alloy design instructions for the alloy to be designed from the terminal via a network path between the server and the terminal; then, it performs an integrity check on the alloy design instructions to obtain the integrity check result; if the integrity check result indicates that the alloy design instructions pass, in response to the alloy design instructions, it determines the candidate alloy composition data of the alloy to be designed based on the element space defined by the alloy design instructions through an optimization algorithm; then, it inputs the candidate alloy composition data into a feature extraction model, and extracts the composition features corresponding to the candidate alloy composition data through the feature extraction model; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy.

[0079] Step S102: Input the composition features into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data.

[0080] Among them, the hydrogen release capacity prediction model refers to a network model that can use compositional features to obtain candidate alloy composition data to predict the hydrogen release capacity, such as the CatBoost model.

[0081] Among them, the predicted hydrogen release capacity is used to characterize the theoretical amount of hydrogen that can be released corresponding to the candidate alloy composition data under the operating condition of 25°C room temperature.

[0082] For example, the server normalizes the composition features to obtain normalized composition features; then, the normalized composition features are input into the trained hydrogen release capacity prediction model to obtain the hydrogen release capacity output by each hydrogen release capacity prediction network in the trained hydrogen release capacity prediction model; then, according to the network weight of each hydrogen release capacity prediction network, the hydrogen release capacity output by each hydrogen release capacity prediction network is summed to obtain the predicted hydrogen release capacity of the candidate alloy composition data.

[0083] Step S103: From the candidate alloy composition data, select the candidate alloy composition data whose predicted hydrogen release capacity is greater than the preset hydrogen release capacity, and use them as the alloy composition data to be verified for the alloy to be designed.

[0084] Among them, the preset hydrogen release capacity refers to the pre-set threshold for hydrogen release capacity.

[0085] Among them, the alloy composition data to be verified refers to the candidate alloy composition data whose predicted hydrogen release capacity is greater than the preset hydrogen release capacity.

[0086] For example, the server queries the correspondence between the alloy identifier and the hydrogen release capacity threshold based on the alloy identifier of the alloy to be designed, and uses the hydrogen release capacity threshold corresponding to the alloy identifier as the preset hydrogen release capacity; then, it filters out the candidate alloy component data with predicted hydrogen release capacity greater than the preset hydrogen release capacity from each candidate alloy component data, and uses these candidate alloy component data as the alloy component data to be verified for the alloy to be designed.

[0087] Step S104: Perform experimental verification processing on the alloy composition data to be verified to obtain the experimental verification results of the alloy composition data to be verified.

[0088] Among them, the experimental verification result refers to the comprehensive judgment information obtained after the experimental verification processing of the alloy composition data to be verified, including verification passed and verification failed.

[0089] For example, the server performs alloy preparation verification processing, phase characterization verification processing, and performance verification processing on the alloy composition data to be verified, and obtains the experimental verification results of the alloy composition data to be verified.

[0090] Step S105: If the experimental verification results indicate that the alloy composition to be verified has passed the verification, the target alloy composition data of the alloy to be designed is obtained based on the alloy composition data to be verified.

[0091] Among them, the target alloy composition data refers to the alloy composition data that matches the alloy to be designed.

[0092] For example, if the experimental verification results indicate that the alloy composition to be verified has passed the verification, the server will use the alloy composition data to be verified as the target alloy composition data of the alloy to be designed.

[0093] In the aforementioned machine learning-based design method for high-capacity body-centered cubic hydrogen storage alloys at room temperature, the following steps are taken: First, in response to the alloy design instructions of the alloy to be designed, candidate alloy composition data are determined, and the compositional features corresponding to the candidate alloy composition data are extracted. The alloy to be designed is used to represent the body-centered cubic hydrogen storage alloy. Then, the compositional features are input into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data. Next, from each candidate alloy composition data, candidate alloy composition data with a predicted hydrogen release capacity greater than the preset hydrogen release capacity are selected as the alloy composition data to be verified for the alloy to be designed. Then, the alloy composition data to be verified is experimentally verified to obtain the experimental verification results. Finally, if the experimental verification results indicate that the alloy composition to be verified has passed verification, the target alloy composition data of the alloy to be designed is obtained based on the alloy composition data to be verified. In this way, when determining the composition of body-centered cubic hydrogen storage alloys, candidate alloy composition data are first obtained and compositional features are extracted. Then, a trained hydrogen release capacity prediction model is used to predict the hydrogen release capacity of the candidate alloys, pre-screening alloy compositions that meet the requirements for predicted hydrogen release capacity. Experimental verification is then carried out on the screened compositions to determine the target alloy composition. Unlike the traditional method of screening alloys based solely on limited empirical criteria of physical properties, this method no longer relies solely on fixed physical property thresholds for rough evaluation. Instead, a machine learning model is used to explore the intrinsic correlation between composition and room temperature hydrogen release capacity, eliminating a large number of candidate compositions that do not meet the room temperature hydrogen release capacity requirements in advance, reducing invalid experimental screening. This solves the technical problems of traditional empirical criteria easily screening out alloy compositions that do not meet the room temperature performance requirements and the low accuracy of composition determination, thereby improving the accuracy of determining the composition of body-centered cubic hydrogen storage alloys.

[0094] In an exemplary embodiment, the hydrogen release capacity prediction model trained in step S102 is obtained by: acquiring sample alloy composition data of the sample alloy and extracting the sample composition features corresponding to the sample alloy composition data; dividing the sample composition features to obtain the training dataset and test dataset of the hydrogen release capacity prediction model to be trained; iteratively training the hydrogen release capacity prediction model to be trained based on the training dataset to obtain the trained hydrogen release capacity prediction model; verifying the trained hydrogen release capacity prediction model based on the test dataset to obtain the verification result of the trained hydrogen release capacity prediction model; and obtaining the trained hydrogen release capacity prediction model based on the trained hydrogen release capacity prediction model when the verification result meets the preset verification conditions.

[0095] Among them, the sample alloy refers to the body-centered cubic hydrogen storage alloy used to train the hydrogen release capacity prediction model.

[0096] Among them, the sample alloy composition data refers to the alloy atomic ratio data corresponding to the sample alloy.

[0097] Among them, the sample composition characteristics refer to the core data characteristics corresponding to the sample alloy composition data, including atomic radius, solid solution formation thermodynamic parameters, shear modulus and room temperature.

[0098] The training dataset refers to the subset of sample component features used for iterative training of the model.

[0099] The test dataset refers to the subset of sample component features used to evaluate the model.

[0100] The validation results are used to represent the performance of the trained hydrogen release capacity prediction model.

[0101] Among them, the preset verification conditions refer to the pre-set criteria for judging the performance of the model.

[0102] For example, in response to a model training instruction for the hydrogen emission capacity prediction model to be trained, the server acquires sample alloy composition data of the sample alloy and extracts the sample composition features corresponding to the sample alloy composition data through a feature extraction model. Then, using stratified sampling, the sample composition features are randomly divided according to a preset ratio (e.g., 8:2) to obtain the training dataset and test dataset of the hydrogen emission capacity prediction model to be trained. Then, based on the training dataset, the hydrogen emission capacity prediction model to be trained is iteratively trained to obtain the trained hydrogen emission capacity prediction model. Next, based on the test dataset, the trained hydrogen emission capacity prediction model is validated to obtain the determination coefficient and mean absolute error of the trained hydrogen emission capacity prediction model. If the determination coefficient is greater than the preset determination coefficient and the mean absolute error is less than the preset mean absolute error, the trained hydrogen emission capacity prediction model is taken as the completed hydrogen emission capacity prediction model.

[0103] In this embodiment, a feature system is constructed by collecting real sample alloy data and divided into independent training datasets and test datasets. The model is iteratively trained based on the training dataset, and the model performance is objectively evaluated using the test dataset that was not used in the training. Preset verification conditions are set to screen qualified models, thereby ensuring that the hydrogen release capacity prediction model has reliable prediction accuracy for unknown alloy components.

[0104] In an exemplary embodiment, extracting sample component features corresponding to sample alloy composition data specifically includes the following steps: extracting initial component features corresponding to sample alloy composition data; determining the correlation coefficients between the initial component features; selecting redundant component features with correlation coefficients greater than a preset correlation coefficient from each initial component feature; deleting redundant component features from the initial component features to obtain first candidate component features corresponding to the sample alloy composition data; determining the importance of the first candidate component features using a random forest algorithm; selecting second candidate component features with importance greater than a preset importance from each first candidate component feature; and progressively eliminating the second candidate component features using a feature recursive elimination algorithm to obtain the sample component features corresponding to the sample alloy composition data.

[0105] The initial composition characteristics refer to all the data characteristics corresponding to the sample alloy composition data, including molar mass, atomic radius, atomic radius mismatch, valence electron concentration, electronegativity, electronegativity mismatch, enthalpy of mixing, configuration entropy, solid solution formation thermodynamic parameters, bulk modulus, bulk modulus mismatch, shear modulus, shear modulus mismatch, and hydrogen release temperature.

[0106] The correlation coefficient refers to the Pearson correlation coefficient, which is used to quantify the strength of the linear correlation between two sets of initial component features.

[0107] The preset correlation coefficient refers to a pre-set threshold for determining correlation, such as 0.8.

[0108] Redundant component features refer to component features in the initial component features whose correlation coefficient is greater than the preset correlation coefficient.

[0109] The first candidate component feature refers to the set of features remaining after removing redundant component features from the initial component features.

[0110] The importance is used to characterize the contribution of the first candidate component features to the predicted hydrogen release capacity.

[0111] Among them, preset importance refers to the importance threshold set in advance.

[0112] Among them, the second candidate component feature refers to the candidate component feature whose importance is greater than the preset importance among the first candidate component features.

[0113] For example, the server calculates the physicochemical parameters of the atomic proportions of each group of elements in the sample alloy composition data to obtain the initial composition features corresponding to the sample alloy composition data. Then, through a pre-constructed correlation coefficient determination model, the correlation coefficients between the initial composition features are determined. From each initial composition feature, initial composition features with correlation coefficients greater than a preset correlation coefficient are selected and treated as redundant composition features. Redundant composition features are then removed from the initial composition features to obtain the first candidate composition features corresponding to the sample alloy composition data. Next, the first candidate composition features are input into a random forest algorithm. The random forest algorithm determines the importance of the first candidate composition features and selects first candidate composition features with importance greater than a preset importance from each first candidate composition feature. These first candidate composition features are then treated as second candidate composition features. Finally, through a feature recursive elimination algorithm, from candidate feature combinations consisting of at least two second candidate composition features, the target feature combination with the highest determination coefficient and the smallest mean absolute error is determined. The candidate composition features contained in the target feature combination are then used as the sample composition features corresponding to the sample alloy composition data.

[0114] For example, a feature importance level bar chart is shown below. Figure 2 As shown, the horizontal axis represents feature importance, and the vertical axis lists four features and their corresponding importance values. The feature importance of r (atomic radius) is 0.42, the feature importance of Ω (solution formation thermodynamic parameter) is 0.33, the feature importance of G (shear modulus) is 0.16, and the feature importance of T... dThe feature importance of (room temperature) is 0.09. It can be seen that the contribution of the four features to the model prediction results from high to low is r, Ω, G, T. d .

[0115] In this embodiment, a three-level progressive feature screening system is constructed by sequentially employing correlation analysis to remove highly collinear redundant features, random forest to coarsely screen variables based on feature contribution, and feature recursion elimination to carry out refined iterative selection. This system can eliminate overlapping features and invalid features with weak predictive effect on room temperature hydrogen release capacity layer by layer, thereby achieving accurate mining of key descriptors affecting the performance of body-centered cubic hydrogen storage alloys.

[0116] In an exemplary embodiment, the hydrogen emission capacity prediction model to be trained is iteratively trained based on the training dataset to obtain the trained hydrogen emission capacity prediction model. Specifically, this includes: iteratively training the hydrogen emission capacity prediction model to be trained based on the training dataset to obtain an initial hydrogen emission capacity prediction model; performing data augmentation processing on the training dataset to obtain an augmented training dataset; and iteratively training the initial hydrogen emission capacity prediction model again based on the augmented training dataset to obtain the trained hydrogen emission capacity prediction model.

[0117] The initial hydrogen release capacity prediction model refers to the hydrogen release capacity prediction model obtained by iteratively training the hydrogen release capacity prediction model to be trained based on the training dataset.

[0118] The augmented training dataset refers to the training dataset after data augmentation processing.

[0119] For example, the server determines a first loss value for the hydrogen emission capacity prediction model to be trained based on the training dataset, and adjusts the model parameters of the hydrogen emission capacity prediction model to be trained based on the first loss value to obtain a hydrogen emission capacity prediction model with adjusted model parameters. The hydrogen emission capacity prediction model with adjusted model parameters is then retrained until the first loss value of the trained hydrogen emission capacity prediction model is less than a loss value threshold. The trained hydrogen emission capacity prediction model is then used as the initial hydrogen emission capacity prediction model. Next, Gaussian noise is used to perform data augmentation processing on the training dataset to obtain an augmented training dataset. Then, based on the augmented training dataset, a second loss value for the initial hydrogen emission capacity prediction model is determined, and the model parameters of the initial hydrogen emission capacity prediction model are adjusted based on the second loss value to obtain an initial hydrogen emission capacity prediction model with adjusted model parameters. The initial hydrogen emission capacity prediction model with adjusted model parameters is then retrained until the second loss value of the trained initial hydrogen emission capacity prediction model is less than a loss value threshold. The trained initial hydrogen emission capacity prediction model is then used as the trained hydrogen emission capacity prediction model.

[0120] For example, Figure 3 This is a scatter plot comparing the predicted and actual values ​​of hydrogen release capacity from the model, with the horizontal axis representing True C. des The vertical axis represents the true reversible hydrogen release capacity, expressed in wt.%, while the vertical axis represents the predicted C. des The value represents the predicted reversible hydrogen release capacity, in wt.%; the solid dots in the figure represent training set samples, labeled Training R. 2 =1.000, the cross symbol represents a test set sample, labeled Testing R 2 =0.838, the solid line is the ideal prediction fitting line, the dashed lines on both sides form the error interval, the samples are distributed along the fitting line, indicating that the model has good predictive ability, the training set has excellent fitting effect, and the test set also has a high coefficient of determination, reflecting that the model has good generalization performance.

[0121] For example, Figure 4 This is a comparison chart of predicted hydrogen release capacity. The horizontal axis represents the actual hydrogen release capacity (True C). des The unit is wt.%, and the vertical axis represents the predicted hydrogen release capacity (Predicted C). des The unit is wt.%. Solid dots in the legend represent training set samples, labeled Training R. 2 =1.000, the cross represents a test set sample, labeled Testing R 2 =0.838. The solid line in the figure is the ideal prediction baseline, and the two dashed lines are the error boundaries. The training set samples closely fit the baseline, and the test set samples are evenly distributed in the error interval around the baseline. This indicates that the model achieves a perfect fit on the training set and has a high coefficient of determination on the test set, demonstrating good generalization prediction ability.

[0122] In this embodiment, a two-stage model training strategy is adopted. First, the initial hydrogen release capacity prediction model is obtained by iterative training using the original training dataset. Then, new virtual samples are generated through data augmentation to construct the enhanced training dataset. This effectively alleviates the overfitting problem that is easy to occur when training with small samples and improves the generalization ability and prediction stability of the hydrogen release capacity prediction model for unknown alloy components.

[0123] In an exemplary embodiment, before obtaining the sample alloy composition data of the sample alloy, the following steps are specifically included: obtaining the element data corresponding to the initial alloy and the frequency of occurrence of the element data in the initial alloy; filtering out target element data with a frequency of occurrence less than a preset frequency from each element data; deleting the initial alloy containing the target element data from each initial alloy to obtain the sample alloy.

[0124] The initial alloy refers to all alloy samples collected from literature and experimental records.

[0125] Among them, elemental data refers to the various elemental identifiers of the chemical composition that make up the initial alloy, such as Ti (titanium) and Fe (iron).

[0126] The frequency of occurrence refers to the proportion of the number of samples in which the element data appears in the initial alloy (or the absolute number of times it appears), which is used to measure the richness of the data for that element.

[0127] The preset occurrence frequency refers to a pre-set frequency threshold.

[0128] Among them, target element data refers to element data whose frequency of occurrence is less than the preset frequency of occurrence.

[0129] For example, the server performs statistical processing on the literature data and experimental record data associated with the initial alloy to obtain the element data corresponding to the initial alloy and the frequency of occurrence of the element data in the initial alloy; then, from each element data, element data with a frequency of occurrence less than a preset frequency of occurrence is selected and these element data are used as target element data; then, from each initial alloy, the initial alloy containing the target element data is deleted to obtain the sample alloy.

[0130] In this embodiment, the target elements that are scarce in the sample are identified by statistically analyzing the frequency of element occurrence, and the initial alloys containing such elements are removed. This achieves pre-processing of the dataset, reduces the data distribution imbalance caused by rare elements, avoids interference from a small number of special alloys in the mining of the general pattern between alloy composition and room temperature hydrogen release capacity, and helps to improve the consistency and reliability of subsequent machine learning model training data.

[0131] In an exemplary embodiment, step S104 above, which involves performing experimental verification processing on the alloy composition data to be verified to obtain experimental verification results, specifically includes the following: performing alloy preparation verification processing on the alloy composition data to be verified to obtain a first verification result; performing phase characterization verification processing on the alloy composition data to be verified to obtain a second verification result; performing performance verification processing on the alloy composition data to be verified to obtain a third verification result; and obtaining the experimental verification result of the alloy composition data based on the first verification result, the second verification result, and the third verification result.

[0132] Among them, alloy preparation verification treatment refers to the experimental process of weighing and batching raw materials based on the element ratio recorded in the alloy composition data to be verified, preparing alloy samples using processes such as vacuum arc melting, and observing and recording the sample forming state and macroscopic defects.

[0133] The first verification result refers to the verification result obtained by performing alloy preparation verification processing on the alloy composition data to be verified.

[0134] Among them, phase characterization and verification processing refers to the detection process of performing X-ray diffraction tests on the alloy corresponding to the alloy composition data to be verified, collecting diffraction patterns and performing phase analysis to identify the types of internal crystal structures of the alloy.

[0135] The second verification result refers to the verification result obtained by performing phase characterization verification on the alloy composition data to be verified.

[0136] Among them, the performance verification process refers to the hydrogen storage test of the alloy sample corresponding to the alloy composition data to be verified using the Sievts volume method, testing the hydrogen absorption and desorption behavior of the alloy under specified temperature conditions, and collecting the actual measured reversible hydrogen desorption capacity at a preset temperature (e.g., 25°C).

[0137] The third verification result refers to the verification result obtained by performing performance verification processing on the alloy composition data to be verified.

[0138] For example, the server weighs various metal raw materials according to the atomic ratio in the alloy composition data to be verified, and performs alloy preparation verification processing through vacuum arc melting process to obtain the first verification result of the alloy composition data to be verified; then, X-ray diffraction test is performed on the alloy corresponding to the alloy composition data to be verified, diffraction pattern is collected and phase analysis is performed to identify the types of internal crystal structures of the alloy to be verified, to obtain the second verification result of the alloy composition data to be verified; then, hydrogen storage test is carried out on the alloy sample corresponding to the alloy composition data to be verified using the Sieverts volume method, the hydrogen absorption and desorption behavior of the alloy is tested under specified temperature conditions, and the measured reversible hydrogen desorption capacity at a preset temperature is collected to obtain the third verification result of the alloy composition data to be verified; finally, the first verification result, the second verification result, and the third verification result are integrated and processed to obtain the experimental verification result of the alloy composition data to be verified.

[0139] In this embodiment, independent experimental verification is carried out in three layers from three dimensions: preparation feasibility, crystal phase structure and room temperature hydrogen storage performance. The first verification result, the second verification result and the third verification result are obtained respectively and then comprehensively evaluated. This avoids judging the quality of the alloy based on a single performance index. It takes into account the feasibility of the process, the stability of the phase structure and the actual performance in actual use, which is conducive to improving the completeness and rigor of alloy screening and evaluation.

[0140] In one exemplary embodiment, such as Figure 5 As shown, another machine learning-based design method for high-capacity body-centered cubic hydrogen storage alloys at room temperature is presented. Taking the application of this method to a server as an example, the specific steps include:

[0141] Step S501: In response to the alloy design instruction of the alloy to be designed, determine the candidate alloy composition data of the alloy to be designed, and extract the composition features corresponding to the candidate alloy composition data; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy.

[0142] Step S502: Input the composition features into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data.

[0143] Step S503: From the candidate alloy composition data, select the candidate alloy composition data whose predicted hydrogen release capacity is greater than the preset hydrogen release capacity, and use them as the alloy composition data to be verified for the alloy to be designed.

[0144] Step S504: Perform alloy preparation verification processing on the alloy composition data to be verified to obtain the first verification result of the alloy composition data to be verified.

[0145] Step S505: Perform phase characterization verification processing on the alloy composition data to be verified to obtain the second verification result of the alloy composition data to be verified.

[0146] Step S506: Perform performance verification processing on the alloy composition data to be verified to obtain the third verification result of the alloy composition data to be verified.

[0147] Step S507: Based on the first verification result, the second verification result, and the third verification result, the experimental verification result of the alloy composition data to be verified is obtained.

[0148] Step S508: If the experimental verification results indicate that the alloy composition to be verified has passed the verification, the target alloy composition data of the alloy to be designed is obtained based on the alloy composition data to be verified.

[0149] In the aforementioned machine learning-based design method for high-capacity body-centered cubic (BCC) hydrogen storage alloys at room temperature, the composition of the BCC is determined by first acquiring candidate alloy composition data and extracting compositional features. Then, a trained hydrogen release capacity prediction model is used to predict the hydrogen release capacity of the candidate alloys, pre-screening alloy compositions whose predicted hydrogen release capacity meets the requirements. Experimental verification is then conducted on the screened compositions to determine the target alloy composition. This method differs from the traditional approach of relying solely on limited empirical property criteria for alloy selection. Instead of simply relying on fixed property thresholds for rough evaluation, it utilizes a machine learning model to uncover the intrinsic correlation between composition and room-temperature hydrogen release capacity, pre-eliminating a large number of candidate compositions with substandard room-temperature hydrogen release capacity. This reduces ineffective experimental screening and solves the technical problems of traditional empirical criteria easily identifying alloy compositions with substandard room-temperature performance and low composition determination accuracy, thereby improving the accuracy of BCC hydrogen storage alloy composition determination.

[0150] In an exemplary embodiment, to more clearly illustrate the machine learning-based room-temperature high-capacity body-centered cubic (BCC) hydrogen storage alloy design method provided in this application, the following specific embodiment will be used to describe this machine learning-based room-temperature high-capacity BCC hydrogen storage alloy design method. In one embodiment, this application also provides a machine learning-based room-temperature high-capacity BCC hydrogen storage alloy design method. Specifically, it includes the following:

[0151] This solution is a materials design-based technical approach that combines machine learning algorithms, feature engineering, intelligent optimization algorithms, and materials preparation and characterization experiments to achieve precise design of high-capacity BCC hydrogen storage alloys at room temperature. The overall technical process consists of seven core steps, executed by a server / terminal equipped with machine learning algorithms and a materials preparation and characterization experimental platform. The sequence of each step is as follows: dataset construction → data preprocessing and feature selection → dataset partitioning → model construction → model optimization → alloy composition design → experimental verification and model iteration. The data flow is: "experimental / literature data → feature calculation and selection → model training and optimization → composition prediction → experimental verification → model iterative optimization," which is described in detail below:

[0152] Step 1: Construct the dataset.

[0153] Data on the composition, hydrogen release temperature, and hydrogen release capacity of BCC hydrogen storage alloys were collected from publicly available literature. Low-frequency element data with a frequency of less than 10 alloy components were removed, resulting in a raw dataset containing more than 120 sets of valid data, which provides basic data support for subsequent model training.

[0154] Step 2: Data preprocessing and feature selection.

[0155] Based on the alloy composition (atomic percentage of components), characteristic parameters reflecting the physicochemical properties of the BCC hydrogen storage alloy were calculated, and the hydrogen release temperature T was also included. d The features are incorporated to form an original feature set containing 14 features. The feature parameters and calculation formulas are as follows (C i Let be the atomic percentage of the i-th component in the alloy, n be the number of alloy components, and R be the gas constant, which is 8.314 J / (mol·K)).

[0156] Equation (1)

[0157] Equation (2)

[0158] Equation (3)

[0159] Equation (4)

[0160] Equation (5)

[0161] Equation (6)

[0162] Equation (7)

[0163] Equation (8)

[0164] Equation (9)

[0165] Equation (10)

[0166] Equation (11)

[0167] Equation (12)

[0168] Equation (13)

[0169] Among them, M i r i VEC i , χ i B i G i Let be the molar mass, atomic radius, valence electron concentration, electronegativity, bulk modulus, and shear modulus of the i-th component, respectively. T is the enthalpy of mixture of the i-th and j-th components. m The melting point of the alloy. For the above 14 original characteristics (molar mass M, atomic radius r, atomic radius mismatch δr, valence electron concentration VEC, electronegativity χ, electronegativity mismatch Δχ, enthalpy of mixing ΔH, configuration entropy ΔS, solid solution formation thermodynamic parameter Ω, bulk modulus B, bulk modulus mismatch δB, shear modulus G, shear modulus mismatch δG), and simultaneously the hydrogen release temperature T... d The data is then standardized to eliminate the influence of units and form a standard dataset.

[0170] A three-step method—Pearson correlation coefficient analysis, random forest feature importance ranking, and recursive feature elimination—was used to screen core features, remove redundant features, and determine the optimal feature subset. This subset contains the core features that serve as the input features for subsequent model training, accurately characterizing the correlation between the composition of BCC hydrogen storage alloys and their hydrogen release performance. The three-step method is as follows:

[0171] 1. Construct a feature correlation heatmap based on the Pearson correlation coefficient to identify redundant feature pairs with a correlation coefficient greater than 0.8.

[0172] 2. Use the random forest algorithm to sort all features by importance and remove redundant features with low importance.

[0173] 3. Using a feature recursive elimination algorithm, features are gradually eliminated and the model performance is verified to determine the optimal feature subset.

[0174] Step 3: Dataset partitioning.

[0175] The standard dataset was randomly divided into a training set (for model training) and a test set (for model performance validation) in an 8:2 ratio. At the same time, 5-fold cross-validation was performed to ensure that the training set and the test set were consistent in terms of alloy composition range and hydrogen release performance range, so as to avoid model overfitting and improve the model's generalization ability.

[0176] Step 4, Machine Learning Model Construction.

[0177] A hydrogen release capacity prediction model for BCC hydrogen storage alloys is constructed based on the CatBoost gradient boosting algorithm. The specific implementation is as follows:

[0178] 1. The model input is the feature subset obtained in step 2, and the model output is the hydrogen release capacity of the BCC hydrogen storage alloy.

[0179] 2. The hyperparameters of the CatBoost model were optimized using a grid search method. The hyperparameters to be optimized included depth (tree depth), iterations (number of iterations), l2_leaf_reg (L2 regularization coefficient of leaf nodes), learning_rate (learning rate), and subsample (sample sampling ratio). The optimal combination of hyperparameters for the CatBoost model was determined to be depth=7, iterations=174, l2_leaf_reg=0.149, learning_rate=0.125, and subsample=0.801.

[0180] 3. The performance evaluation index for the model is the coefficient of determination R. 2 The calculation formula is: ( This is the actual value. The corresponding predicted value, (The average of the true values), the optimization objective is R on the test set. 2 ≥0.80.

[0181] Step 5, Model Optimization.

[0182] The standard dataset is augmented using Gaussian noise perturbation to optimize the model's prediction accuracy and generalization ability. The specific implementation is shown below:

[0183] 1. Add Gaussian noise to the core feature data of the training set, and set the parameters to be optimized as noise level and number of samples n_samples.

[0184] 2. After parameter optimization, the optimal combination of Gaussian noise perturbation parameters was determined to be noise_level=0.0455 and n_samples=1.

[0185] 3. Retrain the CatBoost model using the expanded dataset.

[0186] Step 6, alloy composition design.

[0187] Based on the core features from step 2 and the optimized CatBoost model from step 5, a smart composition design for the BCC hydrogen storage alloy is achieved using a Bayesian optimization algorithm, as shown below:

[0188] 1. Traverse 511 vanadium-containing multi-element alloy combinations (V combined with any one or more elements from Ti, Cr, Al, Fe, Mn, Nb, Ni, Co, and Mo), and use the optimized CatBoost model to predict the hydrogen release capacity of each alloy at 25℃.

[0189] 2. A Bayesian optimization algorithm is used to efficiently search the multi-element alloy composition space. The optimal parameter combination for Bayesian optimization is init_points=15 and n_iter=500.

[0190] 3. The screening target is a single-phase BCC hydrogen storage alloy with a hydrogen release capacity of ≥2.5wt.% at 25℃.

[0191] Step 7: Experimental verification and model iteration.

[0192] The predicted optimal alloy composition was experimentally verified through material preparation, phase characterization, and hydrogen storage performance testing. Simultaneously, a model iteration mechanism was established to ensure the reliability of the design method. Specifically, the following was implemented:

[0193] 1. Alloy Preparation: Optimal alloy composition samples were prepared using vacuum arc melting. The melting process parameters were: vacuum degree maintained at 5.0 × 10⁻⁶. -5 Pa, arc current 120~140A, repeated melting 5 times to ensure uniform alloy composition.

[0194] 2. Phase characterization: The phase structure of the alloy was verified by X-ray diffraction (XRD).

[0195] 3. Performance testing: The hydrogen desorption PCT curve of the alloy at 25℃ was tested using the Sieverts volumetric method to obtain the actual hydrogen desorption capacity.

[0196] 4. Model Iteration: Set the relative error threshold to 5%. If the relative error between the actual hydrogen release capacity and the model prediction exceeds the threshold, readjust the CatBoost model hyperparameters and repeat steps 4 to 7 until the error meets the requirements.

[0197] The specific implementation method is as follows:

[0198] 1. Dataset Construction: Collect data on the composition, hydrogen release temperature, and hydrogen release capacity of BCC hydrogen storage alloy from public literature, remove low-frequency element data with a frequency of less than 10 alloy components, and finally form an original dataset containing more than 134 sets of valid data.

[0199] 2. Data Preprocessing and Feature Screening: Based on the 134 alloy compositions (atomic percentage of each component), 13 characteristic parameters reflecting the physicochemical properties of the BCC hydrogen storage alloy were calculated, including molar mass M, atomic radius r, atomic radius mismatch δr, valence electron concentration VEC, electronegativity χ, electronegativity mismatch Δχ, enthalpy of mixing ΔH, configuration entropy ΔS, solid solution formation thermodynamic parameter Ω, bulk modulus B, bulk modulus mismatch δB, shear modulus G, and shear modulus mismatch δG. Simultaneously, the hydrogen release temperature T was also included. d This incorporates 14 features, forming an initial feature set. These 14 initial features (13 calculated features and the hydrogen release temperature T) are then used to construct the final feature set. d The dataset was standardized to form a standard dataset. A three-step method was used: Pearson correlation coefficient analysis, random forest feature importance ranking, and recursive feature elimination to select core features, remove redundant features, and determine the optimal feature subset containing four core features. For example... Figure 2 As shown, the order of importance is: atomic radius r > thermodynamic parameter of solid solution formation Ω > shear modulus G > hydrogen release temperature T d .

[0200] 3. Dataset splitting: The standard dataset is randomly split into a training set (for model training) and a test set (for model performance validation) in an 8:2 ratio, and 5-fold cross-validation is performed.

[0201] 4. Machine Learning Model Construction: A hydrogen release capacity prediction model for BCC hydrogen storage alloy was constructed based on the CatBoost gradient boosting algorithm. The model input consists of four core features: r, Ω, G, and T. d The model output is the hydrogen release capacity of the BCC hydrogen storage alloy. The CatBoost model hyperparameter combination used is depth=7, iterations=174, l2_leaf_reg=0.149, learning_rate=0.125, subsample=0.801. Figure 3 As shown, the coefficient of determination R0 for the fit between the predicted and actual hydrogen release capacity values ​​is... 2The value of 0.838 indicates that the model has high prediction accuracy.

[0202] 5. Model Optimization: The standard dataset is augmented using Gaussian noise perturbation. The Gaussian noise perturbation parameters are noise_level=0.0455 and n_samples=1. The CatBoost model is then retrained using the augmented dataset. Figure 4 As shown, the coefficient of determination R of the CatBoost model after data augmentation 2 The value of 0.855 indicates that Gaussian noise data augmentation technology effectively improved the model's generalization ability and prediction accuracy.

[0203] 6. Alloy composition design: based on 4 core features r, Ω, G and T d The optimized CatBoost model traverses 511 vanadium-containing multi-element alloy combinations (V combined with any one or more elements from Ti, Cr, Al, Fe, Mn, Nb, Ni, Co, and Mo), and uses the optimized CatBoost model to predict the hydrogen desorption capacity of each alloy at 25℃. Then, a Bayesian optimization algorithm is used to efficiently search the composition space of the multi-element alloys. The optimal parameter combination for Bayesian optimization is init_points=15 and n_iter=500. Figure 7 As shown, the optimal alloy composition obtained through screening is V. 41.35 Ti 21.27 Cr 37.38 The model predicts that the hydrogen release capacity of the alloy at 25°C is 2.51 wt.%.

[0204] 7. Experimental Verification and Model Iteration: V was prepared using the vacuum arc melting method. 41.35 Ti 21.27 Cr 37.38 The alloy sample was kept under a vacuum of 5.0 × 10⁻⁶ during the melting process. -5 Pa, arc current of 130 A, and repeated melting 5 times to ensure uniform composition. Phase characterization of the alloy was performed by X-ray diffraction (XRD), such as... Figure 6 As shown, the sample has a single BCC structure. The Sieverts volumetric method was used to measure V. 41.35 Ti 21.27 Cr 37.38 The hydrogen desorption PCT curve of the alloy at 25°C is shown below. Figure 8 As shown, the alloy exhibits a high hydrogen release capacity at room temperature, with a measured hydrogen release capacity of 2.58 wt.% and a plateau pressure of approximately 0.3 MPa. Compared to the model prediction of 2.51 wt.%, the relative error is only 2.7%, far below the 5% iteration threshold, eliminating the need for model parameter adjustment. This result fully validates the reliability and high accuracy of the proposed design method.

[0205] The core principle of this scheme is to accurately extract the correlation features between the composition and hydrogen release performance of BCC hydrogen storage alloy through feature engineering, fit the nonlinear correlation law using the CatBoost model, and achieve intelligent search of alloy composition by combining Bayesian optimization algorithm. Then, the closed-loop design is completed through experimental verification. This solves the problems of traditional methods being unable to capture the nonlinear correlation of multi-element alloys and having low R&D efficiency. At the same time, the generalization ability of the model is improved through data augmentation and cross-validation to ensure the reliability of the design results.

[0206] In the above embodiments, when determining the composition of body-centered cubic hydrogen storage alloys, candidate alloy composition data of body-centered cubic hydrogen storage alloys are first obtained and composition features are extracted. Then, a trained hydrogen release capacity prediction model is used to predict the hydrogen release capacity of the candidate alloys, pre-screening the alloy compositions to be verified that meet the predicted hydrogen release capacity requirements. Experimental verification is then carried out on the screened compositions to determine the target alloy composition. Unlike the traditional method of screening alloys based solely on limited empirical criteria of physical properties, this method no longer relies solely on fixed physical property thresholds for rough evaluation. Instead, a machine learning model is used to explore the intrinsic correlation between composition and room temperature hydrogen release capacity, eliminating a large number of candidate compositions that do not meet the room temperature hydrogen release capacity requirements in advance, reducing invalid experimental screening. This solves the technical problems of traditional empirical criteria easily screening out alloy compositions that do not meet the room temperature performance requirements and the low accuracy of composition determination, thereby improving the accuracy of determining the composition of body-centered cubic hydrogen storage alloys.

[0207] It should be noted that this scheme has significant technical advantages compared to the existing optimal technology for BCC hydrogen storage alloy design (empirical trial and error method + DFT (Density Functional Theory) / CALPHAD (Calculation of Phase Diagrams) method), as detailed below:

[0208] 1. Significantly improved R&D efficiency and reduced R&D costs: Based on the CatBoost model and Bayesian optimization algorithm, intelligent screening of 511 vanadium-containing multi-element alloy combinations is achieved. Without the need for extensive experimental trial and error, the R&D cycle is shortened directly, and the investment in experimental consumables, equipment and other R&D costs is reduced. This solves the core problem of low efficiency in component screening using traditional methods.

[0209] 2. High model prediction accuracy and strong guidance value for actual R&D: Through targeted feature engineering optimization, optimal model parameter configuration and Gaussian noise data enhancement technology, the model's relative error in predicting hydrogen release capacity is as low as 2.7%, which is far superior to the prediction accuracy of existing machine learning models and DFT / CALPHAD calculation methods. It can provide accurate and reliable technical guidance for the composition design of room temperature high-capacity BCC hydrogen storage alloys.

[0210] 3. The obtained target alloy exhibits excellent room temperature properties and has promising engineering application prospects: The designed V 41.35 Ti 21.27 Cr 37.38 The single-phase BCC hydrogen storage alloy has an actual hydrogen release capacity of 2.58 wt.% at 25°C, which far exceeds the room temperature hydrogen release capacity of commercially available LaNi5 hydrogen storage alloys. Moreover, it has a single BCC structure and a suitable hydrogen absorption and desorption platform pressure (0.3 MPa), which meets the practical application requirements of hydrogen energy storage.

[0211] 4. The method has good versatility and strong generalization ability: The core feature screening method, CatBoost model construction and optimization strategy, and Bayesian optimized composition search method of this scheme can be directly extended to the design and development of other BCC structure multi-element hydrogen storage alloys. The model parameters can be flexibly adjusted according to the needs of different alloy systems. The scope of application covers the entire BCC type hydrogen storage alloy system, which solves the problem that existing machine learning methods are only applicable to specific hydrogen storage alloy systems.

[0212] 5. The design system forms a closed loop with high reliability: A closed-loop design system of "data-model-prediction-experiment-iteration" has been established. The reliability of the design results is ensured through experimental verification and model iteration mechanism. At the same time, all process parameters and model parameters have clear numerical ranges and optimal values, which facilitates industrial promotion and repeated implementation. This solves the problems of uncontrollable design results and poor repeatability of traditional methods.

[0213] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0214] Based on the same inventive concept, this application also provides a machine learning-based room-temperature high-capacity body-centered cubic hydrogen storage alloy design device for implementing the aforementioned machine learning-based room-temperature high-capacity body-centered cubic hydrogen storage alloy design method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more machine learning-based room-temperature high-capacity body-centered cubic hydrogen storage alloy design device embodiments provided below can be found in the limitations of the machine learning-based room-temperature high-capacity body-centered cubic hydrogen storage alloy design method described above, and will not be repeated here.

[0215] In one exemplary embodiment, such as Figure 9 As shown, a machine learning-based design device for high-capacity body-centered cubic hydrogen storage alloys at room temperature is provided, comprising: a feature extraction module 901, a capacity prediction module 902, a component screening module 903, a component verification module 904, and a component determination module 905, wherein:

[0216] The feature extraction module 901 is used to respond to the alloy design command of the alloy to be designed, determine the candidate alloy composition data of the alloy to be designed, and extract the composition features corresponding to the candidate alloy composition data; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy.

[0217] The capacity prediction module 902 is used to input the composition features into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data.

[0218] The component screening module 903 is used to screen out candidate alloy component data from each candidate alloy component data that have a predicted hydrogen release capacity greater than the preset hydrogen release capacity, and use them as the alloy component data to be verified for the alloy to be designed.

[0219] The composition verification module 904 is used to perform experimental verification processing on the composition data of the alloy to be verified, and obtain the experimental verification results of the composition data of the alloy to be verified.

[0220] The composition determination module 905 is used to obtain the target alloy composition data of the alloy to be designed based on the composition data of the alloy to be verified, when the experimental verification results indicate that the composition of the alloy to be verified has been verified.

[0221] In an exemplary embodiment, the machine learning-based room-temperature high-capacity body-centered cubic hydrogen storage alloy design device further includes a model training module, used to acquire sample alloy composition data of sample alloys, extract sample composition features corresponding to the sample alloy composition data; perform segmentation processing on the sample composition features to obtain a training dataset and a test dataset for the hydrogen release capacity prediction model to be trained; perform iterative training on the hydrogen release capacity prediction model to be trained based on the training dataset to obtain the trained hydrogen release capacity prediction model; verify the trained hydrogen release capacity prediction model based on the test dataset to obtain the verification result of the trained hydrogen release capacity prediction model; and, if the verification result meets the preset verification conditions, obtain the trained hydrogen release capacity prediction model based on the trained hydrogen release capacity prediction model.

[0222] In an exemplary embodiment, the model training module is further configured to extract initial component features corresponding to the sample alloy composition data; determine the correlation coefficients between the initial component features; select redundant component features with correlation coefficients greater than a preset correlation coefficient from each initial component feature; delete redundant component features from the initial component features to obtain first candidate component features corresponding to the sample alloy composition data; determine the importance of the first candidate component features using a random forest algorithm; select second candidate component features with importance greater than a preset importance from each first candidate component feature; and gradually eliminate the second candidate component features using a feature recursive elimination algorithm to obtain sample component features corresponding to the sample alloy composition data.

[0223] In an exemplary embodiment, the model training module is further configured to iteratively train the hydrogen emission capacity prediction model to be trained based on the training dataset to obtain an initial hydrogen emission capacity prediction model; perform data augmentation processing on the training dataset to obtain an augmented training dataset; and iteratively train the initial hydrogen emission capacity prediction model again based on the augmented training dataset to obtain a trained hydrogen emission capacity prediction model.

[0224] In an exemplary embodiment, the machine learning-based room-temperature high-capacity body-centered cubic hydrogen storage alloy design device further includes a data acquisition module, used to acquire elemental data corresponding to the initial alloy and the frequency of occurrence of the elemental data in the initial alloy; to filter out target element data with a frequency of occurrence less than a preset frequency from each elemental data; and to delete the initial alloy containing the target element data from each initial alloy to obtain a sample alloy.

[0225] In an exemplary embodiment, the composition verification module is further configured to perform alloy preparation verification processing on the alloy composition data to be verified to obtain a first verification result of the alloy composition data to be verified; perform phase characterization verification processing on the alloy composition data to be verified to obtain a second verification result of the alloy composition data to be verified; perform performance verification processing on the alloy composition data to be verified to obtain a third verification result of the alloy composition data to be verified; and obtain experimental verification results of the alloy composition data to be verified based on the first verification result, the second verification result, and the third verification result.

[0226] The modules in the aforementioned machine learning-based room-temperature high-capacity body-centered cubic hydrogen storage alloy design device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0227] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores data such as compositional characteristics and predicted hydrogen release capacity. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a machine learning-based design method for high-capacity, room-temperature body-centered cubic hydrogen storage alloys.

[0228] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0229] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0230] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above-described method embodiments.

[0231] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0232] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0233] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0234] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A machine learning-based design method for high-capacity room-temperature body-centered cubic hydrogen storage alloys, characterized in that, The method includes: In response to the alloy design instruction of the alloy to be designed, candidate alloy composition data of the alloy to be designed are determined, and the composition features corresponding to the candidate alloy composition data are extracted; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy. The compositional features are input into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data. From the candidate alloy composition data, the candidate alloy composition data with the predicted hydrogen release capacity greater than the preset hydrogen release capacity are selected as the alloy composition data to be verified for the alloy to be designed. The composition data of the alloy to be verified is processed experimentally to obtain the experimental verification results of the composition data of the alloy to be verified. If the experimental verification results indicate that the composition of the alloy to be verified has passed verification, the target alloy composition data of the alloy to be designed is obtained based on the composition data of the alloy to be verified.

2. The method according to claim 1, characterized in that, The hydrogen release capacity prediction model, after training, was obtained through the following method: Obtain the sample alloy composition data of the sample alloy, and extract the sample composition features corresponding to the sample alloy composition data; The sample component features are segmented to obtain the training dataset and test dataset for the hydrogen emission capacity prediction model to be trained. Based on the training dataset, the hydrogen emission capacity prediction model to be trained is iteratively trained to obtain the trained hydrogen emission capacity prediction model. The trained hydrogen emission capacity prediction model was validated based on the test dataset, and the validation results of the trained hydrogen emission capacity prediction model were obtained. If the verification results meet the preset verification conditions, the trained hydrogen release capacity prediction model is obtained based on the trained hydrogen release capacity prediction model.

3. The method according to claim 2, characterized in that, The extraction of sample composition features corresponding to the sample alloy composition data includes: Extract the initial composition features corresponding to the sample alloy composition data; The correlation coefficients between the initial component features are determined. Redundant component features with correlation coefficients greater than a preset correlation coefficient are selected from the initial component features and deleted from the initial component features to obtain the first candidate component features corresponding to the sample alloy composition data. The importance of the first candidate component features is determined by the random forest algorithm, and a second candidate component feature with an importance greater than a preset importance is selected from each of the first candidate component features. The second candidate component features are gradually eliminated using a feature recursive elimination algorithm to obtain the sample component features corresponding to the sample alloy composition data.

4. The method according to claim 2, characterized in that, The step of iteratively training the hydrogen emission capacity prediction model to be trained based on the training dataset to obtain the trained hydrogen emission capacity prediction model includes: Based on the training dataset, the hydrogen emission capacity prediction model to be trained is iteratively trained to obtain an initial hydrogen emission capacity prediction model. The training dataset is augmented to obtain an augmented training dataset. Based on the augmented training dataset, the initial hydrogen emission capacity prediction model is iteratively trained again to obtain the trained hydrogen emission capacity prediction model.

5. The method according to claim 2, characterized in that, Before obtaining the sample alloy composition data, the following steps are also included: Obtain the elemental data corresponding to the initial alloy, and the frequency of occurrence of the elemental data in the initial alloy; From the data of each element, select the target element data whose occurrence frequency is less than the preset occurrence frequency; The sample alloy is obtained by deleting the initial alloys containing the target element data from each of the initial alloys.

6. The method according to any one of claims 1 to 5, characterized in that, The experimental verification process of the alloy composition data to be verified, to obtain the experimental verification results of the alloy composition data to be verified, includes: The alloy composition data to be verified is subjected to alloy preparation verification processing to obtain the first verification result of the alloy composition data to be verified. The composition data of the alloy to be verified is subjected to phase characterization verification processing to obtain a second verification result of the composition data of the alloy to be verified. The alloy composition data to be verified is subjected to performance verification processing to obtain the third verification result of the alloy composition data to be verified. Based on the first verification result, the second verification result, and the third verification result, the experimental verification results of the alloy composition data to be verified are obtained.

7. A machine learning-based design device for high-capacity, room-temperature body-centered cubic hydrogen storage alloys, characterized in that, The device includes: The feature extraction module is used to respond to the alloy design instruction of the alloy to be designed, determine the candidate alloy composition data of the alloy to be designed, and extract the composition features corresponding to the candidate alloy composition data; the alloy to be designed is used to represent a body-centered cubic hydrogen storage alloy. The capacity prediction module is used to input the composition features into the trained hydrogen release capacity prediction model to obtain the predicted hydrogen release capacity of the candidate alloy composition data. The component screening module is used to screen out the candidate alloy component data whose predicted hydrogen release capacity is greater than the preset hydrogen release capacity from each of the candidate alloy component data, and use them as the alloy component data to be verified for the alloy to be designed. The composition verification module is used to perform experimental verification processing on the composition data of the alloy to be verified, and to obtain the experimental verification results of the composition data of the alloy to be verified. The composition determination module is used to obtain the target alloy composition data of the alloy to be designed based on the composition data of the alloy to be verified, when the experimental verification results indicate that the composition of the alloy to be verified has been verified.

8. 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, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images