A method and system for screening long-term battery performance

Abstract

This invention proposes a method and system for long-term battery performance screening. The method involves acquiring charge-discharge capacity increment curves for a training set and a test set; extracting training and test sequence fragments within different first voltage attention windows; extracting features from the training and test sequence fragments to obtain training and test sequence feature sets; screening a second voltage attention window that meets a preset first condition; acquiring the training sequence feature set within the second voltage attention window and training a preset classification model to obtain a performance screening model that meets a preset second condition; acquiring the test sequence feature set within the second voltage attention window and inputting it into the performance screening model to perform long-term battery performance screening on test set battery samples. This invention addresses the problem of low accuracy in long-term battery performance screening caused by the single type of input data and the lack of some key information in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of rapid screening of long-term performance of lithium-ion batteries, and in particular to a method and system for screening the long-term performance of batteries. Background Technology

[0002] Lithium-ion batteries are being used in increasingly wider fields. Batteries with poor long-term performance have shorter cycle life and experience accelerated aging during cycling. Furthermore, batteries with poor long-term performance often suffer severe loss of active materials and active lithium ions. Accelerated aging not only affects the long-term performance of the battery but also poses safety risks during use. Therefore, in the product development stage, it is crucial to select superior batteries to ensure battery performance and safety and improve the overall efficiency and lifespan of the battery pack. From an overall development perspective, battery screening technology is trending towards intelligence and refinement.

[0003] Currently, traditional battery screening technologies typically rely on actual long-term charge-discharge cycle tests. While this method provides relatively accurate data, it is extremely time-consuming, often requiring months or even years to complete. Specific battery screening technologies include accelerated aging early diagnosis methods that combine machine learning, IC curves, DV curves, and discharge curve features; fast-charging battery pack cell screening methods based on three-stage constant current fast-charging data; and three-level screening methods for cascaded utilization batteries based on the open-circuit voltage, voltage variation, and impedance values ​​at different charge levels. The main problems with existing technologies are the long long-term performance testing cycle, the significant consumption of testing time and resources, and the fact that there is no significant difference in capacity between batteries with long and short cycle lives in the early aging stage. Therefore, early cycle capacity data cannot accurately determine the long-term performance of batteries. In summary, existing technologies suffer from limited input data categories and the lack of some key information, resulting in low accuracy in long-term battery performance screening. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a method and system for screening the long-term performance of batteries. This method fully considers the characteristics of sequence fragment data and voltage attention window intervals that have a critical impact on the long-term performance of batteries, thereby improving the accuracy of screening the long-term performance of batteries.

[0005] To achieve the above objectives, embodiments of the present invention provide a method for screening the long-term performance of a battery, comprising:

[0006] Select training set battery samples and test set battery samples to obtain the charge-discharge capacity increment curves of the training set and the charge-discharge capacity increment curves of the test set;

[0007] Based on the preset first voltage attention window, the charge-discharge capacity increment curve of the training set and the charge-discharge capacity increment curve of the test set, the training sequence fragment data and the test sequence fragment data are extracted within different first voltage attention windows;

[0008] Feature extraction is performed on the training sequence fragment data and the test sequence fragment data to obtain the training sequence feature set and the test sequence feature set;

[0009] The preset classification model is trained based on the training sequence feature set, and a second voltage attention window that meets the preset first condition is selected.

[0010] Obtain the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window, and train the preset classification model to obtain a performance screening model that meets the preset second condition.

[0011] Obtain the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window, and input it into the performance screening model to perform long-term battery performance screening on the test set battery samples.

[0012] This invention proposes a method for screening the long-term performance of batteries. It obtains the charge-discharge capacity increment curves of the training set and the test set corresponding to battery samples from both sets. Then, it combines an attention mechanism to obtain different first voltage attention windows, acquiring training sequence fragment data and test sequence fragment data within each first voltage attention window, as well as training sequence feature sets and test sequence feature sets. Finally, it achieves battery long-term performance screening through model training. By using different first voltage attention windows to extract sequence features from the charge-discharge capacity increment curves of the training set and the test set, the method screens for battery long-term performance, increasing the diversity of data input. This allows the machine model to automatically focus on sequence fragment data features and voltage attention window intervals that have a key impact on battery long-term performance, improving the accuracy of battery long-term performance judgment. Furthermore, by screening a second voltage attention window that meets a preset first condition, the machine learning model can perform long-term performance judgment on the test set battery samples based on the second voltage attention window. By processing the key sequence fragment data features within the second voltage window, the method further improves data utilization and judgment accuracy.

[0013] Furthermore, the step of selecting training set battery samples and test set battery samples, and obtaining the charge-discharge capacity increment curves of the training set and the test set includes:

[0014] Select a number of battery samples and divide the battery samples into training set battery samples and test set battery samples;

[0015] By conducting cyclic testing experiments on training set battery samples and test set battery samples, the charge-discharge capacity increment curves of the training set and the test set were obtained.

[0016] Furthermore, the step of obtaining the charge-discharge capacity increment curves of the training set and the test set by conducting cyclic testing experiments on the training set battery samples and the test set battery samples includes:

[0017] A cyclic testing experiment was conducted on the training set battery samples and the test set battery samples;

[0018] Whenever the cyclic test experiment meets the preset number of cycles, obtain the current training set battery samples and the current test set battery samples corresponding to the preset number of cycles.

[0019] Charge and discharge tests are performed on the current training set battery samples and the current test set battery samples at preset cycle intervals to obtain the current training set charge and discharge curves and the current test set charge and discharge curves.

[0020] When the cyclic testing experiment ends, based on several current training set charge-discharge curves and current test set charge-discharge curves, calculate the incremental charge-discharge capacity curves of the training set and the test set.

[0021] Furthermore, the step of extracting training sequence fragment data and test sequence fragment data within different first voltage attention windows based on a preset first voltage attention window, the charge-discharge capacity increment curve of the training set, and the charge-discharge capacity increment curve of the test set includes:

[0022] Based on a preset first voltage attention window, obtain the first starting voltage and the first voltage interval;

[0023] Based on the first starting voltage and the first voltage interval, sequence extraction is performed on the charge-discharge capacity increment curves of the training set and the charge-discharge capacity increment curves of the test set, respectively, to obtain training sequence fragment data and test sequence fragment data within different first voltage attention windows.

[0024] This invention proposes a method for screening the long-term performance of batteries. By combining an attention mechanism and charge-discharge capacity increment curves, machine learning can learn knowledge strongly correlated with the long-term performance of batteries from a portion of the charge-discharge capacity increment curves by changing the width and position of the voltage attention window. This improves the training accuracy and judgment accuracy of the machine learning model in judging accelerated battery aging. At the same time, by combining the attention window, multiple voltage ranges that enable the machine learning model to accurately judge accelerated aging can be selected from the charge-discharge capacity increment curves. Compared with traditional machine learning methods that extract features from complete capacity increment curves, this method improves data utilization and judgment accuracy.

[0025] Furthermore, the step of extracting features from the training sequence fragment data and the test sequence fragment data to obtain the training sequence feature set and the test sequence feature set includes:

[0026] Extract the maximum, minimum, average, kurtosis, skewness, upper and lower limits of the voltage attention window, and average value of the voltage attention window from the training sequence fragment data and the test sequence fragment data, respectively;

[0027] Based on the maximum, minimum, average, kurtosis, skewness, upper and lower limits of the voltage attention window, and average value of the voltage attention window of the training sequence fragment data and the test sequence fragment data, the training sequence feature set and the test sequence feature set are obtained.

[0028] Furthermore, the training sequence feature set and the test sequence feature set also include: training set battery sample labels and test set battery sample labels;

[0029] The labels for the training set battery samples and the test set battery samples are obtained by acquiring the decay trends of the training set battery samples and the test set battery samples based on the charge-discharge capacity increment curves of the training set and the test set battery samples, and then labeling the training set battery samples and the test set battery samples based on the decay trends of the training set battery samples and the test set battery samples.

[0030] Furthermore, the step of training a preset classification model based on a training sequence feature set and selecting a second voltage attention window that meets a preset first condition includes:

[0031] The training sequence feature set and the training set battery sample label extracted from the training set battery sample are input into the preset classification model for classification training and scoring, and the score value of the training sequence fragment data within different first voltage attention windows is output.

[0032] The scores of training sequence fragments within different first voltage attention windows are sorted, and the first voltage attention window corresponding to the training sequence fragments that meet the preset first condition is selected as the second voltage attention window.

[0033] Further, the step of obtaining the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window, and training the preset classification model to obtain a performance screening model that meets the preset second condition, includes:

[0034] Based on the second voltage attention window, obtain the second starting voltage and the second voltage interval;

[0035] Based on the second starting voltage and the second voltage interval, extract the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window;

[0036] Based on the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window and the battery sample labels of the training set, the preset classification model is trained to obtain a performance screening model that meets the preset second condition.

[0037] Furthermore, the step of obtaining the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window and inputting it into the performance screening model to perform long-term battery performance screening on the test set battery samples includes:

[0038] Based on the second voltage attention window and the test sequence fragment data within the second voltage attention window, extract the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window;

[0039] Based on the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window, the long-term performance of the battery samples in the test set is screened and judged by the performance screening model to obtain the long-term performance judgment results of each battery; wherein, the long-term performance judgment results of each battery include: normal aging batteries and accelerated aging batteries.

[0040] This invention also provides a battery long-term performance screening system, including: an incremental curve acquisition module, a sequence fragment extraction module, a sequence feature set extraction module, a second voltage attention window acquisition module, a performance screening model training module, and a battery long-term performance screening module;

[0041] The incremental curve acquisition module is used to select training set battery samples and test set battery samples to obtain the incremental curves of the charging and discharging capacity of the training set and the charging and discharging capacity of the test set.

[0042] The sequence fragment extraction module is used to extract training sequence fragment data and test sequence fragment data within different first voltage attention windows based on a preset first voltage attention window, the training set charge-discharge capacity increment curve and the test set charge-discharge capacity increment curve;

[0043] The sequence feature set extraction module is used to extract features from the training sequence fragment data and the test sequence fragment data to obtain the training sequence feature set and the test sequence feature set.

[0044] The second voltage attention window acquisition module is used to train a preset classification model based on the training sequence feature set and to select a second voltage attention window that meets the preset first condition.

[0045] The performance screening model training module is used to obtain the training sequence feature set corresponding to the training sequence fragment data in the second voltage attention window, and to train the preset classification model to obtain a performance screening model that meets the preset second condition.

[0046] The battery long-term performance screening module is used to obtain the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window, and input it into the performance screening model to perform battery long-term performance screening on the test set battery samples.

[0047] This invention proposes a battery long-term performance screening system. It acquires the incremental charge-discharge capacity curves of the training set and the test set corresponding to battery samples using an incremental curve acquisition module. Then, it uses a sequence fragment extraction module and a sequence feature set extraction module combined with an attention mechanism to obtain different first voltage attention windows. Within these first voltage attention windows, it acquires training sequence fragment data and test sequence fragment data, as well as training sequence feature sets and test sequence feature sets. Finally, a performance screening model training module and a battery long-term performance screening module achieve battery long-term performance screening through model training. The system extracts training set charge-discharge capacity curves using different first voltage attention windows. The sequence characteristics of the capacity increment curve and the charge / discharge capacity increment curve of the test set are used to screen the long-term performance of the battery, which increases the diversity of data input. This allows the machine model to automatically focus on the sequence segment data features and voltage attention window intervals that have a key impact on the long-term performance of the battery, thus improving the accuracy of the long-term performance judgment. In addition, the second voltage attention window acquisition module filters the second voltage attention window that meets the preset first condition, enabling the machine learning model to make long-term performance judgments on the battery samples in the test set based on the second voltage attention window. By judging and processing the key sequence segment data features within the second voltage window, the utilization rate of data and the accuracy of judgment are further improved. Attached Figure Description

[0048] Figure 1 A flowchart illustrating the steps of a battery long-term performance screening method according to a certain embodiment of the present invention;

[0049] Figure 2 A schematic diagram of two long-term degradation trends of a lithium battery provided in a battery long-term performance screening method according to a certain embodiment of the present invention;

[0050] Figure 3 A schematic diagram of the normal aging degradation curve of a lithium battery for long-term performance testing, provided in a certain embodiment of the present invention, for a battery long-term performance screening method.

[0051] Figure 4A schematic diagram of the accelerated aging degradation curve of a lithium battery for long-term performance testing, provided in a certain embodiment of the present invention, for a battery long-term performance screening method.

[0052] Figure 5 A schematic diagram of the charge-discharge IC curve of a lithium battery before cycling, provided in a battery long-term performance screening method according to an embodiment of the present invention.

[0053] Figure 6 A schematic diagram of the IC curve of a lithium battery after 100 cycles of charge and discharge, provided for a battery long-term performance screening method according to a certain embodiment of the present invention.

[0054] Figure 7 A schematic diagram of the IC curve of a lithium battery after 200 cycles of charge and discharge, provided for a battery long-term performance screening method according to an embodiment of the present invention.

[0055] Figure 8 This is a schematic diagram of the module structure of a battery long-term performance screening system provided in one embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] With the rapid development of lithium batteries, selecting superior batteries is crucial to ensuring battery performance and safety. Currently, a common battery screening strategy is to conduct long-term performance screening. Traditional long-term battery performance evaluation and screening methods usually rely on actual long-term charge-discharge cycle tests. Although this method can provide relatively accurate data results, the overall process is time-consuming and not conducive to production and R&D. This is because batteries with poor long-term performance have short cycle lives and are prone to severe loss of active materials and active lithium ions. In addition, accelerated aging occurs during cycling, further affecting the safety performance of the battery pack.

[0058] Based on the problems existing in the current technology and the fact that current battery screening methods mainly consider various battery parameters and characteristics, such as voltage, capacity, charge-discharge curves, and impedance, this invention proposes a long-term battery performance screening method. This method addresses the issues of long long-term performance testing cycles, requiring significant testing time and resources, and the inability to accurately determine the long-term performance of batteries based on early cycle capacity data, as well as the lack of significant capacity difference between batteries with long and short cycle lives in the early aging stage. See details... Figure 2 , Figure 2 This invention provides a schematic diagram illustrating two long-term degradation trends of a lithium battery as part of a battery long-term performance screening method according to a certain embodiment of the present invention; for example... Figure 2 As shown, the capacity characteristics of accelerated-aging and normally-aging batteries are very similar in the early stages of cycling.

[0059] In this embodiment of the invention, the charge / discharge IC curve and the charge / discharge capacity increment curve have the same meaning, which will not be repeated below.

[0060] Example 1

[0061] See Figure 1 , Figure 1 This is a schematic flowchart illustrating the steps of a battery long-term performance screening method according to a certain embodiment of the present invention. Figure 1 As shown in the figure, this embodiment of the invention proposes a method for screening the long-term performance of a battery, including steps 101 to 106, each step of which is as follows:

[0062] Step 101: Select training set battery samples and test set battery samples, and obtain the charge-discharge capacity increment curves of the training set and the charge-discharge capacity increment curves of the test set.

[0063] As an example of this embodiment, several battery samples are selected and divided into training set battery samples and test set battery samples. Cyclic testing experiments are conducted on the training set battery samples and the test set battery samples to obtain the charge-discharge capacity increment curves for the training set and the test set. Specifically, cyclic testing experiments are performed on the training set battery samples and the test set battery samples. Whenever the cyclic testing experiment meets a preset cycle interval, the current training set battery sample and the current test set battery sample corresponding to the preset cycle interval are obtained. Charge-discharge tests are performed on the current training set battery samples and the current test set battery samples corresponding to the preset cycle interval to obtain the current training set charge-discharge curve and the current test set charge-discharge curve. When the cyclic testing experiment ends, the charge-discharge capacity increment curves for the training set and the test set are calculated based on the current training set charge-discharge curves and the current test set charge-discharge curves.

[0064] One possible implementation involves selecting battery samples, dividing them into training and testing sets, and conducting battery cycle degradation experiments under different cycling conditions. These different cycling conditions include: performing a 0.05C low-current charge-discharge test on the battery before cycling and every 100 cycles during the cycle test; the ambient temperature for the low-current charge-discharge test is 25°C; and the upper and lower limit voltages for charge-discharge are 3V and 4.35V, respectively. The degradation trend of the tested batteries includes normal aging and accelerated aging. The degradation trend of normally aged batteries is as follows: Figure 3 As shown, Figure 3This invention provides a schematic diagram of the normal aging degradation curve of a lithium battery for long-term performance testing, illustrating a battery long-term performance screening method according to a certain embodiment of the present invention; the degradation trend of accelerated aging batteries is shown below. Figure 4 As shown, Figure 4 This is a schematic diagram of the accelerated aging degradation curve of a lithium battery for long-term performance testing, provided by a battery long-term performance screening method according to a certain embodiment of the present invention. Through battery cycle degradation experiments, the cycle capacity degradation data of the battery and the 0.05C low-current charge-discharge curves and charge-discharge capacity increment curves at different cycle periods are obtained. Specifically, the low-current charge-discharge curves of the battery before cycling, after 100 cycles, and after 200 cycles are extracted, and the charge-discharge capacity increment curves are calculated. The charge-discharge capacity increment curve before cycling is shown below. Figure 5 As shown, Figure 5 This is a schematic diagram of the charge-discharge IC curve of a lithium battery before cycling, provided by a battery long-term performance screening method according to a certain embodiment of the present invention; the capacity increment curve after 100 charge-discharge cycles is shown below. Figure 6 As shown, Figure 6 A schematic diagram of the IC curve of a lithium battery after 100 charge-discharge cycles, provided in a certain embodiment of the present invention, for a battery long-term performance screening method; the capacity increment curve after 200 charge-discharge cycles is shown below. Figure 7 As shown, Figure 7 A schematic diagram of the IC curve of a lithium battery after 200 charge-discharge cycles, provided for a battery long-term performance screening method according to a certain embodiment of the present invention; wherein, the capacity increment calculation formula at different voltage positions is as follows:

[0065]

[0066] In the formula, dQ / dV is the capacity increment, and Q V1 and Q V2 These represent the capacities at voltages V1 and V2, respectively.

[0067] Step 102: Based on the preset first voltage attention window, the charge-discharge capacity increment curve of the training set and the charge-discharge capacity increment curve of the test set, extract the training sequence fragment data and the test sequence fragment data within different first voltage attention windows;

[0068] As an example of this embodiment, a first starting voltage and a first voltage interval are obtained based on a preset first voltage attention window; based on the first starting voltage and the first voltage interval, sequence extraction is performed on the charge-discharge capacity increment curves of the training set and the charge-discharge capacity increment curves of the test set, respectively, to obtain training sequence fragment data and test sequence fragment data within different first voltage attention windows.

[0069] One specific implementation method involves setting an initial voltage attention window (V0, V0+V). intervalThe initial voltage V0 ∈ (3.05V-4.05V), and the voltage interval V interval Starting with 0.1V, extract sequence fragments of the charge-discharge capacity increment curves of battery samples from the training and test sets within the attention window, and change the initial voltage V0 and the voltage attention window width V. interval Extract sequence fragment data of charge-discharge capacity increment curves within different voltage attention windows {(V0, dQ / dV0), (V1, dQ / dV1), ..., (V n dQ / dV n )}.

[0070] Step 103: Extract features from the training sequence fragment data and the test sequence fragment data to obtain the training sequence feature set and the test sequence feature set;

[0071] As an example of this embodiment, the maximum, minimum, average, kurtosis, skewness, upper and lower limits of the voltage attention window, and average value of the voltage attention window are extracted from the training sequence fragment data and the test sequence fragment data, respectively. Based on the maximum, minimum, average, kurtosis, skewness, upper and lower limits of the voltage attention window, and average value of the voltage attention window, a training sequence feature set and a test sequence feature set are obtained. The training sequence feature set and the test sequence feature set also include: training set battery sample labels and test set battery sample labels; the training set battery sample labels and the test set battery sample labels are obtained by obtaining the decay trend of the training set battery samples and the test set battery samples based on the charge-discharge capacity increment curve of the training set and the charge-discharge capacity increment curve of the test set, and by labeling the training set battery samples and the test set battery samples based on the decay trend of the training set battery samples and the test set battery samples.

[0072] One possible implementation involves extracting the maximum, minimum, average, kurtosis, skewness, upper and lower limits of the voltage attention window, and average value of the charge / discharge capacity increment curve sequence data within different voltage attention windows as feature parameters to construct a training sequence feature set and a test sequence feature set. Then, the normally aged and accelerated aged batteries in the training and test sets are labeled respectively. Normally aged batteries are those that do not experience a capacity drop during the cycle degradation process, while accelerated aged batteries are those that experience a capacity drop during the cycle degradation process.

[0073] Step 104: Train the preset classification model based on the training sequence feature set, and select the second voltage attention window that meets the preset first condition;

[0074] As an example of this embodiment, the training sequence feature set extracted from the training set battery samples and the training set battery sample labels are input into a preset classification model for classification training and scoring, and the score values ​​of the training sequence fragment data in different first voltage attention windows are output; the score values ​​of the training sequence fragment data in different first voltage attention windows are sorted, and the first voltage attention window corresponding to the training sequence fragment data that meets the preset first condition is selected as the second voltage attention window.

[0075] One possible implementation involves training a decision tree classification model or another machine learning classification model using the sequence fragment feature set of the charge / discharge capacity increment curve extracted from the training set and the battery sample labels of the training set. The other machine learning classification model can be a support vector machine classification model, a neural network classification model, or a logistic regression classification model, etc. The classification accuracy of each training set within the decision tree classification model or other machine learning model is output as a score value for the training sequence fragment data within different first voltage attention windows. The formula for calculating the classification accuracy is as follows:

[0076]

[0077] In the formula, N true N is the number of correctly identified battery samples. all This represents the total number of battery samples.

[0078] Then, by sorting the score values ​​of each sequence segment feature set data, the first voltage attention window corresponding to the training sequence segment data that meets the preset first condition is selected as the second voltage attention window, which is equivalent to the optimal voltage attention window (V). 0,best V 0,best +V interval The first preset condition is the maximum score value of the feature set data for each sequence segment; for a more detailed explanation, please refer to Table 1:

[0079] Table 1. Results of Battery Long-Term Performance Screening

[0080] Loop count Lower voltage limit Voltage limit Training set accuracy Before the loop 3.35 3.45 100％ Before the loop 3.85 3.95 100％ 100 3.25 3.35 100％ 100 3.35 3.45 100％ 100 3.65 3.75 100％ 100 3.95 4.05 100％ 200 3.15 3.25 100％

[0081] The voltage range in which the decision tree classification model in the training set achieves 100% accuracy is taken as the optimal voltage attention range (second voltage attention window) for the charge-discharge IC curve. As shown in Table 1, the optimal voltage attention range (second voltage attention window) before the cycle is (3.35V, 3.45V) and (3.85V, 3.95V), and the optimal voltage attention window is different before the cycle, after 100 cycles, and after 200 cycles.

[0082] Step 105: Obtain the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window, and train the preset classification model to obtain a performance screening model that meets the preset second condition.

[0083] As an example of this embodiment, based on the second voltage attention window, a second starting voltage and a second voltage interval are obtained; based on the second starting voltage and the second voltage interval, a training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window is extracted; based on the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window and the battery sample labels of the training set, a preset classification model is trained to obtain a performance screening model that meets the preset second condition.

[0084] One specific implementation method is based on a second voltage attention window (V). 0,best V 0,best +V interval That is, the second starting voltage V of the optimal voltage attention window. 0,best The second voltage interval is the same as the first voltage interval, both being 0.1V. The training sequence feature set and training set battery sample labels corresponding to the training sequence fragment data within the optimal voltage attention window are extracted, and the decision tree classification model is trained to obtain multiple trained decision tree classification models. The performance selection model is obtained based on a preset second condition, wherein the preset second condition is: the optimal decision tree classification model obtained by training within the optimal voltage attention window.

[0085] Step 106: Obtain the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window, and input it into the performance screening model to perform long-term battery performance screening on the test set battery samples.

[0086] As an example of this embodiment, based on the second voltage attention window and the test sequence fragment data within the second voltage attention window, a test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window is extracted; based on the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window, the long-term performance of the test set battery samples is screened and judged through a performance screening model to obtain the long-term performance judgment results of each battery; wherein, the long-term performance judgment results of each battery include: normal aging batteries and accelerated aging batteries.

[0087] One specific implementation method involves obtaining a second voltage attention window (V). 0,best V 0,best +V intervalThe test sequence fragment data of the test set is used, and a corresponding test sequence feature set is constructed based on the test sequence fragment data. The test sequence feature set is input into the performance screening model, and the performance screening model outputs the long-term performance judgment result of the battery to determine whether the battery sample in the test set is a normal aging battery or an accelerated aging battery. Based on the charge and discharge IC curve sequence fragments extracted from the optimal voltage attention window before cycling, after 100 cycles, and after 200 cycles, the decision tree classification model can achieve 100% accuracy on both the training set and the test set, as shown in Table 2.

[0088] Table 2. Battery Long-Term Performance Screening Results

[0089] Loop count Lower voltage limit Voltage limit Training set accuracy Test set accuracy Before the loop 3.35 3.45 100％ 100％ Before the loop 3.85 3.95 100％ 100％ 100 3.25 3.35 100％ 100％ 100 3.35 3.45 100％ 100％ 100 3.65 3.75 100％ 100％ 100 3.95 4.05 100％ 100％ 200 3.15 3.25 100％ 100％

[0090] This invention proposes a method for screening the long-term performance of batteries. It obtains the charge-discharge capacity increment curves of the training set and the test set corresponding to battery samples from both sets. Then, it combines an attention mechanism to obtain different first voltage attention windows, acquiring training sequence fragment data and test sequence fragment data within each first voltage attention window, as well as training sequence feature sets and test sequence feature sets. Finally, it achieves battery long-term performance screening through model training. By using different first voltage attention windows to extract sequence features from the charge-discharge capacity increment curves of the training set and the test set, the method screens for battery long-term performance, increasing the diversity of data input. This allows the machine model to automatically focus on sequence fragment data features and voltage attention window intervals that have a key impact on battery long-term performance, improving the accuracy of battery long-term performance judgment. Furthermore, by screening a second voltage attention window that meets a preset first condition, the machine learning model can perform long-term performance judgment on the test set battery samples based on the second voltage attention window. By processing the key sequence fragment data features within the second voltage window, the method further improves data utilization and judgment accuracy.

[0091] Example 2

[0092] See Figure 8 , Figure 8 This is a schematic diagram of the module structure of a battery long-term performance screening system provided in one embodiment of the present invention. Figure 8 As shown in the figure, this embodiment of the invention proposes a battery long-term performance screening system, including: an incremental curve acquisition module, a sequence fragment extraction module, a sequence feature set extraction module, a second voltage attention window acquisition module, a performance screening model training module, and a battery long-term performance screening module;

[0093] The incremental curve acquisition module is used to select training set battery samples and test set battery samples to obtain the incremental curves of the charging and discharging capacity of the training set and the charging and discharging capacity of the test set.

[0094] The sequence fragment extraction module is used to extract training sequence fragment data and test sequence fragment data within different first voltage attention windows based on a preset first voltage attention window, the training set charge-discharge capacity increment curve and the test set charge-discharge capacity increment curve;

[0095] The sequence feature set extraction module is used to extract features from the training sequence fragment data and the test sequence fragment data to obtain the training sequence feature set and the test sequence feature set.

[0096] The second voltage attention window acquisition module is used to train a preset classification model based on the training sequence feature set and to select a second voltage attention window that meets the preset first condition.

[0097] The performance screening model training module is used to obtain the training sequence feature set corresponding to the training sequence fragment data in the second voltage attention window, and to train the preset classification model to obtain a performance screening model that meets the preset second condition.

[0098] The battery long-term performance screening module is used to obtain the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window, and input it into the performance screening model to perform battery long-term performance screening on the test set battery samples.

[0099] This invention proposes a battery long-term performance screening system. It acquires the incremental charge-discharge capacity curves of the training set and the test set corresponding to battery samples using an incremental curve acquisition module. Then, it uses a sequence fragment extraction module and a sequence feature set extraction module combined with an attention mechanism to obtain different first voltage attention windows. Within these first voltage attention windows, it acquires training sequence fragment data and test sequence fragment data, as well as training sequence feature sets and test sequence feature sets. Finally, a performance screening model training module and a battery long-term performance screening module achieve battery long-term performance screening through model training. The system extracts training set charge-discharge capacity curves using different first voltage attention windows. The sequence characteristics of the capacity increment curve and the charge / discharge capacity increment curve of the test set are used to screen the long-term performance of the battery, which increases the diversity of data input. This allows the machine model to automatically focus on the sequence segment data features and voltage attention window intervals that have a key impact on the long-term performance of the battery, thus improving the accuracy of the long-term performance judgment. In addition, the second voltage attention window acquisition module filters the second voltage attention window that meets the preset first condition, enabling the machine learning model to make long-term performance judgments on the battery samples in the test set based on the second voltage attention window. By judging and processing the key sequence segment data features within the second voltage window, the utilization rate of data and the accuracy of judgment are further improved.

[0100] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0101] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0102] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

Claims

1. A method for screening the long-term performance of a battery, characterized in that, include: Select training set battery samples and test set battery samples to obtain the charge-discharge capacity increment curves of the training set and the charge-discharge capacity increment curves of the test set; Based on the preset first voltage attention window, the charge-discharge capacity increment curve of the training set and the charge-discharge capacity increment curve of the test set, the training sequence fragment data and the test sequence fragment data are extracted within different first voltage attention windows; Feature extraction is performed on the training sequence fragment data and the test sequence fragment data to obtain a training sequence feature set and a test sequence feature set. The training sequence feature set and the test sequence feature set also include: training set battery sample labels and test set battery sample labels. The training set battery sample labels and the test set battery sample labels are obtained by acquiring the decay trend of the training set battery samples and the test set battery samples based on the charge and discharge capacity increment curves of the training set and the test set battery samples, and then labeling the training set battery samples and the test set battery samples based on the decay trend of the training set battery samples and the test set battery samples. Training a preset classification model based on a training sequence feature set, and selecting a second voltage attention window that meets a preset first condition, includes: inputting the training sequence feature set extracted from the battery samples in the training set and the battery sample labels in the training set into the preset classification model for classification training and scoring, and outputting the score values ​​of the training sequence fragment data in different first voltage attention windows; sorting the score values ​​of the training sequence fragment data in different first voltage attention windows, and selecting the first voltage attention window corresponding to the training sequence fragment data that meets the preset first condition as the second voltage attention window; The process involves: acquiring the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window, and training a preset classification model to obtain a performance screening model that meets a preset second condition. This includes: acquiring a second starting voltage and a second voltage interval based on the second voltage attention window; extracting the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window based on the second starting voltage and the second voltage interval; and training a preset classification model based on the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window and the battery sample labels of the training set to obtain a performance screening model that meets a preset second condition. Obtain the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window, and input it into the performance screening model to perform long-term battery performance screening on the test set battery samples.

2. The battery long-term performance screening method as described in claim 1, characterized in that, The step of selecting training set battery samples and test set battery samples, and obtaining the charge-discharge capacity increment curves of the training set and the test set includes: Select a number of battery samples and divide the battery samples into training set battery samples and test set battery samples; By conducting cyclic testing experiments on training set battery samples and test set battery samples, the charge-discharge capacity increment curves of the training set and the test set were obtained.

3. The battery long-term performance screening method as described in claim 2, characterized in that, The step of obtaining the charge-discharge capacity increment curves of the training set and the test set by conducting cyclic testing experiments on the training set and test set battery samples includes: A cyclic testing experiment was conducted on the training set battery samples and the test set battery samples; Whenever the cyclic test experiment meets the preset number of cycles, obtain the current training set battery samples and the current test set battery samples corresponding to the preset number of cycles. Charge and discharge tests are performed on the current training set battery samples and the current test set battery samples at preset cycle intervals to obtain the current training set charge and discharge curves and the current test set charge and discharge curves. When the cyclic testing experiment ends, based on several current training set charge-discharge curves and current test set charge-discharge curves, calculate the incremental charge-discharge capacity curves of the training set and the test set.

4. The battery long-term performance screening method as described in claim 1, characterized in that, The step of extracting training sequence fragment data and test sequence fragment data within different first voltage attention windows based on a preset first voltage attention window, the charge-discharge capacity increment curve of the training set, and the charge-discharge capacity increment curve of the test set includes: Based on a preset first voltage attention window, obtain the first starting voltage and the first voltage interval; Based on the first starting voltage and the first voltage interval, sequence extraction is performed on the charge-discharge capacity increment curves of the training set and the charge-discharge capacity increment curves of the test set, respectively, to obtain training sequence fragment data and test sequence fragment data within different first voltage attention windows.

5. The battery long-term performance screening method as described in claim 1, characterized in that, The step of extracting features from the training sequence fragment data and the test sequence fragment data to obtain the training sequence feature set and the test sequence feature set includes: Extract the maximum, minimum, average, kurtosis, skewness, upper and lower limits of the voltage attention window, and average value of the voltage attention window from the training sequence fragment data and the test sequence fragment data, respectively; Based on the maximum, minimum, average, kurtosis, skewness, upper and lower limits of the voltage attention window, and average value of the voltage attention window of the training sequence fragment data and the test sequence fragment data, the training sequence feature set and the test sequence feature set are obtained.

6. The battery long-term performance screening method as described in claim 1, characterized in that, The step of obtaining the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window and inputting it into the performance screening model to perform long-term battery performance screening on the test set battery samples includes: Based on the second voltage attention window and the test sequence fragment data within the second voltage attention window, extract the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window; Based on the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window, the long-term performance of the battery samples in the test set is screened and judged by the performance screening model to obtain the long-term performance judgment results of each battery; wherein, the long-term performance judgment results of each battery include: normal aging batteries and accelerated aging batteries.

7. A battery long-term performance screening system, characterized in that, include: The module includes an incremental curve acquisition module, a sequence fragment extraction module, a sequence feature set extraction module, a second voltage attention window acquisition module, a performance screening model training module, and a battery long-term performance screening module. The incremental curve acquisition module is used to select training set battery samples and test set battery samples to obtain the incremental curves of the charging and discharging capacity of the training set and the charging and discharging capacity of the test set. The sequence fragment extraction module is used to extract training sequence fragment data and test sequence fragment data within different first voltage attention windows based on a preset first voltage attention window, the training set charge-discharge capacity increment curve and the test set charge-discharge capacity increment curve; The sequence feature set extraction module is used to extract features from the training sequence fragment data and the test sequence fragment data to obtain a training sequence feature set and a test sequence feature set. The training sequence feature set and the test sequence feature set further include: training set battery sample labels and test set battery sample labels. The training set battery sample labels and the test set battery sample labels are obtained by acquiring the decay trend of the training set battery samples and the test set battery samples based on the charge and discharge capacity increment curve of the training set and the charge and discharge capacity increment curve of the test set, and then labeling the training set battery samples and the test set battery samples based on the decay trend of the training set battery samples and the test set battery samples. The second voltage attention window acquisition module is used to train a preset classification model based on a training sequence feature set and to select a second voltage attention window that meets a preset first condition. The module includes: inputting the training sequence feature set extracted from the battery samples in the training set and the battery sample labels in the training set into the preset classification model for classification training and scoring, and outputting the score values ​​of the training sequence fragment data in different first voltage attention windows; sorting the score values ​​of the training sequence fragment data in different first voltage attention windows, and selecting the first voltage attention window corresponding to the training sequence fragment data that meets the preset first condition as the second voltage attention window. The performance screening model training module is used to obtain the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window, and to train a preset classification model to obtain a performance screening model that meets a preset second condition. This includes: obtaining a second starting voltage and a second voltage interval based on the second voltage attention window; extracting the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window based on the second starting voltage and the second voltage interval; and training a preset classification model based on the training sequence feature set corresponding to the training sequence fragment data within the second voltage attention window and the battery sample labels of the training set to obtain a performance screening model that meets the preset second condition. The battery long-term performance screening module is used to obtain the test sequence feature set corresponding to the test sequence fragment data within the second voltage attention window, and input it into the performance screening model to perform battery long-term performance screening on the test set battery samples.

Images