Eutectic electrolyte ionic conductivity prediction method based on machine learning
Through the eutectic electrolyte ion conductivity prediction method based on machine learning, the time-consuming and cost-effective problem of traditional design methods is solved, and the rapid and accurate prediction of ion conductivity is achieved, and the efficiency and accuracy of electrolyte design are improved.
Patent Information
- Application Number
- CN202510225209.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-27
- Publication Date
- 2025-06-17
AI Technical Summary
Traditional methods take a long time and costly to design eutectic electrolytes, and it is difficult to effectively predict ionic conductivity, affecting battery performance.
Using the eutectic electrolyte ion conductivity prediction method based on machine learning, we collect the composition and structural information of known eutectic electrolytes, conduct feature engineering and feature screening, establish a relationship model between element composition and conductivity, and build a visual system for prediction.
It realizes fast and accurate prediction of ionic conductivity, saves experimental time and cost, and improves the efficiency and accuracy of electrolyte design.
Smart Images
Figure CN120164544A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of zinc-ion battery electrolytes, and particularly relates to a method for predicting the ionic conductivity of eutectic electrolytes based on machine learning, which is applied to the rapid and effective prediction of electrolyte ionic conductivity. Background Art
[0002] Energy and environment are two major issues that humanity must face in its survival and social development today. With the depletion of fossil resources such as coal and oil and the increasingly deteriorating environment, the development of renewable energy sources such as solar, wind, and hydropower has become a global trend. Batteries, as a highly efficient electrochemical energy storage device, are widely used in fields such as electric vehicles and mobile phone communications. However, primary batteries cause waste of resources, and traditional lead-acid batteries are prone to serious regional lead pollution. Aqueous zinc-ion batteries have the advantages of high energy density, good safety, and low cost, and have high application value and development prospects in the field of large-scale energy storage.
[0003] However, aqueous zinc-ion batteries still face challenges such as zinc dendrites and parasitic side reactions. To overcome these challenges, various electrolyte design strategies aimed at effectively regulating the Zn2+ solvation structure have been proposed. Among them, eutectic electrolytes have received extensive attention due to their high tunability, easy preparation, and environmental friendliness. While eutectic electrolytes are highly tunable in design, they also introduce complexity. Relying on traditional experimental trial-and-error methods is time-consuming and costly. Therefore, the present invention proposes a visualization system and method for predicting the ionic conductivity of eutectic electrolytes, which can accelerate the design of eutectic electrolytes and reduce the time and resources required by conventional trial-and-error methods. Summary of the Invention
[0004] In view of the deficiencies of the prior art, the present invention proposes a method for predicting the ionic conductivity of eutectic electrolytes based on machine learning, which can save experimental time, reduce experimental costs, and improve the accuracy of conductivity prediction.
[0005] The above object of the present invention is achieved by the following technical solutions:
[0006] A method for predicting the ionic conductivity of eutectic electrolytes based on machine learning, characterized by comprising the following steps:
[0007] Step 1: Collect the ionic conductivity of known eutectic electrolytes from relevant literature, obtain the composition and structure information of the corresponding eutectic electrolytes, and construct a data set, wherein the ionic conductivity is used as the output data of the subsequent machine learning model;
[0008] Step 2: Obtain the molar ratio and molecular structure information of each component of the eutectic electrolyte in the data set constructed in Step 1, wherein the molecular structure information of the eutectic electrolyte includes the atomic information of each component of the eutectic electrolyte;
[0009] Step 3: Perform feature engineering based on the molar ratio and molecular structure information of the eutectic electrolyte to obtain the features corresponding to the eutectic electrolyte;
[0010] Step 4: Use the wrapper feature selection method combined with one of the machine learning algorithms to screen the features, and select the optimal feature subset that has a greater impact on the ionic conductivity according to the evaluation index, which is used as the input data for the subsequent machine learning model;
[0011] Step 5: Use the optimal feature subset screened in Step 4 as the input and the ionic conductivity in the dataset constructed in Step 1 as the output. Train and test the ionic conductivity dataset through the machine learning algorithm, and obtain the minimum error by adjusting the model parameters to establish a relationship model between the element composition and the conductivity;
[0012] Step 6: Based on the relationship model between the element composition and the conductivity established in Step 5, construct a visualization system for the eutectic electrolyte ionic conductivity prediction model. Input the feature information of the eutectic electrolyte to be predicted through the visualization system and output its ionic conductivity.
[0013] Moreover, in Step 3, the implementation method for obtaining the features corresponding to the ionic conductivity of the eutectic electrolyte includes:
[0014] 3.1. Obtain the molecular structure of the eutectic electrolyte;
[0015] 3.2. Determine the number of atoms with a specific coordination number in the molecule;
[0016] 3.3. Combine the number of atoms with a specific coordination number with the molar ratio of each component of the eutectic electrolyte to obtain the features corresponding to the eutectic electrolyte.
[0017] Moreover, in Step 4, the selection of the optimal feature subset adopts one of the wrapper feature selection methods such as sequential forward selection and sequential backward selection; start from an empty feature set using the sequential forward selection method or start from all features using the sequential backward selection method, select features to form a feature subset and evaluate the feature subset. In each iteration of feature selection, according to the evaluation index, continue to add or remove features on the feature subset with the highest score to obtain the optimal feature subset; repeat this process until the set feature target number is reached.
[0018] Moreover, in Step 5, the constructed visualization system for the eutectic electrolyte ionic conductivity prediction model includes: a model selection module, a user input module, a feature acquisition module, an output result module, and a clearing module;
[0019] Model Selection Module: Used to provide custom selection of models. Users can select machine learning models according to their needs. Specifically, the model selection options include: CatBoost, XGBoost, Random Forest;
[0020] User Input Module: Used for users to input relevant information of the eutectic electrolyte to be predicted;
[0021] Feature Acquisition Module: Used to calculate the required features according to the model selection and user input information;
[0022] Output Result Module: Used to output the prediction result of the ionic conductivity of the eutectic electrolyte;
[0023] Clearing Module: Used to clear the content of the input box.
[0024] The advantages and positive effects of the present invention are as follows:
[0025] 1. The method for predicting the ionic conductivity of eutectic electrolytes based on machine learning in the present invention first obtains the molecular structure information of each component of the eutectic electrolyte, determines the number of atoms with specific coordination numbers in the molecule, and does not require DFT calculations, saving a large amount of time and cost; uses a wrapper feature selection method combined with different machine learning models for feature screening, obtains the optimal feature subset for a specific model, and conducts model training, increasing the flexibility of ionic conductivity prediction.
[0026] 2. The method for predicting the ionic conductivity of eutectic electrolytes based on machine learning in the present invention can achieve rapid prediction of ionic conductivity, save experimental time, reduce the cost of trial and error, and provide convenience for selecting eutectic electrolyte systems with high ionic conductivity.
[0027] 3. The visualization system for predicting the ionic conductivity of eutectic electrolytes based on machine learning in the present invention includes a model selection module, a user input module, a feature acquisition module, an output result module, and a clearing module. It provides a visualization auxiliary tool for obtaining the ionic conductivity of eutectic electrolytes, helps researchers save experimental time and research costs, and improves the research efficiency of eutectic electrolytes.
[0028] 4. Through the setting of the model selection module in the present invention, the regression algorithm in the prediction process can be adjusted, increasing the flexibility of the visualization system for predicting the ionic conductivity of eutectic electrolytes.
[0029] 5. Through the setting of the user data input module in the present invention, users can be prevented from collecting redundant feature information, reducing the complexity of the model. The accuracy of predicting the ionic conductivity of eutectic electrolytes is improved.
[0030] 6. By setting the feature acquisition module, the present invention can obtain the optimal input features required by the selected machine learning model, avoid the mismatch between the input features and the model, and improve the accuracy of the prediction of the ionic conductivity of the eutectic electrolyte.
[0031] 7. By setting the clearing module, the present invention can directly clear all the contents of the input and output boxes with one key, which provides convenience for the prediction of multiple eutectic electrolyte systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 is a flowchart of the method for predicting the ionic conductivity of the eutectic electrolyte of the present invention;
[0033] Figure 2 is a scatter plot of the true values and predicted values of CatBoost based on the method for predicting the ionic conductivity of the eutectic electrolyte of the present invention;
[0034] Figure 3 is an overview diagram of the visualization system for predicting the ionic conductivity of the eutectic electrolyte of the present invention. DETAILED DESCRIPTION OF THE INVENTION
[0035] The structure of the present invention will be further described below with reference to the accompanying drawings and through examples. It should be noted that this embodiment is narrative rather than restrictive.
[0036] A method for predicting the ionic conductivity of a eutectic electrolyte based on machine learning, please refer to Figures 1-3 , and the inventive points are as follows:
[0037] Step 1: Collect the ionic conductivity of known eutectic electrolytes from relevant literature, obtain the composition and structure information of the corresponding eutectic electrolytes, and construct a data set. In this embodiment, the constructed data set contains 212 groups of data, and the ionic conductivity is used as the output data of the subsequent machine learning model.
[0038] Step 2: Obtain the molar ratio and molecular structure information of each component of the eutectic electrolyte in the data set constructed in Step 1, wherein the molecular structure information of the eutectic electrolyte includes the atomic information of each component of the eutectic electrolyte;
[0039] Step 3: Perform feature engineering (combine the relevant information extracted from the original data) according to the molar ratio and molecular structure information of the eutectic electrolyte to obtain the corresponding features of the eutectic electrolyte; specifically, the implementation method for obtaining the features corresponding to the ionic conductivity of the eutectic electrolyte includes:
[0040] 3.1. Obtain the molecular structure of the eutectic electrolyte;
[0041] 3.2. Determine the number of atoms with a specific coordination number in the molecule;
[0042] 3.3. Combine the number of atoms with a specific coordination number with the molar ratios of the components of the eutectic electrolyte to obtain the characteristics corresponding to the eutectic electrolyte.
[0043] Step 4: Use the wrapper feature selection method in combination with one of the machine learning algorithms (including one of CatBoost, Random Forest RF, XGBoost) to screen the features, and select the optimal feature subset that has a greater impact on the ionic conductivity according to the evaluation metrics (the evaluation metrics refer to MAE, RMSE, R2, MSE, etc., which are used to measure the accuracy of machine learning model predictions) as the input data for the subsequent machine learning model. Specifically:
[0044] The selection of the optimal feature subset adopts one of the wrapper feature selection methods such as sequential forward selection and sequential backward selection. The specific steps for screening the features by sequential forward selection or sequential backward selection in combination with the machine learning algorithm are as follows: Starting from an empty set of features (using the sequential forward selection method) or all features (using the sequential backward selection method), select features to form a feature subset and evaluate the feature subset. In each iteration of feature selection, according to the evaluation metrics, add or remove features on the feature subset with the highest score to obtain the optimal feature subset; repeat this process until the set feature target number is reached.
[0045] In this embodiment, the optimal feature subset obtained by backward selection and the CatBoost model is: Temperature, Anion mol ratio, Solvent mol ratio, Cl - , C4, O1.
[0046] Step 5: Use the optimal feature subset screened in Step 4 as the input and the ionic conductivity in the dataset constructed in Step 1 as the output, and train and test the ionic conductivity dataset through a machine learning algorithm (CatBoost, RF or XGBoost), and obtain the minimum error by adjusting the model parameters to establish a relationship model between the element composition and the conductivity; the proportion of the training set in the conductivity dataset is 70 - 90%. In the embodiment of the present invention, the training set is 80% and the test set is 20%. See Appendix Figure 2 , and use the CatBoost model to train and test the dataset, and the R2 value of the test set is 0.85, and the root mean square error is as low as 0.01.
[0047] Step 6: Based on the relationship model between the element composition and the conductivity established in Step 5, construct a visualization system for the prediction model of the ionic conductivity of the eutectic electrolyte, and input the characteristic information of the eutectic electrolyte to be predicted through the visualization system and output its ionic conductivity.
[0048] The visualized system for predicting the ionic conductivity of the eutectic electrolyte constructed above, see Figure 3 , including: a model selection module, a user input module, a feature acquisition module, an output result module, and a clearing module
[0049] Model selection module: used to provide custom selection of models. Users can select machine learning models according to their needs. Specifically, the model selection options include: CatBoost, XGBoost, Random Forest (RF).
[0050] User input module: used for users to input relevant information of the eutectic electrolyte to be predicted. In this embodiment, input the relevant information of the system to be predicted Zn(ClO4)2·6H2O:TMU:H2O = 1:0:6.
[0051] Feature acquisition module: used to calculate the required features according to the model selection and user input information. In this embodiment, the features required by the CatBoost model are: Temperature, Anion mol ratio, Solvent mol ratio, Cl - , C4, O1. The features required by the Random Forest model are: Temperature, Anion mol ratio, H2O mol ratio, Solvent mol ratio, OH - , Cl - , C4, N_plus_4. The features required by the XGBoost model are: Temperature, Znmol ratio, H2O mol ratio, Anion mol ratio, Solvent mol ratio, OH - , Cl - , O1.
[0052] Output result module: used to output the prediction result of the ionic conductivity of the eutectic electrolyte.
[0053] Clearing module: used to clear the content of the input box, and users can re-enter information to predict again.
[0054] The Random Forest model is selected in the model selection module. The predicted result of the ionic conductivity of the system to be predicted Zn(ClO4)2·6H2O:TMU:H2O = 1:0:6 is 4.217 mS / cm, and the experimental result is 4.505 mS / cm. The relative error is as low as 6.4%, indicating that the established visualization system can accurately predict the ionic conductivity of the eutectic electrolyte. Through the prediction of the visualization system, electrolyte systems with high ionic conductivity can be quickly screened out, which helps to reduce the dependence on traditional experimental methods, save time and resources, and improve the research efficiency of eutectic electrolytes.
[0055] Although embodiments and drawings of the present invention are disclosed for illustrative purposes, those skilled in the art can understand that various substitutions, changes, and modifications are possible without departing from the spirit of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the content disclosed in the embodiments and drawings.
Claims
1. A method for predicting ionic conductivity of eutectic electrolytes based on machine learning, characterized in that: The steps include: Step 1: Collect the ionic conductivity of known eutectic electrolytes from relevant literature, obtain the composition and structure information of the corresponding eutectic electrolytes, and construct a data set, where the ionic conductivity is used as the output data of the subsequent machine learning model; Step 2: obtaining the molar ratio and molecular structure information of each component of the eutectic electrolyte in the data set constructed in step 1, wherein the molecular structure information of the eutectic electrolyte includes the atomic information of each component of the eutectic electrolyte; Step 3: Perform feature engineering based on the molar ratio and molecular structure information of the eutectic electrolyte to obtain features corresponding to the eutectic electrolyte; Step 4: Using a wrapper feature selection method combined with one of the machine learning algorithms to screen the features, and selecting the optimal feature subset that has a greater impact on ionic conductivity according to the evaluation index, and using it as input data for a subsequent machine learning model; Step 5: Using the optimal feature subset obtained by step 4 as input and the ionic conductivity in the data set constructed in step 1 as output, the ionic conductivity data set is trained and tested by a machine learning algorithm, and the minimum error is obtained by adjusting the model parameters to establish a relationship model between elemental composition and conductivity; Step 6: Based on the relationship model between elemental composition and conductivity established in step 5, a visualization system for predicting the ionic conductivity of eutectic electrolytes is constructed. The characteristic information of the eutectic electrolyte to be predicted is input through the visualization system, and its ionic conductivity is output.
2. The method for predicting eutectic electrolyte ionic conductivity based on machine learning according to claim 1, characterized in that: In step 3, the method for obtaining the corresponding characteristics of the ionic conductivity of the eutectic electrolyte includes: 3.
1. Obtain the molecular structure of the eutectic electrolyte; 3.
2. Determine the number of atoms with a specific coordination number in a molecule; 3.
3. Combine the number of atoms with a specific coordination number with the molar ratio of each component of the eutectic electrolyte to obtain the corresponding characteristics of the eutectic electrolyte.
3. The method for predicting ionic conductivity of eutectic electrolytes based on machine learning according to claim 1, characterized in that: In step 4, the optimal feature subset is selected by one of the wrapping feature selection methods, including the sequential forward selection method and the sequential backward selection method; starting from the feature empty set by the sequential forward selection method or all the features by the sequential backward selection method, features are selected to form feature subsets and the feature subsets are evaluated. In each iteration of feature selection, features are continuously added or removed from the feature subset with the highest score according to the evaluation index to obtain the optimal feature subset; This process is repeated until the target number of features is reached.
4. The method for predicting ionic conductivity of eutectic electrolytes based on machine learning according to claim 1, characterized in that: In step 5, the constructed eutectic electrolyte ionic conductivity prediction model visualization system includes: a model selection module, a user input module, a feature acquisition module, an output result module, and a clearing module; Model selection module: used to provide custom selection of models. Users can select machine learning models according to their needs. Specifically, model selection options include: CatBoost, XGBoost, and Random Forest; User input module: used for users to input relevant information of the eutectic electrolyte to be predicted; Feature acquisition module: used to calculate the required features based on model selection and user input information; Output result module: used to output the prediction results of eutectic electrolyte ion conductivity; Clear module: used to clear the content of the input box.