A method for optimizing electrolyte formulation using artificial intelligence

Abstract

This invention's method effectively removes outlier data by processing the data, resulting in a large amount of valid data sources. Then, using this processed data, a suitable machine learning algorithm is selected for machine model training. The resulting model is then optimized through a correction program, enabling effective model correction. This allows for more accurate capture of the complex interactions between electrolyte components, facilitating better optimization of electrolyte formulations. Furthermore, the method provided by this invention significantly improves the efficiency of electrolyte formulation development, enabling faster provision of optimal electrolyte formulations based on actual application requirements, and significantly reducing the number of experiments and time costs.

Description

Technical Field

[0001] This invention relates to the field of battery electrolyte preparation technology, and in particular to a method for optimizing electrolyte formulation using artificial intelligence. Background Technology

[0002] Electrolytes, as a crucial component of batteries, have a significant impact on battery performance (such as capacity, cycle life, and safety). Traditional electrolyte formulation development relies heavily on extensive experimentation and experience accumulation, resulting in low efficiency and high costs. Utilizing artificial intelligence (AI) for electrolyte formulation optimization can clearly improve efficiency and reduce development costs significantly. In recent years, with breakthroughs in AI research, some technologies have emerged that apply AI to material formulation optimization. However, these technologies still face some challenges.

[0003] Optimizing with artificial intelligence requires training models with large amounts of data. Internet databases can provide a wealth of training data, and generally, the larger the amount of training data, the more accurate the model. However, internet data contains erroneous data that machines cannot effectively identify and remove, leading to biases in the fitting process and reduced accuracy. Manual processing can increase accuracy but reduces efficiency and complexity. In the industry, to reduce the workload of data processing, a small amount of data is often chosen for model training. This results in lower accuracy of the trained model, increases the validation process, and consequently leads to longer execution times and poorer final results. Besides biases caused by erroneous data, missing data also interferes with fitting results; traditionally, this is discarded. However, this method is particularly problematic for niche fields with limited data. Furthermore, due to issues with model selection, training, calibration, and program design, artificial intelligence in material formulation optimization remains largely in the research and development stage and cannot be effectively applied in actual production.

[0004] Electrolyte formulation development involves multiple components, and the type and amount of each component significantly impact electrolyte performance, posing challenges to formulation development. Therefore, utilizing artificial intelligence (AI) to assist in electrolyte formulation development holds great promise. However, current AI applications in electrolyte formulation optimization are not yet fully effective, with results still differing from practical results. Furthermore, existing AI-based optimization techniques for electrolyte formulations are still limited. Therefore, significant development of AI-based electrolyte formulation optimization technologies is expected for a considerable period. In light of this, this invention provides a novel method for optimizing electrolyte formulations using artificial intelligence. Summary of the Invention

[0005] The present invention aims to provide an efficient method for optimizing electrolyte formulations using artificial intelligence, so as to improve R&D efficiency and optimization results.

[0006] The specific technical solution of this invention is as follows:

[0007] In a first aspect, an efficient method for optimizing electrolyte formulations using artificial intelligence is provided, comprising:

[0008] Step 1: Data collection and screening: Collect electrolyte formulations and their corresponding performance data from the laboratory, remove outliers and data with obvious errors to obtain Data 1; collect electrolyte formulations and their corresponding performance data from internet databases to obtain Data 2;

[0009] Step 2, Data Preprocessing: Use data 1 to clean data 2 and fill in missing values ​​to obtain data 3. Standardize the obtained data 3 to a uniform numerical range to eliminate differences in units.

[0010] Step 3, Feature Engineering: Extract multiple features from the electrolyte, including the common types, concentrations, and physicochemical properties of each component. Introduce descriptors based on molecular structure and features obtained through quantum chemical calculations. Extract these features and their corresponding properties to form a feature library. Then, use feature selection and construction techniques to automatically determine the most relevant and representative feature combinations.

[0011] Step 4, AI Algorithm Selection and Model Training: Try various machine learning algorithms and compare and evaluate the results using cross-validation to select a suitable machine learning algorithm; use part of data 3 as the training set to train the model, and use the remaining data 3 as the validation set to validate the model;

[0012] Step 5, Model Evaluation and Optimization: The model obtained in Step 4 is evaluated using the validation set to determine whether the model is overfitting or underfitting; if overfitting or underfitting is found, the model is corrected until it is free from overfitting or underfitting.

[0013] Step 6: Combining optimization algorithms: Input the electrolyte formula constraints, automatically combine the feature library and use optimization algorithms to perform a global search for the electrolyte formula, output the optimized electrolyte formula, and finally conduct experimental verification of the obtained electrolyte formula.

[0014] Furthermore, the Internet database mentioned in step 1 includes public databases of research institutions, academic journal databases, and databases formed by combinations of them;

[0015] Furthermore, the amount of data 1 mentioned in step 1 is not limited here. Within a certain range, the larger the amount of data, the better the effect of processing data 2. However, in order to reduce the workload, the amount of data 1 should not be too large, and can be selected as 10-30% of data 2.

[0016] Furthermore, the specific method for cleaning data 2 by inputting data 1 in step 2 is as follows:

[0017] (1) Data fusion: Integrate data 1 and data 2 and compare the same features and indicators;

[0018] (2) Anomaly detection: Using statistical methods or machine learning algorithms to identify outliers in Internet databases that differ significantly from the distribution of laboratory data;

[0019] (3) Data imputation: For missing values ​​in Internet databases, appropriate imputation methods can be used to imput them based on the characteristics and relationships of laboratory data.

[0020] Furthermore, the data fusion described in step (1) specifically involves:

[0021] 1) Data alignment: Ensure that data from different sources are consistent in terms of measurement units and feature definitions;

[0022] 2) Feature matching: Identify features in two data sources that represent the same concept or attribute, and associate them;

[0023] 3) Data merging: Merging aligned and matched laboratory data with data from internet databases;

[0024] 4) Data standardization and normalization: Standardize or normalize the merged data to eliminate data differences between different data sources;

[0025] 5) Duplicate data processing: Check and process any duplicate data records to ensure the uniqueness and accuracy of the data.

[0026] After the above processing, the final data 3 includes data 1. Furthermore, data fusion allows for full utilization of the advantages of data from different sources, providing a richer and more comprehensive information foundation for subsequent data cleaning, analysis, and modeling.

[0027] Furthermore, the machine learning algorithm mentioned in step (2) includes one of the following: Z-score algorithm, IQR (interquartile range) method, Isolation Forest algorithm, Local Outlier Factor (LOF) algorithm, and One-Class SVM (one-class support vector machine); such algorithms can better identify outliers.

[0028] Furthermore, the interpolation method described in step (3) includes one of the following: mean interpolation, median interpolation, mode interpolation, regression interpolation, multiple interpolation, and nearest neighbor interpolation.

[0029] Furthermore, the descriptor mentioned in step 3 includes at least one of topological descriptors, geometric descriptors, and electronic descriptors.

[0030] Molecular structure-based descriptors are numerical representations used to quantitatively describe the molecular structural characteristics of each component in an electrolyte. These descriptors can capture information such as the size, shape, topology, and electronic properties of each component molecule.

[0031] Furthermore, the quantum computing described in step 3 is implemented using Gaussian, ORCA, or VASP, and the computation results can be appropriately normalized or standardized.

[0032] In electrolyte formulation optimization, quantum chemical calculations can be used to obtain characteristics such as energy and charge distribution related to the molecules of each electrolyte component, providing valuable information for feature engineering.

[0033] Furthermore, specific methods for determining the most relevant and representative feature combinations may include at least one of the following: filtering methods, wrapping methods, embedded methods, and clustering-based methods.

[0034] This invention provides a filter method that calculates the correlation between each feature (such as the types, amounts, structures, and molecular weights of electrolyte substances) and the target variable (such as electrolyte performance: cycle performance, capacity, rate performance, flame retardancy, etc.), using metrics such as Pearson correlation coefficient and mutual information. A threshold is set based on the correlation score, and features with scores higher than the threshold are selected as relevant features.

[0035] This invention employs a wrapper method: using machine learning models (such as random forests and support vector machines), it evaluates the electrolyte performance corresponding to different feature combinations by recursively adding or removing features (such as the types, amounts, structures, and molecular weights of electrolyte substances). For example, starting with an empty feature set, features are added gradually, and the model performance is evaluated after each addition, until the electrolyte performance no longer improves.

[0036] The embedded methods of this invention automatically select key features during model training. For example, Lasso regression can achieve feature selection by compressing the coefficients of key features, reducing some coefficients to zero.

[0037] This invention is based on a clustering method: features (such as the types, amounts, structures, and molecular weights of electrolyte substances) are clustered together, and those with similar features are grouped together. Then, a representative feature is selected from each cluster for combination to reduce the number of key features and maintain the diversity of information.

[0038] Furthermore, the data used as the training set in step 4 can be randomly selected from 75-90% of the data in data 3, with the remaining data used as the validation set; or data 1 can be selected as the validation set, with the corresponding remaining portion of data 3 used as the training set.

[0039] Furthermore, the machine learning algorithm described in step 4 includes random forest, support vector machine, neural network or convolutional neural network; in this step, the machine learning algorithm is used for prediction and modeling, and focuses more on learning and fitting the data.

[0040] Furthermore, when evaluating the model in step 4, if overfitting is found, regularization techniques can be used for adjustment; if underfitting is found, feature engineering can be re-examined or the amount of training data can be increased; depending on the actual needs, if the above methods cannot solve the problem well or quickly, such as when overfitting and underfitting occur more than 4 times in total, the algorithm selection can be re-examined and other algorithms can be selected for retraining.

[0041] Furthermore, in step 5, the obtained model is evaluated by calculating at least two of the following metrics: accuracy, recall, and mean squared error.

[0042] Furthermore, if the training set accuracy reaches 95% or higher, the recall rate is 90% or higher, or the mean squared error is below 0.05, while the validation set accuracy is below 75%, the recall rate is below 70%, or the mean squared error is above 0.15, then overfitting is indicated.

[0043] Furthermore, if the training set accuracy is below 75%, the recall is below 60%, or the mean squared error is above 0.25, and the validation set accuracy reaches above 80%, the recall is above 80%, or the mean squared error is below 0.2, then underfitting is indicated.

[0044] Furthermore, the optimization algorithm described in step 6 includes genetic algorithm, simulated annealing algorithm, and particle swarm optimization algorithm; this optimization algorithm selects the appropriate algorithm to search for the optimal solution based on the constraints of the problem and the optimization objective.

[0045] When there are many features affecting electrolyte performance and the solution space of the problem is very large and complex, the genetic algorithm is preferred; when there is a greater probability of escaping the local optimum of electrolyte performance during the optimization process, the simulated annealing algorithm is preferred; when a good solution is desired quickly and there are few features, the particle swarm optimization algorithm is preferred.

[0046] Furthermore, the electrolyte formulation constraints mentioned in step 6 are the electrolyte formulation to be optimized, a partial electrolyte formulation, and the target performance parameters of the electrolyte, or directly the target performance parameters of the electrolyte.

[0047] Furthermore, the optimized electrolyte formulation output in step 6 can be one or more;

[0048] In this invention, the number of constraints input in step 6 affects the number of optimized electrolyte formulations output; generally, the more constraints, the fewer the output results; therefore, when there are many output electrolyte formulations, which causes inconvenience to testing, constraints can be increased to reduce the output results.

[0049] Furthermore, the electrolyte formulation includes, but is not limited to, lithium-ion battery electrolyte formulations and sodium-ion battery electrolyte formulations.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. The method of the present invention can effectively remove abnormal data and fill in missing values ​​by processing the data to obtain a large number of effective data sources. Then, the processed large number of effective data are used for model training, which is conducive to more accurately capturing the complex interactions between electrolyte components and facilitating the acquisition of a better training model.

[0052] 2. The method of the present invention, through better selection, training and calibration of the model, and reasonable selection and setting of the program for optimizing electrolyte formulation by artificial intelligence, enables the final model to be better matched with the development of electrolyte, and the program to better achieve the optimization of electrolyte formulation;

[0053] 3. The method of this invention utilizes artificial intelligence to assist in electrolyte formulation development, significantly improving the efficiency of electrolyte formulation development and substantially reducing the number of experiments and time costs. Furthermore, the method of this invention also considers production process and cost constraints, ensuring that the optimized formulation has practical application value. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the optimization of electrolyte formulation using artificial intelligence, as described in this invention.

[0055] Figure 2 The graph shows the discharge specific capacity test results for the electrolyte in Example 1. Detailed Implementation

[0056] The technical solution of this application will be clearly and completely described below with reference to the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0057] This invention provides a method for optimizing electrolyte formulation using artificial intelligence, the flowchart of which is shown below. Figure 1 The method includes:

[0058] Step 1: Data collection and screening: Collect electrolyte formulations and their corresponding performance data from the laboratory, remove outliers and data with obvious errors to obtain Data 1; collect electrolyte formulations and their corresponding performance data from internet databases to obtain Data 2;

[0059] Step 2, Data Preprocessing: Use data 1 to clean data 2 and fill in missing values ​​to obtain data 3. Standardize the obtained data 3 to a uniform numerical range to eliminate differences in units.

[0060] Step 3, Feature Engineering: Extract multiple features from the electrolyte, including the common types, concentrations, and physicochemical properties of each component. Introduce descriptors based on molecular structure and features obtained through quantum chemical calculations. Extract these features and their corresponding properties to form a feature library. Then, use feature selection and construction techniques to automatically determine the most relevant and representative feature combinations.

[0061] Step 4, AI Algorithm Selection and Model Training: Try various machine learning algorithms and compare and evaluate the results using cross-validation to select a suitable machine learning algorithm; use a portion of data 3 as the training set to train the model, while retaining the remaining data from data 3 as the validation set to validate the model;

[0062] Step 5, Model Evaluation and Optimization: The model obtained in Step 4 is evaluated using the validation set to determine whether the model is overfitting or underfitting; if overfitting or underfitting is found, the model is corrected until it is free from overfitting or underfitting.

[0063] Step 6: Combining optimization algorithms: Input the electrolyte formula constraints, automatically combine the feature library and use optimization algorithms to perform a global search for the electrolyte formula, output the optimized electrolyte formula, and finally conduct experimental verification of the obtained electrolyte formula.

[0064] The Internet database mentioned in step 1 of this embodiment includes public databases of research institutions, academic journal databases, and databases formed by combinations of them.

[0065] In step 1 of this embodiment, the performance data corresponding to the electrolyte formulation collected in the laboratory must be authentic and valid to ensure that the results used to process data 2 have good effects. For example, when collecting this part of the data, data from authoritative laboratory sources can be selected; and the data should be evaluated by a technician with senior experience in the field to remove data with obvious problems.

[0066] In step 1 of this embodiment, the amount of data 1 can be selected as 10-30% of data 2, for example, 15%, 20%, or 25% of the amount of data 2. Within a certain range, the larger the amount of data 1, the better the effect of processing data 2, the fewer subsequent corrections are needed, and the higher the running efficiency. However, in order to reduce the workload of manual screening, the amount of data 1 should not be too large. Of course, the amount of data 1 should not be too small either. When data 1 is too small, it is not possible to clean data 2 well based on data 1, and it may also lead to over-cleaning of data 2, resulting in greater data loss, which may in turn lead to greater deviation in the results.

[0067] In step 2 of this embodiment, the specific method of cleaning data 2 using data 1 is as follows:

[0068] (1) Data fusion: Integrate data 1 and data 2 and compare the same features and indicators;

[0069] (2) Anomaly detection: Using statistical methods or machine learning algorithms to identify outliers in Internet databases that differ significantly from the distribution of laboratory data;

[0070] (3) Data imputation: For missing values ​​in Internet databases, appropriate imputation methods can be used to imput them based on the characteristics and relationships of laboratory data;

[0071] Specifically, the data fusion described in step (1) is as follows:

[0072] 1) Data alignment: Ensure that data from different sources are consistent in terms of measurement units and feature definitions; 2) Feature matching: Identify features representing the same concepts or attributes in two data sources and associate them; 3) Data merging: Merge the aligned and matched laboratory data with data from internet databases; 4) Data standardization and normalization: Standardize or normalize the merged data to eliminate data differences between different data sources; 5) Duplicate data processing: Check and process any duplicate data records to ensure data uniqueness and accuracy. This invention, through data fusion, can fully utilize the advantages of data from different sources, providing a richer and more comprehensive information foundation for subsequent data cleaning, analysis, and modeling.

[0073] More specifically, the outliers mentioned in (2) are identified by one of the following algorithms: Z-score algorithm, IQR (interquartile range) method, Isolation Forest algorithm, Local Outlier Factor (LOF) algorithm, and One-Class SVM (one-class support vector machine);

[0074] More specifically, the interpolation methods described in (3) include mean interpolation, median interpolation, mode interpolation, regression interpolation, multiple interpolation, and nearest neighbor interpolation.

[0075] As an example, consider the following: For the same electrolyte formulation, Data 1 shows an 85% retention rate after 100 cycles in the same battery. In Data 2, Data a shows a retention rate between 70-90% after 100 cycles, while Data b shows a retention rate below 70% or above 90% after 100 cycles. In this case, the system marks Data a as normal and Data b as an outlier, removing it from Data 2. However, if it's highly unlikely that all results in Data 2 will show a retention rate below 70% or above 90% after 100 cycles, or if a larger portion of these results do, the system will mark Data 2 as normal. Normal data is the cleaned data used to continue with subsequent steps.

[0076] As an example, we can list the following: Data 1 shows that the electrolyte containing 0.8-1.5 mol / L lithium difluorosulfonylimide has a retention rate of 85-90% after 100 cycles and 80-85% after 500 cycles. Data 2 shows an electrolyte with the same formulation as Data 1, where the concentration of lithium difluorosulfonylimide is also 1.0 mol / L, but its performance test data only shows a retention rate of 85-90% after 100 cycles, and the retention rate after 500 cycles was not measured. In order to better obtain the relationship between long-term cycle stability and lithium salt or lithium difluorosulfonylimide concentration in subsequent feature engineering, and to increase the weight of effective data, the system fills in the missing value in Data 2 by selecting the median value of the retention rate of 500 cycles in Data 1 (i.e., 82.5%) based on the retention rate of 500 cycles in Data 1.

[0077] The specific method for determining the most relevant and representative feature combination in this embodiment may include at least one of the following: filtering method, wrapping method, embedded method, and clustering-based method;

[0078] Specifically, filter methods can calculate the correlation measure between each feature (such as the types, amounts, structures, and molecular weights of electrolyte substances) and the target variable (such as electrolyte performance: cycle performance, capacity, rate performance, flame retardancy, etc.), such as Pearson correlation coefficient and mutual information. A threshold is set based on the correlation score, and features with scores higher than the threshold are selected as relevant features.

[0079] Specifically, wrapper methods use machine learning models (such as random forests and support vector machines) to evaluate the performance of electrolytes corresponding to different feature combinations by recursively adding or removing features (such as the types, amounts, structures, and molecular weights of electrolyte substances). For example, starting with an empty feature set, features are added gradually, and the model performance is evaluated after each addition, until the electrolyte performance no longer improves.

[0080] Specifically, the embedded methods of this invention automatically select key features during model training. For example, Lasso regression can compress key feature coefficients, reducing some coefficients to zero, thereby achieving feature selection.

[0081] Specifically, clustering methods are used to cluster features (such as the types, amounts, structures, and molecular weights of electrolyte substances) together, grouping those with similar features into a single group. Then, a representative feature is selected from each cluster for combination to reduce the number of key features and maintain information diversity.

[0082] The molecular structure-based descriptors mentioned in step 3 of this embodiment include at least one of topological descriptors, geometric descriptors, and electronic descriptors. Molecular structure-based descriptors are numerical representations used to quantitatively describe the characteristics of molecular structures. These descriptors can capture information about the size, shape, topological structure, and electronic properties of molecules.

[0083] The quantum computing described in step 3 of this embodiment is implemented using at least one of Gaussian, ORCA, and VASP, and the calculation results can be appropriately normalized or standardized. In electrolyte formulation optimization, quantum chemical calculations can be used to obtain characteristics such as energy and charge distribution related to electrolyte molecules, providing valuable information for feature engineering.

[0084] Specifically, Gaussian is widely used to calculate various molecular properties, such as molecular orbitals, energy, charge distribution, and vibrational frequencies. By inputting the structure and related parameters of each component in the electrolyte, Gaussian can perform complex calculations and simulations based on quantum mechanics principles, providing predictions and analyses of thermodynamic properties, reaction ease, and rates. ORCA, through quantum chemical calculations, can better consider the influence of the solvent environment on molecular properties, more realistically simulating molecular behavior under actual reaction conditions. It can also provide assessments of the reaction kinetics, molecular stability, and thermodynamics of each component in the electrolyte. VASP, ab initio quantum mechanical calculation program based on density functional theory, can provide performance predictions for the mechanical properties (elastic constants, hardness, etc.), thermal properties (thermal conductivity, heat capacity, etc.), and optical properties of each component in the electrolyte.

[0085] The machine learning algorithm described in step 4 of this embodiment includes random forest, support vector machine, neural network, or convolutional neural network; the selection of the machine learning algorithm described in step 4 of this invention can be based on the results of feature engineering.

[0086] As examples, when the amount of data is small (e.g., less than 1000 data points) and the feature combinations are simple (e.g., less than 8 features), the system prioritizes support vector machines, which is beneficial for obtaining results quickly. As examples, when the amount of data is large (e.g., greater than 5000 data points), the feature combinations are complex (e.g., more than 16 features) and there is a certain internal structure, neural networks or convolutional neural networks are preferred. They can automatically learn complex patterns from large amounts of data, which is more conducive to obtaining more accurate results.

[0087] In step 5 of this embodiment, the obtained model is evaluated by calculating at least two of the following metrics: accuracy, recall, and mean squared error.

[0088] Specifically, if the training set accuracy is above 95%, the recall is above 90%, or the mean squared error is below 0.05, while the validation set accuracy is below 75%, the recall is below 70%, or the mean squared error is above 0.15, then overfitting exists. If the training set accuracy is below 75%, the recall is below 60%, or the mean squared error is above 0.25, while the validation set accuracy is above 80%, the recall is above 80%, or the mean squared error is below 0.2, then underfitting exists.

[0089] In step 5 of this embodiment, when validating and evaluating the obtained model, if overfitting is found, regularization techniques can be used for adjustment. Regularization is relatively straightforward and applicable to various types of models. Besides regularization, other methods can also be used for adjustment. For example, increasing the amount of data 1 can enrich the model's learning content and reduce overfitting to existing data. Another example is removing unimportant or redundant key factors from feature combinations to reduce model complexity. If underfitting is found, the feature engineering in step 3 can be re-examined, or the amount of training data can be increased. Depending on actual needs, if the above methods cannot solve the problem well or quickly, such as when overfitting and underfitting occur more than four times cumulatively, the machine learning algorithm in step 4 can be re-selected for retraining.

[0090] In step 6 of this embodiment, random perturbations and diverse initial solutions can be introduced during the optimization process to avoid getting trapped in local optima.

[0091] Specifically, random perturbation refers to applying random changes to the current solution or search direction during the algorithm's execution. For example, in a certain iteration step, slightly and randomly changing the proportion or parameter value of a component in the electrolyte formulation. This allows the algorithm to explore other regions near the current solution, rather than being limited to the vicinity of the current local optimum. Diverse initial solutions involve generating multiple different, random initial electrolyte formulations at the start of the algorithm, instead of starting from a fixed initial point. This increases the number of starting points for the algorithm's search, as different initial solutions may guide the algorithm towards different search paths. Local optima refer to solutions within a certain region of the solution space that are better than solutions in their neighboring regions, even though there may be better solutions in the overall solution space. If the algorithm always starts from the same initial solution and lacks sufficient randomness and diversity during the search, it can easily get trapped in such local optima and fail to find the global optimum. Therefore, by introducing random perturbation and diverse initial solutions, the possibility of the algorithm escaping local optima and discovering better global solutions can be increased.

[0092] The electrolyte formulation constraints mentioned in step 6 of this embodiment are the electrolyte formulation to be optimized, or a partial electrolyte formulation and the target performance of the electrolyte, or the target performance of the electrolyte.

[0093] As an example, the following can be listed: When the electrolyte formula to be optimized and the target performance parameters are input, the system substitutes the corresponding features of the electrolyte formula to be optimized into the obtained model to calculate the deviation. The optimization algorithm adjusts the electrolyte formula around the electrolyte formula to be optimized based on the target performance parameters and the corresponding relationship between the features in the feature library and the performance. Finally, the electrolyte formula that satisfies the model is output. The optimized electrolyte formula may be one or more.

[0094] As an example, the following can be listed: When the target performance parameters of the electrolyte are input, the optimization algorithm combines the features of the feature library and the performance correspondence to select the components and their corresponding features and substitute them into the obtained model to calculate the output electrolyte formula that satisfies the model. The obtained electrolyte formula may be one or more.

[0095] As an example, the following can be listed: When the electrolyte components and target performance parameters are input, the system substitutes the target performance parameters and the corresponding features of the electrolyte components into the model for calculation. The optimization algorithm combines the feature library features and performance correspondence to prioritize the electrolyte components as fixed components and continue to select components around the electrolyte components to substitute into the model for calculation to obtain an electrolyte formula that satisfies the model. The resulting electrolyte formula may be one or more.

[0096] The optimization algorithms described in step 6 of this embodiment include genetic algorithm, simulated annealing algorithm, and particle swarm optimization algorithm.

[0097] Specifically, when there are many parameters and the solution space of the problem is very large and complex, genetic algorithms are preferred; when there is a greater probability of escaping local optima during the optimization process, simulated annealing can be selected; when a good solution is desired quickly and the number of parameters is small, particle swarm optimization can be chosen. The complex electrolyte formulation of this invention involves various types and proportions of solutes and solvents, and each component has multiple selectable concentration ranges. Different combinations of these components and concentrations constitute the solution to the problem. When there are many electrolyte components, and each component has a wide range of selectable values, the number of possible combinations grows exponentially, forming a huge solution space. Furthermore, the interactions between these combinations and their impact on the final performance are not simple and direct. Various nonlinear and interactive effects may exist, making it difficult to determine the optimal solution through intuitive analysis or simple calculations. For example, a small change in the concentration of one solute in the electrolyte may have complex and unpredictable effects on multiple performance indicators of the electrolyte due to interactions with other solutes and solvents. This makes finding the optimal or near-optimal electrolyte formulation in such a huge solution space extremely challenging. Genetic algorithms possess strong global search capabilities, parallel processing capabilities, and the ability to handle complex relationships. Furthermore, their lack of need for derivative information and their good flexibility and adaptability give them unique advantages in dealing with problems with large and complex solution spaces. For complex problems like electrolyte formulation optimization, numerous local optima may exist in the solution space. For example, while an electrolyte formulation may generally satisfy the model, it might have excellent high-temperature resistance but poor cycling performance. Traditional optimization algorithms may easily get trapped in a local optimum. Simulated annealing, by accepting poor solutions during the optimization process, increases the possibility of exploring new regions, thus making it more likely to escape local optima in electrolyte performance and find a globally better solution or a solution close to the global optimum. When the number of features or parameters in the feature combination in step 3 is small, and a certain speed requirement exists, particle swarm optimization can achieve rapid computation and quickly provide optimization results.

[0098] The technical solution of the present invention will be further described below with reference to specific embodiments.

[0099] Example 1: Optimization of lithium-ion battery electrolyte formulation:

[0100] Step 1: Data Collection and Screening: 300 sets of lithium-ion battery electrolyte formulations and their corresponding capacity, cycle performance, and flame retardant performance data were collected from the laboratory. These data were reviewed, and 28 sets with significant errors were removed, resulting in 272 sets of valid data (Data 1). Then, 3000 sets of lithium-ion battery electrolyte formulations and their corresponding capacity, cycle performance, and flame retardant performance data were obtained from internet databases (Data 2).

[0101] Step 2, Data Preprocessing: Data 1 and Data 2 are defined with the same measurement units and features. Features with the same concepts or attributes are associated. Data 1 and Data 2 are merged and then normalized. For the same features and indicators, One-Class SVM is used to identify 17 outliers in Data 2 that have a large distribution difference from Data 1 and removes the outliers in Data 2. Based on the features and relationships of Data 1, the mean imputation method is used to fill in the missing values. Finally, 41 missing values ​​are filled in Data 2 to obtain Data 3. The data is then standardized to the [0,1] interval.

[0102] Step 3, Feature Engineering: The types and amounts of organic solvents, lithium salts, and additives in the electrolyte, as well as the ionic conductivity and decomposition voltage characteristics of the electrolyte, were extracted. Simultaneously, molecular size and shape characteristics obtained from geometric descriptors, along with thermodynamic properties, reaction ease and rate characteristics obtained through Gaussian quantum chemical calculations, and their corresponding capacity, cycle performance, and flame retardant properties, were introduced. Differences in capacity, cycle performance, and flame retardant properties corresponding to different characteristics were also incorporated to form a feature library corresponding to features and properties. A filtering method was used to calculate the correlation between features and target properties to determine feature selection and construction. Nine of the most relevant and representative features were identified (solvent type and concentration, lithium salt concentration and amount, additive type and concentration, electrolyte thermodynamic parameters, ionic conductivity, and decomposition voltage).

[0103] Step 4, AI Algorithm Selection and Model Training: Try three machine learning algorithms: Random Forest, Support Vector Machine, and Neural Network. Compare and evaluate the results using 10-fold cross-validation, and finally select the Random Forest algorithm as the AI ​​algorithm; randomly sample 85% of the data from Data 3 for model training, and keep the remaining 15% of Data 3 as the validation set;

[0104] Step 5, Model Evaluation and Optimization: The trained random forest model was evaluated using the aforementioned 15% validation set. The training set accuracy was 95%, recall was 91%, and mean squared error was 0.05. Slight overfitting was found, and L2 regularization was used for adjustment. The model was evaluated again, and the accuracy was found to be 88%, recall was 87%, and mean squared error was 0.08, which met the requirements.

[0105] Step 6: Combining the optimization algorithm: Using the simulated annealing algorithm, input the electrolyte formula to be optimized and the discharge capacity at 25℃ (10 cycles of 0.2C discharge specific capacity greater than 220mAh g). -1 Using cycle performance (retention rate of over 75% after 500 cycles) and flame retardancy (self-extinguishing time within 5 seconds) as constraints, a global search was conducted on the electrolyte formulations, resulting in 13 optimized lithium-ion electrolyte formulations.

[0106] Of the 13 electrolyte formulations obtained above, one was randomly selected and its performance was tested and compared with that of the electrolyte before optimization under the same conditions. The self-extinguishing time of the electrolyte before optimization was measured to be 4 seconds, and the retention rate after 500 cycles was 71%. The self-extinguishing time of the optimized electrolyte was 5 seconds, and the retention rate after 500 cycles was 78%. Furthermore, from... Figure 2 It can be seen that the optimized electrolyte formulation has a capacity greater than 220 mAh g after 10 cycles at 0.2C. -1 .

[0107] As can be seen from the above results, the method for optimizing electrolyte formulation using artificial intelligence provided by the present invention can effectively optimize the electrolyte formulation, so that the optimized electrolyte formulation meets the set requirements in terms of capacity, 500-cycle retention rate, and flame retardant performance, and the overall performance is significantly improved compared with that before optimization.

Claims

1. A method for optimizing electrolyte formulation using artificial intelligence, characterized in that, include: Step 1, Data Collection and Screening: Collect the electrolyte formulation and its corresponding performance data from the laboratory, remove outliers and data with obvious errors, and obtain Data 1; Data 2 was obtained by collecting electrolyte formulations and corresponding performance data from internet databases. Step 2, Data Preprocessing: Use data 1 to clean data 2 and fill in missing values ​​to obtain data 3. Standardize the obtained data 3 to a uniform numerical range to eliminate differences in units. Step 3, Feature Engineering: Extract multiple features from the electrolyte, including the common types, concentrations, and physicochemical properties of each component. Introduce descriptors based on molecular structure and features obtained through quantum chemical calculations. Extract these features and their corresponding properties to form a feature library. Then, use feature selection and construction techniques to automatically determine the most relevant and representative feature combinations. Step 4, AI Algorithm Selection and Model Training: Try various machine learning algorithms and compare and evaluate the results using cross-validation to select a suitable machine learning algorithm; use part of data 3 as the training set to train the model, and use the remaining data 3 as the validation set to validate the obtained model; Step 5, Model Evaluation and Optimization: The model obtained in Step 4 is evaluated using the validation set to determine whether the model is overfitting or underfitting; if overfitting or underfitting is found, the model is corrected until it is free from overfitting or underfitting. Step 6: Combining optimization algorithms: Input the electrolyte formula constraints, automatically combine the feature library and use optimization algorithms to perform a global search for the electrolyte formula, output the optimized electrolyte formula, and finally conduct experimental verification of the obtained electrolyte formula.

2. The method for optimizing electrolyte formulation using artificial intelligence according to claim 1, characterized in that, The specific methods in step 2 include: (1) Data fusion: Integrate data 1 and data 2 and compare the same features and indicators; (2) Anomaly detection: Use statistical methods or machine learning algorithms to identify outliers in data 2 that differ significantly from the distribution of data 1; (3) Data imputation: For missing values ​​in data 2, appropriate imputation methods are used to imput them based on the characteristics and relationships of data 1.

3. The method for optimizing electrolyte formulation using artificial intelligence according to claim 1, characterized in that, The training set mentioned in step 4 is 75-90% of the randomly selected data 3, and the remaining data 3 is used as the validation set; or data 1 is selected as the validation set, and the corresponding remaining data 3 is used as the training set.

4. The method for optimizing electrolyte formulation using artificial intelligence according to claim 1, characterized in that, The specific methods for determining the feature combination in step 3 include at least one of the following: filtering method, wrapping method, embedded method, and clustering-based method.

5. The method for optimizing electrolyte formulation using artificial intelligence according to claim 1, characterized in that, The descriptor mentioned in step 3 includes at least one of topological descriptor, geometric descriptor, and electronic descriptor; the quantum chemical calculation mentioned in step 3 is at least one of Gaussian, ORCA, and VASP.

6. The method for optimizing electrolyte formulation using artificial intelligence according to claim 1, characterized in that, The machine learning algorithm mentioned in step 4 includes at least one of random forest, support vector machine, neural network or convolutional neural network.

7. The method for optimizing electrolyte formulation using artificial intelligence according to claim 1, characterized in that, In step 4, when evaluating the model, if overfitting is found, regularization techniques are used for correction; if underfitting is found, feature engineering is re-examined or the amount of training data is increased for correction.

8. The method for optimizing electrolyte formulation using artificial intelligence according to claim 1, characterized in that, In step 5, the obtained model is evaluated by calculating at least two of the following metrics: accuracy, recall, and mean squared error.

9. The method for optimizing electrolyte formulation using artificial intelligence according to claim 1, characterized in that, The optimization algorithm mentioned in step 6 includes at least one of the following: genetic algorithm, simulated annealing algorithm, and particle swarm optimization algorithm.

10. The method for optimizing electrolyte formulation using artificial intelligence according to claim 1, characterized in that, The electrolyte formulation constraints mentioned in step 6 are the electrolyte formulation to be optimized or a partial electrolyte formulation and the target performance of the electrolyte, or the target performance of the electrolyte. The electrolyte formulation includes one of a lithium-ion battery electrolyte formulation and a sodium-ion battery electrolyte formulation.

Images