Capacity grading capacity prediction method of new energy battery

Abstract

The invention belongs to the technical field of new energy battery production, and discloses a capacity grading capacity prediction method for a new energy battery, which comprises the following steps of: 1, acquiring multiple groups of test parameters in a new energy battery generation process and the capacity of the new energy battery corresponding to each group of test parameters from a production record according to a time sequence to form an original data set; step 2, screening out key test parameters from the test parameters, and forming a sample set by the key test parameters and the capacities of the corresponding new energy batteries; step 3, constructing a full-connection network model, and training the full-connection network model by using the sample set to obtain a capacity grading capacity prediction network of the new energy battery; and step 4, collecting key test parameters in real time in the production process of the new energy battery as the input of the capacity grading capacity prediction network of the new energy battery, wherein the capacity grading capacity prediction network of the new energy battery outputs the capacity grading capacity of the new energy battery.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy battery production, and in particular relates to a method for predicting the capacity of a new energy battery. Background Art

[0002] With the rapid popularization of electric vehicles and smart phones, the market size of new energy batteries (lithium-ion batteries) has grown rapidly, and related manufacturers have an increasingly urgent need to increase lithium-ion battery production capacity and quality assurance capabilities. In order to seize market opportunities and quickly increase lithium-ion battery production capacity while ensuring quality, companies are also in urgent need of exploring more efficient production models.

[0003] The manufacturing process of lithium-ion batteries includes slurrying, coating, rolling, slitting, winding, assembly, injection, formation and capacity separation. Among them, capacity separation is the process of calibrating the manufactured batteries by measuring their capacity. It is mainly achieved through a capacity separation cabinet. The batteries are placed in the capacity separation cabinet, and the charge and discharge cycles are carried out according to the set program. Finally, they are sorted according to the capacity parameters. This process is not only costly but also time-consuming. Summary of the invention

[0004] The purpose of the present invention is to provide a method for predicting the capacity of a new energy battery, which can utilize the test data in the production process of the new energy battery to accurately predict the capacity of the new energy battery product, simplify the capacity division link, and improve the production efficiency of the new energy battery.

[0005] The technical solution provided by the present invention is: A method for predicting the capacity of a new energy battery comprises the following steps: Step 1: Acquire multiple groups of test parameters in the production process of new energy batteries and the capacity of the new energy batteries corresponding to each group of test parameters from the production records in chronological order to form an original data set; Each group of test parameters includes: total core weight, formation temperature, formation cut-off voltage, DC internal resistance, first OCV test voltage, second OCV test voltage, first OCV test resistance and second OCV test resistance; Step 2: Filter out key test parameters from the test parameters, and form a sample set with the key test parameters and the capacity of the corresponding new energy battery; Step 3: construct a fully connected network model, and use the sample set to train the fully connected network model to obtain a capacity prediction network for new energy batteries; Step 4: During the production process of new energy batteries, key test parameters are collected in real time as inputs to a capacity prediction network for new energy batteries, and the capacity prediction network for new energy batteries outputs the capacity prediction network for new energy batteries.

[0006] In the step 2, the key test parameters are screened out from the test parameters using the MRMR algorithm, and the obtained key test parameters are: formation temperature, DC internal resistance and the second OCV test voltage.

[0007] Preferably, in the step three, the number of hidden layers and the number of neurons in each hidden layer of the fully connected neural network are optimized by a genetic algorithm, including: Step a, taking the number of hidden layers of the fully connected neural network and the number of neurons in each hidden layer as a parameter combination, and randomly initializing the parameter combination to obtain a parameter set; Among them, each parameter combination corresponds to a fully connected neural network structure; Step b, calculating the error of the fully connected neural network corresponding to each parameter combination, and determining the fitness of the parameter combination; Step c: select the parameter combinations with large fitness and directly enter the next generation according to the set ratio; crossover and mutation are performed on the remaining parameter combinations in the parameter set and then enter the next generation; Repeat step b-step c until the maximum number of iterations is reached; select the fully connected network structure corresponding to the parameter combination with the largest fitness for training, and obtain the capacity prediction network for new energy batteries.

[0008] Preferably, in step c, the calculation formula of fitness is: ; in, , For the The actual values ​​of the output parameters of the group experimental samples, For the The output value of the fully connected network corresponding to the group of experimental samples; is the total number of experimental samples.

[0009] Preferably, in the step three and step four, the following steps are further included: The key test parameters are normalized by the following formulas: ; in, Represents the key parameters, =1,2,3 represent the formation temperature, DC internal resistance and the second OCV test voltage respectively. , , Represents key parameters Measured value, minimum value, maximum value; The key parameter The normalized value of To adjust the parameters, The value range is 0.30~0.80.

[0010] Preferably, in step 1, if a test parameter in a group is missing in the original data set, the missing parameter is supplemented by the average value of the test parameter in the two groups adjacent to the missing parameter.

[0011] Preferably, in the step 1, it also includes: If the values ​​of multiple test parameters in a group of test parameters are higher or lower than 1 / 3 of the background value, the group of data is judged to be abnormal and the group of test data is removed from the original data set.

[0012] Preferably, the background value is calculated by: calculating the average value of each test parameter in the original data set, and taking the average value of each parameter as the background value of the test parameter.

[0013] The beneficial effects of the present invention are: The method for predicting the capacity of new energy batteries provided by the present invention can utilize the test data in the production process of new energy batteries to accurately predict the capacity of new energy battery products, simplify the capacity division process, improve the production efficiency of new energy batteries and reduce production costs. DETAILED DESCRIPTION

[0014] The present invention is described in further detail below so that those skilled in the art can implement it according to the description.

[0015] The conventional lithium battery production process is: assembly → baking → primary injection → formation → high temperature static → secondary injection → SOC adjustment → OCV1 → normal temperature static → OCV2 → charging → capacity division. The parameters that need to be tested (detected) during the production process are: weight detection in the assembly link to obtain the core weight, the first electrolyte injection volume, the weight before the first injection, and the weight after the first injection are detected in the primary injection link; temperature detection and voltage detection are performed in the formation link to obtain the formation temperature and formation cut-off voltage respectively; the second electrolyte injection volume, the weight before the second injection, and the weight after the second injection are detected in the secondary injection link; voltage detection and resistance detection are performed in the OCV link to obtain the first OCV test voltage and the first OCV test resistance; voltage detection and resistance detection are performed in the OCV2 link to obtain the second OCV test voltage and the second OCV test resistance, and a DC internal resistance (DCIR) detection is performed.

[0016] The present invention provides a method for predicting the capacity of a new energy battery by capacity division, which can utilize the test data in the production process of the new energy battery to realize the prediction of the capacity of the new energy battery product by capacity division. The specific implementation process is as follows.

[0017] 1. Obtain multiple groups of test parameters in the production process of new energy batteries and the capacity (differential capacity) of the new energy batteries corresponding to each group of test parameters from the production records in chronological order to form an original data set.

[0018] Since there are many test parameters recorded in the production process, the important test parameters that affect the battery capacity are first selected according to experience, including: total core weight, formation temperature, formation cut-off voltage, DC internal resistance, first OCV test voltage, second OCV test voltage, first OCV test resistance and second OCV test resistance.

[0019] If a test parameter in a group is missing in the original data set, the missing parameter will be supplemented by the average value of the test parameter in the two groups before and after the missing parameter. If the same test parameter is missing in more than two adjacent groups, the background value of the test parameter will be used to supplement the missing test parameter in each group.

[0020] Preferably, if the values ​​of multiple test parameters in a certain set of test parameters are higher or lower than 1 / 3 of the background value, the set of data is judged to be abnormal and deleted from the original data set to ensure the stability of the parameters and improve the accuracy of the subsequent training process.

[0021] The background value is calculated by calculating the average value of each test parameter in the original data set, and taking the average value of each parameter as the background value of the test parameter.

[0022] Second, select key test parameters from the selected test parameters, and form a sample set with the key test parameters and the capacity of the corresponding new energy battery. The sample set is divided into a training set and a test set in a ratio of 8:2.

[0023] As a preferred embodiment, the key test parameters are screened out from the test parameters using the MRMR algorithm, and the obtained key test parameters are: formation temperature, DC internal resistance and the second OCV test voltage.

[0024] The process of selecting key test parameters from the test parameters using the MRMR algorithm is as follows.

[0025] The MRMR algorithm aims to strike a balance between minimizing redundancy and maximizing the relevance of a set of input features, thereby effectively describing the response variable. In the MRMR algorithm, the distribution and mutual information between features and between features and response variables are first calculated, where mutual information is used to measure the relationship between two variables. and The relationship between.

[0026] ; In the formula, is the test parameter; It is the capacity of new energy batteries; Representation variables and The mutual information between Representation variables and The joint probability distribution of and Respectively represent variables and The marginal probability distribution of .

[0027] The calculation formula for the maximum correlation is: ; Among them, S represents A collection of .

[0028] Select n features with the largest mutual trust information with the capacity of the new energy battery to obtain a set of n features .gather The minimum redundancy calculation formula is: .

[0029] Finally, we obtain the set with the maximum correlation minus the minimum redundancy. is a function that represents the maximum correlation ( D ) and minimum redundancy ( R ) difference. It is a comprehensive score, based on which the optimal feature subset is selected.

[0031] 3. Construct a fully connected network model, and use the test set in the sample set to train the fully connected network model to obtain a capacity prediction network for new energy batteries.

[0032] The hidden layer of a neural network is located between the input layer and the output layer, and plays the role of connecting and converting between the input and output. The number of hidden layers is one of the key factors affecting the performance of a neural network. Increasing the number of hidden layers can increase the capacity of the network, improve the ability to learn complex relationships, and thus improve prediction accuracy. However, choosing a larger number of hidden layers is also accompanied by increased computational complexity, reduced training efficiency, and an increased risk of overfitting.

[0033] The number of neurons in the hidden layer has a complex effect on network training. If the network model is too simple, it may affect the correlation between input parameters and output parameters. In addition, if the number of neurons in the hidden layer of a fully connected neural network is too large, it may lead to a decrease in the accuracy of model predictions, non-convergence during training, or overfitting.

[0034] As a preferred embodiment, in this embodiment, the number of hidden layers and the number of neurons in each hidden layer of the fully connected neural network are optimized by a genetic algorithm, and the specific process includes: (1) The number of hidden layers of the fully connected neural network and the number of neurons in each hidden layer are used as a parameter combination, and the parameter combination is randomly initialized to obtain a parameter set. Each parameter combination corresponds to a fully connected neural network structure.

[0035] (2) Calculate the error of the fully connected neural network corresponding to each parameter combination and determine the fitness of the parameter combination.

[0036] (3) Select the parameter combinations with the largest fitness and enter the next generation directly according to the set ratio; after crossover and mutation, the remaining parameter combinations in the parameter set enter the next generation.

[0037] Repeat (2) to (3) until the maximum number of iterations is reached; select the fully connected network structure corresponding to the parameter combination with the largest fitness for training, and obtain the capacity prediction network for new energy batteries.

[0038] The calculation formula of fitness is: ; In the formula, , For the The actual values ​​of the output parameters of the group experimental samples, For the The output value of the fully connected network corresponding to the group of experimental samples; is the total number of experimental samples.

[0039] The number of hidden layers and the number of neurons in each hidden layer can improve the computational efficiency and prediction accuracy of the fully connected neural network model, enabling the model to achieve better performance.

[0040] In one embodiment, through iterative optimization, the fully connected network model has two hidden layers, and the number of nodes in both hidden layers is 4.

[0041] The training set is then used to train the fully connected network model after structural optimization to obtain a capacity prediction network for new energy batteries.

[0042] In this embodiment, the network newly trained by the test set is tested. The probability that the battery samples with a capacity of more than 108Ah are correctly judged as more than 108Ah is 95.38%, and the proportion of battery samples with a capacity of less than 540mAh (0.5%) accounts for 92.14% of the total number of samples. This accuracy proves that the obtained capacity prediction network for new energy batteries can be used to predict the capacity of new energy batteries.

[0043] 4. During the production process of new energy batteries, key test parameters are collected in real time as input to a capacity prediction network, and the capacity prediction network outputs the capacity of the new energy battery.

[0044] Among them, during the training and actual application process, the parameters need to be normalized before entering the key parameters into the model (capacity prediction network of new energy batteries).

[0045] As a preference, the key test parameters are normalized by the following formulas: ; in, Represents the key parameters, =1,2,3 represent the formation temperature, DC internal resistance and the second OCV test voltage respectively. , , Represents key parameters Measured value, minimum value, maximum value; The key parameter The normalized value of To adjust the parameters, The value range of is 0.30 to 0.80. In one embodiment, the value of the adjustment parameter is set to 0.6, and the fully connected neural network is trained to obtain more accurate results.

[0046] During the production process, 5 batches of new energy batteries were sampled, and the capacity prediction network of new energy batteries was used to perform real-time capacity prediction. The proportion of battery samples with an average prediction error of less than 540mAh (0.5%) for new energy batteries with a capacity of 108Ah was 92.52% of the total number of samples, which was slightly higher than the test results of the test set, further proving the reliability of the capacity prediction of new energy batteries provided by the present invention.

[0047] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the embodiments shown and described herein.

Claims

1. A method for predicting the capacity of a new energy battery by capacity, characterized in that: The steps include: Step 1: Acquire multiple groups of test parameters in the production process of new energy batteries and the capacity of the new energy batteries corresponding to each group of test parameters from the production records in chronological order to form an original data set; Each group of test parameters includes: total core weight, formation temperature, formation cut-off voltage, DC internal resistance, first OCV test voltage, second OCV test voltage, first OCV test resistance and second OCV test resistance; Step 2: Filter out key test parameters from the test parameters, and form a sample set with the key test parameters and the capacity of the corresponding new energy battery; Step 3: construct a fully connected network model, and use the sample set to train the fully connected network model to obtain a capacity prediction network for new energy batteries; Step 4: During the production process of new energy batteries, key test parameters are collected in real time as inputs to a capacity prediction network for new energy batteries, and the capacity prediction network for new energy batteries outputs the capacity prediction network for new energy batteries.

2. The method for predicting the capacity of a new energy battery according to claim 1, characterized in that: In the step 2, the key test parameters are screened out from the test parameters using the MRMR algorithm, and the obtained key test parameters are: formation temperature, DC internal resistance and the second OCV test voltage.

3. The method for predicting the capacity of a new energy battery according to claim 2, characterized in that: In the step three, the number of hidden layers and the number of neurons in each hidden layer of the fully connected neural network are optimized by a genetic algorithm, including: Step a, taking the number of hidden layers of the fully connected neural network and the number of neurons in each hidden layer as a parameter combination, and randomly initializing the parameter combination to obtain a parameter set; Among them, each parameter combination corresponds to a fully connected neural network structure; Step b, calculating the error of the fully connected neural network corresponding to each parameter combination, and determining the fitness of the parameter combination; Step c: select the parameter combinations with large fitness and directly enter the next generation according to the set ratio; crossover and mutation are performed on the remaining parameter combinations in the parameter set and then enter the next generation; Repeat step b-step c until the maximum number of iterations is reached; select the fully connected network structure corresponding to the parameter combination with the largest fitness for training, and obtain the capacity prediction network for new energy batteries.

4. The method for predicting the capacity of a new energy battery according to claim 3, characterized in that: In step c, the calculation formula of fitness is: ； in, , For the The actual values ​​of the output parameters of the group experimental samples, For the The output value of the fully connected network corresponding to the group of experimental samples; is the total number of experimental samples.

5. The method for predicting the capacity of a new energy battery according to claim 3 or 4, characterized in that: In the step 3 and step 4, it also includes: The key test parameters are normalized by the following formulas: ； in, Represents the key parameters, =1,2,3 represent the formation temperature, DC internal resistance and the second OCV test voltage respectively. , , Represents key parameters Measured value, minimum value, maximum value; The key parameter The normalized value of To adjust the parameters, The value range is 0.30~0.

80.

6. The method for predicting the capacity of a new energy battery according to claim 5, characterized in that: In the step 1, if a test parameter in a group is missing in the original data set, the missing parameter is supplemented by using the average value of the test parameter in the two groups adjacent to the missing parameter.

7. The method for predicting the capacity of a new energy battery according to claim 6, characterized in that: In the step 1, it also includes: If the values ​​of multiple test parameters in a group of test parameters are higher or lower than 1 / 3 of the background value, the group of test parameters is judged to be abnormal and is removed from the original data set.

8. The method for predicting the capacity of new energy batteries according to claim 7, characterized in that: The background value is calculated by: calculating the average value of each test parameter in the original data set, and taking the average value of each parameter as the background value of the test parameter.

Images