Lithium battery capacity grading capacity prediction method

Abstract

The invention discloses a lithium battery capacity grading capacity prediction method, which comprises the following steps of: acquiring data of each process in a battery cell production process, and performing correlation analysis on the process data before a formation process to obtain data field characteristics related to the battery capacity grading capacity; performing feature extraction on the cell formation process data to obtain time sequence features; combining the time sequence features with the data field features to obtain overall features; inputting the overall characteristics into a battery capacity grading capacity prediction model to obtain a predicted capacity value; according to the method, the hidden characteristics related to the capacity grading capacity before the formation process of the battery are extracted and combined with the process data of the formation process to carry out capacity prediction, so that the precision of battery capacity prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium ion batteries, and in particular to a method for predicting the capacity of a lithium battery. Background Art

[0002] At present, when the battery manufacturing industry actually produces lithium batteries, it first performs formation to activate the battery cells; then it performs capacity division and performs multiple charge and discharge cycles to calibrate the capacity of the lithium battery and grade the products. The capacity division of the battery cells is tested according to the standard charge and discharge process, but the complete charge and discharge process takes a lot of time, consumes a lot of electricity, occupies a large factory area, and requires a large number of testing equipment. At present, the leading lithium battery companies all hope to be able to accurately and effectively predict the capacity division of lithium batteries in order to achieve the purpose of eliminating the capacity division process or reducing the capacity division equipment.

[0003] At present, most lithium battery companies obtain the capacity of battery cells by direct measurement. The battery cells are fully charged, left to stand for a period of time, and then discharged. The ampere-hour integration method is used to obtain the capacity of the battery cells. This method requires a long period of charging, discharging and standing, which increases the time cost of the battery cells. When the number of battery cells is large, it will bring huge electricity costs.

[0004] There are also some methods for battery capacity prediction that mainly extract features from battery cell formation data. Although the data in the formation stage can reflect some performance characteristics of the battery, it mainly reflects the electrochemical behavior of the battery after the formation process. Some implicit factors in the battery manufacturing process before formation may not be fully reflected in the formation data. Relying solely on formation data for prediction may result in a certain loss of accuracy, especially in different production batches or under different process conditions. The subtle differences in battery production before formation may cause certain fluctuations in the battery capacity performance. Relying solely on formation data may not be able to fully capture these differences. In addition, there are some methods that use deep learning and machine learning algorithms such as LSTM or Light-GBM to predict battery capacity, but the performance of these algorithms depends largely on their hyperparameter settings, and the optimization of hyperparameters needs to be considered.

[0005] In the related art, the patent application document with publication number CN118917176A proposes to construct a lithium-ion battery capacity prediction model by acquiring production data before the capacity division process, without relying on the production data of the capacity division process, so that the proposed lithium-ion battery capacity prediction model can completely replace the capacity division process, and then by adopting an improved multi-objective particle swarm algorithm to select excellent feature combinations, the prediction accuracy of the lithium-ion battery capacity prediction model is improved. Summary of the invention

[0006] The technical problem to be solved by the present invention is how to improve the prediction accuracy of lithium-ion battery capacity.

[0007] The present invention solves the above technical problems by the following technical means:

[0008] A lithium battery capacity prediction method is proposed, the method comprising:

[0009] Obtain the data of each process of the battery cell production process, and perform correlation analysis on the process data before the formation process to obtain the data field characteristics related to the battery capacity;

[0010] Extract features from the battery cell formation process data to obtain time series features;

[0011] Combining the time series feature with the data field feature to obtain an overall feature;

[0012] The overall characteristics are input into a battery capacity prediction model to obtain a predicted capacity value.

[0013] Furthermore, the data of each process of the battery cell production process includes: slurry mixing process data, coating process data, coil cutting process data, baking process data, liquid injection process data and formation process data;

[0014] The slurry mixing process data includes temperature, solid content, viscosity and fineness related to the battery core;

[0015] The coating process data includes surface density and flatness related to the battery cell;

[0016] The reeling process data includes the positive and negative electrode blanks, diaphragm misalignment and the distance from the diaphragm to the positive and negative electrodes related to the battery cell;

[0017] The baking process data includes the moisture content of the battery cell after baking;

[0018] The injection process data includes the weight of the primary injection battery cell before and after injection, the injection temperature and the injection amount, and the weight of the secondary injection battery cell before and after injection, the injection amount and the replenishment amount;

[0019] The formation process data includes the voltage, current, cell temperature, charging capacity and storage location temperature corresponding to each step of the formation process.

[0020] Furthermore, after obtaining the data of each process of the battery cell production process, the method further includes:

[0021] Calculating statistical parameters of the coating process data and the coil cutting process data respectively;

[0022] Calculate the statistical parameters corresponding to each dimension of each step of the formation process.

[0023] Furthermore, the correlation analysis is performed on the process data before the formation process to obtain data field features related to the battery capacity, including:

[0024] Calculate the Pearson correlation coefficient between the process data before the formation process and the battery capacity;

[0025] The calculated Pearson correlation coefficients are compared with the set thresholds respectively, and the data fields corresponding to the Pearson correlation coefficients greater than the set thresholds are used as data field features related to the battery capacity.

[0026] Furthermore, the feature extraction of the battery cell formation process data to obtain the time series feature includes:

[0027] The LSTM network is used to mine the time series features of the battery cell formation process data to obtain the time series features of the formation process data.

[0028] Furthermore, before using the LSTM network to mine the time series features of the battery cell formation process data, the method also includes:

[0029] The LSTM network is trained, and the GWO optimization algorithm is used to optimize the hyperparameters of the LSTM network during the training process.

[0030] Furthermore, the battery capacity prediction model is trained using an XGBRegressor model.

[0031] Furthermore, the training process of the XGBRegressor model includes:

[0032] Collect historical data of each process in the historical production process of battery cells;

[0033] Conduct correlation analysis on the historical data of each process before the formation process to obtain the historical data field characteristics related to the battery capacity;

[0034] Extract features from the historical data of the chemical process to obtain historical time series features;

[0035] The historical data field features and the historical time series features are combined to obtain the historical overall features, and the historical overall features are used as the input of the XGBRegressor model, and the corresponding historical capacity labels are used as the output of the XGBRegressor model to train the XGBRegressor model.

[0036] Furthermore, after combining the historical data field features and the historical time series features to obtain the historical overall features, the method further includes:

[0037] Normalizing the overall historical features to obtain normalized features;

[0038] Standardizing the historical capacity labels to obtain standardized labels;

[0039] Accordingly, the historical overall features are used as the input of the XGBRegressor model, and the corresponding historical capacity labels are used as the output of the XGBRegressor model to train the XGBRegressor model, including:

[0040] The normalized features are used as input of the XGBRegressor model, and the standardized labels are used as output of the XGBRegressor model to train the XGBRegressor model.

[0041] Furthermore, when training the XGBRegressor model, the method further includes:

[0042] The GWO optimization algorithm is used to optimize the hyperparameters of the XGBRegressor model.

[0043] The advantages of the present invention are:

[0044] (1) The present invention fully considers some implicit factors in the battery manufacturing process before formation, performs correlation analysis on the process data between the formation processes, obtains data field features related to the battery capacity, and combines them with the formation data to calculate the battery capacity, thereby improving the accuracy of the battery capacity value prediction.

[0045] (2) The present invention uses the Grey Wolf Optimization (GWO) algorithm to optimize the hyperparameters of the LSTM network and the XGBRegressor model during the model training process, thereby ensuring the accuracy of the model training and the accuracy of the battery capacity prediction using the model.

[0046] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flow chart of a lithium battery capacity prediction method according to an embodiment of the present invention;

[0048] Figure 2 It is a schematic diagram of the structure of an LSTM network in one embodiment of the present invention;

[0049] Figure 3 It is a schematic diagram of the structure of a shallow neural network model in one embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting the capacity of a lithium battery by capacity, the method comprising the following steps:

[0052] S10, obtaining the data of each process before and during the formation of the battery cell production process, and performing a correlation analysis on the process data before the formation process to obtain data field features related to the battery capacity;

[0053] S20, extracting features from the cell formation process data to obtain time series features;

[0054] S30, combining the time series feature with the data field feature to obtain an overall feature;

[0055] Specifically, the time series features extracted from the formation process are electrical performance characteristics, and the data field features are the attribute features of the battery cells during the production process. The dimensions of the attribute features and the formation time series features are different. For example, each battery cell attribute feature has n dimensions. attribute =[x1, x2, …, x n-1 , x n ], time series feature m dimension, Feat sequence =[y1,y2,…,y m-1 ,y m ], attribute features are longer than time series features, and features are fused to form a new feature Feat new for:

[0056]

[0057] The overall feature is Feat attribute 、Feat sequence and Feat new The splicing forms the overall feature Feat total :

[0058] Feat total =concat([Featattribute ,Feat sequence , Feat new ])

[0059] It should be noted that, in this embodiment, the cell property characteristics and the timing characteristics are integrated to enrich the high-order characteristics of the cell, increase the nonlinear prediction capability of the model, and improve the prediction accuracy of the model.

[0060] S40, inputting the overall characteristics into a battery capacity prediction model to obtain a predicted capacity value.

[0061] At present, lithium battery companies use the full capacity division mode to predict the capacity of battery cells, which has the disadvantages of high energy consumption, increased equipment costs, and the need for a larger site for capacity division experiments. In view of the previous behavior of only considering the extraction of features from the battery formation stage when predicting the capacity of lithium batteries, data from each process before the battery cell formation was collected, and the correlation coefficient was calculated to extract the potential features related to the capacity division before the battery formation process; by making full use of the relevant parameters and process data of each process before the capacity division process of the battery cell, deep learning and machine learning methods are used to mine relevant features and perform capacity prediction, which shortens the production cost of the battery cell, reduces energy consumption and equipment costs, and brings great benefits to lithium battery companies.

[0062] It should be noted that the process data before the formation described in this embodiment include slurry mixing process data, coating process data, coil cutting process data, baking process data, liquid injection process data and formation process data, which are closely related to the capacity of the battery cell. By introducing process prior analysis knowledge, the characteristic factors affecting the capacity can be explored faster, the model convergence speed can be accelerated, and the surface density, battery cell bare weight, battery cell dry weight, and the statistical features of the mined data and the LSTM mining time series features can be integrated to increase the original special diagnosis and deep feature representation of the model, and increase the nonlinear ability of the model prediction. The final mape index of the solution proposed in the patent application with publication number CN118917176A in the related technology is about 13%, while the mape effect of this embodiment in the 20,000 sample test set is 4.8‰, which guarantees the quality service of the battery cell with higher accuracy.

[0063] As a further preferred technical solution, the data of each process of the battery cell production process is obtained, including: slurry mixing process data, coating process data, coil cutting process data, baking process data, liquid injection process data and formation process data;

[0064] The slurry mixing process data includes temperature, solid content, viscosity and fineness related to the battery core;

[0065] The coating process data includes surface density and flatness related to the battery cell;

[0066] The reeling process data includes the positive and negative electrode blanks, diaphragm misalignment and the distance from the diaphragm to the positive and negative electrodes related to the battery cell;

[0067] The baking process data includes the moisture content of the battery cell after baking;

[0068] The injection process data includes the weight of the primary injection battery cell before and after injection, the injection temperature and the injection amount, and the weight of the secondary injection battery cell before and after injection, the injection amount and the replenishment amount;

[0069] The formation process data includes the voltage, current, cell temperature, charging capacity and storage location temperature corresponding to each step of the formation process.

[0070] It should be noted that, in the present embodiment, data of each process in the production process of the battery cell can be collected by correspondingly arranged sensors.

[0071] As a further preferred technical solution, after obtaining the data of each process of the battery cell production process, the method further includes:

[0072] Calculating statistical parameters of the coating process data and the coil cutting process data respectively;

[0073] Calculate the statistical parameters corresponding to each dimension of each step of the formation process.

[0074] It should be noted that since the coating process data include data such as surface density and flatness related to the battery cell, which are long sequence data, this embodiment calculates its statistical parameters; since the slitting process data include data such as the positive and negative electrode blank space, diaphragm misalignment, and the distance from the diaphragm to the positive and negative electrodes related to the battery cell, which are long sequence data, their statistical parameters are calculated.

[0075] Since the data dimensions of the formation process have five characteristic dimensions, namely, voltage, current, cell temperature, charging capacity and storage temperature, and the formation process has multiple steps, first, low-rate constant current charging to a specified voltage, standing for a period of time, repeated multiple times, and then high-rate constant current charging to the cell voltage, and then standing for a period of time, the formation process data is divided into multiple step data by step to obtain the statistical parameters of each dimension of each step. This embodiment analyzes these statistics from an electrochemical perspective, and these statistics have certain physical meanings, such as the characterization of charging capacity and polarization capacity. Mining the physical meaning characteristics from an electrochemical perspective can improve the model's ability to learn capacity influencing factors and improve the model's accuracy in predicting capacity.

[0076] Specifically, the statistical parameters described in this embodiment include minimum value, maximum value, range, mean value, variance and the like.

[0077] As a further preferred technical solution, in step S10, correlation analysis is performed on the process data before the formation process to obtain data field features related to the battery capacity, including the following steps:

[0078] S11, calculating the Pearson correlation coefficient between the data of each process before the formation process and the battery capacity;

[0079] It should be noted that the data of different processes of the battery cell are generated in different time periods. The capacity of the battery cell is generated in the last process. Through the cell code traceability, the relevant data of each process in the front stage of the battery cell can be obtained, organized into structured data and then the Pearson correlation coefficient is calculated.

[0080] Specifically, the Pearson correlation coefficient is used to measure the linear correlation between two variables, and its calculation formula is as follows:

[0081]

[0082] Where r is the Pearson correlation coefficient, the value range of r is between -1 and 1, r = 1 indicates a perfect positive correlation, r = -1 indicates a perfect negative correlation, and r = 0 indicates no linear correlation; x is each observed value of variable X; ∑xy is the sum of the products of each observed value of variable X and Y; ∑y is the sum of all observed values ​​of variable Y; ∑y 2 is the sum of squares of each observed value of variable Y. In the above Pearson correlation coefficient, variable Y is represented as the capacity of the battery cell, and variable X represents the corresponding characteristic variables, such as surface density, bare cell weight and dry cell weight.

[0083] S12. Compare the calculated Pearson correlation coefficients with the set thresholds respectively, and use the data fields corresponding to the Pearson correlation coefficients greater than the set thresholds as data field features related to the battery capacity.

[0084] It should be noted that the threshold value is set to 0.2 in this embodiment. If the absolute value of the correlation is less than 0.2, it means that the correlation between the feature and the cell capacity is low and is not very meaningful as a model input feature. If the absolute value of the correlation is greater than 0.2, these features are retained.

[0085] As a further preferred technical solution, the step S20: extracting features from the cell formation process data to obtain time series features, comprises the following steps:

[0086] The LSTM network is used to mine the time series features of the battery cell formation process data to obtain the time series features of the formation process data.

[0087] It should be noted that if Figure 2As shown in the figure, the core concept of LSTM lies in the cell state and gate structure. The gate structure of LSTM includes forget gate, input gate and output gate. The formulas of each gate and cell state are as follows:

[0088] Forget gate formula: f t =σ(W f ·[h t-1 , x t ]+b f )

[0089] Input gate formula: i t =σ(W i ·[h t-1 , x t ]+b i ); C′ t =tanh(W C ·[h t-1 , x t ]+b C )

[0090] Cell state update formula: C t =f t *C t-1 +i t *C′ t

[0091] Output gate formula: O t =σ(W O ·[h t-1 , x t ]+b O );h t =O t *tanh(C t )

[0092] In the formula, h t-1 represents the hidden state of the model when it is trained to time tl, h t-1 ∈R (N*l*M) , C t ∈R (N*l*M) , N represents the number of hidden layers of the model, M represents the information dimension learned from the d-dimensional data at each moment; W f , W i , W C and W O Respectively represent the forget gate, input gate, cell state update and output gate weight matrix parameters; W f ∈R (4M*d) , W i ∈R (4M*d) , W C ∈R (4M*d) , W O ∈R (4M*d) ; bf , b i , b C and b O They represent the bias of forget gate, input gate, cell state update and output gate, respectively. f ∈R 4M , b i ∈R 4M , b C ∈R 4M , b O ∈R 4M ;σ and tanh represent sigmoid functions, f t represents the output of the forget gate at time t, C t represents the cell state at time t, C t ∈R (N*l*M) , O t Represents the output O of the model at time t t ∈R (b *l*M) .

[0093] Specifically in this embodiment, the input of LSTM is the process data X of the formation process, X = [x1, x2, ..., x L-1 , x L ]; each element x in X i Contains d values, x i =[x i1 , x i2 , ..., x i(d-1) , x id ]; in the above formula, x t represents the data collected during the formation process over a period of time, L represents the length of the data during the formation process, and d represents the data dimension collected during the formation process at each moment; t ∈R (b*l*d) , b represents the number of cells that the model trains at one time, and l represents the length of time it takes to obtain data for each cell during model training. The LSTM model is input into a shallow neural network to obtain the time series features of each cell.

[0094] Specifically, Figure 3 As shown, the LSTM model outputs O t After the reshape operation, the shallow neural network input input is obtained, input∈R(b*(l*M)), and finally the output output is obtained, output∈R(b*1). The output is the discrete feature or time series feature of the entire charging and discharging process in the battery formation stage. H represents the number of neurons in the hidden layer of the shallow neural network.

[0095] As a further preferred solution, before using the LSTM network to mine the time series characteristics of the battery cell formation process data, the method further includes:

[0096] The LSTM network is trained, and the GWO optimization algorithm is used to optimize the hyperparameters of the LSTM network during the training process.

[0097] It should be noted that this embodiment optimizes the hyperparameters of the LSTM network by using the GWO optimization algorithm to ensure the accuracy of LSTM network training.

[0098] As a further preferred technical solution, the battery capacity prediction model is obtained by training with an XGBRegressor model, and the training process of the XGBRegressor model includes the following steps:

[0099] Collect historical data of each process in the historical production process of battery cells;

[0100] Conduct correlation analysis on the historical data of each process before the formation process to obtain the historical data field characteristics related to the battery capacity;

[0101] Extract features from the historical data of the chemical process to obtain historical time series features;

[0102] The historical data field features and the historical time series features are combined to obtain the historical overall features, and the historical overall features are used as the input of the XGBRegressor model, and the corresponding historical capacity labels are used as the output of the XGBRegressor model to train the XGBRegressor model.

[0103] This embodiment optimizes the hyperparameters of the XGBRegressor model by using the GWO optimization algorithm, thereby ensuring the accuracy of XGBRegressor model training and maximizing the performance of the algorithm.

[0104] Specifically, the XGBRegressor model is an application of the XGBoost algorithm for regression prediction. The principles of XGBoost are as follows:

[0105] XGBoost is an additive model consisting of k base models. Assume that the tree model to be trained in the tth iteration is f t (x), then:

[0106]

[0107] in, Represents the prediction result of sample i after the tth iteration; represents the prediction results of the first t-1 trees; ft (x i ) represents the model of the tth tree.

[0108] The prediction accuracy of the model is determined by the deviation and variance of the model. The loss function represents the deviation of the model. If you want to reduce the variance, you need to add a regularization term to the objective function to prevent overfitting. The regularization term is as follows:

[0109]

[0110] Where: represents the model prediction bias, It means summing up the complexity of all t trees and adding it to the objective function as a regularization term to prevent overfitting of the model.

[0111] It should be noted that the weak learners of the Boosting algorithm cannot be iterated in parallel. The most time-consuming process in a single weak learner is the splitting process of the decision tree. XGBoost has made a relatively large parallel optimization for this splitting. For the feature partitioning points of different features, XGBoost selects the maximum gain of the split in parallel in different threads. Each feature of the training is sorted and stored in the memory in a block structure to facilitate subsequent iterations and reduce the amount of calculation.

[0112] As a further preferred technical solution, after combining the historical data field features and the historical time series features to obtain the historical overall features, the method further includes the following steps:

[0113] Normalizing the overall historical features to obtain normalized features;

[0114] Standardizing the historical capacity labels to obtain standardized labels;

[0115] Accordingly, the historical overall features are used as the input of the XGBRegressor model, and the corresponding historical capacity labels are used as the output of the XGBRegressor model to train the XGBRegressor model, including:

[0116] The normalized features are used as input of the XGBRegressor model, and the standardized labels are used as output of the XGBRegressor model to train the XGBRegressor model.

[0117] Specifically, this embodiment performs MinMax normalization on all acquired features and Z-score normalization on Labels to improve the prediction accuracy of the model, where:

[0118] MinMax normalization formula:

[0119]

[0120] Z-score formula:

[0121]

[0122] In the formula, mean(y) represents the mean of y, and std(y) represents the standard deviation of y. The normalized data set is divided into a training set and a test set.

[0123] It should be understood that, in the process of predicting the battery capacity in real time in this embodiment, the overall characteristics may be normalized and then sent to the battery capacity prediction model for capacity prediction.

[0124] As a further preferred technical solution, when training the XGBRegressor model, the method further includes:

[0125] The GWO optimization algorithm is used to optimize the hyperparameters of the LSTM and XGBRegressor models;

[0126] The hyperparameters of GWO optimized XGBRegressor model include hidden layer size hidden_layer_size, dropout rate and learning rate learning_rate;

[0127] The hyperparameters of the GWO optimized XGBRegressor model include the maximum depth of the tree max_depth, the minimum value of the sum of the child node sample weights min_child_weight, the proportion of subsamples of the training model to the entire sample set subsample, the proportion of feature sampling when building the tree colsample_bytree, and the learning rate eta.

[0128] Specifically, the GWO optimized XGBRegressor algorithm is as follows:

[0129] (1) Parameter initialization: The above eight parameters to be optimized are used as the position vector x of the gray wolf individual in the gray wolf pack optimization algorithm to initialize the initialization population X = (X1, X2, …, X N ), and calculate the fitness value corresponding to the initial population individuals through the fitness function calculation formula to perform social level stratification;

[0130] (2) GWO is used to update the parameters of the LSTM and XGBRegressor algorithms to ensure the minimum fitness value. The fitness function is the absolute value of the difference between the predicted capacity value and the actual capacity value. The three wolves with the best fitness are selected and marked as α, β, and δ, and the remaining wolves are marked as ω. In the search space of 8 parameters to be optimized, define the first The location of the solution is Where d<D, Represents the position of the ith solution in d dimensions. It satisfies the following formula:

[0131]

[0132] Where: t represents the current iteration number, To solve the position, is the bracketing step length, where and is defined as follows:

[0133]

[0134]

[0135] Where r1 and r2 are random vectors in [0,1]; the convergence factor a decreases linearly from 2 to 0 with each iteration, and the expression of a is:

[0136] when When , it means that the solutions in the population are relatively dispersed, thus expanding the search range, that is, performing a global search; when When , it means starting to concentrate, thus searching for solutions within a certain range, that is, performing local search. Secondly, in the process of searching for other solutions, the current solution often knows the location of the optimal solution (that is, the optimal location X p ), but in the actual parameter optimization process, the current solution position is unknown.

[0137] (3) The solution in the population is based on the position X of α, β, and δ α , X β , X δ Update the solution position:

[0138]

[0139]

[0140] In the formula, represents the updated solution position vector, ω j (j=α, β, δ) means that the weight coefficient of α, β or δ is 1 / 3.

[0141] This solution uses GWO to optimize the parameters of the two models, LSTM and XGBRegressor, with high optimization efficiency and improved global search capabilities of the two model parameters.

[0142] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0143] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0144] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A lithium battery capacity prediction method, characterized in that: The method comprises: Obtain the data of each process of the battery cell production process, and perform correlation analysis on the process data before the formation process to obtain the data field characteristics related to the battery capacity; Extract features from the battery cell formation process data to obtain time series features; Combining the time series feature with the data field feature to obtain an overall feature; The overall characteristics are input into a battery capacity prediction model to obtain a predicted capacity value.

2. The lithium battery capacity prediction method according to claim 1, characterized in that: The process data of each process of the battery cell production process includes: slurry mixing process data, coating process data, coil cutting process data, baking process data, liquid injection process data and formation process data; The slurry mixing process data includes temperature, solid content, viscosity and fineness related to the battery core; The coating process data includes surface density and flatness related to the battery cell; The reeling process data includes the positive and negative electrode blanks, diaphragm misalignment and the distance from the diaphragm to the positive and negative electrodes related to the battery cell; The baking process data includes the moisture content of the battery cell after baking; The injection process data includes the weight of the primary injection battery cell before and after injection, the injection temperature and the injection amount, and the weight of the secondary injection battery cell before and after injection, the injection amount and the replenishment amount; The formation process data includes the voltage, current, cell temperature, charging capacity and storage location temperature corresponding to each step of the formation process.

3. The lithium battery capacity prediction method according to claim 2, characterized in that: After obtaining the data of each process of the battery cell production process, the method further includes: Calculating statistical parameters of the coating process data and the coil cutting process data respectively; Calculate the statistical parameters corresponding to each dimension of each step of the formation process.

4. The lithium battery capacity prediction method according to claim 1, characterized in that: The correlation analysis of the process data before the formation process is performed to obtain data field features related to the battery capacity, including: Calculate the Pearson correlation coefficient between the process data before the formation process and the battery capacity; The calculated Pearson correlation coefficients are compared with the set thresholds respectively, and the data fields corresponding to the Pearson correlation coefficients greater than the set thresholds are used as data field features related to the battery capacity.

5. The lithium battery capacity prediction method according to claim 1, characterized in that: The feature extraction of the cell formation process data to obtain the time series feature includes: The LSTM network is used to mine the time series features of the battery cell formation process data to obtain the time series features of the formation process data.

6. The lithium battery capacity prediction method according to claim 5, characterized in that: Before using the LSTM network to mine the time series features of the battery cell formation process data, the method further includes: The LSTM network is trained, and the GWO optimization algorithm is used to optimize the hyperparameters of the LSTM network during the training process.

7. The lithium battery capacity prediction method according to claim 1, characterized in that: The battery capacity prediction model is trained using the XGBRegressor model.

8. The lithium battery capacity prediction method according to claim 7, characterized in that: The training process of the XGBRegressor model includes: Collect historical data of each process in the historical production process of battery cells; Conduct correlation analysis on the historical data of each process before the formation process to obtain the historical data field characteristics related to the battery capacity; Extract features from the historical data of the chemical process to obtain historical time series features; The historical data field features and the historical time series features are combined to obtain the historical overall features, and the historical overall features are used as the input of the XGBRegressor model, and the corresponding historical capacity labels are used as the output of the XGBRegressor model to train the XGBRegressor model.

9. The lithium battery capacity prediction method according to claim 8, characterized in that: After combining the historical data field features and the historical time series features to obtain the historical overall features, the method further includes: Normalizing the overall historical features to obtain normalized features; Standardizing the historical capacity labels to obtain standardized labels; Accordingly, the historical overall features are used as the input of the XGBRegressor model, and the corresponding historical capacity labels are used as the output of the XGBRegressor model to train the XGBRegressor model, including: The normalized features are used as input of the XGBRegressor model, and the standardized labels are used as output of the XGBRegressor model to train the XGBRegressor model.

10. The lithium battery capacity prediction method according to claim 8, characterized in that: When training the XGBRegressor model, the method further includes: The GWO optimization algorithm is used to optimize the hyperparameters of the XGBRegressor model.

Images