An early prediction method and system for battery capacity curve knee point

Abstract

The application belongs to the technical field of batteries, and particularly relates to a method and system for early prediction of a capacity curve inflection point of a battery, the method comprising: obtaining a capacity sequence and an internal resistance sequence at the initial stage of the battery life, and extracting multi-dimensional features therefrom. The capacity sequence and the internal resistance sequence are divided into a plurality of charge and discharge cycle stages according to the number of charge and discharge cycles at the initial stage of the battery life, and the cycle statistical features, waveform statistical features and differential waveform statistical features of the battery are calculated according to the capacity features and the internal resistance features of each charge and discharge cycle stage. Finally, the cycle statistical features, the waveform statistical features and the differential waveform statistical features of the battery are input into a trained inflection point prediction model to predict the capacity curve inflection point of the battery. The application can accurately predict the capacity curve inflection point of the battery only by using the capacity and internal resistance data at the initial stage of the battery life, is suitable for a battery management system with limited resources, and helps battery life management and safety evaluation.

Description

Technical Field

[0001] This invention belongs to the field of battery technology, specifically relating to an early prediction method and system for the inflection point of a battery capacity curve. Background Technology

[0002] Batteries (including but not limited to lithium-ion batteries, sodium-ion batteries, quasi-solid-state batteries, and all-solid-state batteries) are core energy storage components, applicable to electrified transportation vehicles such as electric vehicles, rail transit, electric ships, and electric aircraft, as well as energy storage systems such as energy storage power stations, data centers, and intelligent computing centers. However, batteries experience performance degradation during use, with their capacity decay trend typically exhibiting a two-stage non-linear characteristic. The inflection point in this capacity decay trend marks the transition from slow capacity decay to accelerated decay, indicating the beginning of a rapid decline in battery performance. Therefore, accurately predicting the inflection point of the capacity curve is crucial for battery life management and safety assessment.

[0003] Currently, most inflection point prediction methods employ model-based or data-driven approaches. However, model-based methods rely heavily on equivalent circuit or electrochemical models, resulting in numerous parameters, computational complexity, and poor adaptability. Data-driven methods, on the other hand, depend on abundant measurement data (such as voltage, current, and temperature curves) or complex feature engineering (such as differential capacitance curves), making accurate prediction difficult in applications with limited sensor configurations. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a method that can accurately predict the inflection point of the battery capacity curve using only capacity data and internal resistance data at the beginning of the battery life. Under the conditions of limited sensor configuration and short data acquisition cycle, this method can achieve early and accurate prediction of the inflection point of the battery capacity curve, so as to help with battery life management and safety assessment.

[0005] To achieve the above and other related objectives, this invention provides an early prediction method for the inflection point of a battery capacity curve, comprising: acquiring a capacity sequence and an internal resistance sequence at the beginning of the battery's lifespan; extracting multidimensional features from the capacity sequence and the internal resistance sequence, including: dividing the capacity sequence and the internal resistance sequence into several charge-discharge cycle stages based on the number of charge-discharge cycles at the beginning of the battery's lifespan, and calculating the battery's cycle statistical features, waveform statistical features, and differential waveform statistical features based on the battery's capacity characteristics and internal resistance characteristics at each charge-discharge cycle stage; and inputting the battery's cycle statistical features, waveform statistical features, and differential waveform statistical features into a trained inflection point prediction model to predict the inflection point of the battery's capacity curve.

[0006] According to a specific embodiment of the present invention, the step of calculating the cyclic statistical characteristics includes: calculating a first statistical characteristic based on the discharge capacity of each charge-discharge cycle stage; calculating a second statistical characteristic based on the charging capacity of each charge-discharge cycle stage; calculating a third statistical characteristic based on the internal resistance value of each charge-discharge cycle stage; calculating a fourth statistical characteristic based on the ratio of the actual capacity to the corresponding internal resistance value of each charge-discharge cycle stage; and integrating the first statistical characteristic, the second statistical characteristic, the third statistical characteristic, and the fourth statistical characteristic to form the cyclic statistical characteristics.

[0007] According to a specific embodiment of the present invention, the step of calculating the waveform statistical features includes: for any charge-discharge cycle stage, constructing a corresponding discharge capacity time-series waveform based on the capacity characteristics therein, so as to calculate the capacity statistical features and waveform statistical features of the discharge capacity time-series waveform; integrating the capacity statistical features and waveform statistical features of all discharge capacity time-series waveforms to form a fifth statistical feature; performing secondary aggregation on the capacity statistical features of each discharge capacity time-series waveform to calculate a sixth statistical feature; and integrating the fifth statistical feature and the sixth statistical feature to form the waveform statistical features.

[0008] According to a specific embodiment of the present invention, the step of calculating the differential waveform statistical features includes: for any charge-discharge cycle stage, constructing a corresponding linearized capacity waveform based on the capacity characteristics therein, and calculating the capacity statistical features and waveform statistical features of the linearized capacity waveform; integrating the capacity statistical features and waveform statistical features of all linearized capacity waveforms to form a seventh statistical feature; performing secondary aggregation on the capacity statistical features of each linearized capacity waveform to calculate an eighth statistical feature; for any charge-discharge cycle stage, constructing a corresponding differential capacity waveform based on the capacity characteristics therein, and calculating the capacity statistical features and waveform statistical features of the differential capacity waveform; integrating the capacity statistical features and waveform statistical features of all differential capacity waveforms to form a ninth statistical feature; performing secondary aggregation on the capacity statistical features of each differential capacity waveform to calculate a tenth statistical feature; and integrating the seventh statistical feature, the eighth statistical feature, the ninth statistical feature, and the tenth statistical feature to form the differential waveform statistical features.

[0009] According to a specific embodiment of the present invention, the training steps of the inflection point prediction model include: acquiring the capacity sequence and internal resistance sequence of multiple batteries throughout their entire life cycle; constructing corresponding training samples and sample labels based on the capacity sequence and internal resistance sequence of each battery; including: for each battery: dividing the capacity sequence and internal resistance sequence of the early stage of battery life into several charge-discharge cycle stages based on the number of charge-discharge cycles in the early stage of battery life; and calculating the cycle statistical characteristics, waveform statistical characteristics, and differential waveform statistical characteristics of the battery based on the capacity characteristics and internal resistance characteristics of the battery in each charge-discharge cycle stage to construct training samples; and based on the full life cycle of the battery... A capacity curve for the battery is constructed using the capacity sequence of the battery life cycle and the corresponding charge-discharge cycle number. The capacity and cycle index of each point on the capacity curve are normalized. The vertical distance from each point to a preset reference line is calculated based on the normalized point coordinates, and the inflection point of the capacity curve is determined accordingly, serving as the sample label. The inflection point represents the point on the capacity curve that is farthest from the preset reference line. All training samples are input into a pre-constructed inflection point prediction model, and the inflection point prediction model is optimized inversely based on the regression prediction value output by the inflection point prediction model and the sample labels of the training samples.

[0010] According to a specific embodiment of the present invention, the step of inputting all training samples into a pre-built inflection point prediction model and back-optimizing the inflection point prediction model based on the regression prediction value output by the inflection point prediction model and the sample labels of the training samples includes: inputting all training samples into a sample optimization model to determine the global importance of each feature dimension used to construct the training samples; wherein, the sample optimization model adopts an extreme gradient boosting regression model, and the global importance is used to represent the contribution of the current feature dimension type to the global impact, including at least SHAP (SHapleyAdditive exPlanations) contribution and XGBoost (eXtreme Gradient Boosting) contribution; redetermining the multi-dimensional features used to construct the training samples based on the global importance of each feature dimension, and re-optimizing all training samples; inputting the optimized training samples into the pre-built inflection point prediction model, and back-optimizing the inflection point prediction model based on the regression prediction value output by the inflection point prediction model and the sample labels corresponding to the optimized training samples.

[0011] According to a specific embodiment of the present invention, the formula for calculating the SHAP contribution is as follows: , in, This represents the SHAP contribution of the j-th dimension feature used to construct the training samples. Indicates the number of training samples. This represents the SHAP value of the j-th dimension feature in the i-th training sample.

[0012] According to a specific embodiment of the present invention, the formula for calculating the XGBoost contribution is as follows: , in, This represents the XGBoost contribution of the j-th dimension feature used to construct the training samples. This indicates the number of regression trees in the sample optimization model. Let represent the set of all split nodes in the t-th regression tree that use the j-th dimension as the splitting feature. This represents the loss-related value in the objective function of the corresponding parent node before the split at the s-th node. This represents the loss-related value in the objective function at the left child node after the split at the s-th node. This represents the value related to the loss in the objective function on the right child node after the split at the s-th node.

[0013] An early prediction system for the inflection point of a battery capacity curve includes: a data acquisition module for acquiring the capacity sequence and internal resistance sequence at the beginning of the battery's lifespan; a feature extraction module for extracting multi-dimensional features from the capacity sequence and the internal resistance sequence, including: dividing the capacity sequence and the internal resistance sequence into several charge-discharge cycle stages based on the number of charge-discharge cycles at the beginning of the battery's lifespan, and calculating the battery's cycle statistical features, waveform statistical features, and differential waveform statistical features based on the battery's capacity characteristics and internal resistance characteristics at each charge-discharge cycle stage; and an inflection point prediction module for inputting the battery's cycle statistical features, waveform statistical features, and differential waveform statistical features into a trained inflection point prediction model to predict the inflection point of the battery's capacity curve.

[0014] The beneficial effects of this invention are as follows: This invention only requires the collection of capacity data and internal resistance data at the beginning of the battery's lifespan to accurately predict the inflection point of the battery's capacity curve. It is suitable for battery management systems with limited resources, thereby solving the problem of being unable to accurately predict the inflection point of battery capacity due to limited sensor configuration and short data acquisition cycle. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0016] Figure 1This is a flowchart illustrating an early prediction method for the inflection point of a battery capacity curve provided in one embodiment of the present invention. Figure 2 This is a schematic diagram of the inflection point prediction model provided in one embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an early prediction system for the inflection point of a battery capacity curve provided in one embodiment of the present invention; Figure 4 This is a structural block diagram of an electronic device provided in one embodiment of the present invention. Detailed Implementation

[0017] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0018] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0019] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, publicly known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0020] Example 1 Please see Figure 1 The method shown includes an early prediction method for the inflection point of a battery capacity curve, comprising: Step S100: Obtain the capacity sequence and internal resistance sequence at the beginning of the battery life.

[0021] Specifically, in practical applications, after obtaining the capacity sequence and internal resistance sequence at the beginning of the battery life, the raw data can be preprocessed. For example, the raw capacity sequence can be cleaned, smoothed, and outliers removed to ensure that the battery capacity decay trend is monotonically decreasing and to avoid the influence of abnormal noise.

[0022] Correspondingly, the original data can be filtered using a range-based physical threshold method, or outlier detection can be performed using a statistical method based on the interquartile range, or the original data can be smoothed using a Savitzky-Golay filter, and other data preprocessing operations are not limited in this regard.

[0023] It should also be noted that the capacity sequence and internal resistance sequence mentioned here in the early stage of battery life refer to the capacity data and internal resistance data in the battery's historical operating data. For example, it can be the capacity data and internal resistance data from the time the battery was first used until the present. This data is then used to predict the inflection point of the capacity curve.

[0024] Step S200 involves extracting multidimensional features from the capacity sequence and the internal resistance sequence, including: dividing the capacity sequence and the internal resistance sequence into several charge-discharge cycle stages based on the number of charge-discharge cycles in the early stage of the battery life, and calculating the cycle statistical features, waveform statistical features, and differential waveform statistical features of the battery based on the capacity characteristics and internal resistance characteristics of the battery in each charge-discharge cycle stage.

[0025] What can be understood here is that the entire life cycle of a battery includes several charge-discharge cycles, and one charge-discharge cycle refers to the battery being discharged from full capacity to a certain capacity and then charged back to full capacity. Accordingly, as the number of charge-discharge cycles of the battery increases, the battery capacity will also decrease.

[0026] Accordingly, the battery capacity sequence can be divided into multiple charge-discharge cycle stages based on the number of charge-discharge cycles in the early stages of battery life. Furthermore, multi-dimensional features can be extracted based on the capacity and internal resistance characteristics of the battery in each charge-discharge cycle stage, including cycle statistical features, waveform statistical features, and differential waveform statistical features. The extraction methods are as follows: For cyclic statistical characteristics: Based on the discharge capacity of the battery in each charge-discharge cycle, the corresponding first statistical characteristic needs to be calculated, including nine statistical measures: mean, standard deviation, minimum, maximum, median, first value, last value, linear slope, and average rate of change.

[0027] It is understandable that a battery will only undergo one discharge process in one charge-discharge cycle, and the discharge capacity of the battery in this discharge process can be determined accordingly. The above statistics can then be calculated based on the discharge capacity of all charge-discharge cycles.

[0028] Similarly, based on the battery's charging capacity in each charge-discharge cycle, a second statistical characteristic needs to be calculated, including nine statistical measures: mean, standard deviation, minimum, maximum, median, first value, last value, linear slope, and average rate of change. Additionally, based on the battery's internal resistance value in each charge-discharge cycle, a third statistical characteristic needs to be calculated, including nine statistical measures: mean, standard deviation, minimum, maximum, median, first value, last value, linear slope, and average rate of change.

[0029] In addition, for each charge-discharge cycle stage, a ratio can be calculated based on the corresponding actual capacity and internal resistance value. Based on the ratios of all charge-discharge cycle stages, a fourth statistical characteristic is calculated, including six statistical measures: mean, standard deviation, minimum, maximum, linear slope, and area under the curve (AUC).

[0030] Finally, by integrating the first, second, third, and fourth statistical features mentioned above, a 33 (9+9+9+6) dimensional cyclic statistical feature is formed.

[0031] Regarding waveform statistical characteristics: For each charge-discharge cycle stage, the discharge capacity time-series waveform for the current charge-discharge cycle stage can be constructed based on the corresponding capacity characteristics in the capacity sequence. (Waveform). First, the capacity statistical characteristics and waveform statistical characteristics of each discharge capacity time-series waveform can be determined, i.e., feature extraction is performed within a single charge-discharge cycle. Correspondingly, based on the capacity changes represented by the discharge capacity time-series waveform (such as the capacity decay that occurs during battery discharge), the corresponding capacity statistical characteristics need to be calculated, including the mean, standard deviation, minimum, maximum, median, and other statistical quantities. Simultaneously, based on the waveform characteristics represented by the discharge capacity time-series waveform, the corresponding waveform statistical characteristics also need to be calculated, including morphological features such as peak value, area under the curve, and slope. The capacity statistical characteristics and waveform statistical characteristics of a single discharge capacity time-series waveform constitute a fifth statistical characteristic.

[0032] Furthermore, the extracted single-cycle features can be aggregated a second time at the cross-cycle level. That is, the capacity statistical features of each discharge capacity time-series waveform can be aggregated a second time to obtain the sixth statistical feature. For example, a mean value can be calculated for the capacity statistical features of each discharge capacity time-series waveform. The mean values ​​of all discharge capacity time-series waveforms can then be averaged to obtain the mean value of the complete discharge capacity time-series waveform (including all charge-discharge cycle stages). Alternatively, a standard deviation can be calculated for the capacity statistical features of each discharge capacity time-series waveform. The standard deviation of all discharge capacity time-series waveforms can then be averaged to obtain the standard deviation of the complete discharge capacity time-series waveform. Of course, the standard deviation of the complete discharge capacity time-series waveform can also be directly calculated based on the capacity changes represented by all discharge capacity time-series waveforms. Or, a minimum value can be calculated for the capacity statistical features of each discharge capacity time-series waveform. The minimum value among all the minimum values ​​of the discharge capacity time-series waveforms can be selected as the minimum value of the complete discharge capacity time-series waveform, and so on. Ultimately, the calculated sixth statistical feature includes other statistics such as the mean, standard deviation, minimum, maximum, and median of the complete discharge capacity time-series waveform.

[0033] Correspondingly, the fifth statistical feature of all discharge capacity time-series waveforms and the sixth statistical feature of the complete discharge capacity time-series waveform are integrated to form a multidimensional (N*(5+3)+5) waveform statistical feature.

[0034] Regarding the statistical characteristics of differential waveforms: For each charge-discharge cycle stage, a linearized capacity waveform for the current charge-discharge cycle stage can be constructed based on the corresponding capacity characteristics in the capacity sequence. waveform) and differential capacity waveform ( (Waveform). Accordingly, the feature extraction methods described above for discharge capacity time-series waveforms can be used to extract features from the linearized capacity waveform and the differential capacity waveform, respectively. That is, the seventh statistical feature of each linearized capacity waveform (where the statistic is equivalent to the fifth statistical feature of the discharge capacity time-series waveform) and the eighth statistical feature of the complete linearized capacity waveform (where the statistic is equivalent to the sixth statistical feature of the complete discharge capacity time-series waveform) are calculated, as well as the ninth statistical feature of the differential capacity waveform (where the statistic is equivalent to the fifth statistical feature of the discharge capacity time-series waveform) and the tenth statistical feature of the complete differential capacity waveform (where the statistic is equivalent to the sixth statistical feature of the complete discharge capacity time-series waveform). These will not be described in detail here.

[0035] Finally, the seventh statistical feature of all linearized capacity waveforms, the eighth statistical feature of the complete linearized capacity waveform, the ninth statistical feature of all differential capacity waveforms, and the tenth statistical feature of the complete differential capacity waveform are integrated to form a multidimensional (2*(N*(5+3)+5)) differential waveform statistical feature.

[0036] Based on the three features mentioned above, a high-dimensional feature vector can be obtained, which can be used as the input feature of the inflection point prediction model.

[0037] Step S300: Input the battery's cycle statistical characteristics, waveform statistical characteristics, and differential waveform statistical characteristics into the trained inflection point prediction model to predict the inflection point of the battery's capacity curve.

[0038] The inflection point prediction model can employ different machine learning models, including but not limited to tree models, neural network models, etc. Modifications and refinements made by those skilled in the art to the embodiments of the present invention without departing from the spirit of the present invention still fall within the scope of the invention application patent of the present invention.

[0039] Meanwhile, the training method for the inflection point prediction model can be found below: First, it is necessary to obtain the capacity and internal resistance sequences of multiple batteries throughout their entire life cycle in order to construct several training samples and their corresponding sample labels.

[0040] Specifically, for each battery, the capacity sequence and internal resistance sequence at the beginning of the battery life can be extracted from the capacity sequence and internal resistance sequence of its entire life cycle, and multi-dimensional features can be extracted from them in the manner described above to construct the training samples of the model. That is, the multi-dimensional features extracted at the beginning of the battery life can be used as the sample data of a training sample.

[0041] In addition, sample labels for the corresponding samples need to be constructed based on the capacity sequence of the battery's entire life cycle. Specifically, the capacity sequence of the entire life cycle can be divided into multiple charge-discharge cycle stages based on the number of charge-discharge cycles in the battery's entire life cycle, and then the actual capacity of the battery in the current charge-discharge cycle stage can be calculated based on the amount of charge / discharge or other electrical parameters of the battery in each charge-discharge cycle stage.

[0042] It is understandable that as the number of charge and discharge cycles of a battery increases, the actual capacity of the battery will decrease. Accordingly, a capacity curve of the battery can be constructed based on the relationship between the actual capacity of the battery and the number of cycles corresponding to the current charge and discharge cycle stage, and the capacity curve will show a monotonically decreasing trend.

[0043] Furthermore, the corresponding inflection points can be extracted from the capacity curve constructed based on the capacity sequence of the entire battery life cycle using a geometric inflection point detection method based on biaxial normalization, and these inflection points can be used as the sample labels for the corresponding samples.

[0044] Specifically, the capacity and cycle index (the horizontal and vertical coordinates of the corresponding points) of the capacity curve are first normalized to a unit square, as follows: , , in, , Let represent the normalized result of the capacity at the i-th point on the capacity curve and the circular index, respectively. This represents the initial charge-discharge cycle number at the i-th point on the capacity curve. , These represent the minimum and maximum number of charge-discharge cycles in the early stages of battery life, respectively. This represents the original capacity value at the i-th point on the capacity curve. , These represent the minimum and maximum capacity values ​​at the beginning of the battery's lifespan, respectively.

[0045] Then, based on the normalized point coordinates Calculate the perpendicular distance from the point to the preset reference line (y=-x+1). This can be understood as the points with coordinates (0, 1) and (1, 0) forming the reference line y=-x+1.

[0046] The normalized point coordinates Vertical distance from the reference line as follows: , Correspondingly, based on the vertical distance of all points to the reference line, the point that is furthest from the reference line, i.e., the corresponding inflection point, can be identified and used as the sample label of the corresponding sample.

[0047] Secondly, the training samples constructed above are multidimensional features extracted from the capacity sequence and internal resistance sequence at the beginning of the battery life. In order to reduce the dimensionality of the sample data, remove redundancy, and improve the efficiency and interpretability of the model, the training samples can be re-optimized accordingly.

[0048] To address this, all training samples can be input into a sample optimization model to determine the global importance of each feature used to construct the training samples, including SHAP contribution and XGBoost contribution.

[0049] The global importance mentioned here refers to the contribution of a certain feature type to the global picture. For example, the above example extracts ten statistical features from the capacity and internal resistance sequences at the beginning of the battery life to construct training samples. The contribution of these ten features to the global picture varies.

[0050] It is also understandable that the global importance determined here is for this type of feature, rather than the detailed data of that feature in the current sample. That is, it determines the global importance of the first statistical feature, the second statistical feature, etc. used to construct the training samples, rather than the global importance of the first statistical feature, the second statistical feature, etc. in each sample.

[0051] In addition, the preferred sample optimization model is the XGBoost regression model, which is an ensemble learning model based on regression trees. It uses gradient boosting to stack multiple regression trees in rounds to achieve inflection point prediction.

[0052] The XGBoost model consists of T Classification and Regression Trees (CARTs) as base learners, and the overall prediction can be expressed as a weighted sum of the outputs of each base learner:

[0053] in, The input feature vector, For the t-th regression tree, The learning rate is used to scale the contribution of each new regression tree to the total output. The number of base learners (regression trees).

[0054] Every returning tree Internally, the tree consists of two types of nodes: internal nodes and leaf nodes. Internal nodes perform a binary partitioning based on a certain feature dimension and a threshold, routing samples to either the left or right child node. Leaf nodes store a real-valued leaf weight, representing the tree's contribution to the regression output of samples falling into that leaf. The complexity of the tree is constrained by hyperparameters such as maximum depth.

[0055] Accordingly, each regression tree is trained sequentially using a boosting approach. The first tree fits the initial target, and each subsequent tree fits the direction of the current model's residuals (guided by the gradient information of the loss function). The output of this tree is then superimposed onto the existing model according to the learning rate. Therefore, the overall output of the model is a linear superposition of the outputs of each tree.

[0056] Based on the above, after training the XGBoost base model using all training samples, the SHAP contribution of each feature dimension used to construct the training samples can be determined. and XGBoost contribution as follows: , , in, , Let represent the SHAP contribution and XGBoost contribution, respectively, used to construct the j-th dimension feature of the training samples. Indicates the number of training samples. This represents the SHAP value of the j-th dimension feature in the i-th training sample. This indicates the number of regression trees in the model. Let represent the set of all split nodes in the t-th regression tree that use the j-th dimension as the splitting feature. This represents the loss-related value in the objective function of the corresponding parent node before the split at the s-th node. This represents the loss-related value in the objective function at the left child node after the split at the s-th node. This represents the value related to the loss in the objective function on the right child node after the split at the s-th node.

[0057] After determining the global importance of each feature used to construct the training samples, in order to reduce the probability of randomly related features being selected in the training samples and to ensure that the final features are interpretable and can correspond to the degradation mechanism, features related to capacity and internal resistance are retained first, while features with weak correlation or poor interpretability can be removed, that is, the above training samples are re-optimized.

[0058] To address this, the global importance of each feature used to construct the training samples can be ranked accordingly. For example, the ten statistical features used to construct the training samples can be ranked, and the top-ranked statistical features can be selected to reconstruct the training samples, thereby removing other statistical features from the original training samples and thus optimizing the training samples.

[0059] It is understandable that the above examples only illustrate statistical features. In reality, the global importance of each dimension of each statistical feature can be determined, and these dimensions can be ranked to select the optimal one-dimensional or multi-dimensional features to reconstruct the training samples. Meanwhile, the optimized training samples still use the previously determined sample labels.

[0060] Finally, all optimized training samples are used as inputs into the inflection point prediction model, and the inflection point prediction model is optimized in reverse based on the regression prediction values ​​output by the inflection point prediction model and the sample labels corresponding to each optimized training sample.

[0061] In one specific embodiment, the inflection point prediction model can also employ the XGBoost model, and the regression tree within the XGBoost model is as follows: Figure 2 As shown, by inputting several battery samples (i.e., the optimized training samples mentioned above) into the XGBoost model, the regression tree in it is fine-tuned.

[0062] In this model, the root node represents the starting point of the regression tree and contains all battery samples; internal nodes represent splitting the samples based on a certain feature and threshold; multiple branches represent the paths corresponding to the judgment results; and leaf nodes indicate that further splitting has ceased and output a numerical contribution of the regression tree. Finally, the XGBoost model sums the leaf values ​​of all regression trees to obtain the inflection point prediction result.

[0063] It should be added that other regression tree models or neural network models can also be used for inflection point prediction, and there are no restrictions on this. This embodiment only uses XGBoost as an example. Modifications and refinements made by those skilled in the art to the embodiments of the present invention without departing from the spirit of the present invention still fall within the scope of the invention application patent of the present invention.

[0064] It should be noted that the steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they contain the same logical relationship, they are all within the scope of protection of this patent. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this patent.

[0065] Example 2 Please see Figure 3 As shown, this application also provides an early prediction system for the inflection point of a battery capacity curve, comprising: The data acquisition module 10 is used to acquire the capacity sequence and internal resistance sequence at the beginning of the battery life.

[0066] The feature extraction module 20 is used to extract multidimensional features from the capacity sequence and the internal resistance sequence, including: dividing the capacity sequence and the internal resistance sequence into several charge-discharge cycle stages according to the number of charge-discharge cycles in the early stage of battery life, and calculating the cycle statistical features, waveform statistical features and differential waveform statistical features of the battery based on the capacity features and internal resistance features of the battery in each charge-discharge cycle stage.

[0067] The inflection point prediction module 30 is used to input the battery's cycle statistical characteristics, waveform statistical characteristics, and differential waveform statistical characteristics into the trained inflection point prediction model in order to predict the inflection point of the battery's capacity curve.

[0068] It should be noted that the early prediction system for the inflection point of the battery capacity curve provided in the above embodiments and the early prediction method for the inflection point of the battery capacity curve provided in Embodiment 1 belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the early prediction method for the inflection point of the battery capacity curve provided in Embodiment 1 can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above, and no limitation is imposed here.

[0069] Example 3 Please see Figure 4As shown, this application also provides an electronic device, including a memory 2, a processor 1, and a program stored in the memory and executable on the processor, wherein the processor executes the steps of any of the methods described above.

[0070] The memory includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory, magnetic storage, disk, optical disk, etc. In some embodiments, the memory can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory can be an external storage device of the electronic device, such as a plug-in portable hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, Flash Card, etc. Furthermore, the memory can include both internal and external storage units of the electronic device. The memory can be used not only to store application software and various types of data installed on the electronic device, but also to temporarily store data that has been output or will be output.

[0071] In some embodiments, the processor may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory and calls data stored in the memory to perform various functions and process data of the electronic device. The processor executes the operating system and various installed application programs of the electronic device. The processor executes the application programs to implement the steps in the above method embodiments.

[0072] For example, the program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of program instruction segments capable of performing a specific function, which describe the execution process of the program in the electronic device.

[0073] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute some of the functions of the various embodiments of the present invention.

[0074] In summary, this invention only requires collecting capacity and internal resistance data at the beginning of the battery's lifespan to accurately predict the inflection point of the battery's capacity curve. It is suitable for battery management systems with limited resources, thereby solving the problem of being unable to accurately predict the battery capacity inflection point due to limited sensor configuration and short data acquisition cycles.

[0075] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for early prediction of the inflection point of a battery capacity curve, characterized in that, include: Obtain the capacity sequence and internal resistance sequence at the beginning of the battery life; Extracting multidimensional features from the capacity sequence and the internal resistance sequence includes: dividing the capacity sequence and the internal resistance sequence into several charge-discharge cycle stages based on the number of charge-discharge cycles in the early stage of battery life, and calculating the cycle statistical features, waveform statistical features, and differential waveform statistical features of the battery based on the capacity characteristics and internal resistance characteristics of the battery in each charge-discharge cycle stage. The battery's cycle statistics, waveform statistics, and differential waveform statistics are input into a trained inflection point prediction model to predict the inflection point of the battery's capacity curve.

2. The method for early prediction of the inflection point of the battery capacity curve according to claim 1, characterized in that, The steps for calculating the cyclic statistical features include: The first statistical characteristic is calculated based on the discharge capacity of each charge-discharge cycle. The second statistical characteristic is calculated based on the charging capacity of each charge-discharge cycle. The third statistical characteristic is calculated based on the internal resistance value of each charge-discharge cycle stage; The fourth statistical characteristic is calculated based on the ratio of the actual capacity to the corresponding internal resistance value in each charge-discharge cycle. The first statistical feature, the second statistical feature, the third statistical feature, and the fourth statistical feature are integrated to form the cyclic statistical feature.

3. The method for early prediction of the inflection point of the battery capacity curve according to claim 1, characterized in that, The steps for calculating the waveform statistical characteristics include: For any charge-discharge cycle stage, a corresponding discharge capacity time-series waveform is constructed based on the capacity characteristics therein, so as to calculate the capacity statistical characteristics and waveform statistical characteristics of the discharge capacity time-series waveform; The capacity statistical characteristics and waveform statistical characteristics of all discharge capacity time-series waveforms are integrated to form the fifth statistical characteristic; The capacity statistical characteristics of each discharge capacity time-series waveform are aggregated twice to calculate the sixth statistical characteristic; The fifth statistical feature and the sixth statistical feature are combined to form the waveform statistical feature.

4. The method for early prediction of the inflection point of the battery capacity curve according to claim 1, characterized in that, The steps for calculating the statistical characteristics of the differential waveform include: For any charge-discharge cycle stage, a corresponding linearized capacity waveform is constructed based on the capacity characteristics therein, so as to calculate the capacity statistical characteristics and waveform statistical characteristics of the linearized capacity waveform; The capacity statistical features and waveform statistical features of all linearized capacity waveforms are integrated to form the seventh statistical feature; The capacity statistics of each linearized capacity waveform are aggregated twice to calculate the eighth statistical feature; For any charge-discharge cycle stage, a corresponding differential capacity waveform is constructed based on the capacity characteristics therein, so as to calculate the capacity statistical characteristics and waveform statistical characteristics of the differential capacity waveform; The capacity statistical characteristics and waveform statistical characteristics of all differential capacity waveforms are integrated to form the ninth statistical characteristic; The capacity statistical characteristics of each differential capacity waveform are aggregated twice to calculate the tenth statistical characteristic; The seventh statistical feature, the eighth statistical feature, the ninth statistical feature, and the tenth statistical feature are integrated to form the differential waveform statistical feature.

5. The method for early prediction of the inflection point of the battery capacity curve according to claim 1, characterized in that, The training steps for the inflection point prediction model include: Obtain the capacity and internal resistance sequences of multiple batteries throughout their entire lifecycle. Construct corresponding training samples and sample labels based on the capacity and internal resistance sequences of each battery, including: For each battery: Based on the number of charge-discharge cycles in the early stage of battery life, the capacity sequence and internal resistance sequence in the early stage of battery life are divided into several charge-discharge cycle stages. Based on the capacity characteristics and internal resistance characteristics of the battery in each charge-discharge cycle stage, the cycle statistical characteristics, waveform statistical characteristics and differential waveform statistical characteristics of the battery are calculated to construct training samples. A capacity curve of the battery is constructed based on the capacity sequence of the battery's entire life cycle and the corresponding charge-discharge cycle number. The capacity and cycle index of each point on the capacity curve are normalized. The vertical distance from each point to the preset reference line is calculated based on the normalized point coordinates, and the inflection point of the capacity curve is determined accordingly to serve as a sample label. The inflection point represents the point on the capacity curve that is farthest from the preset reference line. All training samples are input into a pre-built inflection point prediction model, and the inflection point prediction model is optimized in reverse based on the regression prediction value output by the inflection point prediction model and the sample labels of the training samples.

6. The method for early prediction of the inflection point of the battery capacity curve according to claim 5, characterized in that, The steps of inputting all training samples into a pre-built inflection point prediction model and then back-optimizing the inflection point prediction model based on the regression prediction values ​​output by the inflection point prediction model and the sample labels of the training samples include: All training samples are input into a sample optimization model to determine the global importance of each feature dimension used to construct the training samples; wherein the sample optimization model adopts an extreme gradient boosting regression model, and the global importance is used to represent the contribution of the current dimension feature type to the global value, including at least SHAP contribution and XGBoost contribution. Based on the global importance of each feature dimension, the multi-dimensional features used to construct the training samples are redefined, and all training samples are re-optimized. The optimized training samples are input into the pre-built inflection point prediction model, and the inflection point prediction model is back-optimized based on the regression prediction values ​​output by the inflection point prediction model and the sample labels corresponding to the optimized training samples.

7. The method for early prediction of the inflection point of the battery capacity curve according to claim 6, characterized in that, The formula for calculating the SHAP contribution is as follows: ， in, This represents the SHAP contribution of the j-th dimension feature used to construct the training samples. Indicates the number of training samples. This represents the SHAP value of the j-th dimension feature in the i-th training sample.

8. The method for early prediction of the inflection point of the battery capacity curve according to claim 6, characterized in that, The formula for calculating the XGBoost contribution is as follows: ， in, This represents the XGBoost contribution of the j-th dimension feature used to construct the training samples. This indicates the number of regression trees in the sample optimization model. Let represent the set of all split nodes in the t-th regression tree that use the j-th dimension as the splitting feature. This represents the loss-related value in the objective function of the corresponding parent node before the split at the s-th node. This represents the loss-related value in the objective function at the left child node after the split at the s-th node. This represents the value related to the loss in the objective function on the right child node after the split at the s-th node.

9. The method for early prediction of the inflection point of the battery capacity curve according to claim 6, characterized in that, The steps of redetermining the multidimensional features used to construct training samples based on the global importance of each feature, and then re-optimizing all training samples, include: The initial multidimensional features used to construct training samples are sorted according to their global importance, and the top N features are selected as the final multidimensional features used to construct training samples.

10. An early prediction system for the inflection point of a battery capacity curve, characterized in that, include: The data acquisition module is used to acquire the capacity sequence and internal resistance sequence at the beginning of the battery life. The feature extraction module is used to extract multidimensional features from the capacity sequence and the internal resistance sequence, including: dividing the capacity sequence and the internal resistance sequence into several charge-discharge cycle stages according to the number of charge-discharge cycles in the early stage of battery life, and calculating the cycle statistical features, waveform statistical features and differential waveform statistical features of the battery based on the capacity features and internal resistance features of the battery in each charge-discharge cycle stage. The inflection point prediction module is used to input the battery's cycle statistical characteristics, waveform statistical characteristics, and differential waveform statistical characteristics into the trained inflection point prediction model in order to predict the inflection point of the battery's capacity curve.

Images