Intelligent battery life prediction method depending on early characteristics

Abstract

The invention discloses an intelligent battery life prediction method depending on early-stage characteristics, and the method comprises the steps: implanting an optical fiber sensor, a thin film sensor or a pressure sensor in a lithium ion battery, and obtaining the internal state information, such as internal strain and internal temperature, of a battery formation and early-stage circulation stage in real time. A battery multi-dimensional feature database containing early internal state features is constructed by combining the external voltage and capacity of the same period, and the internal association between the early internal parameter features and the long-term cycle life of the battery is explored by using an artificial intelligence algorithm to realize life prediction. According to the method, the limitation that the service life is predicted by depending on long-term cycle data or external sensing traditionally is broken through, early-stage accurate prediction of the health state of the battery is realized, and a brand new technical means is provided for production screening, echelon utilization and full-life-cycle management of the battery.

Description

Technical Field

[0001] This invention belongs to the field of battery life prediction technology, specifically relating to an intelligent battery life prediction method that relies on early characteristics. Background Technology

[0002] State of Health (SOH) is a core indicator for evaluating the performance, reliability, and safety of lithium-ion batteries. Accurate prediction of battery life is crucial for ensuring the driving range of electric vehicles, the cost-effective operation of energy storage systems, and the user experience of consumer electronics. Currently, battery SOH prediction methods primarily rely on external electrochemical measurements. For example, online estimation can be performed using ampere-hour integration, open-circuit voltage methods, or algorithms based on equivalent circuit models and Kalman filtering. A common drawback of these methods is that they typically require a long period of charge-discharge cycle data accumulation to obtain accurate predictions, and they struggle to capture early performance variations between individual batteries caused by subtle differences in the manufacturing process. This lag makes rapid and accurate quality sorting during battery production extremely difficult and limits proactive, forward-looking battery health management.

[0003] Therefore, establishing a mapping model from the internal physicochemical characteristics of a battery in its early operation to its state of harmonics (SOH) to enable rapid and forward-looking assessment of battery life is crucial to overcoming the limitations of existing technologies. Summary of the Invention

[0004] To address the problems of long prediction cycles and insufficient accuracy in existing battery life prediction technologies, this invention aims to provide an intelligent battery life prediction method that relies on early features. This method can shorten the prediction cycle, greatly improve the quality inspection efficiency of battery production and the operation and maintenance level of energy storage systems. The method of rapidly predicting the life of lithium-ion batteries through early features is applicable to the quality management and life screening of power batteries and energy storage batteries.

[0005] To solve the technical problem, the technical solution of the present invention is as follows:

[0006] A smart battery lifetime prediction method relying on early features, the method comprising:

[0007] S1: By embedding micro-sensors inside the battery cell, a smart battery that can monitor the internal physical state in real time is constructed.

[0008] S2: During the battery formation stage or the initial few charge-discharge cycles, a high-speed demodulator and an external testing system are used to simultaneously collect internal physical parameters and external electrochemical parameters to construct a multi-dimensional early health dataset.

[0009] S3: Preprocess the collected multi-source health data, including time alignment, filtering and noise reduction and normalization, and associate it with the lifespan labels obtained from long-term aging experiments to build a feature-lifespan database for training artificial intelligence models.

[0010] S4: Based on the database established in step S3, the training set and test set are divided using a five-fold nested cross-validation method. The inner loop is used to search for the optimal hyperparameter combination of gradient boosting GBDT or ExtraTrees model, and the outer loop is used for model evaluation. Finally, the model is trained on the complete training data and its generalization performance is evaluated on the independent test set.

[0011] Furthermore, for new batteries to be tested, only one early charge-discharge test is needed to collect the corresponding internal / external parameters. After the same preprocessing process, these parameters are input into the trained model to quickly predict their future cycle life, enabling efficient screening and application deployment.

[0012] Furthermore, the micro-sensor employs a fiber Bragg grating or a thin-film pressure sensor.

[0013] Furthermore, in step S2, the collected data includes: internal strain, internal temperature, voltage, and capacity.

[0014] Furthermore, step S3 includes:

[0015] S301: Synchronize the internal physical parameter sequence with the external electrochemical parameter sequence in time;

[0016] S302: Employs a filtering algorithm to smooth the data and eliminate random noise interference introduced by the measurement environment or equipment;

[0017] S303: Normalize data of different dimensions, including internal temperature, strain, and voltage;

[0018] S304: Associate the processed early data sequences with lifetime tags determined through long-term aging experiments to form the final database.

[0019] Furthermore, step S4 includes:

[0020] S401 dataset partitioning, five-fold nested cross-validation:

[0021] To ensure the objectivity and robustness of the model evaluation, a nested five-fold cross-validation method is adopted, including: for databases containing early multidimensional features and lifetime labels. Where N is the total number of battery samples, X i Let y be the early feature vector of the i-th battery sample. i Its corresponding actual lifespan label;

[0022] Outer loop: Randomly divide the total dataset D into K-1 similar and disjoint subsets; that is, D = D1∪D2∪…∪D K Where K = 5; perform K iterations, and in the k-th iteration (k = 1, ..., K), the subset D k As the outer test set Used to evaluate the performance of the finally trained sub-models, the remaining K-1 subsets are combined into the outer training set. Used for model training and hyperparameter selection;

[0023] Inner loop: In the k-th round of the outer loop, the current outer training set is processed. J-fold cross-validation is performed again to search for the optimal combination of hyperparameters of the model, i.e., the splitting... Given J mutually exclusive subsets d1, d2, ..., d... J In the j-th round (j=1,...,J) of the inner loop, the subset d j As a verification set The remaining J-1 subsets are combined into the inner training set.

[0024] S402 Model Selection and Hyperparameter Optimization:

[0025] In the inner loop, an ensemble learning model is selected for modeling, with preferred algorithms including gradient boosting decision trees and extreme random trees. In each round of the inner loop, for a set of candidate hyperparameters λ, the inner training set is used... Training model M λ and in the verification set Its performance is evaluated by obtaining an error metric, which is the average verification error E in the inner loop for a fixed set of hyperparameters λ. λ The calculation is as follows:

[0026]

[0027] Strategies such as grid search or Bayesian optimization are employed to find the parameter E within a predefined hyperparameter space. λ Minimize the optimal combination of hyperparameters λ * ,in

[0028] S403 Final Model Training and Performance Evaluation:

[0029] In the k-th round of the outer loop, the optimal combination of hyperparameters λ found in the inner loop is used. * In the complete outer training set A sub-model is obtained through training.

[0030]

[0031] Use the trained sub-model outer layer test set Make predictions and calculate its performance index P. k ;

[0032]

[0033] Repeat the outer loop K times to obtain K performance metrics P1, P2, ..., P K The final generalization performance of the model on unknown data is determined by the average of these K metrics. To evaluate;

[0034]

[0035] After completing the above nested cross-validation process, in order to obtain a final model for actual deployment, the optimal hyperparameter λ is found. * The model is applied to the entire dataset D and trained to obtain the final lifespan prediction model M. final .

[0036] Compared with the prior art, the advantages of the present invention are as follows:

[0037] Real-time monitoring and data richness: Through technologies such as fiber optic sensors and thin-film pressure sensors, multiple internal physical parameters of the battery during formation and early cycling stages can be obtained, providing an important foundation for establishing a comprehensive early health database. This enables predictive models to be trained on richer and more accurate data, improving the reliability of predictions.

[0038] Shorten the life assessment cycle: Traditional battery life assessment usually requires a long cycle, while this invention significantly reduces the reliance on long-term cycle testing by relying on early characteristics for rapid screening, improves the quality inspection efficiency in the production process, and helps to achieve rapid decision-making.

[0039] High-precision prediction: This invention employs advanced machine learning algorithms (such as GBDT and ExtraTrees) to effectively capture the complex correlation between internal physical parameters and long-term degradation, thereby providing high-precision remaining useful life (SOH) prediction and ensuring the safe and reliable application of batteries.

[0040] Supports secondary use and quality management: Through rapid and accurate prediction results, it facilitates battery quality control and subsequent secondary use, achieving efficient resource allocation and sustainable development. Attached Figure Description

[0041] Figure 1The main technology roadmap of the present invention is a smart battery life prediction method that relies on early features. Detailed Implementation

[0042] The specific implementation of the present invention is described below with reference to embodiments:

[0043] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0044] Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0045] Example 1:

[0046] like Figure 1 As shown, the present invention proposes a smart battery lifetime prediction method that relies on early features, and the specific steps are as follows:

[0047] S1. During the stacking or winding process of lithium-ion batteries, encapsulated micro sensors, such as fiber optic sensors and thin-film pressure sensors, are embedded inside the battery cell to obtain internal physical field information.

[0048] To ensure the accuracy and synchronization of multiphysics data acquisition, in this embodiment, taking a fiber optic sensor as an example, a dual-grating fiber Bragg grating (FBG) sensor is preferably used to accurately decouple the temperature and strain signals inside the battery.

[0049] S2. Formation and early cycle testing are performed on the manufactured batteries. During this process, wavelength drift data from the FBG sensor is read using a high-speed demodulator, and internal temperature and strain are decoupled using a pre-calibrated sensitivity coefficient. Simultaneously, the corresponding electrochemical data are recorded by an external testing system.

[0050] S3. Preprocess the early multidimensional data and associate it with long-term aging data to construct a feature lifetime database for training artificial intelligence models.

[0051] In order to improve the quality of data input to the artificial intelligence model, in this embodiment, in S3, a standardized training dataset is constructed by performing synchronous alignment, filtering and denoising on multi-source time series data, and combining it with the final lifetime label obtained from long-term cyclic testing.

[0052] Specific methods for building a database include:

[0053] S31. Synchronize the internal physical parameter sequence with the external electrochemical parameter sequence in time;

[0054] S32. Use filtering algorithms to smooth the data and eliminate random noise interference introduced by the measurement environment or equipment;

[0055] S33. Normalize data of different dimensions, including internal temperature, strain, and voltage.

[0056] S34. Associate the processed early data sequences with lifetime tags determined through long-term aging experiments to form the final database.

[0057] In the data processing process, in this embodiment, in S32, a moving average algorithm can be selected, and the window size can be set according to the frequency domain characteristics of the signal to reduce noise interference while preserving effective signal details.

[0058] S4. Repeat the above steps to create a database for multiple battery samples. Use this database as a training set to build and train an AI model based on ensemble learning, specifically gradient boosting or extreme random tree models. After the model has learned, for any newly manufactured battery, only the same early testing and preprocessed data input are needed to obtain high-accuracy lifespan prediction results.

[0059] After ensuring the early health dataset is complete, the core step S4 of this invention uses an advanced artificial intelligence model to predict and verify battery life. First, the collected multidimensional feature data (including internal strain, temperature, and electrochemical parameters) are integrated with the corresponding long-term life labels to form standardized training samples.

[0060] Next, a nested five-fold cross-validation method is used to split the dataset to enhance the objectivity and reliability of model evaluation. In the inner loop, ensemble learning algorithms, such as Gradient Boosting Decision Tree (GBDT) and ExtraTrees, are selected to optimize the hyperparameters, ensuring the model performs optimally in complex feature spaces. This process effectively avoids overfitting and improves the model's generalization ability.

[0061] In the outer loop phase, the optimized hyperparameters are used to train the model on all training data, and its predictive performance is evaluated on an independent test set. Through multiple rounds of cross-validation, the final output performance metrics provide a reliable estimate of the model's lifetime prediction capability on new batteries.

[0062] The innovation of this step lies in combining early internal characteristics and electrochemical behavior to establish an efficient and accurate battery life prediction model through intelligent algorithms, laying the foundation for battery quality control and effective utilization.

[0063] This invention can quickly and accurately predict the long-term lifespan of batteries in the early stages of battery manufacturing, providing a brand-new technical means for battery quality control and secondary utilization.

[0064] Example 2:

[0065] This embodiment 2 is applied to embodiment 1. In specific implementation, this embodiment includes the following steps: The specific steps are as follows:

[0066] (1) Select 50 soft-pack batteries from the same batch and implant a dual-grid FBG fiber optic sensor into each battery.

[0067] (2) All batteries were formed and tested for 10 standard cycles. Internal temperature and strain data were collected at a sampling frequency of 1 second, while external electrochemical data were recorded.

[0068] (3) Preprocess the early data collected, including aligning the internal physical parameters with the external electrochemical parameters in time series, using digital filters to denoise, and normalizing the multi-dimensional data to form standardized data segments.

[0069] (4) The battery samples were subjected to long-term aging tests until the capacity decayed to 80% of the nominal capacity, and their actual cycle life was recorded as the learning label of the model.

[0070] (5) Use the battery sample data as the total dataset. The performance of the ExtraTrees model is evaluated and hyperparameters are tuned using a five-fold nested cross-validation method. A final lifetime prediction model is trained using all samples and the found optimal hyperparameters.

[0071] (6) After the model is deployed, take another brand new battery from the same batch and test it for 10 cycles, and preprocess it. Input the processed data into the final model, and the model outputs a valid and accurate lifespan prediction result.

[0072] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

[0073] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. A smart battery lifetime prediction method relying on early features, characterized in that, The method includes: S1: By embedding micro-sensors inside the battery cell, a smart battery that can monitor the internal physical state in real time is constructed. S2: During the battery formation stage or the initial few charge-discharge cycles, a high-speed demodulator and an external testing system are used to simultaneously collect internal physical parameters and external electrochemical parameters to construct a multi-dimensional early health dataset. S3: Preprocess the collected multi-source health data, including time alignment, filtering and noise reduction and normalization, and associate it with the lifespan labels obtained from long-term aging experiments to build a feature-lifespan database for training artificial intelligence models. S4: Based on the database established in step S3, the training set and test set are divided using a five-fold nested cross-validation method. The inner loop is used to search for the optimal hyperparameter combination of gradient boosting GBDT or ExtraTrees model, and the outer loop is used for model evaluation. Finally, the model is trained on the complete training data and its generalization performance is evaluated on the independent test set.

2. The intelligent battery lifetime prediction method relying on early features according to claim 1, characterized in that, For new batteries to be tested, only one early charge-discharge test is needed to collect the corresponding internal / external parameters. After the same preprocessing process, these parameters are input into the trained model to quickly predict their future cycle life, enabling efficient screening and application deployment.

3. The intelligent battery lifetime prediction method relying on early features according to claim 1, characterized in that, The miniature sensor employs a fiber Bragg grating or a thin-film pressure sensor.

4. The intelligent battery lifetime prediction method relying on early features according to claim 1, characterized in that, In step S2, the collected data includes: internal strain, internal temperature, voltage, and capacity.

5. The intelligent battery lifetime prediction method relying on early features according to claim 1, characterized in that, Step S3 includes: S301: Synchronize the internal physical parameter sequence with the external electrochemical parameter sequence in time; S302: Employs a filtering algorithm to smooth the data and eliminate random noise interference introduced by the measurement environment or equipment; S303: Normalize data of different dimensions, including internal temperature, strain, and voltage; S304: Associate the processed early data sequences with lifetime tags determined through long-term aging experiments to form the final database.

6. The intelligent battery lifetime prediction method relying on early features according to claim 1, characterized in that, Step S4 includes: S401 dataset partitioning, five-fold nested cross-validation: To ensure the objectivity and robustness of the model evaluation, a nested five-fold cross-validation method is adopted, including: for databases containing early multidimensional features and lifetime labels. Where N is the total number of battery samples, X i Let y be the early feature vector of the i-th battery sample. i Its corresponding actual lifespan label; Outer loop: Randomly divide the total dataset D into K-1 similar and disjoint subsets; that is, D = D1∪D2∪…∪D K Where K = 5; perform K iterations, and in the k-th iteration (k = 1, ..., K), the subset D k As the outer test set Used to evaluate the performance of the finally trained sub-models, the remaining K-1 subsets are combined into the outer training set. Used for model training and hyperparameter selection; Inner loop: In the k-th round of the outer loop, the current outer training set is processed. J-fold cross-validation is performed again to search for the optimal combination of hyperparameters of the model, i.e., the splitting... Given J mutually exclusive subsets d1, d2, ..., d... J In the j-th round (j=1,...,J) of the inner loop, the subset d j As a verification set The remaining J-1 subsets are combined into the inner training set. S402 Model Selection and Hyperparameter Optimization: In the inner loop, an ensemble learning model is selected for modeling, with preferred algorithms including gradient boosting decision trees and extreme random trees. In each round of the inner loop, for a set of candidate hyperparameters λ, the inner training set is used... Training model M λ and in the verification set Its performance is evaluated by obtaining an error metric, which is the average verification error E in the inner loop for a fixed set of hyperparameters λ. λ The calculation is as follows: Strategies such as grid search or Bayesian optimization are employed to find the parameter E within a predefined hyperparameter space. λ Minimize the optimal combination of hyperparameters λ * ,in S403 Final Model Training and Performance Evaluation: In the k-th round of the outer loop, the optimal combination of hyperparameters λ found in the inner loop is used. * In the complete outer training set A sub-model is obtained through training. Use the trained sub-model outer layer test set Make predictions and calculate its performance index P. k ; Repeat the outer loop K times to obtain K performance metrics P1, P2, ..., P K The final generalization performance of the model on unknown data is determined by the average of these K metrics. To evaluate; After completing the above nested cross-validation process, in order to obtain a final model for actual deployment, the optimal hyperparameter λ is found. * The model is applied to the entire dataset D and trained to obtain the final lifespan prediction model M. final .

Images