Magnesium air battery negative electrode component design method and preparation method based on interpretable machine learning

Abstract

The invention discloses a magnesium air battery negative electrode component design method based on interpretable machine learning and a preparation method, and relates to the technical field of magnesium air batteries. The method aims at designing magnesium alloy components capable of improving discharge voltage and negative electrode efficiency at the same time by using an explanatory machine learning model, particularly adopting a light-weight gradient elevator algorithm (LightGBM) model and explainable analysis and combining multi-objective optimization. According to the method, the feature importance of magnesium alloy components is analyzed based on a machine learning model, the Mg-0. 5Zn-1. 5Ga component is preferably selected for contributions of Al, Zn, Ca, Ga, In and other elements to the discharge voltage and the negative electrode efficiency, and compared with a high-purity magnesium negative electrode material, the negative electrode efficiency and the discharge voltage of the magnesium alloy negative electrode material are both improved. The invention further relates to a smelting preparation method of the magnesium alloy negative electrode material, and the method is simple and easy to operate.

Description

Technical Field

[0001] The present invention relates to the technical field of magnesium-air batteries, and in particular to a design method and a preparation method of a magnesium alloy negative electrode composition based on explainable machine learning. Background Art

[0002] Compared with traditional lithium-ion batteries, magnesium-air batteries have the advantages of abundant magnesium resources, environmental friendliness, and high theoretical energy density. These advantages make it show broad application prospects in application fields such as long-endurance drones, electric vehicles, and backup power supplies. Although magnesium-air batteries have many performance advantages, they still face problems of poor discharge voltage and low discharge efficiency in practical applications. These problems directly affect the overall performance of magnesium-air batteries and limit their widespread application in high-performance energy storage devices. Therefore, simultaneously improving the battery discharge voltage and negative electrode efficiency is the key to expanding the application range of magnesium-air batteries.

[0003] Alloying is one of the main means to solve these problems. By introducing other alloying elements (such as Zn, Ca, Al, Ga, etc.) into magnesium, the self-corrosion rate and discharge activity of magnesium alloys can be effectively improved. However, traditional alloy composition design methods often rely on repeated trial and error and experience accumulation in the laboratory, which is not only a lengthy process, but also has high R&D costs, and it is difficult to meet the needs of today's society for the development of efficient materials. With the rapid development of data science and computer technology, material design methods based on machine learning have become an increasing research hotspot.

[0004] Machine learning, especially interpretable machine learning technology, can analyze large amounts of experimental data and obtain key factors that affect material performance, thereby accelerating the process of material design. Compared to traditional black-box machine learning models, interpretable machine learning can not only predict performance but also provide an intuitive explanation of model decisions, allowing researchers to understand the specific contribution of each alloying element to material performance. This not only improves design efficiency, but also helps to clarify the decision-making basis of machine learning and the knowledge laws mined, providing a statistical basis and theoretical guidance for subsequent composition design.

[0005] Based on this background, the present invention combines interpretable machine learning and optimization algorithms, and through intelligent design of the composition of magnesium alloy negative electrode materials, selects the components that can simultaneously improve the discharge voltage and negative electrode efficiency, and explores its smelting preparation process. This design and preparation method can effectively improve the overall performance of magnesium-air batteries, and has significant theoretical value and practical application prospects. Summary of the invention

[0006] The present invention relates to a design method for negative electrode components of magnesium-air batteries based on interpretable machine learning and a preparation method thereof, aiming to solve the problems of low discharge efficiency and poor discharge voltage of existing magnesium alloy negative electrode materials. By combining interpretable machine learning algorithms with multi-objective optimization technology, the present invention can quickly screen out components that can simultaneously improve negative electrode efficiency and discharge voltage, prepare magnesium alloy negative electrode materials through a smelting process based on the optimized alloy components, assemble magnesium-air batteries and test performance.

[0007] The present invention provides a method for designing a composition of a negative electrode material for a magnesium-air battery based on interpretable machine learning and a method for preparing the same, the specific steps of which are as follows:

[0008] Step 1: Data collection and processing

[0009] Through experiments or literature research, a data set is collected, including experimental data on magnesium alloy composition, current density, discharge voltage, and negative electrode efficiency; the features in the data set include but are not limited to: the content of each element in the magnesium alloy (such as Zn, Al, Ca, Ga, In, Sn, Mn, Y, Ce, Sm), current density, and the target variables in the data set are discharge voltage and negative electrode efficiency; the data set only retains negative electrode data prepared by traditional smelting, without deformation or heat treatment after cooling, with a positive electrode catalyst of C / MnO2 and an electrolyte of 3.5wt.% NaCl; for samples with the same composition but different reported performance in different literatures, the average performance is used as the target performance of machine learning;

[0010] The data was preprocessed and the Pearson correlation coefficient was used to test the linear correlation between the descriptors to delete the descriptors with high linear correlation. For example, when the Pearson correlation coefficient of two descriptors was high (>0.9), one of them was removed; the data set was divided into a training set and a test set (7:3).

[0011] Step 2. Construction of machine learning model

[0012] The Lightweight Gradient Boosting Machine (LightGBM) model is used to build prediction models for the target variables discharge voltage and negative electrode efficiency. The parameters that need to be set before training the machine learning model are hyperparameters, and in order to reduce the randomness of data division and improve the generalization prediction ability of the model, a ten-fold cross validation is used to search for hyperparameters; then the hyperparameters are set, and the model is trained using the training set. The input of the model is the alloy composition and current density, and the output is the discharge voltage or negative electrode efficiency to obtain the prediction model; the prediction ability of the prediction model is evaluated using the test set, and indicators such as the mean square error (MSE) are used for evaluation to ensure that the prediction model has a high prediction accuracy;

[0013] Step 3. Feature Importance Evaluation and Interpretability Analysis

[0014] After calculating the LightGBM prediction model in step 2, the built-in feature importance calculation of the model is used to obtain the importance coefficient of each feature, and the contribution of each alloy component to the target performance is analyzed; several alloying elements that are most important to the target performance are screened out; in addition, the trained prediction model is explained using the Shapley Additive exPlanations (SHAP) algorithm, which is applicable to any machine learning model (including LightGBM). The more significant and important feature is that it can quantify the positive and negative correlation of each feature to the model prediction result, i.e., the target, and thus can intuitively explain the influence of the above-mentioned relatively important alloying elements on the target performance, thereby screening out two alloying elements that have a positive effect on the negative electrode efficiency and discharge voltage among several relatively important alloying elements;

[0015] Step 4. Multi-objective optimization design

[0016] Based on the importance of each feature and the interpretable analysis results obtained in step 3, the non-dominated sorting genetic algorithm II (NSGA-II) is used to perform multi-objective optimization design of discharge voltage and efficiency, and a set of Pareto optimal solutions that meet high discharge voltage and high negative electrode efficiency are obtained.

[0017] Step 5. Alloy preparation

[0018] According to the optimized composition and proportion designed in step 4, raw materials such as pure magnesium, pure metal and intermediate alloy are selected, and alloying treatment is performed by smelting process; magnesium and other metals or intermediate alloys are melted under protective atmosphere. The molten alloy is cast into ingots to obtain alloy materials.

[0019] Step 6. Forming and performance testing of magnesium alloy negative electrode

[0020] Processing the prepared magnesium alloy material into a negative electrode plate suitable for magnesium-air batteries, ensuring that its size and shape meet the battery design requirements;

[0021] The prepared magnesium alloy negative electrode was tested for discharge performance, including discharge voltage and negative electrode efficiency. The effect of the optimized alloy composition on improving negative electrode efficiency and discharge voltage was evaluated through experiments.

[0022] application

[0023] The magnesium alloy designed and prepared by the present invention can be used as the screening of negative electrode materials for magnesium-air batteries, and can effectively improve the discharge efficiency and discharge voltage of the battery in a typical sodium chloride solution, thereby improving the battery energy density.

[0024] Advantages of the invention

[0025] The present invention breaks through the limitations of traditional trial-and-error design by introducing interpretable machine learning and multi-objective optimization methods, and realizes efficient design of magnesium alloy negative electrode materials. In addition, the method of the present invention can greatly improve the discharge performance of magnesium-air batteries, especially in terms of discharge voltage and negative electrode efficiency. The magnesium alloy negative electrode material designed by the present invention has high industrial application potential and can be used in the fields of magnesium-air batteries, magnesium-based fuel cells, etc., which will help promote the rapid development of the next generation of efficient and environmentally friendly battery materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flow chart for designing magnesium alloy anode composition and preparing magnesium anode based on interpretable machine learning;

[0027] Figure 2 To explain the prediction performance of the machine learning model on the negative electrode efficiency and discharge voltage, including its performance in the training set and the test set;

[0028] Figure 3 Importance coefficients and SHAP values ​​of features used in interpretable machine learning models;

[0029] Figure 4 Mg-0.5Zn-1.5Ga and high purity magnesium at 1mAcm -2 and 10mAcm -2 The discharge curve. DETAILED DESCRIPTION

[0030] In order to further illustrate the technical solution of the present invention, the following is a detailed description in conjunction with specific embodiments. It is pointed out that these embodiments are only used to illustrate the present invention, rather than to limit the scope of rights of the present invention.

[0031] Example 1

[0032] 1. Data collection and feature engineering

[0033] Figure 1 The flowchart of the whole process is shown in . First, the experimental data of magnesium alloy anode materials are collected, including anode efficiency, discharge voltage and corresponding magnesium alloy components (Al, Zn, Ca, In, Sn, Ga, Mn, Y, Ce, Sm). The data set contains 230 samples (some sample examples are given in Table 1). No highly linearly correlated variables were found in the feature linear correlation. The LightGBM hyperparameters were obtained based on ten-fold cross validation. The collected data set is divided into two parts, which are used for training the anode efficiency model and the discharge voltage model respectively. The ratio of the training set to the test set is 7:3.

[0034] Table 1: Some samples in the data set, the characteristic variables include alloy element content and current density, and the target performance is discharge voltage and negative electrode efficiency

[0035]

[0036]

[0037] 2. Model training and prediction

[0038] Figure 2 The prediction results of the training set and test set for negative electrode efficiency and discharge voltage are shown in Figure 2. It can be seen that the prediction results of the machine learning model on the training set and the test set are relatively close, indicating that the model has a good fitting effect. The mean square error in the test set is 0.008 (% 2 ) and 0.023(V 2 ).

[0039] 3. Feature Importance and Model Parsability Analysis

[0040] Figure 3 Figures a and b show the importance coefficients of each feature to the target performance. The results show that Al, Zn, Ca, Ga and In are the key elements that have the most significant impact on the target performance. Therefore, these elements can be used as the core basis for setting the alloy composition screening space. Further analysis Figure 3 The SHAP value distribution of these elements in c and d reveals their positive and negative correlation with negative electrode efficiency and discharge voltage: ①Al content: Figure 3 d shows that high Al content has a significant positive contribution to the discharge voltage. Figure 3 The results of the study show that high Al content may inhibit the improvement of negative electrode efficiency, which indicates that there is a trade-off in efficiency when Al increases the discharge voltage. ②Ca content: High Ca content has a significant positive effect on negative electrode efficiency ( Figure 3 c), but at the discharge voltage ( Figure 3 d) shows a negative effect. ③In content: Figure 3 Figure c shows that low In content helps improve the negative electrode efficiency. Figure 3 Figure d shows that low In content may significantly reduce the discharge voltage. The above results show that although Al, Ca and In play an important role in regulating the target performance, it is difficult to improve the negative electrode efficiency and discharge voltage at the same time. In contrast: ① Zn content: Low Zn content has a significant effect on negative electrode efficiency ( Figure 3 c) shows a positive effect and has a positive effect on the discharge voltage ( Figure 3 ②Ga content: High Ga content shows a significant positive contribution to the negative electrode efficiency ( Figure 3 c), although the effect on the discharge voltage is small ( Figure 3d), but still shows a positive trend. In summary, the combination of Zn and Ga has a significant synergistic effect, which can improve the negative electrode efficiency while maintaining or slightly increasing the discharge voltage. Therefore, the use of Zn and Ga as alloying elements is a reasonable strategy for preparing magnesium alloys with excellent performance.

[0041] 4. Multi-objective optimization design

[0042] In order to demonstrate the generalization ability of the model in component design, a component space composed of Zn and Ga elements was selected for screening. The space is based on a gradient of 0.5wt.%, from 0.5wt.% to 10wt.%, and all possible combinations are listed. Subsequently, the NSGA-II algorithm was used to search the Pareto frontier. The results show that Mg-0.5Zn-1.5Ga is located on the Pareto frontier and is expected to improve both the negative electrode efficiency and the discharge voltage.

[0043] 5. Alloy preparation

[0044] The composition of the magnesium alloy negative electrode is set to 0.5wt.% Zn and 1.5wt.% Ga. The preheated pure magnesium is heated to 730°C, and a mixed protective gas of nitrogen and sulfur hexafluoride with a volume ratio of 19:1 is introduced at the same time. After the magnesium is completely melted, the temperature is adjusted to 720°C, and the preheated pure Zn and Ga are added in sequence, and kept standing for 15 minutes. Stir the alloy liquid for 2 minutes to ensure that the alloy elements are evenly distributed. Then, after removing the scum on the surface of the alloy liquid, immediately pour it into a metal mold preheated at 200°C, and wait for it to cool and solidify to form an alloy ingot.

[0045] 6. Construction and performance evaluation of magnesium-air batteries

[0046] Mg-0.5Zn-1.5Ga was used as the negative electrode, and the air electrode loaded with commercial MnO2 / C catalyst was used as the positive electrode. The electrolyte was 3.5wt.% NaCl aqueous solution. The test temperature was room temperature. The discharge test was carried out. The discharge voltage was tested using LAND electrical performance monitoring equipment (CT2001A). -2 and 10mA cm -2 ) for 10 hours, and the average voltage measured is taken as the discharge voltage ( Figure 4 ). At the same time, the discharge weight loss was used to test the utilization efficiency; the energy density was calculated by combining the discharge weight loss and the discharge voltage. The results are listed in Table 2.

[0047] Table 2: Comparison of discharge performance parameters at different current densities

[0048]

[0049] As can be seen from Table 2, the discharge performance of Mg-0.5Zn-1.5Ga is better than that of high-purity magnesium, achieving a simultaneous improvement in negative electrode efficiency and discharge voltage. This shows that the effect of designing negative electrode components based on interpretable machine learning and then simultaneously improving negative electrode efficiency and discharge voltage is quite significant.

[0050] Although preferred implementation cases have been listed and described in detail herein, those skilled in the art will appreciate that various improvements, additions, substitutions, etc. may be made without departing from the spirit of the invention, and these contents are considered to be within the scope of the invention as defined by the claims.

Claims

1. A method for designing negative electrode components of magnesium-air batteries based on interpretable machine learning, characterized in that: The following steps are involved: Step 1: Data collection and processing Through experiments or literature research, a data set is collected, including experimental data on magnesium alloy composition, current density, discharge voltage, and negative electrode efficiency; the features in the data set include but are not limited to: the content of each element in the magnesium alloy, current density, and the target variables in the data set are discharge voltage and negative electrode efficiency; Preprocess the data and use the Pearson correlation coefficient to test the linear correlation between the descriptors to remove the highly linearly correlated descriptors, for example, when the Pearson correlation coefficient of two descriptors is high (>0.9), remove one of them; divide the data set into a training set and a test set (for example, 7:3); Step 2. Construction of machine learning model The Lightweight Gradient Boosting Machine (LightGBM) model is used to build prediction models for the target variables discharge voltage and negative electrode efficiency. The parameters that need to be set before training the machine learning model are hyperparameters, and in order to reduce the randomness of data division and improve the generalization prediction ability of the model, a ten-fold cross validation is used to search for hyperparameters; then the hyperparameters are set, and the model is trained using the training set. The input of the model is the alloy composition and current density, and the output is the discharge voltage or negative electrode efficiency to obtain the prediction model; the prediction ability of the prediction model is evaluated using the test set, and indicators such as the mean square error (MSE) are used for evaluation to ensure that the prediction model has a high prediction accuracy; Step 3. Feature Importance Evaluation and Interpretability Analysis After calculating the LightGBM prediction model in step 2, use the built-in feature importance calculation of the model to obtain the importance coefficient of each feature, analyze the contribution of each alloy component to the target performance, and select the alloy elements that are most important to the target performance; In addition, the Shapley Additive exPlanations (SHAP) algorithm is used to interpret the trained prediction model, quantify the positive and negative correlation of each feature on the model prediction result, i.e., the target, and intuitively explain the influence of the above-screened relatively important alloying elements on the target performance, thereby screening out two alloying elements that have a positive effect on the negative electrode efficiency and discharge voltage from among the relatively important alloying elements. Step 4. Multi-objective optimization design Based on the importance of each feature and the interpretable analysis results obtained in step 3, the non-dominated sorting genetic algorithm II (NSGA-II) is used to perform multi-objective optimization design of discharge voltage and efficiency, and a set of Pareto optimal solutions that meet high discharge voltage and high negative electrode efficiency are obtained.

2. The method according to claim 1, characterized in that The elements in the magnesium alloy described in step 1 include one or more of Zn, Al, Ca, Ga, In, Sn, Mn, Y, Ce, and Sm.

3. The method according to claim 1, characterized in that In step 1, only the negative electrode data prepared by traditional smelting, without deformation treatment or heat treatment after cooling, with a positive electrode catalyst of C / MnO2 and an electrolyte of 3.5wt.% NaCl are retained in the collected data set.

4. The method according to claim 1, characterized in that In step 1, for samples with the same composition but different reported performance in different literatures, the average performance is used as the target performance for machine learning.

5. The method according to claim 1, characterized in that The alloy is prepared by optimizing the composition and proportion designed in step 4, selecting raw materials such as pure magnesium, pure metal and master alloy, and performing alloying treatment by smelting process; Under a protective atmosphere, magnesium and other metals or master alloys are melted, and the molten alloy is cast into an ingot to obtain an alloy material.

6. The method according to claim 5, characterized in that Also includes: Processing the prepared magnesium alloy material into a negative electrode plate suitable for magnesium-air batteries, ensuring that its size and shape meet the battery design requirements; The prepared magnesium alloy negative electrode was tested for discharge performance, including discharge voltage and negative electrode efficiency. The effect of the optimized alloy composition on improving negative electrode efficiency and discharge voltage was evaluated through experiments.

Images