A battery life prediction method

Abstract

This invention discloses a battery life prediction method, belonging to the field of power lithium-ion battery technology. The method constructs a dataset by conducting cyclic tests on batteries at different acceleration factor levels with the same acceleration factor type. Users input parameters such as model type and acceleration factor according to prediction requirements. The experimental data is reorganized, and a global optimization strategy based on parameter space discretization search is used to fit the capacity decay model parameters, thereby predicting the capacity decay curve and error range of the target battery at a specified acceleration factor level. Finally, the accuracy of the prediction results is verified by combining measured data, and a visual analysis chart is output. This invention effectively improves prediction efficiency and stability through data modeling across acceleration factor levels and a global optimization strategy. It has the advantages of simple operation, strong applicability, and high prediction accuracy, and has broad application value in lithium-ion battery life assessment and engineering applications.

Description

Technical Field

[0001] This invention belongs to the field of power lithium-ion battery technology, specifically relating to a battery life prediction method. Background Technology

[0002] Lithium-ion batteries, with their high energy density, long lifespan, and environmental advantages, have become one of the core technologies for the global new energy sector. With the rapid rise of numerous new energy battery companies, expectations for lithium-ion battery performance are also increasing. However, the pace of technological advancement and iteration in lithium-ion batteries lags far behind the growing demands for their electrical performance. One major reason for this is the significant cost involved in testing and evaluating battery lifespan under various operating conditions during the research and development process. To address this bottleneck, various technologies for accelerating lifespan prediction have emerged, aiming to effectively improve research and development efficiency by shortening the development cycle and reducing costs. These methods not only accelerate the development process of lithium-ion batteries but also provide new possibilities for meeting the market's demand for high-performance batteries.

[0003] Previous studies have typically used acceleration factors such as temperature, charge / discharge rate, and state-of-charge (SOC) operating range to accelerate the prediction of lithium-ion battery degradation. Accurately describing the relationship between capacity degradation and cycle life is fundamental to exploring efficient methods for accelerating degradation. To this end, researchers have proposed various empirical models, such as power-law and single-term exponential models suitable for linear degradation, and double-logarithmic and double-power-function models for superlinear degradation. Different degradation models and acceleration factors exhibit complex relationships and calculation processes. Furthermore, during the fitting process, issues such as the fitting curve being significantly affected by initial values, failure to fully account for errors caused by battery inconsistencies, and computational errors due to operational complexity often arise. Summary of the Invention

[0004] The purpose of this invention is to provide a battery life prediction method based on multiple acceleration factors, which reasonably quantifies the prediction error caused by battery inconsistencies, and uses a global optimization strategy algorithm based on parameter space discretization search to avoid the influence of initial values, thus efficiently solving the problem of accelerated life prediction under multiple operating conditions in the research and development of lithium-ion batteries.

[0005] The objective of this invention can be achieved through the following technical solution: a battery life prediction method, comprising the following steps:

[0006] S1. Data Acquisition: Select several batteries as battery samples, divide the battery samples into at least two experimental groups, set different levels of the same acceleration factor and conduct charge-discharge cycle tests on the experimental groups respectively, record the cycle test data of the battery samples in each experimental group, and summarize the cycle test data to form a known battery dataset.

[0007] S2. Model parameter configuration: Select the capacity decay model type, the same acceleration factor type and level as in step S1, the known battery dataset, the acceleration factor level of the target battery to be predicted, and the State of Health (SOH) prediction interval of the target battery.

[0008] S3. Model Training and Prediction: Based on parallel sample data of the known battery dataset, data recombination is performed across experimental groups. A global optimization strategy based on parameter space discretization search is used to fit the parameters of the capacity decay model. Based on the fitting results, the capacity decay curve of the target battery at a given acceleration factor level is predicted. Based on the capacity decay curve, the lifetime and error range of the target battery are determined.

[0009] This invention provides a comprehensive battery lifetime prediction method. Its purpose is to establish an executable and scalable lifetime prediction framework, enabling lifetime data obtained under different accelerated testing conditions to be used for predicting the capacity degradation trend of target batteries. Its advantages are twofold: firstly, it allows for forward-looking estimation of battery lifetime under target operating conditions using limited accelerated testing data, shortening the testing cycle and reducing experimental costs; secondly, by introducing acceleration factors and capacity degradation models, the prediction results possess good engineering applicability and transferability, making them suitable for battery R&D, screening, and lifetime management scenarios.

[0010] As a further aspect of the present invention, it also includes S4. Result verification: Under the type and level of the acceleration factor, the measured cycle test data of the target battery is obtained through charge-discharge cycle test; the measured cycle test data is compared with the predicted capacity decay curve, and corresponding comparison charts and verification result data are generated.

[0011] A result verification step has been added. Its purpose is to verify the accuracy and reliability of the model's predictions by actually comparing the predicted capacity decay curve with the measured cycle test data of the target battery. Its advantage lies in enhancing the closed-loop nature and verifiability of the entire method. It not only outputs the prediction results but also visually demonstrates the consistency between the prediction and the actual measurements through comparative charts and verification data. This facilitates technical personnel's evaluation of whether the model meets practical application requirements and also helps with subsequent model correction and parameter optimization.

[0012] As a further embodiment of the present invention: the acceleration factor type in step S1 includes temperature, charge / discharge rate and depth of discharge.

[0013] The types of acceleration factors in step S1 are explicitly listed, which serves to define and explain the typical sources of acceleration factors to which this invention is applicable, including temperature, charge / discharge rate, and depth of discharge. Its advantage lies in clarifying the correspondence between the method and the battery aging mechanism, making the scheme closer to the actual battery degradation process.

[0014] As a further aspect of the present invention: the cycle test data of the battery sample includes discharge capacity and number of cycles or cycle time.

[0015] The specific format of the cyclic test data is explained, clarifying that the data input for this method includes discharge capacity and the number of cycles or cycle time, thus defining multiple optional dimensions for lifetime data acquisition. Its advantage lies in enhancing the method's data compatibility and applicability, enabling it to adapt to different testing platforms, experimental recording methods, and evaluation standards, thereby improving its versatility in laboratory research and industrial testing.

[0016] As a further aspect of the present invention: the general form of the capacity decay model is as follows:

[0017]

[0018] in, Indicates health status This is the ratio of the discharge capacity to the nominal capacity at the current number of cycles or cycle time, i.e. , This represents the battery's discharge capacity during the current cycle. This refers to the battery's nominal capacity. This refers to the number of cycles or the cycle time. This represents the parameter vector of the capacity decay model to be fitted. Common capacity decay models include:

[0019] Power-law model: ;

[0020] Single-index model: ;

[0021] Bi-index model: ;

[0022] Double power function model: ;

[0023] in , , , These are the parameters of the capacity decay model to be fitted.

[0024] The optional types of capacity decay models are listed, and the general form of the model is given. The definition serves to provide a clear mathematical foundation for model building. Its advantage lies in combining theoretical integrity with application flexibility. It can adapt to the degradation behavior of different types of batteries and different aging stages, and it can also form a calculable and fittable relationship between the acceleration factor and the degradation model, thereby improving the pertinence and accuracy of the prediction.

[0025] As a further aspect of the present invention: the capacity decay model parameter vector Level between acceleration factor Functional dependencies exist: This includes exponential relationships, linear or quadratic relationships.

[0026] This invention is not limited to a single model, but can select exponential, linear, or quadratic models according to the battery degradation characteristics, and further establish the functional dependency relationship between model parameters and acceleration factors.

[0027] As a further embodiment of the present invention: the model training and prediction in step S3 includes:

[0028] S31. Data Restructuring:

[0029] Based on the known battery dataset, let A set of levels under the same acceleration factor type, where The total number of the acceleration factor levels. Represents the level set The first in The acceleration factor level;

[0030] From the level set Optional The acceleration factor level ( ), and record it as the selected index set. ,in , indicating the selected first The acceleration factor level exist The index in; each selected index The index level is From the index level Select one parallel sample data independently , forming a combination of training data ,in Indicated at the index level The selected parallel sample data sequence numbers satisfy the following conditions: , Indicates the index level The total number of parallel sample data;

[0031] Iterate through all possible sets of selected indices. and each of the selected index sets The training dataset is constructed by combining all possible parallel sample data indices. ,Right now:

[0032]

[0033] S32. Parameter Fitting:

[0034] For all possible combinations of the training data Each of the following steps involves running a fitting process based on a global optimization strategy using parameter space discretization search, for the combination of training data. traverse each of the internal index levels Using the sample data, fit the globally optimal model parameters. And construct the training data combination. Global optimal model parameter combination By traversing all the aforementioned training data combinations Finally, a complete set of globally optimal model parameters is constructed. :

[0035]

[0036] S33. Predictive Analysis:

[0037] Based on each set of globally optimal model parameter combinations According to the capacity decay model parameter vector There is a functional dependency between it and the acceleration factor. The acceleration factor levels of each of the above are... and the acceleration factor level Optimal model parameters As data points, the functional relationship between the fitting parameters and the level is used to establish a system of equations, which can be solved to obtain the target battery at a given acceleration factor level. Below , This represents the capacity decay model parameter vector obtained for the target battery. By traversing the training data combinations, the capacity decay model parameter set for the target battery is constructed.

[0038]

[0039] Substitute into the capacity decay model This leads to a set of predicted capacity decay curves:

[0040]

[0041] The upper bound of the capacity decay curve and the lower world The error range for the predicted battery sample lifetime. ,Right now;

[0042]

[0043] The upper bound and the lower world The arithmetic mean of the values ​​is used as the benchmark for the predicted capacity decay curve of the battery sample at the acceleration factor level, i.e.:

[0044] .

[0045] The model training and prediction process in step S3 has been refined, explicitly including three sub-steps: data recombination, parameter fitting, and predictive analysis. Its purpose is to combine and reconstruct parallel sample data to form multiple training sample sets, and then generate the prediction curve and error range of the target battery based on each set of parameters. Its advantage lies in fully utilizing the information differences between parallel samples, more comprehensively reflecting sample discreteness and experimental volatility, avoiding random errors caused by relying solely on a single training result, and outputting results through upper and lower bounds and mean curves, providing not only "point estimates" but also "interval estimates," which is more conducive to practical decision-making and risk assessment.

[0046] As a further aspect of the present invention: in step S32, for each set of training data combinations The optimal combination of model parameters is determined by a global optimization strategy of parameter space discretization search. Specifically, this includes: constructing a grid of initial parameter values; and fitting the desired initial values ​​to the desired parameters. dimensional parameter vector ,

[0047] Within the preset parameter range Inside, among them, and They represent the first The lower and upper bounds of the model parameters are set, with a step size. For the first The dimensional parameters are discretized at equal intervals to generate the 3rd dimension. Set of candidate initial values ​​under dimension This allows us to construct a grid of all candidate initial values.

[0048]

[0049] Build a regression model and evaluate its performance: Each set of candidate initial parameters Based on the capacity decay model Simultaneous fitting The different acceleration factor levels The following sample data, for each of the aforementioned acceleration factor levels Establish a regression model:

[0050]

[0051] in, It is the first at that level Actual measurements at a certain number of cycles or cycle time , To measure the number of cycles or cycle time, For residuals, The capacity decay model is used; the acceleration factor level is obtained by minimizing the sum of squared residuals through a fitting optimization algorithm. Based on the candidate initial parameters Obtained candidate model parameters ; Traverse the candidate initial value grid to finally construct the acceleration factor level. Candidate model parameter grid ;

[0052] For each of the aforementioned acceleration factor levels Iterate through the corresponding candidate model parameter grid. Calculate based on the parameters of each candidate model Goodness-of-fit metrics are used to evaluate model performance. Commonly used goodness-of-fit metrics include the coefficient of determination. and mean square error :

[0053]

[0054]

[0055] in, This represents the total amount of data in the sample at this acceleration factor level. For the first acceleration factor level Model prediction under a certain number of cycles or cycle time value, This is the actual measurement at this acceleration factor level. The average value, ;

[0056] Selecting the globally optimal parameter combination: For the candidate model parameter grid at all acceleration factor levels By combining these parameters, a grid of candidate model parameters for all levels can be constructed:

[0057]

[0058] Traversing the candidate model parameter grid The combination of The parameter combination that optimizes the average goodness-of-fit index of the sample data in the training data set is selected as the globally optimal model parameter combination for that data set. :

[0059]

[0060] or

[0061] .

[0062] This further clarifies the specific implementation of the global optimization strategy in step S32, namely, generating candidate parameter combinations through parameter space discretization search, establishing a regression model and evaluating it using the coefficient of determination or mean squared error, and then selecting the optimal parameter combination. Its purpose is to provide a directly implementable parameter solution mechanism, giving the model parameter fitting process a clear operational path. Its advantage lies in the fact that discretized global search is less prone to getting trapped in local optima compared to local optimization methods, and is particularly suitable for nonlinear, non-convex capacity decay model parameter identification problems; simultaneously, it combines... and Common evaluation metrics such as goodness of fit and error magnitude can be used to select the optimal solution and improve the reliability of model parameter estimation.

[0063] As a further embodiment of the present invention: in step S32, during the process of determining the optimal model parameters for each combination using the global optimization strategy, shared constraints are set for the model parameters that do not change with the acceleration factor level, so that the model parameters remain consistent among different battery samples under the combination; while the model parameters that change with the acceleration factor level are independently fitted according to different values ​​of the acceleration factor level.

[0064] The handling of different model parameters during parameter fitting is further defined, distinguishing between "model parameters that do not change with the acceleration factor" and "model parameters that change with the acceleration factor." The former uses shared constraints, while the latter is fitted independently based on different acceleration factor values. Its advantage lies in this structured constraint method, which reduces the number of unnecessary free parameters, prevents overfitting, and enhances the physical consistency and interpretability of model parameters; simultaneously, it preserves the sensitivity of acceleration factor-related parameters to different operating conditions, thus achieving a better balance between model complexity and prediction accuracy.

[0065] As a further aspect of the present invention, the parameter estimation methods for the regression model include: Newton's method, Levenberg-Marquardt algorithm, and gradient descent method.

[0066] The paper lists various possible implementation methods for regression model parameter estimation, which serve to supplement the explanation of the multiple optimization techniques that can be used at the specific solution level, providing suitable solution tools for different model structures and data characteristics. Its advantages lie in enhancing the flexibility and engineering feasibility of the method implementation: for example, Newton's method is suitable for fast convergence, the Levenberg-Marquardt algorithm is suitable for nonlinear least squares problems, and the gradient descent method is convenient for handling large-scale data and iterative optimization. Therefore, the most suitable parameter estimation strategy can be selected according to actual needs to improve fitting efficiency and stability.

[0067] As a further aspect of the present invention: the method is applicable to lithium-ion batteries with various packaging forms and collected materials, including square batteries, pouch batteries and cylindrical batteries, as well as batteries using lithium iron phosphate (LFP), lithium cobalt oxide (LiCoO2), lithium nickel cobalt aluminum oxide (Li(NiCoAl)O2,NCA) or nickel manganese cobalt ternary material (Li(NiMnCo)O2,NCM) cathode materials.

[0068] The applicable objects of the method of this invention have been further explained, clarifying that the method is applicable to lithium-ion batteries with various packaging forms and various cathode material systems. Its purpose is to expand the application boundaries of this invention, indicating that this lifetime prediction method is not limited to a specific battery structure or chemical system, but can cover different packaging types such as prismatic batteries, pouch batteries, and cylindrical batteries, as well as mainstream material systems such as LFP, lithium cobalt oxide, NCA, and NCM. Its advantage lies in significantly enhancing the versatility, compatibility, and industrial adaptability of this invention, enabling the method to meet the lifetime prediction needs of different application scenarios such as power batteries, energy storage batteries, and consumer batteries.

[0069] The beneficial effects of this invention are:

[0070] (1) By introducing an acceleration factor, the present invention obtains cyclic test data of battery samples at different acceleration factor levels, and combines the capacity decay model to perform parameter fitting and lifetime prediction, which can more effectively reveal the decay law of the battery under different working conditions. At the same time, by recombining and training parallel sample data and outputting the error range of the prediction curve, the prediction results not only have high accuracy, but also have good stability and reliability.

[0071] (2) This invention uses accelerated test data to extrapolate and predict the capacity decay behavior of the target battery under specified conditions. It can obtain the life assessment results without relying entirely on long-term natural aging tests, thereby significantly shortening the battery life test time, reducing test resource investment and manpower costs. It is particularly suitable for scenarios with high efficiency requirements such as battery R&D, screening verification and batch testing.

[0072] (3) This invention is applicable to lithium-ion batteries with various packaging forms and various material systems, and supports various capacity decay models and parameter solution methods. It can be flexibly configured according to different battery types, different acceleration factors and different experimental conditions. Therefore, it has good compatibility, scalability and industrial adaptability, and can be widely used in the life prediction and health management of power batteries, energy storage batteries and consumer batteries. Attached Figure Description

[0073] Figure 1 This is a flowchart of the present invention;

[0074] Figure 2 The results of accelerated prediction of lithium iron phosphate batteries at different temperatures in this invention are shown.

[0075] Figure 3 The results show the accelerated prediction of lithium iron phosphate batteries based on different charge and discharge rates in this invention. Detailed Implementation

[0076] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0077] Example 1

[0078] As attached Figure 1 As shown, a battery life prediction method, in Example 1, uses the number of cycles as a lifespan characterization, and includes the following steps:

[0079] S1. Experimental Data Acquisition

[0080] As shown in Table 1, three lithium iron phosphate batteries were selected and divided into two experimental groups. The acceleration factor was set to temperature. One battery was placed in a constant temperature chamber at 45°C, and two batteries were placed in a constant temperature chamber at 55°C. Both batteries were charged and discharged at the same set rate (0.5C). The number of cycles and discharge capacity were recorded to form a known battery dataset.

[0081] Table 1. Detailed information on the battery pack in Example 1

[0082]

[0083] S2. Model Parameter Configuration

[0084] The capacity decay model is set as a double power function model. Acceleration factor type is temperature Input the measured cycle test data and acceleration factor level of the known battery under different acceleration factor conditions (45℃ and 55℃), the acceleration factor level of the target battery (35℃), and the target SOH range (60%~80%) to be predicted.

[0085] Where model parameters , , Its value varies with temperature Changes, determined by the fitted parameters , , , , , Decide; These are model parameters that have no functional relationship with the acceleration factor level and are used to set shared constraints. Fitting parameters. , , , , , It can be solved by solving a set of simultaneous equations.

[0086] S3. Model Training and Prediction

[0087] Based on parallel sample data of known batteries at different acceleration factor levels, this embodiment uses a "pair-up" method to recombine the data. Specifically, one battery is selected from the 45℃ experimental group (1 battery) and the 55℃ experimental group (2 batteries) to form two sets of training data combinations: (45℃-0.5C-#1&55℃-0.5C-#1) and (45℃-0.5C-#1&55℃-0.5C-#2).

[0088] For each data combination, a global optimization strategy based on parameter space discretization search is employed to determine the optimal capacity decay model parameters for the two batteries under each combination. , , , Specifically, regarding the parameters of the capacity decay model... , , , Within a preset parameter range, a set of candidate parameter combinations is generated with a fixed step size. Data from each combination, along with data at acceleration factor levels of 45℃ and 55℃, are then simultaneously fitted, and shared parameters are applied. Given the constraints, the Levenberg-Marquardt algorithm is used to obtain a grid of candidate model parameters in the discretized parameter space. Based on this grid of candidate model parameters, the average value of the sample data at multiple levels in this training data combination is calculated. The largest parameter combination is taken as the globally optimal model parameter combination for that data combination.

[0089] Under this set of globally optimal model parameters, and considering the functional relationship between the acceleration factor level and the model parameters, the fitting parameters can be obtained by solving the system of equations simultaneously. , , , , , By substituting the specified acceleration factor level of 35°C, the capacity decay model parameters at the specified level can be obtained, and thus the predicted capacity decay curve can be obtained.

[0090] The set of capacity decay curves predicted by combining multiple sets of data is summarized, and the maximum and minimum predicted capacities at each cycle point are taken as the upper and lower bounds of the target battery capacity decay curve, respectively, serving as the estimation error range. Then, the arithmetic mean of the upper and lower bounds is used as the benchmark for the predicted capacity decay curve of the target battery at this acceleration factor level.

[0091] S4. Validation of Prediction Results

[0092] As shown in Table 1, under the set acceleration factor conditions (35℃), the target battery was charged and discharged at the same rate (0.5C) to obtain the number of cycles and discharge capacity. The measured data were compared and verified with the lifetime curve output by the prediction model, and corresponding comparison charts and verification results data were generated.

[0093] like Figure 2 As shown, Figure 2The measured lifetime degradation curves of all known batteries and the target battery are presented, and the sources of parallel sample combinations corresponding to the upper and lower boundaries of the prediction interval are clarified. The degradation curves of the target battery predicted by this method all fall within the predicted interval and match the measured lifetime curves.

[0094] Furthermore, Table 2 lists the predicted and experimental values ​​of the target battery's lifespan under 80% SOH conditions. As shown in Table 2, the relative error between the measured value and the baseline value in the predicted value for the target battery is only 0.9% (formula: relative error = |predicted value - measured value| / measured value × 100%). These results clearly demonstrate that this method can predict the lifespan degradation curve of lithium batteries under unknown operating conditions, providing strong support for the development of high-performance lithium batteries.

[0095] Table 2. Comparison of predicted lifetime and experimental lifetime under 80% SOH standard in Example 1

[0096]

[0097] Example 2

[0098] Example 2 is also based on the appendix Figure 1 The process is illustrated in the following steps. Example 2 uses cycle time as a lifetime characterization.

[0099] S1. Experimental Data Acquisition

[0100] As shown in Table 3, five lithium iron phosphate batteries were selected and divided into two experimental groups. The acceleration factor type was set to rate capability. Five batteries were placed in a 55°C constant temperature chamber. Three of them were charged and discharged at a 1C rate, and two were charged and discharged at a 1.5C rate. The cycle time and discharge capacity were recorded to form a known battery dataset.

[0101] Table 3. Detailed information on the battery pack in Example 2

[0102]

[0103] S2. Model Parameter Configuration

[0104] The capacity decay model is set as a double power function model. The acceleration factor type is charge / discharge rate. Input the measured cycle test data of the known battery at different acceleration factor levels (1C and 1.5C) and the corresponding acceleration factor levels, the acceleration factor level of the target battery (0.5C), and the target to be predicted. Range (60%~80%).

[0105] Where model parameters , , Its value varies with the multiplier. Changes, determined by the fitted parameters , , , , , Decide; These are model parameters that have no functional relationship with the acceleration factor level and are used to set shared constraints.

[0106] S3. Model Training and Prediction

[0107] Based on parallel sample data of known batteries at different acceleration factor levels, this embodiment uses a "pair-up" method for data recombination. Specifically, one battery is selected from each of the 1C experimental group (3 batteries) and the 1.5C experimental group (2 batteries), forming a total of six training data combinations: (55℃-1C-#1, 55℃-1.5C-#1), (55℃-1C-#1, 55℃-1.5C-#2), (55℃-1C-#2, 55℃-1.5C-#1), (55℃-1C-#2, 55℃-1.5C-#2), (55℃-1C-#3, 55℃-1.5C-#1), and (55℃-1C-#3, 55℃-1.5C-#2).

[0108] For each data combination, a global optimization strategy based on parameter space discretization search is employed to determine the optimal capacity decay model parameters for the two batteries under each combination. , , , Specifically, regarding the parameters of the capacity decay model... , , , Within a preset parameter range, a set of candidate parameter combinations is generated with a fixed step size. Data from each set of 1C and 1.5C acceleration factor levels are then simultaneously fitted, and shared parameters are applied. To address the constraints, the Levenberg-Marquardt algorithm is employed to obtain a grid of candidate model parameters in the discretized parameter space. Based on this grid, the average value of the multi-level sample data in the training data set is calculated. The largest parameter combination is taken as the globally optimal model parameter combination for that data combination.

[0109] Under this set of globally optimal model parameters, and considering the functional relationship between the acceleration factor level and the model parameters, the fitting parameters can be obtained by solving the system of equations simultaneously. , , , , , By substituting the specified acceleration factor level of 0.5C, the capacity decay model parameters at the specified acceleration factor level can be obtained, and thus the predicted capacity decay curves for this dataset can be obtained.

[0110] The set of capacity decay curves predicted by combining multiple sets of data is summarized, and the maximum and minimum predicted capacities at each cycle point are taken as the upper and lower bounds of the target battery capacity decay curve, respectively, serving as the estimation error range. Then, the arithmetic mean of the upper and lower bounds is used as the benchmark for the predicted capacity decay curve of the target battery under the acceleration factor condition.

[0111] S4. Validation of Prediction Results

[0112] As shown in Table 4, under the set acceleration factor (0.5C), the target battery was charged and discharged at the same temperature (55℃) to obtain the cycle time and discharge capacity. The measured data were compared and verified with the lifetime curve output by the prediction model, and corresponding comparison charts and verification results data were generated.

[0113] like Figure 3 As shown, Figure 3 The measured lifetime degradation curves of all known batteries and the target battery are presented, and the sources of parallel sample combinations corresponding to the upper and lower boundaries of the prediction interval are clarified. The degradation curves of the target battery predicted by this method all fall within the predicted interval and match the measured lifetime curves.

[0114] Furthermore, Table 4 lists the predicted and experimental values ​​of the target battery's lifespan under 80% SOH conditions. The data shows that the measured values ​​of the target battery are very close to the baseline values ​​in the predictions, with an error of only 4.2% at 55℃-0.5C-#2. These results fully demonstrate that this method can predict the lifespan degradation curve of lithium batteries under unknown operating conditions, providing strong support for the development of high-performance lithium batteries.

[0115] Table 4. Comparison of predicted lifetime and experimental lifetime under 80% SOH standard in Example 2

[0116]

[0117] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0118] It should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0119] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0120] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a connection that allows communication between them; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0121] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

Claims

1. A method for predicting battery life, characterized in that, Includes the following steps: S1. Data Acquisition: Select several batteries as battery samples, divide the battery samples into at least two experimental groups, set different levels of the same acceleration factor for each experimental group, and conduct charge-discharge cycle tests on the experimental groups respectively, record the cycle test data of the battery samples in each experimental group, and summarize the cycle test data to form a known battery dataset. S2. Model parameter configuration: Select the capacity decay model type, the same acceleration factor type and level as in step S1, the known battery dataset, the acceleration factor level of the target battery to be predicted, and the health state prediction interval of the target battery; S3. Model Training and Prediction: Based on parallel sample data of the known battery dataset, data recombination is performed across the experimental groups. A global optimization strategy based on parameter space discretization search is used to fit the parameters of the capacity decay model. Based on the fitting results, the capacity decay curve of the target battery at a given acceleration factor level is predicted. Based on the capacity decay curve, the lifetime and error range of the target battery are determined.

2. The method according to claim 1, characterized in that, It also includes S4. Result verification: Under the type and level of the acceleration factor, the measured cycle test data of the target battery is obtained through charge-discharge cycle test; the measured cycle test data is compared with the predicted capacity decay curve, and corresponding comparison charts and verification result data are generated.

3. The method according to claim 1, characterized in that, The acceleration factor types in step S1 include temperature, charge / discharge rate, and depth of discharge.

4. The method according to claim 1, characterized in that, The cycle test data for the battery samples includes discharge capacity and number of cycles or cycle time.

5. The method according to claim 1, characterized in that, The general form of the capacity decay model is as follows: in, Indicates the battery's health status This is the ratio of the discharge capacity to the nominal capacity at the current number of cycles or cycle time, i.e. , This represents the battery's discharge capacity during the current cycle. This refers to the battery's nominal capacity. This refers to the number of cycles or the cycle time. This is the parameter vector of the capacity decay model to be fitted.

6. The method according to claim 5, characterized in that, The capacity decay model parameter vector Level between acceleration factor Functional dependencies exist: This includes exponential relationships, linear or quadratic relationships.

7. The method according to any one of claims 1-6, characterized in that, The model training and prediction in step S3 includes: S31. Data Restructuring: Based on the known battery dataset, let A set of levels under the same acceleration factor type, where The total number of the acceleration factor levels. Represents the level set The first in The acceleration factor level; From the level set Optional The acceleration factor level And record it as the selected index set. ,in , indicating the selected first The acceleration factor level exist The index in; each selected index The index level is From the index level Select one parallel sample data independently , forming a combination of training data ,in Indicated at the index level The selected parallel sample data sequence numbers satisfy the following conditions: , Indicates the index level The total number of parallel sample data; Iterate through all possible sets of selected indices. and each of the selected index sets The training dataset is constructed by combining all possible parallel sample data indices. ,Right now: S32. Parameter Fitting: For all possible combinations of the training data Each of the following processes executes a fitting procedure based on a global optimization strategy using parameter space discretization search, for the combination of training data. traverse each of the internal index levels Using the sample data, fit the globally optimal model parameters. And construct the training data combination. Global optimal model parameter combination By traversing all the aforementioned training data combinations Finally, a complete set of globally optimal model parameters is constructed. : S33. Predictive Analysis: Based on each set of globally optimal model parameter combinations According to the capacity decay model parameter vector There is a functional dependency between it and the acceleration factor. The acceleration factor levels of each of the above are... and the optimal model parameters at the aforementioned acceleration factor level As data points, the functional relationship between the fitting parameters and the level is used to establish a system of equations, which can be solved to obtain the target battery at a given acceleration factor level. Below , This represents the capacity decay model parameter vector obtained for the target battery. By traversing the training data combinations, the capacity decay model parameter set for the target battery is constructed. Substitute into the capacity decay model This leads to a set of predicted capacity decay curves: The upper bound of the capacity decay curve and the lower world The error range for the predicted battery sample lifetime. ,Right now: The upper bound and the lower world The arithmetic mean of the values ​​is used as the benchmark for the predicted capacity decay curve of the battery sample at the acceleration factor level, i.e.:

8. The method according to claim 7, characterized in that, In step S32, for each set of training data combinations The optimal combination of model parameters is determined by a global optimization strategy of parameter space discretization search. Specifically, this includes: constructing a grid of initial parameter values; and fitting the desired initial values ​​to the desired parameters. dimensional parameter vector Within the preset parameter range Inside, among them, and They represent the first The lower and upper bounds of the model parameters are set, with a step size. For the The dimensional parameters are discretized at equal intervals to generate the 3rd dimension. Set of candidate initial values ​​under dimension This allows us to construct a grid of all candidate initial values. Build a regression model and evaluate its performance: Each set of candidate initial parameters Based on the capacity decay model Simultaneous fitting The different acceleration factor levels The following sample data, for each of the aforementioned acceleration factor levels Establish a regression model: in, It is the first at that level Actual measurements at a certain number of cycles or cycle time , To measure the number of cycles or cycle time, For residuals, The capacity decay model is used; the acceleration factor level is obtained by minimizing the sum of squared residuals through a fitting optimization algorithm. Based on the candidate initial parameters Obtained candidate model parameters ; Traverse the candidate initial value grid to finally construct the acceleration factor level. Candidate model parameter grid ; For each of the aforementioned acceleration factor levels Iterate through the corresponding candidate model parameter grid. Calculation based on each candidate parameter Goodness-of-fit metrics are used to evaluate model performance. Commonly used goodness-of-fit metrics include the coefficient of determination. and mean square error : in, This represents the total amount of data in the sample at this acceleration factor level. For the first acceleration factor level Model prediction under a certain number of cycles or cycle time value, For the measured values ​​at this acceleration factor level The average value, ; Selecting the globally optimal parameter combination: for all acceleration factor levels Candidate model parameter grid By combining these parameters, a grid of candidate model parameters for all levels can be constructed: (16) Traversing the candidate model parameter grid The combination of The candidate parameter combination that optimizes the average goodness-of-fit index of the sample data in the training data set is selected as the globally optimal model parameter combination for that data set. : or 。 9. The method according to claim 7, characterized in that, In step S32, during the process of determining the optimal model parameters for each combination using the global optimization strategy, shared constraints are set for model parameters that do not change with the acceleration factor level, so that the model parameters remain consistent across different battery samples under that combination; while model parameters that change with the acceleration factor level are independently fitted according to different values ​​of the acceleration factor level.

10. The method according to claim 8, characterized in that, The parameter estimation methods for the regression model include: Newton's method, Levenberg-Marquardt algorithm, and gradient descent method.

Images