Method and apparatus for optimizing battery electrolyte formulation, and device and storage medium
By constructing a structured dataset and machine learning model for battery electrolytes, and optimizing electrolyte formulation recommendations, the problems of time-consuming and costly battery electrolyte formulation development have been solved, achieving efficient battery electrolyte formulation research and development.
Patent Information
- Application Number
- PCT/CN2025/147802
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-09
- Filing Date
- 2025-12-31
- Publication Date
- 2026-11-12
AI Technical Summary
Existing technologies for generating battery electrolyte formulations are time-consuming and costly, and high-throughput computing methods suffer from high computational resource requirements and discrepancies between computational results and actual values.
By acquiring a structured dataset of battery electrolytes, and based on electrolyte formulation attributes, electrical performance indicators, and correlation scores, we construct a potential feature matrix of electrical performance and a potential feature matrix of electrolyte formulation attributes using alternating least squares and machine learning models. We then optimize the prediction scoring matrix, select the electrolyte formulation attribute with the highest prediction correlation, and combine a generalized linear model and a deep neural network model to determine the recommended battery electrolyte formulation.
This greatly improves the R&D efficiency of battery electrolyte formulations, reduces the economic costs of computing power and trial and error, and improves the R&D efficiency of battery performance.
Smart Images

Figure CN2025147802_12112026_PF_FP_ABST
Abstract
Description
A method, apparatus, device, and storage medium for optimizing battery electrolyte formulations.
[0001] Cross-reference to related applications
[0002] This application claims priority to Chinese Patent Application No. 202510600523.8, filed on May 9, 2025, entitled "A Method, Apparatus, Device and Storage Medium for Recommending Battery Electrolyte Formulations", the entire contents of which are incorporated herein by reference. Technical Field
[0003] This application relates to the field of battery technology, and in particular to a method, apparatus, device and storage medium for optimizing battery electrolyte formulation. Background Technology
[0004] With the continuous advancement of technology, lithium batteries have been widely used in numerous fields, such as portable electronic devices, electric vehicles, and large-scale grid energy storage. These applications place increasingly higher demands on the performance of lithium batteries. Currently, most methods for researching lithium battery electrolyte formulations are based on the controlled variable method to explore the types and proportions of lithium salts, solvents, and additives. These methods test the performance of lithium salts in the battery cell in terms of energy density, cycle life, and safety, and then select the best solvents and additives based on their performance. Alternatively, high-throughput computational methods can be used to study electrolyte formulations. However, high-throughput computation requires powerful computing resources and involves a large amount of data calculation, resulting in long processing times and high costs for generating battery electrolyte formulations. Summary of the Invention
[0005] Based on this, this application provides a method, apparatus, device, and storage medium for optimizing battery electrolyte formulations, in order to solve the technical problems of long time consumption and high cost in generating battery electrolyte formulations.
[0006] Firstly, a method for optimizing a battery electrolyte formulation is provided, the method comprising:
[0007] Obtain a structured dataset of battery electrolytes. The structured dataset includes electrolyte formulation attributes, electrical performance indicators, and correlation scores in the structured dataset. The correlation scores are the ratings given by users on the degree of correlation between electrolyte formulation attributes and electrical performance indicators.
[0008] Based on electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the structured dataset, a prediction scoring matrix for electrical performance indicators and electrolyte formulation attributes is determined. The prediction scoring matrix includes the predicted correlation scores for all electrolyte formulation attributes corresponding to each electrical performance indicator.
[0009] Based on the prediction scoring matrix, the electrolyte formulation attribute with the highest predicted relevance score for each electrical performance index is selected as the recommended feature list to construct a battery electrolyte formulation recommendation set; and
[0010] Based on the recommended set of battery electrolyte formulations, a recommended battery electrolyte formulation is determined using a pre-set model.
[0011] According to one feasible method in an embodiment of this application, a prediction scoring matrix for the electrical performance indicators and electrolyte formulation attributes is determined based on the electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the structured dataset, including:
[0012] Based on electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the structured dataset, an initial matrix between electrolyte formulation attributes and electrical performance indicators is constructed.
[0013] The initial matrix is decomposed into a latent feature matrix of electrical performance and a latent feature matrix of electrolyte formulation properties based on the alternating least squares algorithm; and
[0014] Based on the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes, a prediction scoring matrix for electrical performance indicators and electrolyte formulation attributes is determined.
[0015] According to one feasible method in an embodiment of this application, a predictive scoring matrix for electrical performance indicators and electrolyte formulation attributes is determined based on the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes, including:
[0016] Randomly initialize the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation properties;
[0017] By fixing the potential feature matrix of electrical properties, updating the potential feature matrix of electrolyte formulation attributes, the latest potential feature matrix of electrolyte formulation attributes is obtained.
[0018] By fixing the latent feature matrix of electrolyte formulation properties, updating the latent feature matrix of electrical properties, and obtaining the latest latent feature matrix of electrical properties;
[0019] Based on the latest potential feature matrix of electrolyte formulation attributes and the latest potential feature matrix of electrical performance, a predictive scoring matrix for electrical performance indicators and electrolyte formulation attributes is determined.
[0020] According to one feasible method in the embodiments of this application, based on the prediction scoring matrix, the electrolyte formulation attribute with the highest predicted correlation score among each electrical performance index is selected as a recommended feature list to construct an electrolyte formulation recommendation set, including:
[0021] Based on the prediction correlation scores in the prediction scoring matrix, all electrolyte formulation attributes corresponding to each electrical performance index are sorted.
[0022] The electrolyte formulation attribute with the highest predicted correlation score among each electrical performance index is selected as the recommended feature list; and
[0023] A recommended electrolyte formulation set is constructed by compiling a list of recommended features for each electrical performance indicator.
[0024] According to one feasible method in an embodiment of this application, a recommended battery electrolyte formulation is determined based on a preset model according to a set of recommended battery electrolyte formulations, including:
[0025] The numerical features of the battery electrolyte formulation recommendation set are cross-combined and standardized to obtain the first standardized feature.
[0026] The category field in the battery electrolyte formulation recommendation set is uniquely thermally encoded, and the numerical features are standardized to obtain the second standardized features.
[0027] Based on the first and second standardized features, the recommendation score for each battery electrolyte formulation in the recommended set is determined using a pre-defined model; and
[0028] Recommended battery electrolyte formulations are determined based on the recommended score for each formulation.
[0029] According to one achievable method in an embodiment of this application, the preset model includes a generalized linear model and a deep neural network model; based on the first standardized feature and the second standardized feature, the recommendation score of each battery electrolyte formulation in the battery electrolyte formulation recommendation set is determined based on the preset model, including:
[0030] The first standardized feature is input into the generalized linear model to obtain the first output feature;
[0031] The second standardized feature is input into the deep neural network model to obtain the second output feature; and
[0032] The first and second output features are concatenated and combined, and then input into the fully connected layer for fusion to obtain the recommendation score for each battery electrolyte formula in the battery electrolyte formula recommendation set.
[0033] According to one achievable method in an embodiment of this application, before obtaining the structured dataset of the battery electrolyte, the method further includes:
[0034] A generalized linear model is constructed by setting linear parameters based on the linear model;
[0035] A deep neural network model is constructed by setting up an input layer, two hidden layers, an output layer, and activation functions; and
[0036] A pre-defined model is constructed based on a generalized linear model, a deep neural network model, and a fully connected layer.
[0037] Secondly, a battery electrolyte formulation optimization device is provided, the device comprising:
[0038] The acquisition module is used to acquire a structured dataset of battery electrolytes. The structured dataset includes electrolyte formulation attributes, electrical performance indicators, and correlation scores in the structured dataset. The correlation scores are the ratings given by users on the degree of correlation between electrolyte formulation attributes and electrical performance indicators.
[0039] The determination module is used to determine the prediction scoring matrix of the electrolytic performance index and the electrolyte formulation attribute based on the electrolyte formulation attributes, electrolytic performance index and the correlation scores existing in the structured dataset. The prediction scoring matrix includes the prediction correlation score of each electrolytic performance index for all electrolyte formulation attributes.
[0040] The selection module is used to select the electrolyte formulation attribute with the highest predicted relevance score for each electrical performance index from the predicted scoring matrix, and construct a recommended set of battery electrolyte formulations as a recommended feature list; and
[0041] The determination module is also used to determine the recommended battery electrolyte formulation based on a preset model and a set of recommended battery electrolyte formulations.
[0042] Thirdly, a computer device is provided, comprising:
[0043] At least one processor; and
[0044] A memory communicatively connected to the at least one processor; wherein,
[0045] The memory stores computer instructions that can be executed by the at least one processor to enable the at least one processor to perform the method involved in the first aspect above.
[0046] Fourthly, a computer-readable storage medium is provided, having stored thereon computer instructions, wherein the computer instructions are used to cause a computer to perform the methods involved in the first aspect above.
[0047] According to the technical content provided in the embodiments of this application, a structured dataset of battery electrolyte is obtained. The structured dataset includes electrolyte formulation attributes, electrical performance indicators, and correlation scores existing in the structured dataset. Based on the electrolyte formulation attributes, electrical performance indicators, and correlation scores existing in the structured dataset, a prediction scoring matrix of electrical performance indicators and electrolyte formulation attributes is determined. According to the prediction scoring matrix, the electrolyte formulation attribute with the highest predicted correlation score for each electrical performance indicator is selected as a recommended feature list to construct a battery electrolyte formulation recommendation set. Based on the battery electrolyte formulation recommendation set, a recommended battery electrolyte formulation is determined based on a preset model, which greatly improves the R&D efficiency of battery electrolyte formulations and reduces the economic costs of computing power and trial and error. Attached Figure Description
[0048] Figure 1 is a diagram illustrating the application environment of the battery electrolyte formulation optimization method in one or more embodiments;
[0049] Figure 2 is a flowchart illustrating the battery electrolyte formulation optimization method in one or more embodiments;
[0050] Figure 3 is a structural block diagram of the battery electrolyte formulation optimization device in one or more embodiments;
[0051] Figure 4 is a schematic structural diagram of a computer device in one or more embodiments. Detailed Implementation
[0052] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the scope of the present application.
[0053] Lithium-ion batteries are widely used in numerous fields, while increasingly stringent performance requirements are being placed on them. For example, electric vehicles require lithium-ion batteries with higher energy density to increase driving range, as well as fast charging capabilities, good high and low temperature performance, and long cycle life. These performance characteristics are closely related to the electrolyte formulation. The electrolyte is a key component of the battery, directly affecting its performance. Choosing a suitable electrolyte formulation to improve the energy density, cycle life, and safety of the battery cell is crucial. Currently, lithium-ion battery electrolyte formulation development is still in a stage of experience-based enumeration and trial-and-error, and its research efficiency cannot meet the needs of battery cell development companies.
[0054] Currently, testing in battery cells focuses on electrical performance indicators such as energy density, cycle life, and safety. Based on these performance indicators, better solvents and additives are selected according to their types and proportions. This method relies on experience, is time-consuming, and has high material costs. Another approach involves high-throughput computing for electrolyte formulation research. However, high-throughput computing requires powerful computing resources, and the cost of purchasing and maintaining such hardware is high. It also involves significant computation time and energy consumption, further increasing computational costs, especially when dealing with complex electrolyte systems and large-scale datasets. High-throughput computing is typically based on theoretical models and assumptions, such as molecular dynamics simulations and quantum mechanical calculations. However, these theoretical models may have limitations in describing the complex chemical environment and physical processes of electrolytes, leading to discrepancies between calculated results and actual conditions. In some machine learning-based high-throughput computing methods, training the potential function is a crucial step. However, constructing a representative and accurate potential function dataset is often difficult, affecting the generalization ability and prediction accuracy of the machine learning model.
[0055] To address the problems in the prior art, embodiments of this application provide a method, apparatus, device, and computer storage medium for optimizing battery electrolyte formulations.
[0056] For ease of understanding, the system to which this application applies is first described. The battery electrolyte formulation optimization method provided in this application can be applied to the system architecture shown in Figure 1. In this system, terminal 110 communicates with server 120 via a network. Terminal 110 acquires a structured dataset of battery electrolytes. This structured dataset includes electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the dataset. The correlation scores represent the user's rating of the degree of correlation between electrolyte formulation attributes and electrical performance indicators. Based on the electrolyte formulation attributes, electrical performance indicators, and the correlation scores in the structured dataset, a prediction scoring matrix for electrical performance indicators and electrolyte formulation attributes is determined. This prediction scoring matrix includes the predicted correlation scores for all electrolyte formulation attributes corresponding to each electrical performance indicator. According to the prediction scoring matrix, the electrolyte formulation attribute with the highest predicted correlation score for each electrical performance indicator is selected as a recommended feature list to construct a battery electrolyte formulation recommendation set. Based on this battery electrolyte formulation recommendation set, a recommended battery electrolyte formulation is determined using a preset model. Among them, terminal 110 can be, but is not limited to, various personal computers, laptops, smartphones and tablets, and server 120 can be implemented by independent servers or server clusters composed of multiple servers.
[0057] The following describes the battery electrolyte formulation optimization method provided in the embodiments of this application. Figure 2 shows a flowchart of the battery electrolyte formulation optimization method provided in one embodiment of this application. As shown in Figure 2, the method may include the following steps:
[0058] S210, Obtain the structured dataset of battery electrolyte.
[0059] A large amount of structural and property data on electrolyte molecules is collected and automatically organized into a structured dataset. This structured dataset includes electrolyte formulation attributes, electrical performance indicators, and correlation scores. The correlation score is a user's rating of the degree of correlation between electrolyte formulation attributes and electrical performance indicators. It is typically obtained through expert scoring based on accumulated experimental data, but not all correlation scores between electrical performance indicators and electrolyte formulation attributes are available. Electrical performance indicators may include energy density, cycle life, and safety performance, while electrolyte formulation attributes may include electrolyte salts, organic solvents, and film-forming additives.
[0060] S220, based on electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the structured dataset, determines the prediction scoring matrix for electrical performance indicators and electrolyte formulation attributes.
[0061] Based on the data structure <electrolyte performance index, electrolyte formulation attribute, score>, an initial matrix is constructed by combining the electrolyte formulation attribute, electrolyte performance index, and correlation scores present in the structured dataset. This initial matrix is typically sparse, meaning most correlation scores are missing. The initial matrix is then decomposed into a latent feature matrix for electrolyte performance and a latent feature matrix for electrolyte formulation attributes. These two matrices are continuously optimized. The correlation scores between electrolyte formulation attributes and electrolyte performance indexes are predicted based on the product of these two matrices, resulting in a prediction score matrix. This prediction score matrix includes the predicted correlation scores for each electrolyte performance index across all electrolyte formulation attributes.
[0062] S230. Based on the prediction scoring matrix, select the electrolyte formulation attribute with the highest prediction correlation score for each electrical performance index as a recommended feature list to construct a battery electrolyte formulation recommendation set.
[0063] The predicted correlation score between each electrical performance index and different electrolyte formulation attributes can be obtained by using the predicted scoring matrix. Based on the magnitude of the predicted correlation score, the electrolyte formulation attribute with the highest predicted correlation score for each electrical performance index is selected as the recommended feature list. This is to mine and obtain the most relevant electrolyte formulation attribute features for each electrical performance. The recommended feature lists corresponding to each electrical performance index are then combined to construct a battery electrolyte formulation recommendation set.
[0064] S240, based on the battery electrolyte formulation recommendation set, determine the recommended battery electrolyte formulation based on a preset model.
[0065] The preset model is a pre-trained model, which may include a generalized linear Wide model, a deep neural network Deep model, and a fully connected layer. Based on data from the battery electrolyte formulation recommendation set, Wide model features and Deep model features are constructed and input into the Wide and Deep models respectively. The output features of the Wide and Deep models are then concatenated and combined, and finally fused through a fully connected layer to obtain a recommendation score for each battery electrolyte formulation in the recommendation set. Based on the recommendation score, the battery electrolyte formulations are ranked, and the battery electrolyte formulation with the appropriate ranking is selected as the recommended battery electrolyte formulation.
[0066] As can be seen, this embodiment of the application obtains a structured dataset of battery electrolytes. The structured dataset includes electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the structured dataset. Based on the electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the structured dataset, a prediction scoring matrix for the electrical performance indicators and electrolyte formulation attributes is determined. According to the prediction scoring matrix, the electrolyte formulation attribute with the highest predicted correlation score for each electrical performance indicator is selected as a recommended feature list to construct a battery electrolyte formulation recommendation set. Based on the battery electrolyte formulation recommendation set, a recommended battery electrolyte formulation is determined based on a preset model. This greatly improves the R&D efficiency of battery electrolyte formulations and reduces the economic costs of computing power and trial-and-error consumption.
[0067] As an feasible approach, a predictive scoring matrix for the electrical performance indicators and electrolyte formulation attributes is determined based on the electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the structured dataset. This includes:
[0068] Based on electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the structured dataset, an initial matrix between electrolyte formulation attributes and electrical performance indicators is constructed.
[0069] The initial matrix is decomposed into a potential feature matrix of electrical performance and a potential feature matrix of electrolyte formulation properties based on the alternating least squares algorithm.
[0070] Based on the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes, a prediction scoring matrix for electrical performance indicators and electrolyte formulation attributes is determined.
[0071] Based on the data structure <electrolyte performance index, electrolyte formulation attribute, score>, an initial matrix D is constructed between electrolyte formulation attributes and electrolyte performance index. Using the alternating least squares algorithm, the initial matrix is decomposed into a latent feature matrix U for electrolyte performance and a latent feature matrix V for electrolyte formulation attributes, where D∈R.m×n ,U∈R m×k ,V∈R n×k That is, D≈U×V T In this context, each row of U represents a latent vector of electrical performance, reflecting the degree of influence of different latent features on electrical performance, and each column of V represents a latent vector of electrolyte formulation attributes, reflecting the degree of various latent features contained in the formulation.
[0072] After obtaining the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes, the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes are continuously optimized to predict the correlation scores of all electrolyte formulation attributes and electrical performance indicators.
[0073] Specifically, based on the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes, a predictive scoring matrix for electrical performance indicators and electrolyte formulation attributes is determined, including:
[0074] Randomly initialize the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation properties;
[0075] By fixing the potential feature matrix of electrical properties, updating the potential feature matrix of electrolyte formulation attributes, the latest potential feature matrix of electrolyte formulation attributes is obtained.
[0076] By fixing the potential feature matrix of electrolyte formulation properties, updating the potential feature matrix of electrical performance, the latest potential feature matrix of electrical performance is obtained;
[0077] Based on the latest potential feature matrix of electrolyte formulation attributes and the latest potential feature matrix of electrical performance, a predictive scoring matrix for electrical performance indicators and electrolyte formulation attributes is determined.
[0078] Typically, the latent feature matrices of electrical performance and electrolyte formulation attributes are randomly initialized using a normal or uniform distribution. The loss function for constructing the latent feature matrices of electrical performance U and electrolyte formulation attributes V is expressed as follows:
[0079] Where, r i,j This represents the score of electrical performance index i for electrolyte formulation attribute j, where λ is the regularization coefficient, and u i The eigenvector representing electrical performance index i, v j The feature vector representing the electrolyte formulation attribute j.
[0080] With a fixed electrical performance latent characteristic matrix U, L(U, V) with respect to v j Taking the partial derivative, we obtain the update formula v. j =(U T U+λI) -1 U T rj , where r j Let I represent the score vector of electrolyte formulation attribute j, and let I represent the identity matrix. Update the latent feature matrix V of electrolyte formulation attributes to obtain the latest latent feature matrix of electrolyte formulation attributes.
[0081] With a fixed electrolyte formulation property latent feature matrix V, L(U,V) with respect to u i Taking the partial derivative, we obtain the update formula u. i =(V T V+λI) -1 V T r i Update the potential feature matrix U of electrical performance to obtain the latest potential feature matrix of electrical performance.
[0082] Among them, the potential feature matrix U of electrical performance and the potential feature matrix V of electrolyte formulation attributes in the update formula both have correlation scores in the structured dataset.
[0083] Calculate the product of the latest electrical performance latent feature matrix and the latest electrolyte formulation attribute latent feature matrix to obtain the initial scoring matrix of electrical performance and electrolyte formulation attributes. The elements in the initial scoring matrix are all predicted correlation scores. Calculate the error L between the correlation scores existing in the structured dataset and the corresponding correlation scores in the initial scoring matrix. Repeat the above steps until convergence, that is, the update of the electrical performance latent feature matrix U and the electrolyte formulation attribute latent feature matrix V no longer leads to a significant reduction in the total error. The scoring matrix obtained at this time is determined as the predicted scoring matrix of electrical performance index and electrolyte formulation attribute.
[0084] As one feasible approach, based on the prediction scoring matrix, the electrolyte formulation attribute with the highest predicted relevance score for each electrical performance index is selected as a recommended feature list to construct an electrolyte formulation recommendation set, including:
[0085] Based on the prediction correlation scores in the prediction scoring matrix, all electrolyte formulation attributes corresponding to each electrical performance index are sorted.
[0086] The electrolyte formulation attribute with the highest prediction correlation score among each electrical performance index is used as the recommended feature list;
[0087] A recommended electrolyte formulation set is constructed by compiling a list of recommended features for each electrical performance indicator.
[0088] Based on the magnitude of the predicted relevance scores in the predicted scoring matrix, all electrolyte formulation attributes corresponding to each electrical performance index are sorted to obtain a sequence of electrolyte formulation attributes. The electrolyte formulation attribute ranked first in this sequence has the highest predicted relevance score. The electrolyte formulation attribute with the highest predicted relevance score is used as the recommended feature list for that electrical performance index. The recommended feature lists of each electrical performance index are then compiled to construct an electrolyte formulation recommendation set.
[0089] As one feasible approach, a recommended battery electrolyte formulation is determined based on a pre-defined model from a set of recommended battery electrolyte formulations, including:
[0090] The numerical features of the battery electrolyte formulation recommendation set are cross-combined and standardized to obtain the first standardized feature.
[0091] The category field in the battery electrolyte formulation recommendation set is uniquely thermally encoded, and the numerical features are standardized to obtain the second standardized features.
[0092] Based on the first and second standardized features, the recommendation score of each battery electrolyte formulation in the battery electrolyte formulation recommendation set is determined based on the preset model.
[0093] Recommended battery electrolyte formulations are determined based on the recommended score for each formulation.
[0094] Numerical features from the battery electrolyte formulation recommendation set can be cross-combined using methods such as multinomial cross-combination and Cartesian product cross-combination to form new features. These new features are then standardized; for example, the Z-Score standardization algorithm can be used to convert the features into a distribution with a mean of 0 and a standard deviation of 1, or the Min-Max normalization algorithm can be used to scale the features to the range [0,1]. Based on these methods, a first standardized feature is obtained, which serves as the input to the Wide model.
[0095] For the categorical fields (classification features) in the battery electrolyte formulation recommendation set, one-hot encoding is performed to convert them into numerical values. For example, a categorical field about electrolyte solvent type has three possible values: ethylene carbonate (EC), diethyl carbonate (DEC), and dimethyl carbonate (DMC). Three new binary features are created to correspond to these three solvent types. If an electrolyte formulation uses ethylene carbonate (EC) as the solvent, the one-hot encoded vector is [1,0,0]; if it's diethyl carbonate (DEC), it's [0,1,0]; and if it's dimethyl carbonate (DMC), it's [0,0,1]. One-hot encoding helps improve the model's accuracy and generalization ability.
[0096] Next, the numerical features in the battery electrolyte formulation recommendation set are standardized using the same method as that used to generate the first standardized features. The second standardized features include standardized features of both categorical features and numerical features.
[0097] The preset models include a Wide model and a Deep model; based on the first and second normalized features, the recommendation score for each battery electrolyte formulation in the battery electrolyte formulation recommendation set is determined based on the preset models, including:
[0098] The first standardized feature is input into the generalized linear model to obtain the first output feature;
[0099] The second standardized feature is input into the deep neural network model to obtain the second output feature;
[0100] The first and second output features are concatenated and combined, and then input into the fully connected layer for fusion to obtain the recommendation score for each battery electrolyte formula in the battery electrolyte formula recommendation set.
[0101] Before obtaining a structured dataset of battery electrolyte, the method also includes: building a generalized linear model based on setting linear parameters of the linear model;
[0102] Construct a deep neural network model by setting up an input layer, two hidden layers, an output layer, and an activation function;
[0103] A pre-defined model is constructed based on a generalized linear model, a deep neural network model, and a fully connected layer.
[0104] The Wide model is built upon a linear model using logistic regression. It sets the linear parameters, which can include weights for each standardized feature and bias terms. The linear model algorithm can be expressed as follows: Where, x i To represent the standardized features, w i represents the weights corresponding to the standardized features, b represents the bias term, and m represents the number of standardized features. The Wide model is used to learn the simple and clear interaction relationships and rules of battery electrolyte formulation features in the battery electrolyte formulation recommendation set. It can capture some strongly correlated recommendation signals and play a role in memorization.
[0105] A simple deep neural network is constructed, consisting of an input layer, two hidden layers, an output layer, and an activation function, serving as the Deep model. The input layer has k neurons; if the dimension after One-Hot Encoding is 'a' and the dimension of the numerical features is 'c', then k = a + c. The first hidden layer has 64 neurons, the second hidden layer has 32 neurons, and the output layer has 1 neuron. The ReLU activation function is chosen to introduce non-linearity and enhance the Deep model's ability to non-linearly fit the battery electrolyte formulation characteristics to electrical performance. The ReLU function can be expressed as:
[0106] The first standardized feature is input into the Wide model, directly calculating a linear output, i.e., the first output feature. The second standardized feature is input into the Deep model, obtaining the second output feature. Then, the first and second output features are concatenated and combined through a multi-layer neural network and input into a fully connected layer for fusion. The final output is the recommendation score for each battery electrolyte formulation in the recommended battery electrolyte formulation set. The recommendation scores of each battery electrolyte formulation are sorted, and the number of recommended battery electrolyte formulations can be selected according to actual needs. For example, the top 3 battery electrolyte formulations with the highest recommendation scores are determined as recommended battery electrolyte formulations, and the electrical performance of the recommended battery electrolyte formulations is verified.
[0107] It should be understood that although the steps in the flowchart of Figure 2 are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated in this application, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in Figure 2 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.
[0108] Figure 3 is a schematic diagram of a battery electrolyte formulation optimization device provided in an embodiment of this application. This device can be installed at terminal 110 in the system shown in Figure 1 to execute the method flow shown in Figure 2. As shown in Figure 3, the battery electrolyte formulation optimization device 300 may include: an acquisition module 310, a determination module 320, and a selection module 330, and may further include: a model building module. The main functions of each component module are as follows:
[0109] The acquisition module 310 is used to acquire a structured dataset of battery electrolyte. The structured dataset includes electrolyte formulation attributes, electrical performance indicators, and correlation scores in the structured dataset. The correlation scores are the scores given by the user on the degree of correlation between electrolyte formulation attributes and electrical performance indicators.
[0110] The determination module 320 is used to determine the prediction scoring matrix of the electrolytic performance index and the electrolyte formulation attribute based on the electrolyte formulation attributes, the electrolytic performance index and the correlation scores existing in the structured dataset. The prediction scoring matrix includes the prediction correlation scores of all electrolyte formulation attributes corresponding to each electrolytic performance index.
[0111] Module 330 is used to select the electrolyte formulation attribute with the highest predicted correlation score in each electrical performance index as a recommended feature list to construct a battery electrolyte formulation recommendation set based on the prediction scoring matrix.
[0112] The determination module 320 is also used to determine the recommended battery electrolyte formula based on a preset model according to the recommended set of battery electrolyte formulas.
[0113] As one feasible approach, module 320 is specifically used to: construct an initial matrix between electrolyte formulation attributes and electrical performance indicators based on electrolyte formulation attributes, electrical performance indicators, and correlation scores present in the structured dataset; decompose the initial matrix into a potential feature matrix of electrical performance and a potential feature matrix of electrolyte formulation attributes based on the alternating least squares algorithm; and determine the prediction scoring matrix of electrical performance indicators and electrolyte formulation attributes based on the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes.
[0114] As one feasible approach, module 320 is specifically used for: randomly initializing the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes; fixing the potential feature matrix of electrical performance and updating the potential feature matrix of electrolyte formulation attributes to obtain the latest potential feature matrix of electrolyte formulation attributes; fixing the potential feature matrix of electrolyte formulation attributes and updating the potential feature matrix of electrical performance to obtain the latest potential feature matrix of electrical performance; and determining the prediction scoring matrix of electrical performance indicators and electrolyte formulation attributes based on the latest potential feature matrix of electrolyte formulation attributes and the latest potential feature matrix of electrical performance.
[0115] As one possible approach, module 330 is specifically used to: sort all electrolyte formulation attributes corresponding to each electrical performance index based on the predicted relevance scores in the predicted scoring matrix; take the electrolyte formulation attribute with the highest predicted relevance score for each electrical performance index as a recommended feature list; and compile the recommended feature lists for each electrical performance index to construct an electrolyte formulation recommendation set.
[0116] As one feasible approach, module 320 is further specifically used for: cross-combining and standardizing the numerical features in the battery electrolyte formulation recommendation set to obtain a first standardized feature; performing unique thermal encoding on the category fields in the battery electrolyte formulation recommendation set and standardizing the numerical features to obtain a second standardized feature; determining the recommendation score of each battery electrolyte formulation in the battery electrolyte formulation recommendation set based on a preset model according to the first and second standardized features; and determining the recommended battery electrolyte formulation based on the recommendation score of each battery electrolyte formulation.
[0117] As an achievable approach, the preset model includes a generalized linear model and a deep neural network model; the determination module 320 is also specifically used for: inputting the first standardized feature into the generalized linear model to obtain the first output feature; inputting the second standardized feature into the deep neural network model to obtain the second output feature; and concatenating and combining the first output feature and the second output feature and then inputting them into a fully connected layer for fusion to obtain the recommendation score of each battery electrolyte formula in the battery electrolyte formula recommendation set.
[0118] As one possible approach, the device also includes a model building module, which is used to build a generalized linear model based on linear parameters before acquiring a structured dataset of battery electrolyte; build a deep neural network model by setting an input layer, two hidden layers, an output layer and an activation function; and build a preset model based on the generalized linear model, the deep neural network model and the fully connected layer.
[0119] It is understood that implementing any method or product of this application does not necessarily require achieving all of the advantages described above simultaneously.
[0120] The same or similar parts among the above embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments.
[0121] According to embodiments of this application, this application also provides a computer device and a computer-readable storage medium.
[0122] Figure 4 is a block diagram of a computer device according to an embodiment of this application. The term "computer device" is intended to represent various forms of digital computers or mobile devices. The digital computer may include a desktop computer, a portable computer, a workbench, a personal digital assistant, a server, a mainframe computer, and other suitable computers. The mobile device may include a tablet computer, a smartphone, a wearable device, etc.
[0123] As shown in Figure 4, the computer device 400 includes a computing unit 401, a ROM 402, a RAM 403, a bus 404, and an input / output (I / O) interface 405. The computing unit 401, ROM 402, and RAM 403 are interconnected via the bus 404. The input / output (I / O) interface 405 is also connected to the bus 404.
[0124] The computing unit 401 can execute various processes in the method embodiments of this application according to computer instructions stored in the read-only memory (ROM) 402 or computer instructions loaded from the storage unit 408 into the random access memory (RAM) 403. The computing unit 401 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. The computing unit 401 can include, but is not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. In some embodiments, the methods provided in the embodiments of this application can be implemented as computer software programs, which are tangibly contained in a computer-readable storage medium, such as the storage unit 408.
[0125] RAM 403 can also store various programs and data required for the operation of computer device 400. Part or all of the computer program can be loaded and / or installed on computer device 400 via ROM 402 and / or communication unit 409.
[0126] The input unit 406, output unit 407, storage unit 408, and communication unit 409 in the computer device 400 can be connected to the I / O interface 405. The input unit 406 can be, for example, a keyboard, mouse, touchscreen, or microphone; the output unit 407 can be, for example, a monitor, speaker, or indicator light. The computer device 400 can exchange information and data with other devices through the communication unit 409.
[0127] It should be noted that the device may also include other components necessary for normal operation. It may also include only the components necessary for implementing the solution of this application, without necessarily including all the components shown in the figures.
[0128] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof.
[0129] The computer instructions used to implement the methods of this application may be written in any combination of one or more programming languages. These computer instructions may be provided to the computing unit 401 such that when executed by the computing unit 401, such as a processor, the computer instructions cause the execution of the steps involved in the embodiments of the methods of this application.
[0130] The computer-readable storage medium provided in this application can be a non-volatile computer-readable storage medium. It can be a tangible medium that can contain or store multiple computer instructions readable by a computer device for performing the steps involved in the method embodiments of this application. The computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, and other forms of storage media.
[0131] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for optimizing a battery electrolyte formulation, characterized in that, The method includes: Obtain a structured dataset of battery electrolytes, the structured dataset including electrolyte formulation attributes, electrical performance indicators, and correlation scores existing in the structured dataset, the correlation scores being the ratings given by users on the degree of correlation between the electrolyte formulation attributes and the electrical performance indicators; Based on the electrolyte formulation attributes, the electrical performance indicators, and the correlation scores present in the structured dataset, a prediction scoring matrix for the electrical performance indicators and the electrolyte formulation attributes is determined. The prediction scoring matrix includes the predicted correlation scores for all electrolyte formulation attributes corresponding to each electrical performance indicator. Based on the predicted scoring matrix, the electrolyte formulation attribute with the highest predicted relevance score for each electrical performance index is selected as a recommended feature list to construct a battery electrolyte formulation recommendation set; and Based on the battery electrolyte formulation recommendation set, a recommended battery electrolyte formulation is determined using a preset model.
2. The method according to claim 1, characterized in that, The step of determining the prediction scoring matrix for the electrical performance indicators and electrolyte formulation attributes based on the electrolyte formulation attributes, the electrical performance indicators, and the correlation scores present in the structured dataset includes: Based on the electrolyte formulation attributes, the electrical performance indicators, and the correlation scores present in the structured dataset, an initial matrix is constructed between the electrolyte formulation attributes and the electrical performance indicators. The initial matrix is decomposed into a latent feature matrix of electrical performance and a latent feature matrix of electrolyte formulation properties based on the alternating least squares algorithm; and Based on the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes, a prediction scoring matrix for the electrical performance index and the electrolyte formulation attributes is determined.
3. The method according to claim 2, characterized in that, The step of determining the prediction scoring matrix of the electrical performance index and the electrolyte formulation attribute based on the potential feature matrix of the electrical performance and the potential feature matrix of the electrolyte formulation attribute includes: Randomly initialize the potential feature matrix of electrical performance and the potential feature matrix of electrolyte formulation attributes; Fix the potential feature matrix of electrical properties, update the potential feature matrix of electrolyte formulation attributes, and obtain the latest potential feature matrix of electrolyte formulation attributes; By fixing the latent feature matrix of the electrolyte formulation attributes and updating the latent feature matrix of electrical performance, the latest latent feature matrix of electrical performance is obtained; and Based on the latest potential feature matrix of electrolyte formulation attributes and the latest potential feature matrix of electrical performance, a prediction scoring matrix for the electrical performance index and the electrolyte formulation attributes is determined.
4. The method according to claim 1, characterized in that, The step involves selecting the electrolyte formulation attribute with the highest predicted relevance score among each electrical performance index from the predicted scoring matrix to construct a recommended electrolyte formulation set, including: Based on the prediction correlation scores in the prediction scoring matrix, all electrolyte formulation attributes corresponding to each electrical performance index are sorted. The electrolyte formulation attribute with the highest predicted correlation score among each of the aforementioned electrical performance indicators is selected as a recommended feature list; and An electrolyte formulation recommendation set is constructed by compiling a list of recommended features for each of the aforementioned electrical performance indicators.
5. The method according to claim 1, characterized in that, The step of determining a recommended battery electrolyte formulation based on a preset model according to the battery electrolyte formulation recommendation set includes: The numerical features in the battery electrolyte formulation recommendation set are cross-combined and standardized to obtain the first standardized feature. The category field in the battery electrolyte formulation recommendation set is uniquely thermally encoded, and the numerical features are standardized to obtain the second standardized features. Based on the first and second standardized features, and using the preset model, determine the recommendation score for each battery electrolyte formulation in the recommended set; and Recommended battery electrolyte formulations are determined based on the recommended score for each of the battery electrolyte formulations.
6. The method according to claim 5, characterized in that, The preset model includes a generalized linear model and a deep neural network model; the step of determining the recommendation score of each battery electrolyte formulation in the battery electrolyte formulation recommendation set based on the preset model according to the first standardized feature and the second standardized feature includes: The first standardized feature is input into the generalized linear model to obtain the first output feature; The second standardized feature is input into the deep neural network model to obtain the second output feature; and The first output feature and the second output feature are concatenated and combined, and then input into a fully connected layer for fusion to obtain the recommendation score of each battery electrolyte formula in the battery electrolyte formula recommendation set.
7. The method according to claim 1, characterized in that, Before obtaining a structured dataset of battery electrolytes, the method further includes: A generalized linear model is constructed by setting linear parameters based on the linear model; A deep neural network model is constructed by setting up an input layer, two hidden layers, an output layer, and activation functions; and Based on the generalized linear model, the deep neural network model, and the fully connected layer, a preset model is constructed.
8. A battery electrolyte formulation optimization device, characterized in that, The device includes: The acquisition module is used to acquire a structured dataset of battery electrolyte. The structured dataset includes electrolyte formulation attributes, electrical performance indicators, and correlation scores in the structured dataset. The correlation scores are the ratings given by users on the degree of correlation between the electrolyte formulation attributes and the electrical performance indicators. The determination module is used to determine a prediction scoring matrix of the electrolytic performance index and the electrolyte formulation attribute based on the electrolyte formulation attribute, the electrical performance index and the correlation score present in the structured dataset. The prediction scoring matrix includes the predicted correlation score of each electrical performance index for all electrolyte formulation attributes. The selection module is used to select the electrolyte formulation attribute with the highest predicted relevance score for each of the electrical performance indicators as a recommended feature list to construct a battery electrolyte formulation recommendation set; and The determining module is further configured to determine a recommended battery electrolyte formula based on a preset model according to the battery electrolyte formula recommendation set.
9. A computer device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores computer instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 1-7.
10. A computer-readable storage medium storing computer instructions thereon, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.
