A lithium battery SOH estimation model and its construction system and application method

Abstract

The present invention provides a lithium battery SOH estimation model and its construction system and application method. The construction process of the estimation model includes data collection process, data processing and aggregation process, feature engineering, model construction process, model training process and model optimization process. The core is that the battery only needs to be charged to a remaining power (State of Charge, SOC) greater than 15%, and the battery charging data is collected during this process. By building and training a high-precision neural network model, the complex nonlinear relationship in the battery charging process can be learned, thereby achieving accurate prediction of SOH. Compared with traditional methods, the present invention not only significantly reduces the difficulty and cost of operation, but also improves the accuracy and efficiency of estimation.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery health management, and in particular to a lithium battery SOH estimation model and a construction system and application method thereof. Background Art

[0002] With the rapid progress of electronic science and technology and the continuous improvement of people's living standards, lithium-ion batteries have become an indispensable core component of various mobile devices and electric vehicles. As a key indicator for its performance evaluation, the state of health (SOH) of lithium-ion batteries not only reveals the comparative relationship between the current working performance of the battery and the initial or rated state, but is also directly related to the stable operation of the equipment, the extension of battery life, and the optimization of overall energy management.

[0003] For stable equipment operation, accurate SOH estimation can help predict the trend of battery performance degradation and take timely measures to prevent equipment downtime or performance degradation due to battery failure, thereby ensuring the continued efficient operation of the equipment.

[0004] To extend the battery life, by monitoring SOH, more accurate battery management strategies can be implemented, such as adjusting the charging strategy, limiting the depth of discharge, etc., to slow down the battery aging and extend the overall battery life.

[0005] For optimizing energy management, SOH information is essential for energy distribution, charging strategy planning, and maintenance cost control in electric vehicles and large energy storage systems. By accurately estimating SOH, energy can be managed more efficiently and operating costs can be reduced.

[0006] However, the estimation of lithium battery SOH is a complex process, which is affected by many factors, including the number of battery charge and discharge cycles, working environment temperature, charge and discharge current and voltage, etc. Therefore, data-driven methods, especially neural network methods, have gradually become a research hotspot in the field of lithium battery SOH estimation due to their powerful nonlinear fitting and self-learning capabilities.

[0007] Currently, some traditional methods for estimating the SOH of lithium batteries are known for their high accuracy. However, these methods often rely on strict laboratory environments and require the battery to undergo a complete charge and discharge cycle (i.e., full charge and discharge). This process not only significantly increases the complexity of the operation, but also brings a high cost burden. Specifically, the full charge and discharge process may accelerate the aging of the battery and shorten its service life. At the same time, the construction and maintenance of the laboratory environment also require considerable investment. In addition, for new energy vehicles, frequent full charge and discharge operations not only affect the normal use of the vehicle, but may also have an adverse effect on the overall performance of the battery pack. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide a lithium battery SOH estimation model and its construction system and application method. By introducing neural network technology to construct a SOH estimation model, which is applied to the estimation of lithium battery SOH, the problems of cumbersome operation and high cost in the existing lithium battery SOH estimation method are solved, and a more efficient, accurate and economical solution is provided for battery management of new energy vehicles.

[0009] In a first aspect, the present invention provides a lithium battery SOH estimation model, which is constructed by the following process:

[0010] Data collection process: Collect charging data of new energy vehicles from different manufacturers and models. The charging data is the detailed charging data recorded at a certain frequency during the full charging process, including key parameters such as current SOC value, charging voltage, charging current, charging capacity, and rated parameters of each vehicle;

[0011] Data processing and aggregation process: The charging data is grouped and aggregated according to the current SOC value to obtain charging statistics in each SOC interval; a sliding window method of 15% SOC is used to generate each sample in the training set;

[0012] Feature Engineering: For continuous features with values ​​in a fixed interval, the min-max normalization method is used to map the value range to [0, 1] to eliminate dimensional differences; for continuous features with values ​​that do not have a fixed interval, the z-score normalization method is used to convert them into a standard normal distribution; for categorical features with fixed categories, the one-hot encoding method is used to convert categorical data into a numerical form that can be processed by the machine learning model; for categorical features with non-fixed categories, the embedding encoding method is used;

[0013] Model building process: A neural network model is constructed using the MLP model and the LSTM model; the MLP model is used to encode the rated parameter data of the vehicle and extract key information; the LSTM model is used to encode the charging details data of the vehicle and add a Self-Attention mechanism to capture the timing dependency; the key information and the timing dependency are connected so that the output layer of the neural network model can output the SOH estimation value;

[0014] Model training process: using the charging data collected in the data collection process to train the estimation model of the lithium battery SOH, to obtain the estimation model of the lithium battery SOH.

[0015] In a second aspect, the present invention provides a system for constructing a lithium battery SOH estimation model, comprising:

[0016] The data collection module is used to collect charging data of new energy vehicles from different manufacturers and models. The charging data is the detailed charging data recorded at a certain frequency during the full charging process, including key parameters such as the current SOC value, charging voltage, charging current, charging capacity, and rated parameters of each vehicle;

[0017] A data processing and aggregation module is used to group and aggregate the charging data according to the SOC value to obtain charging statistics in each SOC interval; a sliding window method of 15% SOC is used to generate each sample in the training set;

[0018] Feature engineering module, for continuous features with values ​​in a fixed interval, uses the min-max standardization method to map the value range to [0, 1] to eliminate dimensional differences; for continuous features with values ​​without a fixed interval, uses the z-score standardization method to convert them into a standard normal distribution; for categorical features with fixed categories, uses the one-hot encoding method to convert categorical data into a numerical form that can be processed by the machine learning model; for categorical features with non-fixed categories, uses the embedding encoding method;

[0019] A model building module, used to build a neural network model using an MLP model and an LSTM model; the MLP model is used to encode the rated parameter data of the vehicle and extract key information; the LSTM model is used to encode the charging detail data of the vehicle and add a Self-Attention mechanism to capture the timing dependency; the key information and the timing dependency are connected so that the output layer of the neural network model can output the SOH estimation value;

[0020] A model training module, used to train the estimation model of the lithium battery SOH using the charging data collected by the data collection module to obtain the estimation model of the lithium battery SOH;

[0021] The model optimization module is used to adjust the model's hyperparameters using a grid search method during model training to better adapt to different data sets and scenarios.

[0022] In the third aspect, the present invention provides an application method of a lithium battery SOH estimation model, which records charging detail data with SOC values ​​greater than 15% at a preset frequency, records the rated parameters of each vehicle, processes the charging detail data in a manner of constructing a training set, loads the trained lithium battery SOH estimation model as described in the first aspect, and inputs the processed charging detail data. The lithium battery SOH estimation model infers and calculates the feature data through the network structure of the model, and outputs a predicted SOH value.

[0023] One or more technical solutions provided by the present invention have at least the following technical effects or advantages:

[0024] (1) Higher estimation accuracy: By combining the neural network architecture of LSTM (Long Short-Term Memory Network) and MLP (Multi-layer Perceptron), it is possible to simultaneously capture the timing dependency and non-timing characteristics of the battery charging process, thereby more accurately estimating the battery's SOH (State of Health). Compared with the existing technology that only focuses on one aspect of the timing data or static data and ignores the interaction between the two, the accuracy of the estimation result is greatly improved.

[0025] (2) Easy operation and low cost: When collecting data, the charging details are recorded at a fixed frequency only during the full charging process, avoiding frequent full charging and discharging operations and simplifying the battery testing process; reducing test time and energy consumption, reducing test costs; shortening the test cycle and improving the overall efficiency of the battery management system; avoiding damage to the battery caused by frequent full charging and discharging, which helps to extend the battery life.

[0026] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0028] Figure 1 It is a flow chart of the implementation process of the estimation model of lithium battery SOH of the present invention;

[0029] Figure 2 Schematic diagram of the principle of the sliding window data acquisition method of 15% SOC in an embodiment of the present invention;

[0030] Figure 3 A flowchart of the model building process of the present invention;

[0031] Figure 4 It is a structural schematic diagram of the model building device of the present invention. DETAILED DESCRIPTION

[0032] The embodiments of the present application provide a lithium battery SOH estimation model and an application method thereof, and construct a SOH estimation model by introducing neural network technology, which is applied to the estimation of lithium battery SOH, thereby solving the problems of cumbersome operation and high cost in the existing lithium battery SOH estimation method, and providing a more efficient, accurate and economical solution for battery management of new energy vehicles.

[0033] The technical solution in the embodiment of the present application has the following general idea: an innovative neural network-based lithium battery SOH estimation method is proposed, the core of which is that it abandons the cumbersome operation of fully charging and discharging the battery in the traditional method, and instead adopts a simpler and more cost-effective shallow charging strategy. Specifically, it is only necessary to charge the battery to a remaining power (State of Charge, SOC) greater than 15%, and collect the battery charging data in the process, including but not limited to key parameters such as voltage, current, and charging capacity. By constructing and training a high-precision neural network model, the present invention can use these collected data to accurately estimate the SOH of the battery. The neural network model can learn the complex nonlinear relationship in the battery charging process, thereby realizing accurate prediction of SOH. Compared with traditional methods, the present invention not only significantly reduces the difficulty and cost of operation, but also improves the accuracy and efficiency of estimation. In addition, since frequent full charging and discharging operations are avoided, it also helps to extend the service life of the battery and improve the overall performance of new energy vehicles. In summary, the present invention aims to solve the problems of cumbersome operation and high cost in existing lithium battery SOH estimation methods by introducing neural network technology, and to provide a more efficient, accurate and economical solution for battery management of new energy vehicles. Embodiment 1

[0034] like Figure 1 As shown, this embodiment provides a lithium battery SOH estimation model, and its construction process includes data collection process, data processing and aggregation process, feature engineering, model construction process, model training process and model optimization process.

[0035] Data collection process: Collect charging data from new energy vehicles of different manufacturers and models. The charging data is detailed charging data recorded at a certain frequency during the full charging process, including key parameters such as current SOC value, charging voltage, charging current, charging capacity, and rated parameters of each vehicle; the frequency is not fixed, but data should be collected according to equal frequency intervals, for example, data should be recorded at a frequency of every 1 second or every 5 seconds.

[0036] By collecting charging data from new energy vehicles of different manufacturers and models, a more accurate battery SOH estimation model can be constructed. The rated parameters of each vehicle, such as rated capacity and rated voltage, can provide necessary references for subsequent model construction. Since the full charging process data is collected, the calculation method for the actual SOH of the vehicle is: The true SOH is used as the label value of the model and can be used for the back propagation of loss during subsequent model training.

[0037] Data processing and aggregation process: The charging data is grouped and aggregated according to the current SOC value to obtain charging statistics in each SOC interval; a 15% SOC sliding window method is used to generate each sample in the training set; this method enables us to capture the charging behavior of the battery in a continuous SOC interval, thereby improving the generalization ability of the model. Among them, the 15% SOC sliding window method is specifically: after the charging data is grouped and aggregated, a gradually increasing SOC sequence is obtained, such as Figure 2 As shown, by using a window size of 15 and a step size of 1, sliding data is taken on the SOC sequence to obtain multiple SOC sequences of the same length.

[0038] Feature engineering: For continuous features with values ​​in a fixed interval, the min-max normalization method is used to map the value range to [0, 1] to eliminate dimensional differences; for continuous features with values ​​that do not have a fixed interval, the z-score normalization method is used to convert them into a standard normal distribution; for categorical features with fixed categories, the one-hot encoding method is used to convert categorical data into a numerical form that can be processed by the machine learning model; for categorical features with non-fixed categories, the embedding encoding method is used; these feature processing methods help the model better understand and utilize the information in the input data.

[0039] The specific steps of using the min-max normalization method to map the range of continuous features with values ​​in a fixed interval to between [0, 1] are as follows: the minimum min and maximum max of each feature in the data set are calculated; each feature value x in the original data set is replaced by its standardized value x' by applying the following conversion formula: x' = (x -min) / (max- min);

[0040] For continuous features whose values ​​do not have a fixed interval, the z-score standardization method is used to convert them into a standard normal distribution. Specifically, the mean μ and standard deviation σ of each feature in the data set are calculated; and the following conversion formula is applied to replace each feature value x in the original data set with its standardized value x': y' = (y -μ) / σ.

[0041] The one-hot encoding method is used to convert the categorical data into a numerical form that can be processed by the machine learning model for the categorical features with fixed categories. Specifically, for each categorical feature, the possible number of categories is determined; for each data point, a binary vector with a length equal to the number of categories is created, in which the position corresponding to the category in the vector is 1, and the other positions are 0; each categorical feature in the original data set is replaced by its corresponding One-Hot encoding vector.

[0042] The specific steps of using the embedding encoding method to process classification features with non-fixed categories are: selecting a suitable embedding dimension; creating a randomly initialized embedding matrix with the number of rows being the number of categories and the number of columns being the embedding dimension; during the model training process, optimizing the embedding matrix through the back propagation algorithm, and the embedding matrix is ​​trained as part of the network; during the model training and prediction process, using the trained embedding vector to replace the original category identifier.

[0043] Model building process: Figure 3 As shown, a neural network model is constructed using an MLP (multi-layer perceptron) model and an LSTM (long short-term memory network) model; the MLP model is used to encode the rated parameter data of the vehicle and extract key information; the LSTM model is used to encode the charging detail data of the vehicle and add a Self-Attention mechanism to capture the timing dependency; the key information and the timing dependency are connected so that the output layer of the neural network model can output the SOH estimation value;

[0044] The construction process of the MLP model includes:

[0045] Define the input layer: The size of the input layer is equal to the number of nominal parameters;

[0046] Construct hidden layers: Use one or more fully connected layers as hidden layers. Each hidden layer is followed by a batch normalization layer to accelerate training and improve the generalization ability of the model. Then, an activation function is used to introduce nonlinear factors into the model, so that the neural network can handle and solve more complex problems.

[0047] Construct output layer: The output layer is used to extract key information and its size can be adjusted as needed.

[0048] The construction process of the LSTM model includes:

[0049] Define the input layer: The LSTM input layer has a time dimension and a feature dimension. The size of the feature dimension is equal to the number of features of the charging details data, and the size of the time dimension is equal to the length of the time series.

[0050] Build LSTM layer: Use LSTM layer or bidirectional LSTM layer to process time series data;

[0051] Introducing the Self-Attention mechanism: The Self-Attention mechanism is applied to the output of the LSTM layer to capture the key information in the time series. The obtained output is summed along the time dimension to achieve dimensionality reduction, which is consistent with the dimension of the MLP output.

[0052] Construct the output layer: The size of the output layer is consistent with the output layer dimension of the MLP model, which is used to extract the key information of the charging details data.

[0053] The output of the MLP model is two-dimensional data, namely samples and features, while the output of the LSTM model is three-dimensional data, namely samples, time series, and features. When constructing the LSTM model, the present invention sums the output according to the time dimension to achieve dimensionality reduction, so the output dimension becomes samples and features, which is consistent with the output of the MLP. The two outputs are then concatenated according to the feature dimension and then passed to the output layer to obtain the final SOH.

[0054] Model training process: using the charging data collected in the data collection process to train the estimation model of the lithium battery SOH, to obtain the estimation model of the lithium battery SOH.

[0055] Model optimization process: During the model training process, in order to improve the accuracy of SOH estimation, a grid search method is used to adjust the model's hyperparameters to better adapt to different data sets and scenarios;

[0056] The grid search method includes the following steps:

[0057] Define the hyperparameters that need to be adjusted, including learning rate, batch size, number of LSTM units, number of hidden layers, and dropout rate;

[0058] Setting up a grid search, including defining a grid and cross-validation; defining a grid is to define a possible value range for each of the hyperparameters and create a grid containing all possible combinations; cross-validation is to use k-fold cross-validation, in each cross-validation, the training set is divided into k subsets, k-1 subsets are used for training each time, and the remaining 1 subset is used for validation;

[0059] Grid search execution includes traversing the grid and selecting the optimal combination; traversing the grid is to train and evaluate the model for each hyperparameter combination in the grid using a cross-validation method, and record the average validation performance of each combination; selecting the optimal combination is to select the optimal hyperparameter combination according to the average validation performance;

[0060] Training and evaluation: Use the selected optimal hyperparameter combination to perform final training on the entire training set to obtain the optimized model. Use the test set to evaluate the performance of the optimized model to ensure that the model has good generalization ability. Embodiment 2

[0061] Based on the same inventive concept, the present application also provides a device corresponding to the method in Example 1, see Example 2 for details.

[0062] like Figure 4 As shown, in this embodiment, a system for constructing a lithium battery SOH estimation model is provided, including a data collection module, a data processing and aggregation module, a feature engineering module, a model construction module, a model training module and a model optimization module.

[0063] The data collection module is used to collect charging data of new energy vehicles from different manufacturers and models. The charging data is the detailed charging data recorded at a certain frequency during the full charging process, including key parameters such as the current SOC value, charging voltage, charging current, charging capacity, and rated parameters of each vehicle;

[0064] A data processing and aggregation module is used to group and aggregate the charging data according to the SOC value to obtain charging statistics in each SOC interval; a sliding window method of 15% SOC is used to generate each sample in the training set;

[0065] Feature engineering module, for continuous features with values ​​in a fixed interval, uses the min-max standardization method to map the value range to [0, 1] to eliminate dimensional differences; for continuous features with values ​​without a fixed interval, uses the z-score standardization method to convert them into a standard normal distribution; for categorical features with fixed categories, uses the one-hot encoding method to convert categorical data into a numerical form that can be processed by the machine learning model; for categorical features with non-fixed categories, uses the embedding encoding method;

[0066] A model building module, used to build a neural network model using an MLP model and an LSTM model; the MLP model is used to encode the rated parameter data of the vehicle and extract key information; the LSTM model is used to encode the charging detail data of the vehicle and add a Self-Attention mechanism to capture the timing dependency; the key information and the timing dependency are connected so that the output layer of the neural network model can output the SOH estimation value;

[0067] The model training module is used to train the estimation model of the lithium battery SOH using the charging data collected by the data collection module to obtain the estimation model of the lithium battery SOH.

[0068] The model optimization module is used to adjust the model's hyperparameters using a grid search method to improve the accuracy of SOH estimation during model training to better adapt to different data sets and scenarios.

[0069] The connection relationship between the above modules is as follows:

[0070] The data output by the data collection module is passed to the data processing module for preprocessing and aggregation.

[0071] The processed data enters the feature engineering module for feature extraction and conversion.

[0072] The converted features are input into the SOH estimation model module for training and prediction.

[0073] The model training module and the model optimization module are responsible for model training and adjusting model parameters and hyperparameters to improve the accuracy of SOH estimation.

[0074] The operation process of building the system in this embodiment is as follows:

[0075] (1) Data collection: The data collection module records the charging details of new energy vehicles at a frequency of one second, and collects the rated parameters of each vehicle. The specific data collected can be shown in Table 1.

[0076] Table 1

[0077]

[0078] 2) Data processing: Use the data processing module to group and aggregate the data according to the SOC value to generate training samples. The constructed training set data can be shown in Table 2.

[0079] Table 2

[0080]

[0081] (3) Feature engineering: Through the feature engineering module, the collected data is extracted and transformed to form a feature representation suitable for model input. The processing method of each feature can be shown in Table 3.

[0082] Table 3

[0083]

[0084] (4) Model construction:

[0085] Input layer: receives feature data processed by feature engineering, including non-time series features and time series features.

[0086] LSTM layer: used to process time series features and capture the time series dependencies during battery charging. A DropOut layer is added after the LSTM layer to reduce overfitting.

[0087] MLP layer: used to process non-time series features and extract key information. A BN layer is added after the MLP layer to accelerate training and improve model performance.

[0088] Connection layer: connects the outputs of the LSTM layer and the MLP layer to form a joint feature representation.

[0089] Output layer: Output the estimated SOH value of the battery based on the joint feature representation.

[0090] The architecture of the obtained neural network model is shown in Figure 3 shown.

[0091] (5) Model training and optimization: The processed data is input into the neural network model and trained using the Adam optimizer and the MSE loss function. The hyperparameters are adjusted through grid search to optimize the model performance and obtain the final lithium battery SOH estimation model.

[0092] Since the device introduced in the second embodiment of the present invention is a device used to implement the method of the first embodiment of the present invention, those skilled in the art can understand the specific structure and deformation of the device based on the method introduced in the first embodiment of the present invention, so it is not described here in detail. All devices used in the method of the first embodiment of the present invention belong to the scope of protection of the present invention. Embodiment 3

[0093] Based on the same inventive concept, the present application also provides an application method of the lithium battery SOH estimation model of Example 1, see Example 3 for details.

[0094] The present embodiment provides an application method of a lithium battery SOH estimation model, which records charging detail data with an SOC value greater than 15% at a preset frequency, records the rated parameters of each vehicle, processes the charging detail data in a manner of constructing a training set, loads the lithium battery SOH estimation model trained by the method of Embodiment 1 or the system of Embodiment 2, and inputs the processed charging detail data. The lithium battery SOH estimation model infers and calculates the feature data through the network structure of the model, and outputs a predicted SOH value.

[0095] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A lithium battery SOH estimation model, characterized in that: Constructed through the following process: Data collection process: Collect charging data from new energy vehicles of different manufacturers and models. The charging data is detailed charging data recorded at a certain frequency during the full charging process, including the current SOC value, charging voltage, charging current, charging capacity and rated parameters of each vehicle; Data processing and aggregation process: grouping and aggregating the charging data according to the current SOC value to obtain charging statistics in each SOC interval; A 15% SOC sliding window method is used to generate each sample in the training set. The 15% SOC sliding window method is as follows: the charging data is grouped and aggregated to obtain a gradually increasing SOC sequence, and a sliding method is performed on the SOC sequence with a window size of 15 and a step size of 1 to obtain multiple SOC sequences of the same length; Feature Engineering: For continuous features with values ​​in a fixed interval, the min-max standardization method is used to map the value range to [0, 1]. For continuous features with values ​​that do not have a fixed interval, the z-score standardization method is used to convert them into a standard normal distribution. For categorical features with fixed categories, the one-hot encoding method is used to convert categorical data into a numerical form that can be processed by the machine learning model. For categorical features with non-fixed categories, the embedding encoding method is used. Model building process: A neural network model is built using the MLP model and the LSTM model; the MLP model is used to encode the vehicle's rated parameter data and extract key information; the LSTM model is used to encode the vehicle's charging details data and add a Self-Attention mechanism to capture temporal dependencies; Connecting the key information and the timing dependency so that an output layer of the neural network model can output an SOH estimation value; Model training process: using the charging data collected in the data collection process to train the estimation model of the lithium battery SOH, to obtain the estimation model of the lithium battery SOH.

2. The estimation model of lithium battery SOH according to claim 1, characterized in that: Also includes: Model optimization process: During the model training process, the grid search method is used to adjust the model's hyperparameters; The grid search method includes the following steps: Define the hyperparameters that need to be adjusted, including learning rate, batch size, number of LSTM units, number of hidden layers, and dropout rate; Setting up a grid search, including defining a grid and cross-validation; defining a grid is to define a possible value range for each of the hyperparameters and create a grid containing all possible combinations; cross-validation is to use k-fold cross-validation, in each cross-validation, the training set is divided into k subsets, k-1 subsets are used for training each time, and the remaining 1 subset is used for validation; Grid search execution includes traversing the grid and selecting the optimal combination; traversing the grid is to train and evaluate the model for each hyperparameter combination in the grid using a cross-validation method, and record the average validation performance of each combination; selecting the optimal combination is to select the optimal hyperparameter combination according to the average validation performance; Training and evaluation,Use the selected optimal hyperparameter combination to perform final training on the entire training set to obtain the optimized model, and use the test set to evaluate the performance of the optimized model.

3. The estimation model of lithium battery SOH according to claim 1, characterized in that: For continuous features whose values ​​are within a fixed interval, the min-max normalization method is used to map the value range to [0, 1] as follows: the minimum min and maximum max of each feature in the data set are calculated; each feature value x in the original data set is replaced by its standardized value x' by the following conversion formula: x' = (x -min) / (max - min); For continuous features whose values ​​do not have a fixed interval, the z-score standardization method is used to convert them into a standard normal distribution. Specifically, the mean μ and standard deviation σ of each feature in the data set are calculated; and the following conversion formula is applied to replace each feature value y in the original data set with its standardized value y': y' = (y -μ) / σ.

4. The estimation model of lithium battery SOH according to claim 1, characterized in that: The one-hot encoding method is used to convert the categorical data into a numerical form that can be processed by the machine learning model for the categorical features with fixed categories. Specifically, for each categorical feature, the possible number of categories is determined; for each data point, a binary vector with a length equal to the number of categories is created, in which the position corresponding to the category in the vector is 1, and the other positions are 0; each categorical feature in the original data set is replaced by its corresponding One-Hot encoding vector.

5. The estimation model of lithium battery SOH according to claim 1, characterized in that: The specific steps of using the embedding encoding method to process classification features with non-fixed categories are: selecting a suitable embedding dimension; creating a randomly initialized embedding matrix with the number of rows being the number of categories and the number of columns being the embedding dimension; during the model training process, optimizing the embedding matrix through the back propagation algorithm, and the embedding matrix is ​​trained as part of the network; during the model training and prediction process, using the trained embedding vector to replace the original category identifier.

6. The estimation model of lithium battery SOH according to claim 1, characterized in that: The construction process of the MLP model includes: Define the input layer: The size of the input layer is equal to the number of nominal parameters; Construct hidden layers: Use one or more fully connected layers as hidden layers. Each hidden layer is followed by a batch normalization layer and then an activation function to introduce nonlinear factors into the model. Construct output layer: The output layer is used to extract key information and its size can be adjusted as needed.

7. The estimation model of lithium battery SOH according to claim 1, characterized in that: The construction process of the LSTM model includes: Define the input layer: The LSTM input layer has a time dimension and a feature dimension. The size of the feature dimension is equal to the number of features of the charging details data, and the size of the time dimension is equal to the length of the time series. Build LSTM layer: Use LSTM layer or bidirectional LSTM layer to process time series data; Introducing the Self-Attention mechanism: Apply the Self-Attention mechanism to the output of the LSTM layer to capture key information in the time series and sum the obtained output along the time dimension; Construct the output layer: The size of the output layer is consistent with the output layer dimension of the MLP model, which is used to extract the key information of the charging details data.

8. A system for constructing a lithium battery SOH estimation model, characterized in that: include: The data collection module is used to collect charging data of new energy vehicles from different manufacturers and models. The charging data is the charging detailed data recorded at a certain frequency during the full charging process, including the current SOC value, charging voltage, charging current, charging capacity and rated parameters of each vehicle; A data processing and aggregation module, used for grouping and aggregating the charging data according to the current SOC value to obtain charging statistics in each SOC interval; A 15% SOC sliding window method is used to generate each sample in the training set. The 15% SOC sliding window method is as follows: the charging data is grouped and aggregated to obtain a gradually increasing SOC sequence, and a sliding method is performed on the SOC sequence with a window size of 15 and a step size of 1 to obtain multiple SOC sequences of the same length; The feature engineering module is used to map the value range of continuous features within a fixed interval to [0, 1] using the min-max standardization method; for continuous features with values ​​that do not have a fixed interval, the z-score standardization method is used to convert them into a standard normal distribution; for categorical features with fixed categories, the one-hot encoding method is used to convert categorical data into a numerical form that can be processed by the machine learning model; for categorical features with non-fixed categories, the embedding encoding method is used; A model building module, used to build a neural network model using an MLP model and an LSTM model; the MLP model is used to encode the rated parameter data of the vehicle and extract key information; the LSTM model is used to encode the charging detail data of the vehicle and add a Self-Attention mechanism to capture the timing dependency; the key information and the timing dependency are connected so that the output layer of the neural network model can output the SOH estimation value; A model training module, used to train the estimation model of the lithium battery SOH using the charging data collected by the data collection module to obtain the estimation model of the lithium battery SOH; The model optimization module is used to adjust the model's hyperparameters using a grid search method during model training to better adapt to different data sets and scenarios.

9. An application method of a lithium battery SOH estimation model, characterized in that: Record charging detail data with SOC values ​​greater than 15% at a preset frequency, record the rated parameters of each vehicle, process the charging detail data in a manner of constructing a training set, load the trained lithium battery SOH estimation model as described in any one of claims 1 to 7, and input the processed charging detail data, and the lithium battery SOH estimation model will infer and calculate the characteristic data through the network structure of the model to output a predicted SOH value.

Images