Method for regulating and controlling components of multi-element vanadium-based hydrogen storage alloy in small range

Abstract

The invention discloses a method for regulating and controlling components of a multi-element vanadium-based hydrogen storage alloy in a small range, and relates to the technical field of hydrogen storage alloy materials. The method aims at solving the problem that the traditional method mostly depends on limited experience adjustment or simple univariate experiment to screen the vanadium-based hydrogen storage alloy with high-performance components, and the screening efficiency is low. Obtaining a feature set of various vanadium-based hydrogen storage alloys, training the constructed feature prediction model, calculating an evaluation index of each vanadium-based hydrogen storage alloy, selecting the vanadium-based hydrogen storage alloy corresponding to the best evaluation index, selecting each element from the alloy, and presetting a component interval for each selected element; and values are taken from the intervals to form a plurality of new alloys, the effective hydrogen storage capacity of each new alloy is predicted by utilizing the model, and the new alloy corresponding to the maximum value is selected as the optimal vanadium-based hydrogen storage alloy. The method is used for obtaining the optimal vanadium-based hydrogen storage alloy.

Description

Technical Field

[0001] This invention relates to the field of hydrogen storage alloy materials technology. Background Technology

[0002] Hydrogen energy, due to its clean, efficient, and renewable characteristics, is considered an important component of the future energy system. Hydrogen storage technology is one of the key links in hydrogen energy applications. Among them, solid-state hydrogen storage materials have attracted much attention due to their high volumetric hydrogen storage capacity and high safety. Vanadium-based hydrogen storage alloys, with their high theoretical hydrogen storage capacity, suitable hydrogen absorption plateau pressure, and good cycle stability, have become a key research direction for medium- and low-temperature hydrogen storage materials.

[0003] However, vanadium-based hydrogen storage alloys are typically multi-element systems (such as V-Ti-Cr-Fe, V-Ti-Cr-Mn, etc.), with complex compositions and unclear interaction mechanisms between elements, making alloy performance extremely sensitive to composition. Traditional methods rely on limited empirical adjustments or simple single-variable experiments, making it difficult to achieve synergistic control of multiple elements and to quickly identify high-performance vanadium-based hydrogen storage alloys within a broad compositional space. Therefore, traditional screening methods are inefficient and costly. In particular, there is a lack of effective theoretical prediction for structural transformations such as lattice distortion and charge redistribution caused by hydrogen dynamic migration. It is urgent to clarify the compositional structure design concept, establish the structure-property relationship between composition and hydrogen storage capacity and performance, and elucidate the performance regulation mechanism to provide theoretical guidance for the development of room-temperature high-capacity, long-life vanadium-based hydrogen storage alloys. Summary of the Invention

[0004] The purpose of this invention is to address the problem of low screening efficiency in traditional methods that rely heavily on limited empirical adjustments or simple single-variable experiments to screen vanadium-based hydrogen storage alloys with high-performance components. This invention proposes a method for small-scale control of the composition of multi-component vanadium-based hydrogen storage alloys.

[0005] A method for controlling the composition of a multi-component vanadium-based hydrogen storage alloy within a small range, the method comprising the following:

[0006] Step 1: Obtain the feature set: Obtain m features of various vanadium-based hydrogen storage alloys from the database, and the values ​​of these m features meet the requirements of the application scenario. The m features include alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption pressure, hydrogen release pressure, hydrogen absorption entropy, hydrogen release entropy, and effective hydrogen storage capacity. The various vanadium-based hydrogen storage alloys are composed of at least two elements.

[0007] Step 2: Establish sample and feature prediction model: Take one feature (excluding alloy composition) of each vanadium-based hydrogen storage alloy as the output data of one sample, and take the corresponding remaining features as the input data of one sample, forming a total of m-1 types of samples; and construct a feature prediction model.

[0008] Step 3: Select a vanadium-based hydrogen storage alloy with a certain number of components: Train the feature prediction model sequentially using each type of sample to obtain the trained feature prediction model, and use the evaluation index to evaluate the prediction performance of the trained feature prediction model for various vanadium-based hydrogen storage alloys. Select the vanadium-based hydrogen storage alloy with the best prediction performance according to the evaluation index, select each element from the vanadium-based hydrogen storage alloy, and preset the composition range for each selected element.

[0009] Step 4: Predict the effective hydrogen storage capacity of the alloy: Sequentially extract one fixed value from the preset composition range corresponding to each element. Multiple elements are composed of their respective fixed value compositions to form a new alloy. Obtain m-1 features of the new alloy, including alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption pressure, hydrogen release pressure, hydrogen absorption entropy, and hydrogen release entropy. Input these features into the trained feature prediction model and output the effective hydrogen storage capacity of each element.

[0010] Step 5: Screening the optimal vanadium-based hydrogen storage alloy: Select the new alloy corresponding to the maximum value from multiple effective hydrogen storage capacities as the optimal vanadium-based hydrogen storage alloy. If there are multiple new alloys corresponding to multiple maximum values, select an alloy composition from the alloy compositions corresponding to the multiple maximum values ​​based on the cost of the alloy as the optimal vanadium-based hydrogen storage alloy.

[0011] Preferably, the application scenario in step 1 refers to an on-board hydrogen storage system, a stationary energy storage system, or a hydrogen fuel cell buffer tank.

[0012] Preferably, the hydrogen absorption temperature, hydrogen release temperature, and hydrogen absorption / release pressure required by the on-board hydrogen storage system are as follows: hydrogen absorption below 10°C / 5MPa, and hydrogen release above 60°C / 0.5MPa.

[0013] Preferably, the various vanadium-based hydrogen storage alloys include VTiCrFe vanadium-based hydrogen storage alloy, VTiCrNi vanadium-based hydrogen storage alloy, VTiCrMn vanadium-based hydrogen storage alloy, and VTiCrFeAi vanadium-based hydrogen storage alloy.

[0014] Preferably, in step 3, the evaluation index refers to R. 2 Or RMSE.

[0015] Preferably, the feature prediction model includes one input layer, one hidden layer, and one output layer;

[0016] The input layer receives m-1 features, performs high-dimensional feature processing on the m-1 features to generate a high-dimensional feature vector, and transmits it to the hidden layer.

[0017] The high-dimensional feature processing is as follows: The Lasso regression method is used to calculate the coefficients of m-1 features. Features with coefficients equal to 0 are removed. Each remaining feature interacts with the component. The logarithm of the regularization parameter and the MAE loss value of each interaction feature are calculated. Interaction features with a logarithm of the regularization parameter less than a corresponding preset threshold and an MAE loss value less than a preset loss value are selected as high-order interaction features. The SHAP value of each high-order interaction feature is calculated. The SHAP values ​​of all high-order interaction features are sorted from largest to smallest. The high-order interaction features corresponding to the top k SHAP values ​​are selected to form a high-dimensional feature vector.

[0018] Hidden layers are used to map high-dimensional feature vectors to an abstract feature space and then transmit the obtained abstract feature vectors to the output layer.

[0019] The output layer is used to cluster abstract feature vectors into a single value for output.

[0020] The beneficial effects of this invention are:

[0021] This invention first obtains m features of vanadium-based hydrogen storage alloys composed of different element groups from a database. Based on these m features, m-1 classes of samples are constructed and input into the feature prediction model constructed in this invention. Evaluation metrics are used to assess the predictive performance of the trained feature prediction model for various vanadium-based hydrogen storage alloys. Based on the evaluation metrics, the vanadium-based hydrogen storage alloy with the best predictive performance is selected. Composition ranges are preset for each element in this vanadium-based hydrogen storage alloy with these composition ranges, forming alloys with these ranges. All possible alloys are extracted from these ranges to predict their corresponding effective hydrogen storage capacity. The alloy with the maximum value is selected as the optimal vanadium-based hydrogen storage alloy. This invention's constructed feature prediction model has high prediction accuracy and efficiency, especially for vanadium-based hydrogen storage alloys with the selected single element group; therefore, the optimal alloy selected using the effective hydrogen storage alloy is more accurate.

[0022] Investigating the effects of material composition, unit cell geometry, and valence electron concentration on hydrogen absorption / desorption capacity and thermodynamic properties, and developing high-performance vanadium-based hydrogen storage alloys, is a key scientific issue. This invention explores the influence of material composition on hydrogen storage performance, screens and optimizes the elemental composition of vanadium-based alloys, and determines the optimal elemental combination; in the quaternary and pentagonal vanadium-based hydrogen storage alloys, the V content is between 70-85%, the Ti:Cr ratio is between 0.5-0.8, and the Fe content is within 3%. When the Fe content is within 3%, the effective hydrogen storage capacity is 2.15-2.7 wt.%. Attached Figure Description

[0023] Figure 1 A flowchart of a method for controlling the composition of a multi-component vanadium-based hydrogen storage alloy within a small range;

[0024] Figure 2 For alloy feature maps within the database;

[0025] Figure 3(a) shows the R obtained by predicting the maximum hydrogen storage capacity using six models. 2 Resulting image;

[0026] Figure 3(b) shows the R obtained by predicting the effective hydrogen storage capacity using six models. 2 Resulting image;

[0027] Figure 3(c) shows the R obtained by predicting hydrogen absorption pressure using six models. 2 Resulting image;

[0028] Figure 3(d) shows the R values ​​obtained by predicting hydrogen release pressure using six models. 2 Resulting image;

[0029] Figure 3(e) shows the R obtained by predicting hydrogen absorption entropy using six models. 2 Resulting image;

[0030] Figure 3(f) shows the R obtained by predicting the hydrogen exothermic entropy using six models. 2 Resulting image;

[0031] Figure 4(a) shows the evaluation results generated by using RF to predict hydrogen absorption pressure. The horizontal axis represents the true value, the vertical axis represents the predicted value, the red diagonal line represents the training set, and the blue dots represent the test set.

[0032] Figure 4(b) shows the evaluation results generated by using RF to predict hydrogen release pressure;

[0033] Figure 4(c) shows the evaluation results generated by using RF to predict the effective hydrogen storage capacity;

[0034] Figure 5(a) shows the composition of VTiCrFe vanadium-based hydrogen storage alloy predicted by the RF model based on different hydrogen absorption pressures;

[0035] Figure 5(b) shows the composition of VTiCrFe vanadium-based hydrogen storage alloy predicted by the RF model based on different hydrogen release pressures;

[0036] Figure 5(c) shows the composition of VTiCrFe vanadium-based hydrogen storage alloy predicted by the RF model based on different effective hydrogen storage capacities;

[0037] Figure 6(a) shows the composition of VTiCrFeAi vanadium-based hydrogen storage alloy predicted by the RF model based on different hydrogen absorption pressures;

[0038] Figure 6(b) shows the composition of the VTiCrFeAi vanadium-based hydrogen storage alloy predicted by the RF model based on different hydrogen release pressures;

[0039] Figure 6(c) shows the composition of VTiCrFeAi vanadium-based hydrogen storage alloy predicted by the RF model based on different effective hydrogen storage capacities;

[0040] Figure 7(a) shows the composition of VTiCrFeMn vanadium-based hydrogen storage alloy predicted by the RF model based on different hydrogen absorption pressures;

[0041] Figure 7(b) shows the composition of VTiCrFeMn vanadium-based hydrogen storage alloy predicted by the RF model based on different hydrogen release pressures;

[0042] Figure 7(c) shows the composition of VTiCrFeMn vanadium-based hydrogen storage alloy predicted by the RF model based on different effective hydrogen storage capacities;

[0043] Figure 8(a) shows V 82 Ti7Cr9Fe 1.2 Al 0.8 In the PCT performance curve at 10°C, the horizontal axis represents the effective mass hydrogen storage density, and the vertical axis represents the hydrogen absorption pressure.

[0044] Figure 8(b) shows V 82 Ti7Cr9Fe 1.2 Al 0.8 In the PCT performance curve at 60°C, the horizontal axis represents the effective mass hydrogen storage density, and the vertical axis represents the hydrogen release pressure.

[0045] Figure 9 The coefficient for each feature varies. A trajectory diagram of the changes;

[0046] Figure 10(a) shows the relationship between Log(alpha) of effective hydrogen capacity and MAE loss value;

[0047] Figure 10(b) shows the relationship between the hydrogen release pressure (Log(alpha)) and the MAE loss value;

[0048] Figure 10(c) shows the relationship between the hydrogen absorption pressure (Log(alpha)) and the MAE loss value.

[0049] Figure 11 This is a schematic diagram of the feature prediction model. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0052] Example:

[0053] Combination Figure 1 This embodiment of the specification discloses a method for controlling the composition of a multi-component vanadium-based hydrogen storage alloy within a small range. The method includes the following:

[0054] Step 1: Obtain the feature set: Obtain m features of various vanadium-based hydrogen storage alloys from the database, and the values ​​of these m features meet the requirements of the application scenario. The m features include alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption pressure, hydrogen release pressure, hydrogen absorption entropy, hydrogen release entropy, and effective hydrogen storage capacity. The various vanadium-based hydrogen storage alloys are composed of at least two elements.

[0055] Step 2: Establish sample and feature prediction model: Take one feature (excluding alloy composition) of each vanadium-based hydrogen storage alloy as the output data of one sample, and take the corresponding remaining features as the input data of one sample, forming a total of m-1 types of samples; and construct a feature prediction model.

[0056] like: Figure 2 The database includes vanadium-based hydrogen storage alloys with different components, and multiple characteristics of each alloy. Each alloy has the same number of characteristics. Based on the application scenario, such as an alloy required for an on-board hydrogen storage system where the absorption temperature, release temperature, and pressure are below 10°C / 5MPa for absorption and above 60°C / 0.5MPa for release, alloys meeting these conditions are first selected from the database. For example, if V... 70 Ti2Cr5Fe3 vanadium-based hydrogen storage alloy and V 72 Ti3Cr9Fe3Ai 0.8 The alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption / desorption pressure, hydrogen absorption / desorption entropy, and effective hydrogen storage capacity of vanadium-based hydrogen storage alloys;

[0057] The process of forming m-1 type samples is as follows: The hydrogen absorption temperature of the two alloys is used as output data, and the alloy composition, hydrogen release temperature, hydrogen absorption / desorption pressure, hydrogen absorption / desorption entropy, and effective hydrogen storage capacity are used as input data to form the first sample; the hydrogen release temperature of the two alloys is used as output data, and the alloy composition, hydrogen absorption temperature, hydrogen absorption / desorption pressure, hydrogen absorption / desorption entropy, and effective hydrogen storage capacity are used as input data to form the second sample; the hydrogen absorption pressure of the two alloys is used as output data, and the alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen release pressure, hydrogen absorption / desorption entropy, and effective hydrogen storage capacity are used as input data to form the third sample; the hydrogen release .... The first sample consists of four parts: a fourth sample, using the hydrogen absorption temperature, hydrogen absorption pressure, hydrogen absorption / desorption entropy, and effective hydrogen storage capacity as input data; a fifth sample, using the hydrogen absorption entropy of the two alloys as output data and the alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption pressure, hydrogen release pressure, hydrogen release entropy, and effective hydrogen storage capacity as input data; a sixth sample, using the hydrogen release entropy of the two alloys as output data and the alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption pressure, hydrogen release pressure, hydrogen absorption entropy, and effective hydrogen storage capacity as input data; and a seventh sample, using the effective hydrogen storage capacity of the two alloys as output data and the alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption pressure, hydrogen release pressure, hydrogen absorption entropy, and hydrogen release entropy as input data.

[0058] Step 3: Select a vanadium-based hydrogen storage alloy with a certain number of components: Train the feature prediction model sequentially using each type of sample to obtain the trained feature prediction model, and use the evaluation index to evaluate the prediction performance of the trained feature prediction model for various vanadium-based hydrogen storage alloys. Select the vanadium-based hydrogen storage alloy with the best prediction performance according to the evaluation index, select each element from the vanadium-based hydrogen storage alloy, and preset the composition range for each selected element.

[0059] For example: based on the feature prediction model and m-1 types of samples, calculate m-1 evaluation indicators for each vanadium-based hydrogen storage alloy, select the vanadium-based hydrogen storage alloy with the most evaluation indicators exceeding the threshold, select each element from the vanadium-based hydrogen storage alloy, and preset the composition range for each selected element.

[0060] Specifically, a portion of the m-1 class samples is selected as the training set to train the model. After the model is trained, each sample of each alloy is sequentially input into the trained model to output the predicted value. Based on the predicted value and the output data in each sample, the evaluation index is calculated. If each alloy obtains m-1 evaluation indices and all of them exceed the threshold, it indicates that the alloy has the best prediction performance and is selected.

[0061] To demonstrate the good prediction performance of the feature prediction model constructed in this embodiment, the following verification methods were used:

[0062] By simultaneously training n different models, including the SVM, boosting model, ensemble model, and feature prediction model shown in Figure 3, on each class of samples, multiple trained network models are obtained. The R-values ​​of these n trained network models for each class of samples are then calculated. 2 Or RMSE, if calculating R 2 n R values ​​obtained from each class of samples 2 Choose one R from each 2 The trained network model corresponding to the maximum value is used as the preferred model for each type of sample. The preferred model with the largest proportion among the preferred models of the eight types of samples is selected as the optimal network model. That is, as can be seen from Figure 3(a) to Figure 3(f), the blue feature prediction model has the highest fitting degree for each output data. Therefore, the feature prediction model is selected as the optimal network model. From the R2 analysis, it can be seen that compared with SVM, boosting and ensemble models, the feature prediction model has the highest prediction evaluation of the alloy hydrogen storage performance, which can reach 0.935.

[0063] The process of selecting a vanadium-based hydrogen storage alloy with a certain number of components is as follows: the hydrogen absorption pressure, hydrogen release pressure and effective hydrogen storage capacity are predicted by the characteristic prediction model, and R2 and RMSE are calculated. The characteristic prediction model is evaluated by the fitting degree R2 and RMSE evaluation index, and the evaluation results are shown in Figures 4(a) to 4(c). As can be seen from the figures, the R2 value is between 0.69 and 0.7, indicating that the model is suitable for predicting the composition of vanadium-based quaternary hydrogen storage alloys.

[0064] Step 4: Predict the effective hydrogen storage capacity of the alloy: Sequentially extract one fixed value from the preset composition range corresponding to each element. Multiple elements are composed of their respective fixed value compositions to form a new alloy. Obtain m-1 features of the new alloy, including alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption pressure, hydrogen release pressure, hydrogen absorption entropy, and hydrogen release entropy. Input these features into the trained feature prediction model and output the effective hydrogen storage capacity of each element.

[0065] Based on the prediction results of the feature prediction model, the composition distribution diagrams of quaternary V-based hydrogen storage alloys with different hydrogen absorption / desorption pressures and different effective hydrogen storage capacities were plotted. Analysis of the figures shows that the alloy with V content between 0.75 and 0.85%, TiCr content between 0.25 and 0.30%, and Fe content between 0 and 0.1% exhibits excellent effective capacity (Figures 5(a) to 5(c), 6(a) to 6(c), and 7(a) to 7(c)). Based on this, the vanadium-based hydrogen storage alloy composition with excellent effective hydrogen storage capacity was selected as V78-82Ti8-10Cr10-12Fe2-3.

[0066] Step 5: Screening the optimal vanadium-based hydrogen storage alloy: Select the new alloy corresponding to the maximum value from multiple effective hydrogen storage capacities as the optimal vanadium-based hydrogen storage alloy. If there are multiple new alloys corresponding to multiple maximum values, select an alloy composition from the alloy compositions corresponding to the multiple maximum values ​​based on the cost of the alloy as the optimal vanadium-based hydrogen storage alloy.

[0067] Specifically, cost can refer to the cost of producing the alloy.

[0068] Based on 240 articles and over forty characteristic statistical data from databases, a series of characteristic data for vanadium-based hydrogen storage alloys were extracted and data quality controlled, ultimately constructing a dataset of key parameters for vanadium-based hydrogen storage alloys (Table 1). To predict the hydrogen storage capacity of V-Ti-Cr-M (M=Mn, Fe, Ni) and V-Ti-Fe-Cr alloys, a targeted hydrogen storage dataset containing the aforementioned elemental systems was constructed. The retrieved and included data included phase, effective hydrogen capacity, temperature, hydrogen absorption pressure, and hydrogen release pressure.

[0069] Table 1 Key Parameter Dataset for Vanadium-Based Hydrogen Storage Alloys

[0070]

[0071]

[0072] Studies have found that Al (1.61) has a higher electronegativity than both V (1.63) and Ti (1.54). The addition of Al alters the density of electronic states near the Fermi level, enhancing the bonding force between the metal and hydrogen atoms. The rare earth element Ce can effectively improve the hydrogen desorption plateau pressure of vanadium-based alloys. Based on this, alloys containing Al can be selected from the database, and the hydrogen absorption capacity and desorption plateau pressure can be further improved by adding Al and Ce. A high-capacity, long-life vanadium-based hydrogen storage alloy with room temperature capability has been developed, capable of absorbing hydrogen below 10°C / 5MPa and desorbing hydrogen above 60°C / 0.5MPa, with an effective mass hydrogen storage density of 2.32 wt.%.

[0073] To further define, the application scenarios in step 1 refer to on-board hydrogen storage systems, stationary energy storage systems, or hydrogen fuel cell buffer tanks.

[0074] Further specifying, the various vanadium-based hydrogen storage alloys include VTiCrFe vanadium-based hydrogen storage alloy, VTiCrNi vanadium-based hydrogen storage alloy, VTiCrMn vanadium-based hydrogen storage alloy, and VTiCrFeAi vanadium-based hydrogen storage alloy.

[0075] Further specifying, in step 3, the evaluation indicator refers to R. 2 Or RMSE.

[0076] Further specifying, the feature prediction model includes one input layer 1, one hidden layer 2, and one output layer 3;

[0077] Input layer 1 is used to receive m-1 features, perform high-dimensional feature processing on the m-1 features, generate high-dimensional feature vectors, and transmit them to hidden layer 2;

[0078] The high-dimensional feature processing is as follows: The Lasso regression method is used to calculate the coefficients of m-1 features. Features with coefficients equal to 0 are removed. Each remaining feature interacts with the component. The logarithm of the regularization parameter and the MAE loss value of each interaction feature are calculated. Interaction features with a logarithm of the regularization parameter less than a corresponding preset threshold and an MAE loss value less than a preset loss value are selected as high-order interaction features. The SHAP value of each high-order interaction feature is calculated. The SHAP values ​​of all high-order interaction features are sorted from largest to smallest. The high-order interaction features corresponding to the top k SHAP values ​​are selected to form a high-dimensional feature vector.

[0079] Hidden layer 2 is used to map high-dimensional feature vectors to an abstract feature space and transmit the obtained abstract feature vectors to output layer 3;

[0080] Output layer 3 is used to cluster the abstract feature vectors into a single value for output.

[0081] Specifically, Lasso regression, as a regularization technique, introduces an L1 regularization term into the regression model, which compresses the coefficients of higher-order features to zero, thereby achieving automatic variable selection.

[0082] Lasso regression is used to select features for m-1 classes. Figure 9 Regularization parameters in Lasso regression The effect of (logarithmic form Log(alpha)) on the model coefficients.

[0083] Figure 9 The horizontal axis represents the regularization parameter. The logarithm of . With As the value increases (from right to left), the strength of regularization increases.

[0084] Figure 9 The vertical axis represents the coefficients of each feature in the model. The magnitude and sign of the coefficients reflect the direction and strength of the feature's influence on the target variable.

[0085] Figure 9 The curve represents the coefficient of each feature as... The trajectory of change; when the coefficient curves of certain higher-order features eventually reach zero, this indicates that these features are considered unimportant by Lasso regression and are thus excluded from the final model. From this, appropriate higher-order features can be selected.

[0086] As shown in Figures 10(a) to 10(c), the optimal regularization parameter with the minimum MAE loss is further found. This enables better simulation of the data.

[0087] We can see the optimal regularization parameter for the interaction between effective hydrogen capacity and composition. The MAE loss is minimized when the value is around 0.05. The optimal regularization parameter for plateau pressure release, plateau pressure absorption, and component interaction is determined. The MAE loss is minimized when the value is around 0.5.

[0088] Experimental verification:

[0089] The optimal vanadium-based hydrogen storage alloy was prepared using smelting technology. The effective hydrogen storage capacity of the prepared optimal vanadium-based hydrogen storage alloy was tested. This capacity was the same as the predicted capacity value, thereby verifying the accuracy of the feature prediction model in this embodiment.

[0090] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A method for controlling the composition of a multi-component vanadium-based hydrogen storage alloy within a small range, characterized in that, The method includes the following: Step 1: Obtain the feature set: Obtain m features of various vanadium-based hydrogen storage alloys from the database, and the values ​​of these m features meet the requirements of the application scenario. The m features include alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption pressure, hydrogen release pressure, hydrogen absorption entropy, hydrogen release entropy, and effective hydrogen storage capacity. Various vanadium-based hydrogen storage alloys are composed of at least two elements. Step 2: Establish sample and feature prediction model: Take one feature (excluding alloy composition) of each vanadium-based hydrogen storage alloy as the output data of one sample, and take the corresponding remaining features as the input data of one sample, forming a total of m-1 types of samples; and construct a feature prediction model. Step 3: Select a vanadium-based hydrogen storage alloy with a certain number of components: Train the feature prediction model sequentially using each type of sample to obtain the trained feature prediction model, and use the evaluation index to evaluate the prediction performance of the trained feature prediction model for various vanadium-based hydrogen storage alloys. Select the vanadium-based hydrogen storage alloy with the best prediction performance according to the evaluation index, select each element from the vanadium-based hydrogen storage alloy, and preset the composition range for each selected element. Step 4: Predict the effective hydrogen storage capacity of the alloy: Sequentially extract one fixed value from the preset composition range corresponding to each element. Multiple elements are composed of their respective fixed value compositions to form a new alloy. Obtain m-1 features of the new alloy, including alloy composition, hydrogen absorption temperature, hydrogen release temperature, hydrogen absorption pressure, hydrogen release pressure, hydrogen absorption entropy, and hydrogen release entropy. Input these features into the trained feature prediction model and output the effective hydrogen storage capacity of each element. Step 5: Screening the optimal vanadium-based hydrogen storage alloy: Select the new alloy corresponding to the maximum value from multiple effective hydrogen storage capacities as the optimal vanadium-based hydrogen storage alloy. If there are multiple new alloys corresponding to multiple maximum values, select an alloy composition from the alloy compositions corresponding to the multiple maximum values ​​based on the cost of the alloy as the optimal vanadium-based hydrogen storage alloy.

2. The method for small-range control of the composition of a multi-component vanadium-based hydrogen storage alloy according to claim 1, characterized in that, The application scenarios in step 1 refer to on-board hydrogen storage systems, stationary energy storage systems, or hydrogen fuel cell buffer tanks.

3. The method for small-scale control of the composition of a multi-component vanadium-based hydrogen storage alloy according to claim 2, characterized in that, The required hydrogen absorption temperature, hydrogen release temperature, and hydrogen absorption / release pressure for on-board hydrogen storage systems are as follows: hydrogen absorption below 10°C / 5MPa, and hydrogen release above 60°C / 0.5MPa.

4. The method for small-scale control of the composition of a multi-component vanadium-based hydrogen storage alloy according to claim 1, characterized in that, Various vanadium-based hydrogen storage alloys include VTiCrFe vanadium-based hydrogen storage alloy, VTiCrNi vanadium-based hydrogen storage alloy, VTiCrMn vanadium-based hydrogen storage alloy, and VTiCrFeAi vanadium-based hydrogen storage alloy.

5. The method for small-range control of the composition of a multi-component vanadium-based hydrogen storage alloy according to claim 1, characterized in that, In step 3, the evaluation index refers to R. 2 Or RMSE.

6. The method for small-range control of the composition of a multi-component vanadium-based hydrogen storage alloy according to claim 1, characterized in that, The feature prediction model consists of one input layer (1), one hidden layer (2), and one output layer (3). The input layer (1) is used to receive m-1 features, perform high-dimensional feature processing on the m-1 features, generate high-dimensional feature vectors, and transmit them to the hidden layer (2). The high-dimensional feature processing is as follows: The Lasso regression method is used to calculate the coefficients of m-1 features. Features with coefficients equal to 0 are removed. Each remaining feature interacts with the component. The logarithm of the regularization parameter and the MAE loss value of each interaction feature are calculated. Interaction features with a logarithm of the regularization parameter less than a corresponding preset threshold and an MAE loss value less than a preset loss value are selected as high-order interaction features. The SHAP value of each high-order interaction feature is calculated. The SHAP values ​​of all high-order interaction features are sorted from largest to smallest. The high-order interaction features corresponding to the top k SHAP values ​​are selected to form a high-dimensional feature vector. Hidden layer (2) is used to map high-dimensional feature vectors to an abstract feature space and transmit the obtained abstract feature vectors to the output layer (3). The output layer (3) is used to cluster the abstract feature vectors into a single value and output it.

Images