Battery capacity inflection point prediction method and device, equipment and storage medium

Abstract

The invention discloses a battery capacity inflection point prediction method and device, equipment and a storage medium, relates to the technical field of batteries, and discloses a battery capacity inflection point prediction method comprising the following steps: obtaining cycle data of a to-be-predicted battery in a calendar aging test stage; wherein the cycle data comprises a corresponding relation between different storage times collected in the calendar aging test stage and the maximum discharge capacity of the battery to be predicted; and inputting the cycle data into a pre-trained battery capacity inflection point prediction model, and predicting a battery capacity inflection point of the to-be-predicted battery based on the cycle data by the battery capacity inflection point prediction model. The battery capacity inflection point is predicted based on the cycle data by using the prediction model, compared with manual prediction, the accuracy is high, the method is suitable for a scene in which the position of the battery capacity inflection point is rapidly obtained in a battery research and development stage, additional testing is not needed, testing acceleration can be provided for the battery research and development process, and the research and development process is remarkably accelerated.

Description

Technical Field

[0001] The present application relates to the field of battery technology, and in particular to a method, device, equipment and storage medium for predicting a battery capacity inflection point. Background Art

[0002] In recent years, with the rapid development of new energy vehicles and energy storage industries, the performance stability and reliability of new energy batteries have received more and more attention. Among them, battery life is a key issue of concern to consumers, which directly affects the cost of use and customer experience. Ideally, the battery life is infinite, but in reality, the battery will not only experience capacity decay during use, but even capacity diving. Capacity diving means that after the battery capacity decays to a certain extent (it can also be understood as reaching the inflection point of the battery capacity), it suddenly accelerates the decay, and the capacity decays to the end of life in a short period of time, making the battery pack unable to work normally, which has a significant impact on the battery life and performance.

[0003] At present, the capacity drop phenomenon not only seriously affects the user experience, but also causes certain safety hazards. Therefore, how to accurately predict the inflection point of battery capacity is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The main purpose of the present application is to provide a method, device, equipment and storage medium for predicting the inflection point of battery capacity, aiming to solve the problem of how to accurately predict the inflection point of battery capacity.

[0005] To achieve the above objectives, the present application proposes a method for predicting a battery capacity inflection point, the method comprising: Acquire cycle data of the battery to be predicted during the calendar aging test phase; wherein the cycle data includes a correspondence between different storage times collected during the calendar aging test phase and a maximum discharge capacity of the battery to be predicted; The cycle data is input into a pre-trained battery capacity inflection point prediction model, and the battery capacity inflection point prediction model predicts the battery capacity inflection point of the battery to be predicted based on the cycle data.

[0006] In one embodiment, the step of inputting the cycle data into a pre-trained battery capacity inflection point prediction model, and predicting the battery capacity inflection point of the battery to be predicted based on the cycle data by the battery capacity inflection point prediction model, comprises: Inputting the cycle data and the auxiliary prediction information into a pre-trained battery capacity inflection point prediction model, and having the battery capacity inflection point prediction model predict the battery capacity inflection point of the battery to be predicted based on the cycle data and the auxiliary prediction information; The auxiliary prediction information includes at least one of a storage sampling interval, a storage temperature and a battery formula parameter.

[0007] In this embodiment, when the model is used to predict the battery capacity inflection point, the influence of different storage temperatures and storage sampling intervals on batteries with different formulations is fully considered, thereby improving the rationality, reliability and versatility of the battery capacity inflection point prediction.

[0008] In one embodiment, before inputting the cycle data into a pre-trained battery capacity inflection point prediction model, the method further includes: Obtain the maximum discharge capacity of the test battery corresponding to different storage times during the calendar aging test phase; Determine a capacity decay curve of the test battery based on the maximum discharge capacity corresponding to different storage times of the test battery; wherein the capacity decay curve represents the corresponding relationship between different storage times and the battery health state SOH; In the capacity decay curve, determining a target battery capacity inflection point of the test battery; Based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, constructing a training data set; The training data set is used to train a preset candidate model, and the trained model is obtained as the battery capacity inflection point prediction model.

[0009] In this embodiment, a specific implementation method of how to train a battery capacity inflection point prediction model is provided. Specifically, the maximum discharge capacity corresponding to different storage times of the test battery in the calendar aging test phase is first obtained, and then the capacity decay curve of the test battery is determined based on the maximum discharge capacity corresponding to different storage times of the test battery, and the target battery capacity inflection point of the test battery is determined in the capacity decay curve, and then a training data set can be constructed based on the correspondence between the capacity decay curve and the target battery capacity inflection point, and the target battery capacity inflection point is used as the prediction label of the capacity decay curve, and the candidate model is supervisedly trained to obtain the trained model as the battery capacity inflection point prediction model.

[0010] In one embodiment, before constructing a training data set based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, the method further includes: Obtaining storage sampling interval, storage temperature and battery formula parameters of the test battery during the calendar aging test phase; The constructing of a training data set based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point includes: A training data set is constructed based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, storage sampling interval, storage temperature and battery formula parameters.

[0011] In this embodiment, when training the battery capacity inflection point prediction model, the influence of different storage temperatures and storage sampling intervals on batteries with different formulations is fully considered. Specifically, based on the correspondence between the capacity decay curve and the target battery capacity inflection point, the storage sampling interval, the storage temperature and the battery formulation parameters, a training data set is constructed to train the candidate model, thereby improving the rationality, reliability and versatility of the trained model in predicting the battery capacity inflection point, and making the prediction more accurate.

[0012] In one embodiment, determining the capacity decay curve of the test battery based on the maximum discharge capacity corresponding to different storage times of the test battery includes: Calculate the ratio between the maximum discharge capacity corresponding to different storage times and the maximum discharge capacity of the initial storage as the SOH of the test battery at different storage times; Based on the SOH of the test battery at different storage times, determining a first capacity decay curve corresponding to the test battery; wherein the first capacity decay curve represents a corresponding relationship between the SOH and different storage times; Using the box plot corresponding to the first capacity decay curve, filtering the discrete points in the first capacity decay curve to obtain a second capacity decay curve; Based on the second capacity fade curve, a capacity fade curve of the test battery is determined.

[0013] In this embodiment, a specific implementation method for determining the capacity decay curve of the test battery is provided. First, by calculating the ratio between the maximum discharge capacity corresponding to different storage times and the maximum discharge capacity of the initial storage, a first capacity decay curve characterizing the SOH corresponding to different storage times is determined, and then the discrete points in the first capacity decay curve are filtered out using a box plot to obtain a second capacity decay curve, and then based on the second capacity decay curve, the capacity decay curve of the test battery is determined. By filtering the box plot, the influence of discrete points on subsequent model training can be reduced, so as to reduce the interference of noise data on model training, thereby improving the prediction effect of the trained battery capacity inflection point prediction model.

[0014] In one embodiment, the using of the box plot corresponding to the first capacity decay curve to filter discrete points in the first capacity decay curve to obtain the second capacity decay curve includes: Calculate the quartile and upper quartile corresponding to the first capacity decay curve; Based on the quartile and the upper quartile, calculating the interquartile range corresponding to the first capacity decay curve; Determining a first threshold based on the quartile and the interquartile range; Determining a second threshold based on the upper quartile and the interquartile range; In the first capacity decay curve, points where the SOH is less than the first threshold or the SOH is greater than the second threshold are taken as discrete points; The discrete points are filtered out from the first capacity decay curve, and linear interpolation filling is performed at the storage time corresponding to the discrete points to obtain the second capacity decay curve.

[0015] In this embodiment, a specific implementation method for filtering discrete points through a box plot is provided. Specifically, the quartiles and upper quartiles corresponding to the first capacity decay curve are first calculated, and the corresponding interquartile range is calculated, and then the first threshold is calculated based on the quartiles and the interquartile range, and the second threshold is calculated based on the upper quartile and the interquartile range, and the discrete points that are not between the first threshold and the second threshold are filtered out, and then the positions of the filtered discrete points are filled by linear interpolation to ensure the number of points in the capacity decay curve while ensuring the smoothness of the curve, which is helpful for the subsequent effective training of the model.

[0016] In one embodiment, determining the capacity decay curve of the test battery based on the second capacity decay curve includes: Calculating a variance value of SOH in the second capacity decay curve; When the variance value is greater than a preset third threshold, filtering the second capacity decay curve using a one-dimensional Gaussian filter to obtain a capacity decay curve of the test battery; When the variance value is less than or equal to the third threshold, the second capacity decay curve is filtered by using an SG filter to obtain the capacity decay curve of the test battery.

[0017] In this embodiment, after determining the second capacity decay curve, if the variance value of SOH in the second capacity decay curve is large, the second capacity decay curve can be further filtered for smoothing. Specifically, when the variance value is greater than the third threshold, filtering with an SG filter may cause anomalies, so a one-dimensional Gaussian filter is used for filtering. When the variance value is less than or equal to the third threshold, the SG filter is used for filtering. The second capacity decay curve after smoothing is helpful for subsequent effective training of the model.

[0018] In one embodiment, the training data set is constructed based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, the storage sampling interval, the storage temperature and the battery formula parameters, including: Based on the correspondence between the capacity decay curve and the target battery capacity inflection point, the storage sampling interval, the storage temperature and the battery formula parameters, a two-dimensional array corresponding to the test battery is generated; wherein the two-dimensional array includes N groups of SOH corresponding to different storage times, storage sampling intervals, storage temperatures, battery formula parameters and time from the inflection point, the time from the inflection point represents the difference between different storage times and the time corresponding to the target battery capacity inflection point, and N is an integer greater than 2; The time from the inflection point is used as a prediction label, and the N groups of data in the two-dimensional array are slid sequentially according to a preset sliding window length to generate a time series data set as the training data set.

[0019] In this embodiment, after generating a two-dimensional array corresponding to the test battery based on the relevant data of the test battery, a sliding window can be used to slide on each row of the two-dimensional array to generate a time series data set to increase the data volume of the training data set, which helps to improve the efficiency and effectiveness of model training.

[0020] In one embodiment, the candidate model uses a Transformer algorithm; Before using the training data set to train a preset candidate model to obtain the trained model as the battery capacity inflection point prediction model, the method further includes: A function obtained by weighted summing the mean absolute error function and at least one regularization penalty term is used as a loss function for training the candidate model; The regularization penalty term is used to control the range of the battery capacity inflection point predicted by the trained battery capacity inflection point prediction model.

[0021] In this embodiment, the candidate model using the Transformer algorithm has an attention mechanism, and uses time series regression with attention to slide and predict the inflection point of battery capacity, which can capture the potential relationship of battery aging and perform real-time prediction. As the test data increases, it can further help the model improve the effect; in addition, the new loss function helps to improve the prediction accuracy and robustness of the algorithm.

[0022] In one embodiment, the using the training data set to train a preset candidate model to obtain the trained model as the battery capacity inflection point prediction model includes: Using the training data set to train a preset candidate model to obtain a trained model; The mean absolute error function is set as the optimization target of the trained model, and at least one of the learning rate, number of iterations, batch size, dropout value and dropout value of the fully connected layer is set as the optimization parameter of the trained model. The trained model is optimized using the Bayesian algorithm to obtain the battery capacity inflection point prediction model.

[0023] In this embodiment, a Bayesian algorithm is used to optimize the optimization parameters of the trained model to obtain the best model, thereby improving the prediction accuracy of the battery capacity inflection point prediction model.

[0024] In addition, to achieve the above purpose, the present application also proposes a device for predicting a battery capacity inflection point, the device comprising: An acquisition module, used to acquire cycle data of the battery to be predicted during the calendar aging test phase; wherein the cycle data includes a correspondence between different storage times collected during the calendar aging test phase and the maximum discharge capacity of the battery to be predicted; The prediction module is used to input the cycle data into a pre-trained battery capacity inflection point prediction model, and the battery capacity inflection point prediction model predicts the battery capacity inflection point of the battery to be predicted based on the cycle data.

[0025] In addition, to achieve the above-mentioned purpose, the present application also proposes a battery capacity inflection point prediction device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the battery capacity inflection point prediction method as described above.

[0026] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the method for predicting the battery capacity inflection point as described above are implemented.

[0027] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the method for predicting the battery capacity inflection point as described above are implemented.

[0028] One or more technical solutions proposed in this application have at least the following technical effects: 1) Input the cycle data of the battery to be predicted during the calendar aging test phase into a pre-trained battery capacity inflection point prediction model, so that the prediction model can predict the battery capacity inflection point based on the cycle data, which is more accurate than manual prediction.

[0029] 2) By using the historical cycle data of the calendar aging stage, the battery capacity inflection point can be predicted through the model. This is suitable for application scenarios where the battery capacity inflection point position can be quickly obtained during the battery R&D stage. No additional testing is required, and it can provide test acceleration for the battery R&D process, significantly speeding up the R&D process. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0032] Figure 1 This is one of the flow charts of the method for predicting the battery capacity inflection point provided in this application; Figure 2 This is the second flow chart of the method for predicting the inflection point of battery capacity provided by the present application; Figure 3 This is the third flow chart of the method for predicting the inflection point of battery capacity provided by the present application; Figure 4 This is the fourth flow chart of the method for predicting the battery capacity inflection point provided by the present application; Figure 5 This is the fifth flow chart of the method for predicting the inflection point of battery capacity provided by the present application; Figure 6 This is the sixth flow chart of the method for predicting the battery capacity inflection point provided by the present application; Figure 7 This is the seventh flow chart of the method for predicting the inflection point of battery capacity provided by the present application; Figure 8 It is a schematic diagram of a discharge capacity curve of a storage battery in the method for predicting the battery capacity inflection point provided in the present application; Fig. 9 It is a schematic diagram of a capacity decay curve of a storage battery in the method for predicting the battery capacity inflection point provided in the present application; Fig.10 It is a schematic diagram of the process of generating a training data set in the method for predicting the battery capacity inflection point provided in the present application; Fig.11 It is a schematic diagram of a process of generating a data set by using a sliding window in the method for predicting the inflection point of battery capacity provided by the present application; Fig.12It is a schematic diagram of the process of Bayesian optimization of hyperparameters in the method for predicting the battery capacity inflection point provided in this application; Fig.13 It is a schematic diagram of the structure of the device for predicting the inflection point of battery capacity provided by the present application; Fig.14 It is a structural schematic diagram provided by the device for predicting the inflection point of battery capacity in this application.

[0033] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0034] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" in the specification and claims of this application and the above-mentioned figure descriptions and any variations thereof are intended to cover non-exclusive inclusions.

[0036] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0037] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0038] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0039] In the related art, battery aging is divided into calendar aging and cycle aging. Calendar aging of a battery refers to the phenomenon that the performance of a battery gradually decreases during long-term storage without use; cycle aging of a battery refers to the phenomenon that the performance of a battery gradually decreases during repeated charge and discharge cycles.

[0040] It can be seen that the time consumption of calendar aging test is mainly storage rather than charge and discharge test. No charge and discharge test is performed during storage, and there is no lithium ion deintercalation and intercalation process. Many curves required to predict the inflection point of battery capacity in related technologies (such as capacity differential) do not have a complete life cycle in calendar aging test. In addition, the calendar aging test is performed for 1 to 3 cycles after storage to characterize the capacity. Since the storage time may be of variable length, the test interval after each storage may also be of variable length. During the entire storage battery test cycle, the total number of times the capacity needs to be measured is the number of storage times plus the first measurement, and the total number is relatively small. Because each storage requires human intervention, the small number of measurements is objectively unavoidable, and the number of cycles in the cyclic aging test is very large (lithium-ion batteries generally have more than 500 cycles), and the number of cycles is increasing and equally spaced. The technology for predicting the inflection point of battery capacity in the related art requires conducting aging test experiments under different battery capacity influencing parameters and calculating model parameters, which takes a long time and is not suitable for quickly obtaining the inflection point position of the battery in the battery research and development stage.

[0041] In addition, the techniques for predicting the inflection point of battery capacity in the related art do not take into account the impact of the battery storage temperature on the capacity inflection point, nor the impact of different battery formulations on the capacity inflection point. The selection of positive electrode materials directly affects the energy density, cycle stability and safety of the battery; different positive electrode materials will undergo different chemical reactions during the charging and discharging process, thus affecting the life of the battery; the electrolyte plays the role of transmitting ions in the battery, and its type and properties directly affect the cycle performance and life of the battery; different electrolytes have different ion conductivity, chemical stability and thermal stability, thus affecting the charging and discharging efficiency and cycle life of the battery; adjusting the ratio of positive and negative electrode materials, changing the composition and concentration of the electrolyte, etc., can have a positive impact on the life of the battery; in addition, at high temperatures, the activity of the chemical substances inside the battery is enhanced, the collision frequency between molecules increases, and the reaction rate is accelerated, which will accelerate the consumption and corrosion of the substances inside the battery, leading to the aging and damage of the battery plates, thereby shortening the battery life. In addition, high temperatures will also cause the evaporation of chemical substances inside the battery and the expansion of the electrolyte liquid, further damaging the structure and performance of the battery.

[0042] In response to the above technical problems, the present application provides a method for predicting the inflection point of battery capacity, aiming to find a method for accurately predicting the position of the inflection point of battery capacity with different formulations and storage temperatures without conducting additional testing experiments.

[0043] It should be noted that the execution subject of the embodiment of the present application may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a battery capacity inflection point prediction device, etc. The following takes the battery capacity inflection point prediction device as an example to illustrate the embodiment of the present application and the following embodiments.

[0044] The present application embodiment provides a method for predicting the inflection point of battery capacity. Figure 1 , Figure 1 This is one of the flow charts of the method for predicting the battery capacity inflection point provided in this application.

[0045] In this embodiment, the method for predicting the battery capacity inflection point includes steps S101-S102: Step S101, obtaining cycle data of a battery to be predicted during a calendar aging test phase; The cycle data includes the corresponding relationship between different storage times collected in the calendar aging test phase and the maximum discharge capacity of the battery to be predicted.

[0046] It should be noted that in the calendar aging test stage, the above maximum discharge capacity is extracted from the cycle test performed by the operator to calibrate the capacity after each storage. The data of this cycle test is only 3 cycles of charge and discharge data at most (charge and discharge process: rest → discharge → charge → rest → discharge). Usually, the maximum discharge capacity of the second cycle can be taken, and this value is considered to be the more accurate actual available capacity of the storage battery. Each storage is accompanied by a cycle test, that is, a point data can be obtained.

[0047] Step S102, inputting the cycle data into a pre-trained battery capacity inflection point prediction model, and having the battery capacity inflection point prediction model predict the battery capacity inflection point of the battery to be predicted based on the cycle data.

[0048] Specifically, the existing cycle data of the battery to be predicted in the calendar aging test stage is used and input into a trained battery capacity inflection point prediction model. The battery capacity inflection point prediction model can first determine the SOH (State of Health) corresponding to different storage times based on the correspondence between different storage times and the maximum discharge capacity in the cycle data, and then predict the battery capacity inflection point of the battery to be predicted.

[0049] Among them, SOH is used to measure the degree of battery degradation and remaining service life, usually expressed as a percentage.

[0050] In some embodiments, the battery capacity inflection point may be output as a specific time or as a time from the inflection point (Day to Knee Point: DTK). The specific time corresponding to the battery capacity inflection point may also be derived through the DTK.

[0051] For example, if the battery capacity inflection point prediction model outputs a DTK of 45 days, it is believed that the battery capacity will reach the inflection point 45 days later. The current date plus 45 days is considered to be the specific date corresponding to the battery capacity inflection point.

[0052] In the method for predicting the battery capacity inflection point provided in the embodiment of the present application, the cycle data of the battery to be predicted in the calendar aging test stage is input into a pre-trained battery capacity inflection point prediction model, so that the prediction model predicts the battery capacity inflection point based on the cycle data, which has higher accuracy than manual prediction; in addition, the present application uses the cycle data of the calendar aging stage acquired historically to predict the battery capacity inflection point through the model, which is suitable for application scenarios in the battery research and development stage to quickly obtain the battery capacity inflection point position, without the need for additional testing, and can provide test acceleration for the battery research and development process, significantly speeding up the research and development process.

[0053] The following describes a specific implementation method for predicting the battery capacity inflection point by referring to the auxiliary prediction information in combination with a feasible implementation method: In one possible implementation, Figure 1 Based on the corresponding embodiments, refer to Figure 2 , Figure 2 This is the second flow chart of the method for predicting the battery capacity inflection point provided by the present application. The above step S102 includes the following steps S1021: Step S1021, inputting the cycle data and the auxiliary prediction information into a pre-trained battery capacity inflection point prediction model, and the battery capacity inflection point prediction model predicts the battery capacity inflection point of the battery to be predicted based on the cycle data and the auxiliary prediction information; The auxiliary prediction information includes at least one of a storage sampling interval, a storage temperature and a battery formula parameter.

[0054] It should be noted that the storage sampling interval refers to the difference in days between two storage days, for example, 9 days; the storage temperature is, for example, 60°C; the battery formula parameters can be multiple, for example, including the positive electrode physical and chemical content, electrolyte formula and binder content, etc., which are only used as examples in this application and are not limited to this.

[0055] In the embodiment of the present application, when using the model to predict the inflection point of battery capacity, the influence of different storage temperatures and storage sampling intervals on batteries with different formulations is fully considered, thereby improving the rationality, reliability and versatility of the prediction of the inflection point of battery capacity.

[0056] The following describes how to train a battery capacity inflection point prediction model in combination with feasible implementation methods: In one possible implementation, refer to Figure 3 , Figure 3 This is the third flow chart of the method for predicting the battery capacity inflection point provided by the present application. Before the above step S102, the following steps S301 to S305 are also included: Step S301, obtaining the maximum discharge capacity of the test battery corresponding to different storage times in the calendar aging test phase.

[0057] In some embodiments, the second cycle data may be selected, and the maximum discharge capacity corresponding to the data at different storage times may be obtained, for example, 0.019 Ah.

[0058] Step S302, determining a capacity decay curve of the test battery based on the maximum discharge capacity corresponding to different storage times of the test battery; The capacity decay curve represents the corresponding relationship between different storage times and SOH.

[0059] It should be noted that the capacity decay curve can also be called a SOH curve.

[0060] Step S303: determining a target battery capacity inflection point of the test battery in the capacity decay curve.

[0061] Step S304: constructing a training data set based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point.

[0062] In some embodiments, for the SOH curve, multiple people can independently mark the potential inflection point position, and obtain the final inflection point position through multi-person voting, and make the inflection point positions of parallel samples of the same formula as close as possible to ensure that the labels have a certain consistency. This is because different people have different definitions of the inflection point position of the same SOH curve, which is a human error. Finally, the DTK value of each point is calculated as the predicted label of the regression algorithm.

[0063] Step S305, using the training data set to train a preset candidate model, and obtaining the trained model as the battery capacity inflection point prediction model.

[0064] In the embodiment of the present application, a specific implementation method of how to train a battery capacity inflection point prediction model is provided. Specifically, the maximum discharge capacity corresponding to different storage times of the test battery in the calendar aging test phase is first obtained, and then the capacity decay curve of the test battery is determined based on the maximum discharge capacity corresponding to different storage times of the test battery, and the target battery capacity inflection point of the test battery is determined in the capacity decay curve, and then a training data set can be constructed based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, and the target battery capacity inflection point is used as the prediction label of the capacity decay curve, and the candidate model is supervisedly trained to obtain the trained model as the battery capacity inflection point prediction model.

[0065] The following describes how to construct a training data set in conjunction with feasible implementation methods: In one possible implementation, Figure 3 Based on the corresponding embodiments, refer to Figure 4 , Figure 4 This is the fourth flow chart of the method for predicting the battery capacity inflection point provided by the present application. Before the above step S304, the following step S306 is also included: Step S306, obtaining the storage sampling interval, storage temperature and battery formula parameters of the test battery during the calendar aging test phase.

[0066] It should be noted that the present application does not limit step S306 to be performed only before step S304. Figure 4 The order of executing the various steps in the process is only an illustrative example and is not limited in this application.

[0067] Accordingly, the above step S304 includes step S3041-1: Step S3041-1, constructing a training data set based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, storage sampling interval, storage temperature and battery formula parameters.

[0068] In the embodiment of the present application, when training the battery capacity inflection point prediction model, the influence of different storage temperatures and storage sampling intervals on batteries with different formulations is fully considered. Specifically, based on the correspondence between the capacity decay curve and the target battery capacity inflection point, the storage sampling interval, the storage temperature and the battery formulation parameters, a training data set is constructed to train the candidate model, thereby improving the rationality, reliability and versatility of the trained model in predicting the battery capacity inflection point, and making the prediction more accurate.

[0069] In another feasible implementation manner, the above step S3041-1 includes the following steps: Based on the correspondence between the capacity decay curve and the target battery capacity inflection point, the storage sampling interval, the storage temperature and the battery formula parameters, a two-dimensional array corresponding to the test battery is generated; wherein the two-dimensional array includes N groups of SOH corresponding to different storage times, storage sampling intervals, storage temperatures, battery formula parameters and time from the inflection point, the time from the inflection point represents the difference between different storage times and the time corresponding to the target battery capacity inflection point, and N is an integer greater than 2; The time from the inflection point is used as a prediction label, and the N groups of data in the two-dimensional array are slid sequentially according to a preset sliding window length to generate a time series data set as the training data set.

[0070] Specifically, each test battery corresponds to a two-dimensional array, where each column is a model input dimension / feature and each row is the primary storage data of the battery.

[0071] It should be noted that the storage temperature of the battery is the ambient temperature set before the calendar aging test begins. This temperature remains unchanged during the test. The storage sampling interval is actually the current storage time minus the next storage time.

[0072] For example, the first storage time is 8.5 days, and the first capacity is tested; the second storage time is 15.5 days, and the second capacity is tested; the third storage time is 26 days, and the third capacity is tested. The sampling intervals are 7 days and 10.5 days respectively. After three storage tests, three discharge capacity data can be obtained and calculated as SOH, and the rest are analogous. Since there is no next storage in the last interval, the interval can be set to 0, so the final interval values ​​are 7 days, 10.5 days, 9 days, 15 days, 19 days, and 0 days respectively. The data of each battery is a two-dimensional array, and its structure is shown in Table 1.

[0073] Table 1 Two-dimensional array corresponding to the battery

[0074] When generating a time series data set, first traverse the two-dimensional array of all batteries, then use DTK as the label of the regression algorithm and use a sliding window length of 3 to generate the data set: 1) When sliding for the first time, the label is DTK 45, and the first sequence is obtained, whose content is the data of the 1st, 2nd, and 3rd rows of the two-dimensional array.

[0075] 2) When sliding for the second time, the label is DTK 36, and the second sequence is obtained, whose content is the data of the 2nd, 3rd, and 4th rows of the two-dimensional array.

[0076] 3) When sliding for the third time, the label is DTK 21, and the third sequence is obtained, whose content is the data of the 3rd, 4th and 5th rows of the two-dimensional array. And so on, until the inflection point.

[0077] It should be noted that the above sliding window length can be set according to actual conditions.

[0078] In some embodiments, if the two-dimensional data is too little and the quality is uneven, the sliding window can continue to slide after the inflection point, but the data after the inflection point cannot be used when predicting the inflection point, nor can the prediction results after the inflection point be used (meaningless). It is only used to verify the model.

[0079] In an embodiment of the present application, after generating a two-dimensional array corresponding to the test battery based on the relevant data of the test battery, a sliding window can be used to slide on each row of the two-dimensional array to generate a time series data set, so as to increase the data volume of the training data set, which helps to improve the efficiency and effectiveness of model training.

[0080] In some embodiments, the candidate model may use the Transformer algorithm. A specific implementation method for determining the loss function of the candidate model is provided. Specifically, before the above step S305, the following steps are also included: A function obtained by weighted summing a mean absolute error function and at least one regularization penalty term is used as a loss function for training the candidate model; The regularization penalty term is used to control the range of the battery capacity inflection point predicted by the trained battery capacity inflection point prediction model.

[0081] It should be noted that the Transformer algorithm is a neural network model based on the self-attention mechanism, which is mainly used to process sequence data. This application may use the multivariate time series Transformer, referred to as TST, instead of the original Transformer.

[0082] In some embodiments, the mean absolute error function is, for example: .

[0083] in, Characterize the predicted DTK value, Characterize the true DTK value, Characterize the total number of samples.

[0084] In some embodiments, the regularization penalty term includes, for example: .

[0085] Specifically, a custom function is selected as the loss function of the training model, which is the weighted sum of the above three parts. The mean absolute error (MAE) is chosen as the loss function when training the model, mainly because MAE is less sensitive to outliers. When there is a lot of noise and uneven storage time in the data, which leads to a large error in the predicted label (the value of the number of days from the inflection point), using MAE as the loss function can make the model easier to fit. This is because MAE does not amplify the impact of outlier errors on the total loss like the root mean square error (RMSE).

[0086] At the same time, a regularization penalty term is embedded in the neural network training to observe the effect of the training validation set. When it is less than 0 or greater than 200 (the interval [0,200] should be selected according to the data situation, and this is just an example), a regularization term is added on the basis of MAE, and the weight of the regularization term is dynamically controlled by introducing the α and β parameters. Among them, ReLU (Rectified Linear Unit) is a widely used activation function, especially in deep learning and neural networks.

[0087] In an embodiment of the present application, the candidate model using the Transformer algorithm has an attention mechanism, and uses time series regression with attention to slide and predict the inflection point of battery capacity, which can capture the potential relationship of battery aging and perform real-time prediction. As the test data increases, it can further help the model improve the effect; in addition, the new loss function helps to improve the prediction accuracy and robustness of the algorithm.

[0088] In some other embodiments, a specific implementation method of further optimizing the battery capacity inflection point prediction model is provided. Specifically, the above step S305 includes the following steps: Using the training data set to train a preset candidate model to obtain a trained model; The mean absolute error function is set as the optimization target of the trained model, and at least one of the learning rate, number of iterations, batch size, dropout value and dropout value of the fully connected layer is set as the optimization parameter of the trained model. The trained model is optimized using the Bayesian algorithm to obtain the battery capacity inflection point prediction model.

[0089] It should be noted that Bayesian algorithm optimization (also known as Bayesian hyperparameter optimization) is a method used in the field of machine learning to find the optimal hyperparameter combination. Hyperparameters refer to parameters that need to be manually set before training a machine learning model. They are not adjusted through the learning of the optimization algorithm during the training process, but are determined by developers or researchers based on experience, experiments, or automatic tuning methods.

[0090] It should also be noted that the mean absolute error function in this embodiment is similar to the formula of the mean absolute error function in the above embodiment. The reason for using MAE here is the same as in the above embodiment, but the role of this error and loss function is completely different.

[0091] In some embodiments, when using the Bayesian algorithm for optimization, the number of optimization iterations can also be set. The more times, the slower the optimization process, but better parameters may be found. The range of parameters should be narrowed here, otherwise the time cost will be huge, especially for large data sets.

[0092] In an embodiment of the present application, a Bayesian algorithm is used to optimize the optimization parameters of the trained model in order to obtain the best model. Specifically, the best hyperparameters of the model are quickly found within a limited time, and the final model is trained using the hyperparameters, which can effectively improve the prediction accuracy of the battery capacity inflection point prediction model.

[0093] The following describes how to determine the capacity decay curve of the test battery in combination with a feasible implementation method: In one possible implementation, Figure 3 Based on the corresponding embodiments, refer to Figure 5 , Figure 5 This is the fifth flow chart of the method for predicting the battery capacity inflection point provided by the present application. The above step S302 includes the following steps S3021~S3024: Step S3021, calculating the ratio between the maximum discharge capacity corresponding to different storage times and the maximum discharge capacity of the initial storage as the SOH of the test battery at different storage times.

[0094] Step S3022, determining a first capacity decay curve corresponding to the test battery based on the SOH of the test battery at different storage times; The first capacity decay curve represents the corresponding relationship between the SOH and different storage times.

[0095] Step S3023: using the box plot corresponding to the first capacity decay curve, filtering the discrete points in the first capacity decay curve to obtain a second capacity decay curve.

[0096] Step S3024: determining a capacity decay curve of the test battery based on the second capacity decay curve.

[0097] It should be noted that the box plot, also known as the box-and-whisker plot, is a statistical chart used to show the dispersion of a set of data. It describes the distribution of a data set through five key values ​​(minimum value, first quartile Q1, median Q2, third quartile Q3 and maximum value).

[0098] It should also be noted that the above-mentioned outliers are individual values ​​in a set of battery data or performance parameters that are significantly different from other data or parameters. These values ​​may be caused by measurement errors, improper operation, abnormalities of the battery itself, or performance under specific conditions.

[0099] In the embodiment of the present application, a specific implementation method for determining the capacity decay curve of the test battery is provided. First, by calculating the ratio between the maximum discharge capacity corresponding to different storage times and the maximum discharge capacity of the initial storage, a first capacity decay curve characterizing the SOH corresponding to different storage times is determined, and then the discrete points in the first capacity decay curve are filtered out using a box plot to obtain a second capacity decay curve, and then based on the second capacity decay curve, the capacity decay curve of the test battery is determined. By filtering the box plot, the influence of discrete points on subsequent model training can be reduced, so as to reduce the interference of noise data on model training, thereby improving the prediction effect of the trained battery capacity inflection point prediction model.

[0100] In another possible implementation, referring to Figure 6 , Figure 6 This is the sixth flow chart of the method for predicting the battery capacity inflection point provided by the present application. The above step S3023 includes the following steps S3023-1~S3023-6: Step S3023-1, calculate the quartile and upper quartile corresponding to the first capacity decay curve.

[0101] Step S3023-2: Calculate the interquartile range corresponding to the first capacity decay curve based on the quartile and the upper quartile.

[0102] In some embodiments, the above step S3023-2 can calculate the interquartile range IQR by the following formula: IQR=Q3-Q1; wherein, Q3 represents the above upper quartile, Q1 represents the above quartile, and Q1 and Q3 can be calculated using the np.percentile function from the NumPy library of Python, which is widely used in vectorized calculations.

[0103] Step S3023-3: determine a first threshold based on the quartile and the interquartile range.

[0104] Step S3023-4: Determine a second threshold based on the upper quartile and the interquartile range.

[0105] It should be noted that the present application does not limit the order of executing step S3023-3 and step S3023-4, that is, the first threshold may be determined first, or the second threshold may be determined first, or the first threshold and the second threshold may be determined in parallel.

[0106] In some embodiments, the first threshold value in step S3023-3 may be determined by the following formula: First threshold = Q1-k*IQR; The second threshold value in step S3023-4 can be determined by the following formula: Second threshold = Q3 + k*IQR; Here, k represents the weight. The larger k is, the more discrete points are filtered out. It can usually be set to an empirical value, such as 1.5.

[0107] Step S3023-5: In the first capacity decay curve, points where the SOH is less than the first threshold or the SOH is greater than the second threshold are taken as discrete points.

[0108] Step S3023-6, filtering out the discrete points from the first capacity decay curve, and performing linear interpolation filling at the storage time corresponding to the discrete points to obtain the second capacity decay curve.

[0109] It should be noted that linear interpolation is a mathematical method used to estimate or construct a new data point between two known data points, whose value is calculated based on the values ​​of the two known points and their relative positions through a linear relationship (i.e. a straight line).

[0110] In the embodiment of the present application, a specific implementation method for filtering discrete points through a box plot is provided. Specifically, the quartiles and upper quartiles corresponding to the first capacity decay curve are first calculated, and the corresponding interquartile range is calculated, and then the first threshold is calculated based on the quartiles and the interquartile range, and the second threshold is calculated based on the upper quartile and the interquartile range, and the discrete points that are not between the first threshold and the second threshold are filtered out, and then the positions of the filtered discrete points are filled by linear interpolation to ensure the number of points in the capacity decay curve while ensuring the smoothness of the curve, which is helpful for the subsequent effective training of the model.

[0111] In another feasible implementation manner, the above step S3024 includes the following steps: The variance value of SOH in the second capacity decay curve is calculated.

[0112] When the variance value is greater than a preset third threshold, the second capacity decay curve is filtered by a one-dimensional Gaussian filter to obtain the capacity decay curve of the test battery.

[0113] When the variance value is less than or equal to the third threshold, the second capacity decay curve is filtered by using an SG filter to obtain the capacity decay curve of the test battery.

[0114] It should be noted that the SG filter, the full name of which is Savitzky-Golay filter, is a filtering method based on polynomial least squares fitting in the time domain.

[0115] In an embodiment of the present application, after determining the second capacity decay curve, if the variance value of SOH in the second capacity decay curve is large, the second capacity decay curve can be further filtered for smoothing. Specifically, when the variance value is greater than the third threshold, filtering with an SG filter may cause anomalies, so a one-dimensional Gaussian filter is used for filtering. When the variance value is less than or equal to the third threshold, an SG filter is used for filtering. The second capacity decay curve after smoothing is helpful for subsequent effective training of the model.

[0116] The following example illustrates the prediction method of the battery capacity inflection point provided in the embodiment of the present application. It is specifically implemented through the following steps: obtain the maximum discharge capacity of the battery after storage → calculate the capacity decay curve (SOH curve) of the battery → use a variety of data processing techniques to process the SOH curve and mark the position of the battery capacity inflection point → obtain the storage temperature, storage sampling interval, positive and negative electrode formula and other design information of the battery, and build a complete data set in combination with SOH → use a sliding window to slide on the SOH curve to generate a time series data set → use the Transformer algorithm to model the data, and the label of the model is how many storage days are left from the current storage day to the inflection point (that is, the time from the inflection point) → use a five-fold cross-validation to evaluate the model → find the optimal model hyperparameters based on Bayesian hyperparameter optimization technology.

[0117] In addition, the obvious abnormal sampling points in the SOH curve (that is, the above-mentioned discrete points) can also be processed to improve data quality and enhance the robustness of the model: obtain the maximum discharge capacity of the battery after storage → calculate the battery capacity decay curve (SOH curve) → use the box plot to filter the outliers on the SOH curve → use linear interpolation technology to fill the filtered SOH curve points → use SG filter and Gaussian filter to smooth the SOH curve and obtain the final SOH curve for constructing a complete data set. Finally, in order to improve the robustness of the model to abnormal data, a new loss function was developed, and the improved loss function can help the model improve the prediction effect.

[0118] Figure 7 This is the seventh flow chart of the method for predicting the inflection point of battery capacity provided by the present application, such as Figure 7 As shown, steps S701 to S706 are included: Step S701, reading cycle data and obtaining the maximum discharge capacity of the battery during the discharge phase.

[0119] Specifically, the maximum discharge capacity is extracted from the cycle test performed by the operator to calibrate the capacity after each storage. The data of the cycle test is only 3 cycles of charge and discharge data at most (charge and discharge process: rest → discharge → charge → rest → discharge), and the maximum discharge capacity of the second cycle is usually taken. This value is the actual available capacity of the storage battery recognized by the battery cell engineer. Each storage is accompanied by a cycle test, that is, a point data can be obtained.

[0120] For example, to obtain the maximum discharge capacity of the battery expressed as DQ, the calculation formula is: DQ=max(abs(Qs)); The DQ in the formula is defined as: the maximum capacity of the second discharge phase when the operator performs a cycle test after the battery is stored. Qs is the capacity sequence of the second discharge phase of the cycle test. The abs function represents the absolute value, and the max function represents the maximum value. The length of the capacity sequence depends on the rated capacity, discharge current, discharge time, and sampling frequency of the battery. After being processed by the max function in this step, each stored cycle sequence is converted into a discharge capacity. Within a storage life cycle, the data of all storage points constitute a capacity curve, such as Figure 8 As shown, Figure 8 It is a schematic diagram of a discharge capacity curve of a storage battery in the method for predicting the battery capacity inflection point provided in the present application.

[0121] It should be noted that the storage time here is equivalent to the storage days, both of which represent the total number of days accumulated from the start of storage to the end of this storage. The storage time of the first point is 0 days. The first one is only used as a capacity benchmark. The first point can be discarded in the SOH curve.

[0122] Step S702, calculate the SOH curve and process it, multiple people mark and vote to get the number of days at the inflection point, and calculate the DTK value.

[0123] Specifically, the maximum discharge capacity of each storage is calculated, and the SOH is calculated with the initial capacity for the first time, and all storages constitute the SOH curve. The outliers of the SOH curve are filtered using a box plot, and then the filtered SOH curve is filled using linear interpolation, and finally the SOH curve is smoothed using an SG filter.

[0124] For the SOH curve, multiple people independently mark the potential inflection point positions, and the final inflection point position is obtained through a vote by multiple people, and the inflection point positions of parallel samples with the same formulation are made as close as possible to ensure a certain consistency of the labels. This is because there are differences in the definition of the inflection point position of the same SOH curve by different people, which belongs to human error. Finally, the DTK value of each point is calculated as the prediction label of the regression algorithm.

[0125] For example, the formula for calculating SOH is: SOH = DQ / DQ 0 , where DQ represents the discharge capacity and DQ 0 represents the first discharge capacity.

[0126] The steps for filtering discrete points of the SOH curve using a box plot are as follows: 1) Calculate Q1 (the first quartile) and Q3 (the third quartile); then calculate the IQR (interquartile range), where IQR = Q3 - Q1.

[0127] 2) Calculate the outlier detection range, usually from Q1 - k * IQR to Q3 + k * IQR, that is, when SOH > Q3 + k * IQR or SOH < Q1 - k * IQR, it is an outlier, and here k = 1.5 is set.

[0128] 3) Identify and mark the index positions of the outliers, and filter out the outliers through the index.

[0129] 4) Set the filtered outliers to NA and fill them using linear interpolation.

[0130] In some embodiments, the fitting and smoothing process is optional and aims to make the curve smoother while ensuring the curve trend remains unchanged. Its implementation is as follows: when the variance value of the SOH curve exceeds the variance threshold, a 1D Gaussian filter is used, and when it is small, a Savitzky - Golay filter is used. They are calculated using the gaussian_filter1d function from the scipy.ndimage library and the savgol_filter function from the scipy.signal library in Python respectively. The number of points on the processed SOH curve does not change, so there is no need to perform interpolation filling again. As Fig. 9 shown, Fig. 9 is a schematic diagram of the capacity decay curve of a certain storage battery in the method for predicting the inflection point of the battery capacity provided by this application.

[0131] Among them, the outliers have been filtered. Suppose the inflection point position finally confirmed manually is 71.0 days, which is framed by a rectangle. Now calculate the DTK value of each point before (including the inflection point) to the inflection point.

[0132] The DTK value is defined as: how many storage days are needed from the current storage days to reach the inflection point; the calculation formula is: DTK = inflection point storage time (days) - storage time at each point (days). Fig. 9 There are 6 points before the turning point, and the DTK values ​​are: 71-8.5 (62.5 days), 71-15.5 (55.5 days), 71-26 (45 days), 71-35 (36 days), 71-50 (21 days), and 71-71 (0 days).

[0133] It should be noted that the above variance threshold depends on the specific data set, and the threshold set in this embodiment may be 0.09.

[0134] It should also be noted that the input parameters of the savgol_filter function are x (SOH curve), window (window width), and polyorder (order), where the window width calculation formula is: window=min(len(x), 15), and polyorder is set to 3. The min function in the formula means selecting the minimum number from two numbers, and the len function means the number of points to obtain the SOH curve, that is, the length of the curve composed of discrete points.

[0135] The input parameters of the gaussian_filter1d function are input (SOH curve) and sigma (controls the width of the Gaussian kernel). A larger sigma will result in more smoothing effects. The sigma set in this embodiment is 0.45.

[0136] Step S703, obtaining the design information of the battery such as storage temperature, storage sampling interval, positive and negative electrode formula, etc.

[0137] Specifically, the storage temperature of each battery is obtained, and the storage sampling interval refers to the difference in days between two storage days. Combined with the recipe information, each battery gets a two-dimensional array, where each column is a model input dimension / feature and each row is a storage of the battery.

[0138] For example, the above two-dimensional array is, for example, Table 1 above. Table 1 is only an example because the formula ingredients are usually secret. However, the solution of the present application is not limited to a specific formula, so it is also applicable to other calendar-aged and cycle-aged batteries. In practical applications, the number of storage times may be between 3 and 50 times (considering the balance between the accuracy of capacity calibration and labor costs). The number of storage times should not be less than the sliding window length mentioned above. The sliding window length should be continuously tried and adjusted as needed. If the data and hardware allow, the longer the sliding window length, the more information it contains, and the better the effect may be.

[0139] The data listed in Table 1 only shows 6 storages, in which the storage temperature and formula are fixed values. Because the temperature and formula remain unchanged during the storage process. It is mainly used to assist model prediction. In addition to the positive electrode physicochemical content, electrolyte, binder content provided in Table 1, there may also be negative electrode physicochemical content, diaphragm, battery cell type and other formula parameters, which are not restricted in this application. Here are the effective parameters obtained after preliminary screening by battery experts. Among them, the positive electrode physicochemical content can be divided into multiple more detailed data, such as the carbon content and water content of the positive electrode of the battery, and other formulas are similar.

[0140] Step S704: Using the DTK value as the prediction label, a sliding window is used to slide on the SOH curve to generate a time series data set.

[0141] Specifically, based on the M1 two-dimensional arrays obtained above (where M1 is the number of batteries), a time series data set can be generated by sliding a sliding window on the rows of each two-dimensional array. For example, if the sliding window length is 3, the window slides one point each time, so the battery data should have at least 3 points or more to be used. The shape of the time series data set is: (number of records, number of feature dimensions, sequence length), where the number of records is the total number of records generated by all batteries after the sliding window operation, the number of feature dimensions is the number of features including all inputs such as temperature, SOH, formula, etc., and the sequence length is the size of the sliding window.

[0142] For example, Fig.10 is a schematic diagram of the process of generating a training data set in the method for predicting the battery capacity inflection point provided in this application, such as Fig.10 As shown, the core process of this step is to generate a data set for time series regression, which specifically includes steps S1001 to S1004: Step S1001, traverse the two-dimensional array of all batteries. The two-dimensional array structure is shown in Table 1 above.

[0143] Step S1002: Sliding window generates a sequence.

[0144] Specifically, DTK is used as the label of the regression algorithm and a sliding window length of 3 is used to generate the data set. Fig.11 is a flow chart of using a sliding window to generate a data set in the method for predicting the inflection point of battery capacity provided by the present application, such as Fig.11 As shown, each time a dotted line connects three data, it represents a slide. The following is a detailed description: 1) When sliding for the first time, the label is DTK 45, and the first sequence is obtained, whose content is the data of the 1st, 2nd, and 3rd rows of the two-dimensional array.

[0145] 2) When sliding for the second time, the label is DTK 36, and the second sequence is obtained, whose content is the data of the 2nd, 3rd, and 4th rows of the two-dimensional array.

[0146] 3) When sliding for the third time, the label is DTK 21, and the third sequence is obtained, whose content is the data of the 3rd, 4th and 5th rows of the two-dimensional array. And so on, until the inflection point.

[0147] Step S1003, merging the data generated by sliding each battery. Specifically, merging the data generated by sliding each battery into one set.

[0148] Step S1004, divide the model data set into a training set, a validation set, and a test set in a ratio of 6:2:2.

[0149] Step S705, using the Transformer algorithm to build a model and predict how many storage days are left from the current storage day to the turning point.

[0150] Specifically, Transformer is a very popular deep learning algorithm that can be used to capture the potential relationships in complex time series. Based on this, a time series regression model is established to predict a DTK value based on 3 points of data for a battery each time. Because the time window length is 3, 3 points of data are needed each time for prediction.

[0151] For example, when using the Transformer algorithm to build a model, it is only necessary to use the training set divided in step S1004 to train the model, the validation set to validate the model, and the test set to test the final effect of the model.

[0152] The loss function of the training model selects a custom function, which is the weighted sum of the following three parts: ; .

[0153] Step S706, finding the optimal model hyperparameters through Bayesian hyperparameter optimization technology.

[0154] Specifically, after the modeling is completed, based on the Bayesian optimization method, the optimal hyperparameters of the model are quickly found within a limited time, and the final model is trained using the hyperparameters.

[0155] For example, this step uses Bayesian techniques to optimize hyperparameter optimization in order to obtain the best model. Fig.12 : is a schematic diagram of the process of Bayesian optimization of hyperparameters in the battery capacity inflection point prediction method provided in this application, such as Fig.12 As shown, the following steps are included: Step S1201, constructing a Bayesian optimization framework.

[0156] Specifically, the open source framework optuna is used as the implementation of Bayesian optimization technology.

[0157] Step S1202, setting the optimization target: validation set evaluation error.

[0158] Specifically, the error is set as the optimization target, and the error calculation formula is: .

[0159] It should be noted that the reason for using MAE is the same as the content related to loss function. However, the role of this error and loss function is completely different.

[0160] Among them, MAE is used to measure the mean absolute error between the predicted value and the true value. The smaller the MAE, the better the model. is the predicted value, is the true value, is the total number of samples.

[0161] Step S1203, setting the parameters and ranges to be optimized.

[0162] Specifically, the parameters that need to be optimized in the model are: learning rate, number of iterations, batch size, dropout value, and fully connected layer dropout value.

[0163] Step S1204, set the number of iterations and start training the model.

[0164] Specifically, the number of optimization iterations is set. The more iterations, the slower the optimization process, but better parameters may be found. The range of parameters should be narrowed, otherwise the time cost is huge, especially for large data sets.

[0165] It should be noted that the learning rate and the number of iterations have a great impact. Their value ranges can be 0.0005~0.004 and 100~150 respectively. Giving a suitable range can speed up the optimization process. After optimizing the best parameters, retrain the model to get the best model. If it is deployed, the full dataset can be used to train the model without dividing it into training and test sets.

[0166] In the embodiments of the present application, there are at least the following beneficial effects: 1) Fully consider the impact of different temperatures on batteries with different formulas, ensure the rationality, reliability and versatility of the prediction of the battery inflection point, provide test acceleration for the battery research and development process, and significantly speed up the research and development process, that is, the battery does not need to wait for storage all the time.

[0167] 2) Using time series regression with attention to slide the prediction inflection point can capture the potential relationship of battery aging and make real-time predictions. As the test data increases, it can further help the model improve its performance.

[0168] 3) Data quality is very important for small samples. Considering that the battery has measurement errors during the test, and the storage test is that the operator places the battery in a designated area, takes it out of the designated area after the storage time is up, and performs a cycle test to characterize the capacity, there may be problems such as non-standard operation, measurement accuracy differences between different machines, and different storage time intervals between different operators during the cycle process. After cleaning and smoothing the data, the interference of noise data on the model can be reduced, and the new loss function can improve the prediction accuracy and robustness of the algorithm.

[0169] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the method for predicting the inflection point of battery capacity of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0170] This application also provides a device for predicting the inflection point of battery capacity. Please refer to Fig.13 , Fig.13 : is a schematic diagram of the structure of a device for predicting the inflection point of battery capacity provided in the present application, and the device for predicting the inflection point of battery capacity includes: The acquisition module 1301 is used to acquire the cycle data of the battery to be predicted in the calendar aging test stage; wherein the cycle data includes the corresponding relationship between different storage times collected in the calendar aging test stage and the maximum discharge capacity of the battery to be predicted; The prediction module 1302 is used to input the cycle data into a pre-trained battery capacity inflection point prediction model, and the battery capacity inflection point prediction model predicts the battery capacity inflection point of the battery to be predicted based on the cycle data.

[0171] The device for predicting the inflection point of battery capacity provided in the present application adopts the method for predicting the inflection point of battery capacity in the above-mentioned embodiment, and can solve the problem of how to accurately predict the inflection point of battery capacity.

[0172] Compared with the related art, the beneficial effects of the battery capacity inflection point prediction device provided in the present application are the same as the beneficial effects of the battery capacity inflection point prediction method provided in the above-mentioned embodiment, and other technical features in the battery capacity inflection point prediction device are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0173] The present application provides a device for predicting a battery capacity inflection point, the device for predicting a battery capacity inflection point comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for predicting a battery capacity inflection point in any of the above-mentioned embodiments.

[0174] Reference below Fig.14 , Fig.14 The structure diagram provided by the prediction device of the battery capacity inflection point of the present application shows the structure diagram of the prediction device of the battery capacity inflection point suitable for realizing the embodiment of the present application. The prediction device of the battery capacity inflection point in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions: tablet computers), PMPs (Portable Media Players: portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.14 The battery capacity inflection point prediction device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0175] like Fig.14 As shown, the prediction device for the battery capacity inflection point may include a processing device 1401 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1402 or a program loaded from a storage device 1403 to a random access memory (RAM: Random Access Memory) 1404. In RAM 1404, various programs and data required for the operation of the prediction device for the battery capacity inflection point are also stored. The processing device 1401, ROM 1402, and RAM 1404 are connected to each other via a bus 1405. An input / output (I / O) interface 1406 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1406: input devices 1407 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1408 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1403 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1409. The communication device 1409 can allow the prediction device of the battery capacity inflection point to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a prediction device of the battery capacity inflection point with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0176] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1403, or installed from a ROM 1402. When the computer program is executed by the processing device 1401, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0177] The battery capacity inflection point prediction device provided by the present application adopts the battery capacity inflection point prediction method in the above embodiment, which can solve the problem of how to accurately predict the battery capacity inflection point. Compared with the related art, the beneficial effects of the battery capacity inflection point prediction device provided by the present application are the same as the beneficial effects of the battery capacity inflection point prediction method provided by the above embodiment, and the other technical features of the battery capacity inflection point prediction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0178] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0179] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0180] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the battery capacity inflection point prediction method in the above-mentioned embodiment.

[0181] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0182] The computer-readable storage medium may be included in the battery capacity inflection point prediction device; or may exist independently without being assembled into the battery capacity inflection point prediction device.

[0183] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by a battery capacity inflection point prediction device, the battery capacity inflection point prediction device performs the following steps: obtaining cycle data of the battery to be predicted during the calendar aging test phase; wherein the cycle data includes the correspondence between different storage times collected during the calendar aging test phase and the maximum discharge capacity of the battery to be predicted; inputting the cycle data into a pre-trained battery capacity inflection point prediction model, and the battery capacity inflection point prediction model predicts the battery capacity inflection point of the battery to be predicted based on the cycle data.

[0184] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0185] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0186] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0187] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned method for predicting the inflection point of battery capacity, and can solve the problem of how to accurately predict the inflection point of battery capacity. Compared with the related art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the method for predicting the inflection point of battery capacity provided in the above-mentioned embodiment, and will not be elaborated here.

[0188] The present application also provides a computer program product, including a computer program, which implements the steps of the battery capacity inflection point prediction method as described above when the computer program is executed by a processor.

[0189] The computer program product provided in this application can solve the problem of how to accurately predict the inflection point of battery capacity. Compared with the related art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the method for predicting the inflection point of battery capacity provided in the above embodiment, which will not be repeated here.

[0190] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for predicting a battery capacity inflection point, characterized in that: The method comprises: Acquire cycle data of the battery to be predicted during the calendar aging test phase; wherein the cycle data includes a correspondence between different storage times collected during the calendar aging test phase and a maximum discharge capacity of the battery to be predicted; The cycle data is input into a pre-trained battery capacity inflection point prediction model, and the battery capacity inflection point prediction model predicts the battery capacity inflection point of the battery to be predicted based on the cycle data.

2. The method according to claim 1, characterized in that The step of inputting the cycle data into a pre-trained battery capacity inflection point prediction model, and using the battery capacity inflection point prediction model to predict the battery capacity inflection point of the battery to be predicted based on the cycle data, comprises: Inputting the cycle data and the auxiliary prediction information into a pre-trained battery capacity inflection point prediction model, and having the battery capacity inflection point prediction model predict the battery capacity inflection point of the battery to be predicted based on the cycle data and the auxiliary prediction information; The auxiliary prediction information includes at least one of a storage sampling interval, a storage temperature and a battery formula parameter.

3. The method according to claim 1 or 2, characterized in that Before inputting the cycle data into a pre-trained battery capacity inflection point prediction model, the method further includes: Obtain the maximum discharge capacity of the test battery corresponding to different storage times during the calendar aging test phase; Determine a capacity decay curve of the test battery based on the maximum discharge capacity corresponding to different storage times of the test battery; wherein the capacity decay curve represents the corresponding relationship between different storage times and the battery health state SOH; In the capacity decay curve, determining a target battery capacity inflection point of the test battery; Based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, constructing a training data set; The training data set is used to train a preset candidate model, and the trained model is obtained as the battery capacity inflection point prediction model.

4. The method according to claim 3, characterized in that Before constructing a training data set based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, the method further includes: Obtaining storage sampling interval, storage temperature and battery formula parameters of the test battery during the calendar aging test phase; The constructing of a training data set based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point includes: A training data set is constructed based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, storage sampling interval, storage temperature and battery formula parameters.

5. The method according to claim 3, characterized in that The determining the capacity decay curve of the test battery based on the maximum discharge capacity corresponding to different storage times of the test battery includes: Calculate the ratio between the maximum discharge capacity corresponding to different storage times and the maximum discharge capacity of the initial storage as the SOH of the test battery at different storage times; Based on the SOH of the test battery at different storage times, determining a first capacity decay curve corresponding to the test battery; wherein the first capacity decay curve represents a corresponding relationship between the SOH and different storage times; Using the box plot corresponding to the first capacity decay curve, filtering the discrete points in the first capacity decay curve to obtain a second capacity decay curve; Based on the second capacity fade curve, a capacity fade curve of the test battery is determined.

6. The method according to claim 5, characterized in that The filtering of discrete points in the first capacity decay curve using the box plot corresponding to the first capacity decay curve to obtain a second capacity decay curve includes: Calculate the quartile and upper quartile corresponding to the first capacity decay curve; Based on the quartile and the upper quartile, calculating the interquartile range corresponding to the first capacity decay curve; Determining a first threshold based on the quartile and the interquartile range; Determining a second threshold based on the upper quartile and the interquartile range; In the first capacity decay curve, points where the SOH is less than the first threshold or the SOH is greater than the second threshold are taken as discrete points; The discrete points are filtered out from the first capacity decay curve, and linear interpolation filling is performed at the storage time corresponding to the discrete points to obtain the second capacity decay curve.

7. The method according to claim 6, characterized in that The determining the capacity decay curve of the test battery based on the second capacity decay curve includes: Calculating a variance value of SOH in the second capacity decay curve; When the variance value is greater than a preset third threshold, filtering the second capacity decay curve by a one-dimensional Gaussian filter to obtain a capacity decay curve of the test battery; When the variance value is less than or equal to the third threshold, the second capacity decay curve is filtered by using an SG filter to obtain the capacity decay curve of the test battery.

8. The method according to claim 4, characterized in that The training data set is constructed based on the corresponding relationship between the capacity decay curve and the target battery capacity inflection point, the storage sampling interval, the storage temperature and the battery formula parameters, including: Based on the correspondence between the capacity decay curve and the target battery capacity inflection point, the storage sampling interval, the storage temperature and the battery formula parameters, a two-dimensional array corresponding to the test battery is generated; wherein the two-dimensional array includes N groups of SOH corresponding to different storage times, storage sampling intervals, storage temperatures, battery formula parameters and time from the inflection point, the time from the inflection point represents the difference between different storage times and the time corresponding to the target battery capacity inflection point, and N is an integer greater than 2; The time from the inflection point is used as a prediction label, and the N groups of data in the two-dimensional array are slid sequentially according to a preset sliding window length to generate a time series data set as the training data set.

9. The method according to claim 3, characterized in that The candidate model adopts the Transformer algorithm; Before using the training data set to train a preset candidate model to obtain the trained model as the battery capacity inflection point prediction model, the method further includes: A function obtained by weighted summing a mean absolute error function and at least one regularization penalty term is used as a loss function for training the candidate model; The regularization penalty term is used to control the range of the battery capacity inflection point predicted by the trained battery capacity inflection point prediction model.

10. The method according to claim 3, characterized in that The method of using the training data set to train a preset candidate model to obtain the trained model as the battery capacity inflection point prediction model includes: Using the training data set to train a preset candidate model to obtain a trained model; The mean absolute error function is set as the optimization target of the trained model, and at least one of the learning rate, number of iterations, batch size, dropout value and dropout value of the fully connected layer is set as the optimization parameter of the trained model. The trained model is optimized using the Bayesian algorithm to obtain the battery capacity inflection point prediction model.

11. A device for predicting a battery capacity inflection point, characterized in that: The device comprises: An acquisition module, used to acquire cycle data of the battery to be predicted during the calendar aging test phase; wherein the cycle data includes a correspondence between different storage times collected during the calendar aging test phase and the maximum discharge capacity of the battery to be predicted; The prediction module is used to input the cycle data into a pre-trained battery capacity inflection point prediction model, and the battery capacity inflection point prediction model predicts the battery capacity inflection point of the battery to be predicted based on the cycle data.

12. A device for predicting a battery capacity inflection point, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the method for predicting a battery capacity inflection point according to any one of claims 1 to 10.

13. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the method for predicting the battery capacity inflection point according to any one of claims 1 to 10 are implemented.

14. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method for predicting a battery capacity inflection point according to any one of claims 1 to 10 are implemented.

Images