A lithium-ion battery classification prediction method

Abstract

The present invention relates to the field of battery technology and discloses a lithium-ion battery classification prediction method, comprising the steps of data collection, label processing, feature extraction, data cleaning, feature screening, and SVM modeling. The method can classify batteries into three categories and achieve a high accuracy rate using a small amount of cycle data, thereby improving classification efficiency while having a good classification effect. The method has great application value for battery quality inspection and control after battery manufacturing and capacity formation.

Description

Technical Field

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

[0002] At present, lithium-ion batteries are widely used in new energy electric vehicles, 3C portable electronic devices, and terminal energy storage devices due to their advantages such as high energy density, long cycle life, recyclability, and cleanliness and environmental protection. As the market develops more and more fiercely, new requirements are put forward for the performance of lithium-ion batteries, such as requiring lithium-ion batteries to have a high cycle life, high energy density, and excellent safety characteristics. Among them, the life of lithium-ion batteries has always been the focus of attention of manufacturers and terminal products.

[0003] The life prediction of lithium-ion batteries is generally achieved using empirical models, data-driven methods, or a combination of the two. When predicting the remaining useful life (RUL), a threshold is set and the number of cycles based on the threshold is obtained. The result returned is a numerical variable. However, if the battery can be classified and evaluated to determine whether the battery life belongs to one of long cycle life, medium cycle life, and short cycle life, such work can be of greater value in battery quality inspection and control.

[0004] Although battery life metrics can more accurately classify batteries, lithium-ion cycle prediction generally requires a training set of 50 cycles or more, which is inefficient. Therefore, designing an efficient and accurate battery classification prediction method can be of great value in the battery manufacturing process. Summary of the Invention

[0005] The purpose of the present invention is to provide a lithium-ion battery classification prediction method with high efficiency and accurate results.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A lithium-ion battery classification prediction method, comprising:

[0008] Step S1: formulate a battery charge and discharge protocol, then perform a cycle test until the battery reaches a set capacity threshold, and obtain data from the battery recording layer and the cycle layer;

[0009] Step S2: After the cycle test is completed, a classification standard is set according to the cycle life of the battery, and a classification label is marked. The classification label includes short cycle life, medium cycle life, and long cycle life;

[0010] Step S3, using the aggregation method to extract the features of the recording layer data of the first N cycles of discharge, 3≤N≤10, and forming QV, dQ / dV-V, dV / dQ-Q, Q in each cycle cycle-cycle0 and V cycle-cycle0 Function variable curve, and obtain the characteristic value of each curve;

[0011] Step S4: perform data cleaning on the eigenvalues ​​obtained in step S3, delete the features whose feature columns are all 0, use the SMOT algorithm to average the samples, then split the test set and the training set, and finally standardize the eigenvalues ​​of the training set and the test set;

[0012] Step S5, using RFE to screen features, and then standardizing the screened features;

[0013] Step S6: Use the SVW algorithm to perform classification modeling. After the modeling is completed, the model is evaluated based on indicators such as the confusion matrix to verify the classification accuracy.

[0014] Optionally, in step S1:

[0015] The battery includes several commercial lithium iron phosphate / graphite cells, which are cycled at different charge rates until the capacity decays to 70%-90% of the rated capacity;

[0016] The data of the battery recording layer and the cycle layer include the temperature, current, charge and discharge capacity measurement data within the cycle, as well as the capacity, internal resistance and charging time measurement data of each cycle.

[0017] Optionally, in step S2:

[0018] When the cycle life of the battery is less than X weeks, it is defined as short cycle life;

[0019] When the cycle life of the battery is between X weeks and Y weeks, it is defined as medium cycle life;

[0020] When the cycle life of a battery is greater than Y weeks, it is defined as long cycle life.

[0021] Optionally, in step 2, 300≤X≤700, 1000≤Y≤1500.

[0022] Optionally, in step 3, N=5.

[0023] Optionally, in step 3, the characteristic values ​​include:

[0024] Minimum, maximum, median, mean, quartile, std, var, AUC, skewness, and kurtosis.

[0025] Optionally, in step 4, the splitting of the test set and the training set specifically includes:

[0026] Randomly split all the classification label samples, use 70%-80% of them as training set and 20%-30% as test set; or, use 80% of them as training set and 20% as test set.

[0027] Optionally, in step S5:

[0028] The classifier is svm.SVC;

[0029] The classifier uses the kernel function kernel as the linear kernel function linear;

[0030] The penalty coefficient is 0.01 to 100;

[0031] The multi-classification method is decision_function_shape = 'ovo';

[0032] The weight is class_weight = 'balanced';

[0033] The number of eigenvalues ​​was chosen to be 5-50.

[0034] Optionally, step S6 specifically includes:

[0035] After using the SVM algorithm for classification modeling, the parameters in the SVM are adjusted using grid optimization;

[0036] The kernel function uses Gaussian kernel function BRF;

[0037] The penalty coefficient C is 0.01 to 100;

[0038] gamma is 0.01-100;

[0039] Multi-classification method decision_function_shape = 'ovo';

[0040] weight class_weight = 'balanced';

[0041] After tuning the parameters and optimizing them, the model is tested using the test set, and the model is evaluated based on indicators such as the confusion matrix.

[0042] Optionally, in step S6:

[0043] After the modeling is completed, the model can also be evaluated by ROC to verify the classification accuracy.

[0044] The beneficial effects of the present invention are that it can classify batteries into three categories, and a high accuracy rate can be achieved using the first five cycle data. This improves classification efficiency while achieving excellent classification results, and has great application value in battery quality inspection and control after battery manufacturing and capacity division. In addition, by using the SMOT algorithm to average samples for sample imbalance, aggregating the first few cycle data of the recording layer to obtain multiple eigenvalues, and then using RFE to screen the optimal eigenvalues, and finally using SVM grid optimization to obtain the optimal model, a complete modeling system has been formed. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings and the accompanying drawings, which constitute part of this application, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. Among them:

[0046] Figures 1(a)-(e) are schematic diagrams of characteristic function variable curves in the lithium-ion battery classification prediction method provided by an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of an eigenvalue matrix in a lithium-ion battery classification prediction method provided by an embodiment of the present invention;

[0048] Figure 3 A schematic diagram of a confusion matrix in a lithium-ion battery classification prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments. Each example is provided by way of explanation of the present invention and is not intended to limit the present invention. Indeed, it will be apparent to those skilled in the art that modifications and variations may be made in the present invention without departing from the scope or spirit of the present invention. For example, a feature shown or described as part of one embodiment may be used in another embodiment to produce yet another embodiment. Therefore, it is intended that the present invention encompasses such modifications and variations as come within the scope of the appended claims and their equivalents.

[0050] In the description of the present invention, the terms "longitudinal", "transverse", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and do not require that the present invention must be constructed and operated in a specific orientation. Therefore, they should not be understood as limitations on the present invention. The terms "connected", "connected", and "set" used in the present invention should be understood in a broad sense. For example, they can be fixed connections or detachable connections; they can be directly connected or indirectly connected through intermediate components. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0051] An embodiment of the present invention provides a lithium-ion battery classification prediction method, comprising:

[0052] Step S1: formulate a battery charge and discharge protocol, then perform a cycle test until the battery reaches a set capacity threshold, and obtain data from the battery recording layer and the cycle layer;

[0053] Step S2: After the cycle test is completed, a classification standard is set according to the cycle life of the battery, and a classification label is marked. The classification label includes short cycle life, medium cycle life, and long cycle life;

[0054] Step S3, using the aggregation method to extract the features of the recording layer data of the first N cycles of discharge, 3≤N≤10, and forming QV, dQ / dV-V, dV / dQ-Q, Q in each cycle cycle-cycle0 and V cycle-cycle0 Function variable curve, and obtain the characteristic value of each curve;

[0055] For example, in actual operation, N=5 may be selected for feature extraction. It should be noted that this embodiment is merely an example, and in actual recording, an appropriate number of cycles may be recorded according to actual conditions.

[0056] Step S4: perform data cleaning on the eigenvalues ​​obtained in step S3, delete the features whose feature columns are all 0, use the SMOT algorithm to average the samples, then split the test set and the training set, and finally standardize the eigenvalues ​​of the training set and the test set;

[0057] Step S5, using RFE to screen features, and then standardizing the screened features;

[0058] Step S6: Use the SVW algorithm to perform classification modeling. After the modeling is completed, the model is evaluated based on indicators such as the confusion matrix to verify the classification accuracy.

[0059] Figures 1(a)-(e) are schematic diagrams of characteristic function variable curves in the lithium-ion battery classification prediction method provided by an embodiment of the present invention. Referring to Figures 1(a)-(e), in this embodiment, data of the battery recording layer and the cycle layer are first obtained to divide the battery cycle life into three categories: short cycle life, medium cycle life, and long cycle life; then, an aggregation method is used to extract features from the recording layer data of the first five cycle discharges and form a function variable curve, as shown in Figures 1(a)-(e); then, characteristic value data cleaning and sample averaging are performed, and the test set and training set are split and standardized; then, RFE is used to screen features and standardize the screened features; finally, the SVW algorithm is used to perform classification modeling and verify the classification accuracy.

[0060] This method uses the SMOT algorithm to average samples for the sample imbalance problem, then aggregates the first 5 cycle data of the recording layer to obtain multiple eigenvalues, then uses RFE to screen out the optimal eigenvalues, and finally uses SVM set grid optimization to obtain the best model, forming a complete modeling idea system; and a high accuracy rate can be obtained with a small amount of cycle data, which improves the classification efficiency and has a good classification effect. It has great application value for battery quality inspection and control after battery manufacturing and capacity division.

[0061] As an optional embodiment, in step S1:

[0062] The battery, comprising several commercial lithium iron phosphate / graphite batteries, was cycled at various charge rates until the capacity decayed to 70%-90% of the rated capacity. For example, during the cycle test, the battery capacity could be decayed to 80% of the rated capacity. It should be noted that the 80% in this embodiment is merely an example; in actual cycle testing, the capacity decay threshold can be adjusted to meet the specific requirements, as long as the battery capacity decays to 70%-90% of the rated capacity. Data from the battery recording layer and the cycling layer includes temperature, current, and charge / discharge capacity measurements during the cycle, as well as capacity, internal resistance, and charge time measurements for each cycle.

[0063] As an optional embodiment, in step S2:

[0064] When the cycle life of the battery is less than X weeks, it is defined as short cycle life;

[0065] When the cycle life of the battery is between X weeks and Y weeks, it is defined as medium cycle life;

[0066] When the cycle life of a battery is greater than Y weeks, it is defined as long cycle life.

[0067] Specifically, for example:

[0068] When the cycle life of the battery is less than 500 cycles, it is defined as short cycle life;

[0069] When the cycle life of the battery is between 500 and 1100 cycles, it is defined as medium cycle life;

[0070] When the cycle life of a battery is greater than 1100 cycles, it is defined as long cycle life.

[0071] It should be noted that this embodiment is merely an example. In actual operation, appropriate X and Y values ​​may be selected according to actual conditions to define the cycle life.

[0072] Specifically, as an optional embodiment, in step 3, the characteristic value includes:

[0073] Minimum, maximum, median, mean, quartile, std, var, AUC, skewness, and kurtosis.

[0074] Among them, the quartile value includes the 25% value and the 75% value. Thus, taking N=5 as an example, since the first 5 cycles are recorded in total, and five function variable curves are recorded in each cycle, and each function variable curve obtains 11 eigenvalues, step S3 can obtain a total of 275 eigenvalues.

[0075] Specifically, as an optional embodiment, in step 4, the splitting of the test set and the training set specifically includes:

[0076] Randomly split all the classification label samples, use 70%-80% of them as training set and 20%-30% as test set.

[0077] Specifically, for example, 70% of all classified label samples can be used as the training set and 30% as the test set; or 80% of all classified label samples can be used as the training set and 20% as the test set. Again, this embodiment is only an example. When splitting the data, the ratio of the training set to the test set can be determined according to the specific situation. It is sufficient to meet the requirement that the training set accounts for 70%-80% and the test set accounts for 20%-30%. There is no specific limitation.

[0078] Figure 2 Schematic diagram of the eigenvalue matrix in the lithium-ion battery classification prediction method provided by the embodiment of the present invention, reference Figure 2 Taking N=5 as an example, since step S3 can obtain a total of 275 eigenvalues, the features with all 0s in the feature column are deleted, and 253 eigenvalues ​​are obtained. Due to the imbalance of the number of samples, the SMOT algorithm is used to process the samples and the reverse average, so that the number of samples of the three labels is 1:1:1, totaling 270 samples.

[0079] Specifically, as an optional embodiment, in step S5:

[0080] The classifier is svm.SVC;

[0081] The classifier uses the kernel function kernel as the linear kernel function linear;

[0082] The penalty coefficient is 0.01 to 100;

[0083] The multi-classification method is decision_function_shape = 'ovo';

[0084] The weight is class_weight = 'balanced';

[0085] The number of eigenvalues ​​was chosen to be 5-50.

[0086] Specifically, as an optional embodiment, step S6 specifically includes:

[0087] After using the SVM algorithm for classification modeling, the parameters in the SVM are adjusted using grid optimization;

[0088] The kernel function uses Gaussian kernel function BRF;

[0089] The penalty coefficient C is 0.01 to 100;

[0090] gamma is 0.01-100;

[0091] Multi-classification method decision_function_shape = 'ovo';

[0092] weight class_weight = 'balanced';

[0093] After tuning the parameters and optimizing them, the model is tested using the test set, and the model is evaluated based on indicators such as the confusion matrix. Figure 3 A schematic diagram of a confusion matrix in the lithium-ion battery classification prediction method provided by an embodiment of the present invention, such as Figure 3 As shown, specifically, in the feature recursive elimination stage, the penalty coefficient C=0.8 and the number of features are 20. In the remodeling stage, the penalty coefficient C=0.05 and gamaa=10 are selected, and the accuracy of the model is 95.06%.

[0094] As an optional embodiment, in step S6:

[0095] After the modeling is completed, the model can also be evaluated by ROC to verify the classification accuracy.

[0096] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A lithium-ion battery classification prediction method, characterized in that: The following steps are involved: Step S1: formulate a battery charge and discharge protocol, then perform a cycle test until the battery reaches a set capacity threshold, and obtain data from the battery recording layer and the cycle layer; Step S2, after the cycle test is completed, setting classification standards according to the cycle life of the battery and marking classification labels, the classification labels including short cycle life, medium cycle life and long cycle life; Step S3, using the aggregation method to extract the features of the recording layer data of the first N cycles of discharge, 3≤N≤10, and forming QV, dQ / dV-V, dV / dQ-Q, Q in each cycle cycle-cycle0 and V cycle-cycle0 Function variable curve, and obtain the characteristic value of each curve; Step S4: perform data cleaning on the eigenvalues ​​obtained in step S3, delete the features whose feature columns are all 0, use the artificial synthetic data SMOT algorithm to average the samples, then split the test set and the training set, and finally standardize the eigenvalues ​​of the training set and the test set; Step S5, using recursive feature elimination (RFE) to filter features, and then standardizing the filtered features; Step S6: Use the support vector machine (SVM) algorithm to perform classification modeling. After the modeling is completed, the model is evaluated based on the confusion matrix index to verify the classification accuracy.

2. The lithium-ion battery classification prediction method according to claim 1, characterized in that: In step S1: The battery includes several commercial lithium iron phosphate / graphite batteries, which are cycled at different charge rates until the capacity decays to 70%-90% of the rated capacity; The data of the battery recording layer and the cycle layer include the temperature, current, charge and discharge capacity measurement data within the cycle, as well as the capacity, internal resistance and charging time measurement data of each cycle.

3. The lithium-ion battery classification prediction method according to claim 1, characterized in that: In step S2: When the cycle life of the battery is less than X weeks, it is defined as short cycle life; When the cycle life of the battery is between X weeks and Y weeks, it is defined as medium cycle life; When the cycle life of a battery is greater than Y weeks, it is defined as long cycle life.

4. The lithium-ion battery classification prediction method according to claim 3, characterized in that: In step 2, 300≤X≤700, 1000≤Y≤1500.

5. The lithium-ion battery classification prediction method according to claim 1, characterized in that: In step 3, N=5.

6. The lithium-ion battery classification prediction method according to claim 1, characterized in that: In step 3, the characteristic values ​​include: Minimum, maximum, median, mean, quartile, std, var, AUC, skewness, and kurtosis.

7. The lithium-ion battery classification prediction method according to claim 1, characterized in that: In step 4, the splitting of the test set and the training set specifically includes: Randomly split all the classification label samples, use 70%-80% of them as training set and 20%-30% as test set.

8. The lithium-ion battery classification prediction method according to claim 1, characterized in that: In step S5: The classifier is svm.SVC; The classifier uses the kernel function kernel as the linear kernel function linear; The penalty coefficient is 0.01~100; The multi-classification method is decision_function_shape='ovo'; The weight is class_weight='balanced'; The number of eigenvalues ​​was chosen to be 5-50.

9. The lithium-ion battery classification prediction method according to claim 1, characterized in that: The step S6 specifically includes: After using the SVM algorithm for classification modeling, the parameters in the SVM are adjusted using grid optimization; The kernel function uses Gaussian kernel function RBF; The penalty coefficient C is 0.01~100; gamma is 0.01-100; Multi-classification method decision_function_shape='ovo'; weight class_weight = 'balanced'; After tuning the parameters and optimizing them, the model is tested using the test set, and the model is evaluated based on the confusion matrix indicators.

10. The lithium-ion battery classification prediction method according to any one of claims 1 to 9, characterized in that: In step S6: After the modeling is completed, the model can also be evaluated using the receiver operating characteristic curve (ROC) to verify the classification accuracy.

Images