Lithium ion battery life prediction method based on machine learning
By constructing a ternary lithium battery dataset and an XGBoost model, and combining temperature and battery type characteristics, the prediction error problem of lithium-ion batteries under multi-chemical systems and variable temperature conditions is solved, achieving efficient and accurate battery life prediction, which is suitable for real-time management of electric vehicles and energy storage systems.
Patent Information
- Application Number
- CN202511281099.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-09
- Publication Date
- 2026-01-16
AI Technical Summary
Existing lithium-ion battery life prediction methods lack generalization ability under multi-chemical systems and variable temperature conditions, resulting in large prediction errors and making them difficult to apply to the real-time management of electric vehicles and energy storage systems.
A ternary lithium battery dataset was constructed, and battery physical and operating condition features were extracted. The XGBoost model was used for training, and hyperparameters were optimized through grid search to determine the optimal data window for remaining life prediction. Temperature and battery type one-hot encoding were combined to improve the model's generalization ability and data efficiency.
Achieved an 11.84% reduction in prediction error on the NCA/NCM/NCM_NCA battery dataset, a 46% improvement over traditional methods, making it suitable for real-time remaining lifetime assessment of electric vehicles and energy storage systems while reducing computational costs.
Smart Images
Figure CN121348093A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of lithium-ion battery health management, and in particular to a data-driven method for predicting remaining life (RUL), applicable to dynamic operating scenarios such as electric vehicles and energy storage power stations. Background Technology
[0002] As a core component of electric vehicles (accounting for 67% of global power battery installations) and energy storage systems, lithium-ion batteries face significant safety and economic challenges due to their lifespan degradation. Traditional battery life prediction methods typically rely on complex physical models, which not only struggle to accurately describe the battery degradation process under different usage conditions but also incur high computational costs, making them difficult to apply in real-time to practical battery management systems.
[0003] In recent years, data-driven methods have attracted widespread attention because they can handle complex nonlinear relationships. Machine learning algorithms, such as Support Vector Machine (SVM), Random Forest (RF), and Deep Learning (DL), can provide highly accurate and reliable battery RUL predictions by learning from historical data, and have become one of the most commonly used and promising tools in battery health management.
[0004] While mainstream data-driven methods (such as the elastic networks proposed by Severson et al.) can utilize early-cycle data, their generalization ability is insufficient in multi-chemical systems and under varying temperature conditions. Specifically: (1) Cross-system failure: Existing models are mostly developed for single battery types (such as lithium iron phosphate LFP). When they are transferred to ternary systems (nickel cobalt aluminum / nickel cobalt manganese, NCA / NCM), the prediction error increases sharply to 21.6% due to the difference in material aging mechanisms (such as NCM high temperature phase transition accelerating decay). (2) Blind spot of dynamic working conditions: The laboratory constant current charge and discharge data cannot fully cover the dynamic stresses such as sudden temperature changes (20℃~60℃) and rapid acceleration / braking in the actual scenario, which leads to the performance degradation of the model in the actual vehicle or energy storage system environment. (3) Low data efficiency: Elastic networks and other models are extremely sensitive to the amount of early cycle data. The prediction error of the first 30 cycles is 180% higher than that of the first 100 cycles, and the error increases by 3 to 5 percentage points when the sample size is reduced by 50%, which seriously restricts the application in low data scenarios. Summary of the Invention
[0005] To address the shortcomings of existing technologies in generalizing lithium-ion battery lifespan across multiple chemical systems and varying temperature conditions (20℃~60℃), this invention provides a generalized prediction method that integrates multi-source features, adapts to dynamic operating conditions, and is data-efficient, thus meeting the industry's demand for precise battery health management. The specific technical solution is as follows.
[0006] In a first aspect, this disclosure proposes a method for predicting the lifespan of ternary lithium-ion batteries, comprising: constructing a ternary lithium battery dataset with a preset operating temperature range and charge / discharge rate range; based on the ternary lithium battery dataset, extracting battery physical characteristics and operating condition characteristics, training several XGBoost models with different hyperparameter combinations, and obtaining an XGBoost model with the optimal hyperparameter combination, wherein the hyperparameters include the learning rate and the number of trees; using the battery cycle count range as a data window, quantitatively analyzing the nonlinear impact of the data scale on the prediction error, and determining the optimal data window; during inference, acquiring the physical characteristics and operating condition characteristics within an optimal data window for a ternary battery, and using the XGBoost model with the optimal hyperparameter combination to predict the remaining lifespan.
[0007] In one embodiment of the above technical solution, the physical characteristics include the logarithmic value of the variance of the discharge capacity difference, the logarithmic value of the minimum discharge capacity difference, the linear fitting slope of the capacity decay curve from the second cycle to the N1th cycle and the linear fitting intercept of the capacity decay curve, the discharge capacity of the second cycle and the average charging time of the first N2 cycles, wherein N, N1, and N2 are set values.
[0008] In one embodiment of the above technical solution, the operating condition characteristics include battery temperature and battery type unique thermal encoding.
[0009] In one embodiment of the above technical solution, the evaluation indicators include root mean square error and average percentage error.
[0010] In one embodiment of the above technical solution, the optimal data window is a cycle count of 30 to 100.
[0011] In one embodiment of the above technical solution, grid search based on the hold-out method is used to train and tune the hyperparameters of the XGBoost model in order to obtain an XGBoost model with the optimal combination of hyperparameters.
[0012] In one embodiment of the above technical solution, during the grid search, the search space consists of 100 to 500 trees, the learning rate is 0.05 to 0.15, the maximum number of splits per tree is 10 to 100, and the minimum number of samples per leaf node is 1 to 5.
[0013] In one embodiment of the above technical solution, the preset operating temperature range and charge / discharge rate range are 20℃-60℃ and 0.25C~4C, respectively.
[0014] Secondly, this disclosure proposes a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods described above.
[0015] Thirdly, this disclosure proposes a ternary lithium-ion battery life prediction system, comprising: a preprocessing module configured to construct a ternary lithium battery dataset with an operating temperature range of 20℃-60℃ and a charge / discharge rate range of 0.25C-4C; based on the ternary lithium battery dataset, extracting battery physical features and operating condition features, training several models, and selecting the final XGBoost model based on evaluation metrics; an optimal model determination module configured to extract battery physical features and operating condition features based on the ternary lithium battery dataset, train several XGBoost models with different hyperparameter combinations, and obtain an XGBoost model with the optimal hyperparameter combination, wherein the hyperparameters include the learning rate and the number of trees; and an inference application module configured to acquire physical features and operating condition features in an optimal data window of a ternary battery, and use the XGBoost model with the optimal hyperparameter combination to predict the remaining life.
[0016] The beneficial technical effects of this invention are as follows: Breaking through cross-system generalization bottlenecks: Achieving a prediction error of 11.84% on the NCA / NCM / NCM_NCA battery dataset, a 46% improvement over traditional elastic networks; Improved data efficiency: Requires only the first 100 cycles (XGBoost) or the first 90 cycles (Random Forest), enabling relatively accurate prediction of the remaining cycle life of different types of lithium-ion batteries under dynamic operating conditions, demonstrating strong generalization ability; Lower computational cost: Suitable for rapid assessment of battery health status, providing accurate prediction data for battery management systems, extending battery life, and reducing maintenance costs. This solution is applicable to real-time remaining life assessment of electric vehicles and large-scale energy storage systems, and is of great significance for promoting the engineering application of battery health management technology. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This diagram illustrates the prediction error obtained by substituting battery data of different cycle counts and data of a battery reduced by half into the elastic network model.
[0019] Figure 2 This is a scatter plot of the battery life prediction from the XGBoost model when using the first 100 cycles of cyclic aging data from a ternary lithium battery.
[0020] Figure 3 This diagram illustrates the trend of XGBoost model prediction error as a function of the number of iterations used.
[0021] Figure 4 This section analyzes key features used in XGBoost model predictions.
[0022] Figure 5 This represents the distribution of prediction errors in the XGBoost model.
[0023] Figure 6 A comparison of the errors in predicting lithium-ion battery life using three machine learning algorithms: XGBoost, Random Forest, and Elastic Network.
[0024] Figure 7 This is a schematic diagram of the life prediction method for ternary lithium batteries. Detailed Implementation
[0025] As is known from the background art, existing technologies for predicting the lifespan of lithium-ion batteries with multiple chemical systems suffer from generalization defects and data dependence problems. Therefore, this invention proposes a lithium-ion battery lifespan prediction method based on machine learning.
[0026] (a) Data preprocessing A ternary lithium battery dataset was constructed, which includes 130 batteries of three types: NCA, NCM, and NCM_NCA. Ternary lithium batteries are lithium-ion batteries that use nickel, cobalt, and manganese (or aluminum) oxides as cathode materials. NCA represents nickel-cobalt-aluminum batteries, and NCM represents batteries composed of nickel (Ni), cobalt (Co), and manganese (Mn).
[0027] The battery dataset contains cycle data for multiple batteries, with each battery's data including key parameters such as discharge capacity and charging time. The operating conditions for each battery are: temperature 20℃-60℃, charge / discharge rate 0.25C-4C. In some embodiments, extreme temperatures, such as -50℃ to -40℃, or above 60℃, may also be considered.
[0028] Real-time lifespan assessment can be performed using early-cycle data from lithium batteries, such as the data from the first N cycles of each battery. This data can then be used for feature extraction and analysis. The specific value of N is set according to the characteristics of each battery, with a reference range of N=30 to 100.
[0029] In one implementation, the data collected for each battery includes overall information for each cycle and detailed voltage and time data measured during the cycle. The overall information includes battery type, discharge capacity, capacity decay, cycle life, remaining life, temperature, etc.
[0030] (II) Feature Extraction Based on a ternary lithium battery dataset, battery physical features and operating condition features were extracted to train several machine learning models, including Elastic Network, Random Forest, and XGBoost models.
[0031] The physical characteristics selected are mainly those related to battery health, including discharge capacity, capacity decay, cycle life, and remaining life. In addition, this application also considers operating condition characteristics, including battery temperature and unique thermal encoding of battery type.
[0032] In one embodiment, the physical characteristics of each battery, such as discharge capacity, capacity decay, and battery temperature, are obtained from the data of the first N cycles and further processed to obtain the following six features: the logarithmic value of the variance of the discharge capacity difference, the logarithmic value of the minimum discharge capacity difference, the slope of the linear fit of the capacity decay curve from the 2nd cycle to the N1st cycle, the intercept of the linear fit of the capacity decay curve from the 2nd cycle to the N1st cycle, the discharge capacity of the 2nd cycle, and the average charging time of the first N2 cycles. These are then combined with two operating condition features: battery temperature and battery type unique thermal encoding, resulting in a total of eight features. N1 and N2 are set values. For example, N1=100 and N2=5. In this embodiment, the above features comprehensively reflect the battery's health status and decay trend in the early cycles, providing important information for predicting remaining lifespan.
[0033] In one implementation, the extracted physical features are stored in a matrix, which determines the number of batteries (numBatteries), the number of cycles used (numCyclesToUse), and the number of features. The matrix size is numBatteries × numCyclesToUse × number of features. This matrix is then combined with operating condition features to construct a comprehensive feature matrix, thereby improving data processing speed.
[0034] (III) Model Training and Hyperparameter Tuning In one implementation, a hold-out-based grid search is used for model training and hyperparameter tuning to ensure the model's generalization ability and prediction accuracy. The specific implementation logic is as follows: (a) Explicit hyperparameter mesh definition Outer parameters: Set up 3 models, and set the learning rates to [0.05, 0.1, 0.15] respectively.
[0035] Inner layer parameters: Set the number of trees for each model to [100, 200, 500].
[0036] Under this hyperparameter grid definition, the total number of combinations is 3×3=9, which constitutes the standard grid search. (b) Hold-out validation (non-cross-validation) Validation method: The loss is calculated directly using an independent validation set, rather than cross-validation. That is, the lifetime predicted by features extracted from the validation set and the actual cycle lifetime of the validation set are used to calculate the loss. Each hyperparameter combination is trained only once, evaluated on a fixed validation set, and the combination with the minimum loss is selected. (c) Dynamically update the optimal model: In each iteration, if the validation loss is less than the current optimal loss, the optimal model is updated, and the hyperparameter combination that performs best on the validation set is ultimately retained. (d) Full data retraining: After finding the optimal hyperparameters, retrain the final model using the training and validation sets to improve generalization with more data.
[0037] In one implementation, the XGBoost algorithm is used to train the regression model, and the hyperparameters are optimized through grid search, including the number of trees, the minimum number of leaf nodes, and the maximum tree depth.
[0038] In one implementation, the hyperparameters of the XGBoost model are tuned by grid search, with the search space including 100 to 500 trees, a learning rate of 0.05 to 0.15, a maximum number of splits per tree of 10 to 100, and a minimum number of samples per leaf node of 1 to 5.
[0039] In one embodiment, the range of battery cycle counts is used as a data window to quantify the nonlinear impact of data scale (cycle count, battery sample size) on prediction error and determine the optimal data window. The data scale includes measurements from the perspectives of cycle count and battery sample size. The optimal data window is the usage cycle interval where the prediction error is less than a preset threshold. For example, if the prediction error is ≤15.37% for the first 40-90 cycles, and increases sharply by 180% for the first 30 cycles, then the first 40-90 cycles are considered the optimal data window. Furthermore, the nonlinear impact of battery sample size on prediction error can also be considered to determine the optimal model parameters.
[0040] In one embodiment, a multi-source degradation feature synergistic extraction method is adopted for ternary lithium batteries (NCA / NCM / NCM_NCA). First, temperature features (dynamic range of 20℃~60℃) and unique thermal encoding of battery type are introduced to address the generalization failure problem caused by differences in chemical systems. Second, early physical indicators such as the logarithm of the variance of discharge capacity difference, capacity decay slope, and second-cycle discharge capacity are integrated to construct a cross-scale feature matrix. Experimental data show that this reduces the multi-condition prediction error of ternary batteries from 21.92% with an elastic network to 11.84%, a 46% improvement over the traditional elastic network, and reduces the standard deviation of the error distribution by 40%.
[0041] (iv) Model testing and performance evaluation After selecting the optimal hyperparameters, the final model is trained using the entire dataset and evaluated on the test set. Evaluation metrics include root mean square error (RMSE) and mean percentage error. The specific steps are as follows: 1. Training the final model: The final XGBoost model is trained on all feature data using the optimal hyperparameter configuration.
[0042] 2. Model Performance Evaluation: Using the trained model to make predictions on the test set, calculate the RMSE and relative error between the predicted and actual values to evaluate the model's predictive performance. RMSE reflects the average deviation between the predicted and actual values; a smaller value indicates a more accurate prediction. Relative error represents the proportion of the deviation between the predicted and actual values, providing a relative measure of prediction accuracy.
[0043] In one implementation, for LFP / graphite batteries, when the training sample size is reduced by 50%, the prediction error only increases by 3 to 5 percentage points, and the model stability decreases slightly; for ternary lithium batteries, the final model performance evaluation shows an RMSE of 105.6 and an average relative error of 11.7%.
[0044] By generating a scatter plot, the predicted values are compared with the actual values, and an ideal reference line is added. This visual comparison intuitively demonstrates the model's predictive performance and accuracy. Experimental results show that the optimal RMSE on the test set is 105.60 cycles, with an average relative error controlled at 11.84%. This indicates that the XGBoost model of this invention, combining physical and operating condition characteristics, possesses efficient and accurate battery life prediction capabilities, while also exhibiting good generalization ability and stability.
[0045] Comparing the XGBoost model with Elastic Networks and Random Forests, the results show that the XGBoost model significantly outperforms the other two methods in predicting lithium-ion battery life due to its regularization mechanism and excellent handling of sparse features. Furthermore, XGBoost features a dynamic optimization mechanism that captures the nonlinear interaction effect between temperature and battery type through derivative optimization (based on the Hessian matrix). Nonlinear interaction effects refer to the combined effect of two or more independent variables acting on the dependent variable in a nonlinear model; their effects are not simply additive but exhibit nonlinear superposition characteristics.
[0046] (v) Application Based on a pre-trained XGBoost model, the system takes early-cycle data of the target battery as input and outputs a predicted remaining cycle life (RUL). This solution supports real-time lifetime assessment using early-cycle data of lithium batteries under dynamic operating conditions. The dynamic operating conditions range from 20℃ to 60℃ and from 0.25C to 4C.
[0047] In one implementation, only the data from the first 100 cycles (XGBoost) or the first 90 cycles (Random Forest) are needed to accurately predict the remaining cycle life of different types of lithium-ion batteries under dynamic operating conditions, demonstrating strong generalization ability. Moreover, the computational cost is low, making it suitable for quickly assessing battery health status, providing accurate predictive data for battery management systems, extending battery life, and reducing maintenance costs.
[0048] The implementation of the technical solution in this case will be described below with reference to the accompanying drawings.
[0049] Figure 1 The prediction error is calculated by substituting battery data with different cycle counts and data from half the battery count into the elastic network model. The vertical axis represents the average percentage error, and the horizontal axis represents the number of cycles used. It can be seen that the prediction error exhibits a typical non-linear growth characteristic. Experimental data shows that when the input data is reduced from the first 100 cycles to the first 30 cycles, the prediction error increases sharply by 180% (from 10.16% to 28.15%), while the error fluctuation is relatively gentle (11.33%–15.37%) within the data window of the first 40–90 cycles. When using half the battery data for model training, compared to using all the battery data, the model's prediction error increases under different conditions (except for the first 30 cycles), and the model becomes more unstable, indicating that more data helps improve the model's prediction accuracy and stability.
[0050] Figure 2This is a scatter plot showing the battery life predictions of the XGBoost model using the first 100 cycles of ternary lithium battery aging data. The horizontal axis represents the actual cycle life of the battery, and the vertical axis represents the predicted cycle life. It can be seen that the predicted values have a high fitting accuracy with the actual remaining life, with a prediction error of 11.84% and a root mean square error of 105.60. This indicates that using the battery features described in this invention, the XGBoost model has a strong ability to capture complex nonlinear degradation modes during battery aging (such as the dynamic inflection point of the accelerated decay stage and abrupt responses under abnormal operating conditions). Furthermore, the prediction error exhibits a highly concentrated normal distribution, with the mean approaching zero, the standard deviation significantly reduced, and almost no significant long-tailed outliers. This reflects that the model not only possesses excellent prediction accuracy but also demonstrates outstanding prediction stability, effectively suppressing the occurrence of extreme errors.
[0051] Figure 3 The chart shows the trend of XGBoost model prediction error as the number of iterations used changes. It can be seen that as the number of iterations gradually expands from the first 30 to the first 100, the prediction error exhibits a non-linear evolution characteristic of "first fluctuating upwards and then continuously decreasing": In the first 30 to 60 iterations, the prediction error rate rises from 17.00% to 19.82% (peaking at 10-50 iterations) and then falls back to 17.21%, with the root mean square error (RMSE) fluctuating between 131.11 and 157.34. This reflects that the data in this stage contains mixed characteristics of battery aging, from initial linear decay to mid-term non-linear transition, and XGBoost's gradient boosting mechanism is prone to fitting bias when dealing with feature space instability. However, when the number of iterations expands to the first 70 to 100, the prediction error rate gradually decreases to 11.84%, and the RMSE simultaneously decreases to 105.60, achieving the model's optimal performance.
[0052] Figure 4 This analysis examines the importance of parameters in the optimal XGBoost model. It reveals that the most important parameters are: the average charging time (AvgChargeTimeFirst5Cycles), the logarithm of the minimum discharge capacity difference (DeltaQMin_log), the linear intercept of the capacity decay curve from the 2nd to the 100th cycle (CapacityFadeIntercept), and the second discharge capacity (DischargeCapacityCycle2). Other parameters, such as the logarithm of the variance of the discharge capacity difference (DeltaQVar_log), temperature, the linear slope of the capacity decay curve from the 2nd to the 100th cycle (CapacityFadeSlope), and battery type (Is_NCM, Is_NCA), also have some influence.
[0053] Figure 5 The distribution of battery life prediction errors shows that the prediction errors for most batteries are concentrated between -50 and 50 cycles, with only a few batteries having errors greater than -200 cycles. This indicates that the machine learning method has high accuracy and low error in predicting battery life.
[0054] Figure 6 This graph compares the lifetime prediction accuracy and mean error of three different machine learning methods: Elastic Net, Random Forest, and Extreme Gradient Boosting (XGBoost). As can be seen from the graph, XGBoost's lifetime prediction accuracy (prediction error 11.8%) is significantly better than the other two methods, and its root mean square error (RMSE) for lifetime prediction (105.8 cycles) is also significantly lower.
[0055] In summary, this invention proposes a method for predicting the lifespan of ternary lithium-ion batteries, comprising: constructing a ternary lithium battery dataset with a preset operating temperature range and charge / discharge rate range; based on the ternary lithium battery dataset, extracting battery physical characteristics and operating condition characteristics, training several XGBoost models with different hyperparameter combinations, and obtaining an XGBoost model with the optimal hyperparameter combination, wherein the hyperparameters include the learning rate and the number of trees; using the battery cycle count range as a data window, quantifying and analyzing the nonlinear impact of data size on prediction error, and determining the optimal data window; during inference, acquiring the physical characteristics and operating condition characteristics within an optimal data window for a ternary battery, and using the XGBoost model with the optimal hyperparameter combination to predict the remaining lifespan. The above method steps are described in detail below. Figure 7 As shown in the image. This solution can be used to monitor battery usage in real time, improving battery safety.
[0056] Based on the above method, the corresponding system can perform information processing on the method. For example, a ternary lithium-ion battery life prediction system includes: a preprocessing module configured to construct a ternary lithium battery dataset with a preset operating temperature range and charge / discharge rate range; an optimal model determination module configured to extract battery physical characteristics and operating condition characteristics based on the ternary lithium battery dataset, train several XGBoost models, and select the final XGBoost model based on evaluation metrics; an optimal data window determination module configured to use the battery cycle number range as a data window, quantitatively analyze the nonlinear impact of data size on prediction error, and determine the optimal data window; and an inference application module configured to acquire the physical characteristics and operating condition characteristics within an optimal data window of a ternary battery, and use the trained XGBoost model to predict the remaining life.
[0057] The present invention can also be a computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the present invention.
[0058] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.
[0059] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.
[0060] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, Python, etc., and conventional procedural programming languages such as "C" or similar languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.
[0061] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.
[0062] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.
[0063] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions that execute on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.
[0064] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It will be well known to those skilled in the art that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are equivalent.
[0065] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.
Claims
1. A method for predicting the lifetime of a ternary lithium-ion battery, characterized in that, The method comprises the following steps: constructing a ternary lithium battery dataset with a preset working temperature range and a charge / discharge rate range; extracting battery physical features and working condition features based on the ternary lithium battery dataset, training a plurality of XGBoost models with different hyperparameter combinations, and obtaining an XGBoost model with an optimal hyperparameter combination, wherein the hyperparameters include a learning rate and the number of trees; quantitative analysis of the nonlinear influence of data size on prediction error by taking the battery cycle number range as a data window, and determining the optimal data window; in inference, obtaining the physical features and working condition features in the optimal data window of a ternary battery, and using the XGBoost model with the optimal hyperparameter combination to predict the remaining life.
2. The method of claim 1, wherein: the physical features include the logarithmic value of the variance of the discharge capacity difference, the logarithmic value of the minimum value of the discharge capacity difference, the linear fitting slope of the capacity attenuation curve from the 2nd cycle to the N1th cycle and the linear fitting intercept of the capacity attenuation curve, the discharge capacity of the 2nd cycle and the average charging time of the previous N2 cycles, wherein N, N1 and N2 are set values.
3. The method of claim 1, wherein, the working condition features include battery temperature and battery type one-hot encoding.
4. The method of claim 1, wherein, the evaluation indicators include root mean square error and average percentage error.
5. The method of claim 1, wherein, The optimal data window is 30-100 cycle number data.
6. The method of claim 1, wherein, The grid search based on the leave-one-out method is used for model training and hyperparameter optimization to obtain the XGBoost model with the optimal hyperparameter combination.
7. The method of claim 6, wherein, In the grid search, the search space is 100-500 trees, the learning rate is 0.05-0.15, the maximum number of splits for a single tree is 10-100, and the minimum number of leaf nodes is 1-5.
8. The method of claim 1, wherein, The preset working temperature range and the charge / discharge rate range are 20-60°C and 0.25C-4C, respectively.
9. A computer-readable storage medium, characterized in that: The computer program capable of being loaded and executed by the processor to perform the method of any one of claims 1-8 is stored.
10. A ternary lithium-ion battery life prediction system, characterized by, The method comprises the following steps: a preprocessing module configured to construct a ternary lithium battery dataset with a preset working temperature range and a charge / discharge rate range; an optimal model determination module configured to extract battery physical features and working condition features based on the ternary lithium battery dataset, train a plurality of XGBoost models with different hyperparameter combinations, and obtain an XGBoost model with an optimal hyperparameter combination, wherein the hyperparameters include a learning rate and the number of trees; an optimal data window determination module configured to take the battery cycle number range as a data window, quantitatively analyze the nonlinear influence of data size on prediction error, and determine the optimal data window; an inference application module configured to obtain the physical features and working condition features in the optimal data window of a ternary battery, and use the XGBoost model with the optimal hyperparameter combination to predict the remaining life.