Lithium battery life decline quantitative prediction and health state intelligent evaluation method

Abstract

The invention relates to the technical field of lithium batteries, and discloses a lithium battery life decline quantitative prediction and health state intelligent evaluation method, which comprises the following steps: S1, multi-source data acquisition: arranging strain gauge sensors at battery electrode plates and a shell to acquire stress-strain data, integrating a gas sensor in a battery module to acquire gas data, and calculating the data of the stress-strain data; battery data and environment data are acquired by a high-precision synchronous acquisition system through nanosecond clock signals and are processed and stored, and the data quality is monitored at the same time; s2, data preprocessing: generating an extended data set through data transformation and combination and GAN-based data, and processing abnormal data by using a method based on statistics and machine learning; according to the invention, through multi-source data fusion acquisition, a stress-strain sensor and a plurality of gas sensors are added to monitor the internal mechanical change and gas component concentration of the battery, so that the problem that the accuracy and practicability of the model are greatly reduced due to a traditional method based on an electrochemical model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium batteries, and in particular to a method for quantitatively predicting the life decline of lithium batteries and intelligently evaluating their health status. Background Art

[0002] With the rapid development of modern science and technology, lithium batteries have been widely used in many fields, especially in electric vehicles, portable electronic devices and large-scale energy storage systems, and have become an indispensable key energy component. For example, in the field of electric vehicles, the performance of lithium batteries is directly related to the vehicle's cruising range, charging time and overall safety; in energy storage systems, they play a vital role in balancing energy supply and demand and improving energy efficiency. However, during the use of lithium batteries, their performance will inevitably decline gradually, and this phenomenon is affected by a variety of complex factors. From the perspective of internal factors, the structural stability of electrode materials faces challenges during repeated charging and discharging. For example, the insertion and extraction of lithium ions may cause changes in the lattice structure of electrode materials, volume expansion or contraction, and thus affect the interface stability between the electrode and the electrolyte, accelerating the loss of active substances; the electrolyte itself will also undergo chemical changes such as decomposition and deterioration over time, changing its ion conductivity and compatibility with the electrode. From the perspective of external factors, the size of the charge and discharge rate is directly related to the diffusion rate and reaction activity of lithium ions in the electrode material. Excessively high charge and discharge rates may cause the polarization phenomenon to intensify, resulting in battery heating, increased internal resistance and other problems. The ambient temperature also has a significant impact on the performance of lithium batteries. Low temperature environments will reduce the ion migration rate of the electrolyte and the reaction activity of the electrode materials, while high temperature environments may accelerate the rate of side reactions inside the battery and shorten the battery life. In addition, long-term cyclic use and deep charge and discharge operations will gradually accumulate damage to the battery, leading to performance degradation such as battery capacity decay and increased internal resistance.

[0003] Although the traditional method based on electrochemical model can describe the electrochemical reaction process inside the battery in principle, the physical and chemical processes inside the lithium battery are extremely complex, involving many parameters, some of which are difficult to accurately measure. Therefore, the accuracy and practicality of the model are greatly reduced. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method for quantitatively predicting the life decline and intelligently evaluating the health status of lithium batteries, which solves the problem that although the traditional method based on electrochemical model can describe the electrochemical reaction process inside the battery in principle, the physical and chemical processes inside the lithium battery are extremely complex, involving many parameters and some parameters are difficult to accurately measure, which will cause the accuracy and practicality of the model to be greatly reduced.

[0005] To achieve the above objectives, the present invention is implemented by the following technical scheme: A method for quantitatively predicting the life decline of lithium batteries and intelligently evaluating their health status, comprising the following steps:

[0006] S1. Multi-source data acquisition: strain gauge sensors are installed at the battery electrode sheet and the shell to obtain stress-strain data, gas sensors are integrated in the battery module to collect gas data, and battery data and environmental data are collected and processed and stored through a high-precision synchronous acquisition system with nanosecond clock signals, while monitoring data quality;

[0007] S2, data preprocessing, expanding the data set through data transformation and combination and GAN-based data generation, and processing abnormal data using statistical and machine learning-based methods;

[0008] S3, model construction and training, select appropriate machine learning or deep learning models, divide the data set for training and adjust the parameters until convergence;

[0009] S4. Model evaluation and application: After evaluation with the test set, it is applied to the actual system and the lithium battery is managed based on the results.

[0010] Preferably, in step S1, the strain gauge sensor is based on the piezoresistive effect, and the conversion relationship of the Wheatstone bridge circuit is: and Gas sensors such as electrochemical hydrogen sensors depend on I= To work back on the concentration, the system collects data and checks its characteristics.

[0011] Preferably, in step S2, data enhancement adopts data transformation and combination and GAN-based data generation, and GAN-based data generation includes constructing a specific structure generator network G and a discriminator network D, and the generator minimizes As the goal, the discriminator maximizes The target is the Wasserstein distance metric, and the discriminator loss function in WGAN is Generator loss function And the discriminator weight gradient is clipped; data transformation and combination include translation, scaling, flipping, time series stretching and compression of the original data and local perturbation operation and combination, and PCA is used to extract and filter features, PCA calculates the original data matrix X covariance matrix After decomposition, the principal component Y = XU (m is the number of samples, U is the eigenvector matrix), and the vector corresponding to the large eigenvalue is selected to form U new Get the reduced dimension principal component data Y new =XU new .

[0012] Preferably, in step S2, the abnormal data processing uses an abnormality detection method based on statistical models and machine learning, and the abnormality detection based on the statistical model adopts a multivariate Gaussian mixture model, and its initial parameters are determined according to historical normal data. Substitute real-time data into the calculated probability density value According to the probability threshold ε, the abnormality (α j , μ j ,∑ j are the weight, mean, and covariance matrix of the jth Gaussian component, k is the number of Gaussian components, is the Gaussian distribution probability density function); anomaly detection based on machine learning adopts an ensemble learning method, combining random forest and isolation forest algorithms. Random forest consists of multiple decision trees T i The isolation forest is based on the degree of isolation of data points and also uses the autoencoder network to calculate the reconstruction error. E is the encoder, D is the decoder) judges the anomaly, corrects the correctable abnormal data, and removes the uncorrectable marks.

[0013] Preferably, in step S3, a neural network and a support vector machine model or a convolutional neural network, a recurrent neural network model are selected, the preprocessed data set is divided into a training set, a validation set and a test set, the model is trained using the training set, the model hyperparameters are adjusted according to the validation set performance, and the stochastic gradient descent and Adam optimization algorithms are used for training until the model converges.

[0014] Preferably, in step S4, a test set is used to evaluate model performance, and evaluation indicators include accuracy, recall, F1 value, mean square error, and mean absolute error.

[0015] Preferably, in step S1, the battery data includes but is not limited to current and voltage, and the environmental data includes but is not limited to temperature, humidity, vibration, and air pressure.

[0016] Preferably, in step S2, data enhancement also includes verifying data validity, calculating the difference in basic statistical characteristics between generated data and original data, using a dynamic time warping algorithm to measure time series similarity, applying the enhanced data set to preliminary training tests of multiple machine learning models and performing comparative analysis on the performance of models trained with the original data set, and using cross-validation technology to evaluate the impact of data enhancement on the generalization ability of the model.

[0017] The present invention provides a method for quantitatively predicting the life decline of lithium batteries and intelligently evaluating their health status. It has the following beneficial effects:

[0018] 1. In the present invention, through multi-source data fusion acquisition, not only the traditional voltage, current and temperature data are obtained, but also stress strain sensors and multiple gas sensors are added to monitor the mechanical changes and gas component concentrations inside the battery. Stress strain data can reflect the microscopic changes of the electrode materials and structures inside the battery during the charging and discharging process, and the gas sensor can capture the gas information generated by electrolyte decomposition, side reactions, etc., and warn of potential internal faults of the battery in advance. Multiple data complement each other and verify each other, overcoming the limitations of traditional single data acquisition, and providing a more comprehensive and in-depth information basis for accurately evaluating the health status of the battery, thereby greatly improving the comprehensiveness and accuracy of the data, thereby improving the traditional method based on electrochemical models. Although it can describe the electrochemical reaction process inside the battery in principle, the physical and chemical process inside the lithium battery is extremely complex, involving many parameters and some parameters are difficult to accurately measure, which will cause the accuracy and practicality of the model to be greatly reduced.

[0019] 2. In the present invention, the problem of scarce and uneven data distribution in the field of lithium battery life prediction is effectively solved through data enhancement technology. GAN-based data generation can generate a large amount of simulated data according to the requirements of different battery types and application scenarios. These data have similar distribution characteristics to real data, effectively expanding the size of the data set. Data transformation and combination operations further enrich the diversity of data, including time series stretching and compression to simulate different usage rates, local perturbations to simulate external interference, and cross-combinations of different parameter data, so that the data covers a wider range of working conditions and fault conditions. After data enhancement, it is applied to machine learning model training, which allows the model to fully learn the inherent laws and characteristics of the data, reduce overfitting, and significantly improve the generalization ability of the model, so that it can more accurately predict actual data samples that have not been seen, and provide more reliable model support for the quantitative prediction of lithium battery life degradation, adapting to various complex and changeable actual application environments.

[0020] 3. In the present invention, an abnormal data processing method based on statistical models and machine learning is adopted, which has significant advantages over traditional simple threshold judgment. The statistical model based on the multivariate Gaussian mixture model (GMM) can learn the statistical distribution law of data based on historical normal data, so as to more accurately identify abnormal data points that deviate greatly from the normal distribution, and is suitable for processing abnormalities caused by internal factors such as battery material heterogeneity and manufacturing process defects. The integrated learning method (such as random forest combined with isolated forest, autoencoder network) can analyze the data from different angles, and effectively deal with complex abnormal situations caused by external factors such as rapid temperature changes, electromagnetic interference, and unstable charging and discharging equipment. By accurately identifying abnormal data and correcting or marking it for elimination, the interference of abnormal data on subsequent model training and evaluation is avoided, the reliability and effectiveness of the data are guaranteed, and the accuracy and stability of the entire lithium battery life decline prediction and health status assessment are improved, which helps to timely discover potential safety hazards and performance problems of the battery, take corresponding measures in advance, and ensure the safe and stable operation of the lithium battery system.

[0021] 4. In the present invention, based on comprehensive and accurate data collection, preprocessing, and reliable model construction and evaluation, this method can achieve quantitative prediction of lithium battery life decline and intelligent evaluation of health status, and provide scientific and reasonable decision-making basis for the use, maintenance and management of lithium batteries based on the evaluation results. For example, according to the predicted degree of life decline and the current health status, the charge and discharge strategy can be intelligently adjusted, and the charge and discharge rate can be appropriately increased when the battery performance is good to improve the use efficiency, and the charge and discharge rate can be reduced when the battery is near the end of its life or in poor health to extend the battery life and ensure safety. At the same time, it can warn of faults in advance, and promptly notify users or related systems to perform maintenance or replace batteries to avoid equipment downtime or safety accidents caused by battery failure. This intelligent decision-making and optimization management function helps to improve the use efficiency of lithium batteries, reduce maintenance costs, and extend service life. It has important economic and safety significance in many lithium battery application fields such as electric vehicles and energy storage systems, and promotes the intelligent development process of the entire lithium battery industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0023] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Please refer to the attached Figure 1 The embodiment of the present invention provides a method for quantitatively predicting the life decline of a lithium battery and intelligently evaluating its health status, comprising the following steps:

[0025] S1. Multi-source data acquisition: strain gauge sensors are installed at the battery electrode sheet and the shell to obtain stress-strain data, gas sensors are integrated in the battery module to collect gas data, and battery data and environmental data are collected and processed and stored through a high-precision synchronous acquisition system with nanosecond clock signals, while monitoring data quality;

[0026] S2, data preprocessing, expanding the data set through data transformation and combination and GAN-based data generation, and processing abnormal data using statistical and machine learning-based methods;

[0027] S3, model construction and training, select appropriate machine learning or deep learning models, divide the data set for training and adjust the parameters until convergence;

[0028] S4. Model evaluation and application: After evaluation with the test set, it is applied to the actual system and the lithium battery is managed based on the results.

[0029] Specifically, through multi-source data collection, we can comprehensively obtain the internal stress and strain of the battery, gas composition concentration, and conventional voltage, current, temperature and other information. Compared with relying solely on traditional parameters, it can more deeply reflect the complex physical and chemical changes in the battery, thereby achieving a more accurate quantitative prediction of the health status and life decline of lithium batteries. Through data transformation and combination in data preprocessing and data generation based on GAN, the limited original data set is effectively expanded, the data diversity is enriched, and the model can learn more potential features and patterns, improving the generalization ability of the model. The processed data is applied to model training, which can reduce the overfitting problem caused by insufficient data, make the model more adaptable to different working conditions and individual differences of batteries, improve the accuracy and reliability of prediction, select the appropriate model through data characteristics and optimize the training, and obtain an efficient and accurate evaluation and prediction model. In the model evaluation and application stage, models with excellent performance are screened out based on the test set evaluation results and applied to the actual system. The lithium battery charging and discharging strategy can be adjusted in real time based on the evaluation results to avoid excessive charging and discharging and other behaviors that damage the battery. Faults can be warned in advance to prevent the deterioration of faults and cause safety accidents. The timing of battery replacement can be reasonably determined to optimize the battery usage cost and system operation efficiency, and realize the intelligent management and efficient utilization of lithium batteries throughout their life cycle.

[0030] In step S1, the strain gauge sensor is based on the piezoresistive effect, and the conversion relationship of the Wheatstone bridge circuit is: and Gas sensors such as electrochemical hydrogen sensors rely on To work back on the concentration, the system collects data and checks its characteristics.

[0031] Specifically, by utilizing the piezoresistive effect of the strain gauge sensor and the Wheatstone bridge circuit, the tiny deformation of the electrode sheet and the outer shell caused by charging and discharging inside the battery can be accurately converted into measurable electrical signal changes, thereby accurately calculating the stress and strain data. This helps to gain a deeper understanding of the mechanical changes in the internal structure of the battery during operation.

[0032] In step S2, data enhancement uses data transformation and combination and GAN-based data generation. GAN-based data generation includes constructing a specific structure generator network G and a discriminator network D. The generator minimizes As the goal, the discriminator maximizes The target is the Wasserstein distance metric, and the discriminator loss function in WGAN is Generator loss function And the discriminator weight gradient is clipped; data transformation and combination include translation, scaling, flipping, time series stretching and compression of the original data and local perturbation operation and combination, and PCA is used to extract and filter features, PCA calculates the original data matrix X covariance matrix After decomposition, the principal component Y = XU (m is the number of samples, U is the eigenvector matrix), and the vector corresponding to the large eigenvalue is selected to form U new Get the reduced dimension principal component data Y new =XU new .

[0033] Specific:

[0034] In step S2, the abnormal data processing uses an abnormality detection method based on statistical models and machine learning. The abnormality detection based on the statistical model uses a multivariate Gaussian mixture model, and its initial parameters are determined according to historical normal data. Substitute real-time data into the calculated probability density value According to the probability threshold ε, the abnormality (α j , μ j ,∑ j are the weight, mean, and covariance matrix of the jth Gaussian component, k is the number of Gaussian components, is the Gaussian distribution probability density function); machine learning-based anomaly detection uses an ensemble learning method, combining random forest and isolation forest algorithms. Random forest consists of multiple decision trees T. Isolation forest is based on the degree of isolation of data points and also uses an autoencoder network to calculate the reconstruction error. E is the encoder, D is the decoder) judges the anomaly, corrects the correctable abnormal data, and removes the uncorrectable marks.

[0035] In step S3, a neural network and a support vector machine model or a convolutional neural network, a recurrent neural network model are selected, the preprocessed data set is divided into a training set, a validation set and a test set, the model is trained using the training set, the model hyperparameters are adjusted according to the validation set performance, and the stochastic gradient descent and Adam optimization algorithms are used for training until the model converges.

[0036] Specifically, by selecting a neural network and a support vector machine model or a convolutional neural network, a recurrent neural network model, and dividing the preprocessed data set into a training set, a validation set, and a test set, the model can accurately learn the data rules during training, monitor the performance with the help of the validation set to prevent overfitting, and objectively evaluate the performance with the test set; it is beneficial for the model to learn data diversity and adapt to changes in data distribution, and it can also provide a basis for model selection and fusion, thereby effectively realizing the quantitative prediction of lithium battery life degradation and intelligent evaluation of health status, improving the accuracy, reliability and adaptability of the evaluation, and providing strong support for the rational use, maintenance and management strategy formulation of lithium batteries. By using the training set to train the model, the model can be prompted to learn features and rules from the data. Adjusting the hyperparameters according to the performance of the validation set can effectively balance the model fitting and generalization capabilities, avoid overfitting and underfitting, and use the stochastic gradient descent and Adam optimization algorithms to accelerate the training process and ensure the stable convergence of the model, and finally obtain an accurate and adaptable model, which can realize the accurate quantitative prediction of lithium battery life degradation and reliable intelligent evaluation of health status, and provide key support for efficient management and safety assurance in lithium battery related applications.

[0037] In step S4, the test set is used to evaluate the model performance, and the evaluation indicators include accuracy, recall, F1 value, mean square error, and mean absolute error.

[0038] Specifically, by using the test set and related evaluation indicators to evaluate the model performance, the accuracy, stability and reliability of the model in predicting the life degradation of lithium batteries and assessing their health status can be accurately quantified, and the performance of the model on unseen data can be intuitively displayed, thereby effectively judging the pros and cons and applicability of the model, and providing key basis and decision-making guidance for optimizing the model or determining its application value in actual lithium battery management scenarios.

[0039] In step S1 , the battery data includes but is not limited to current and voltage, and the environmental data includes but is not limited to temperature, humidity, vibration, and air pressure.

[0040] Specifically, by collecting multi-dimensional information such as battery data such as current and voltage, and environmental data such as temperature, humidity, vibration, air pressure, etc., it is possible to comprehensively and deeply characterize the operating status of lithium batteries under different working conditions and environmental conditions, and accurately capture the details of the interaction between the internal electrochemical reactions of the battery and the external environment, providing rich and critical basis for accurately predicting battery life degradation and assessing health status, effectively improving the battery management system's ability and reliability in monitoring lithium battery performance, fault warning, and optimizing usage strategies, and ensuring the efficient, safe, and stable operation of lithium batteries in various application scenarios.

[0041] In step S2, data enhancement also includes verifying data validity, calculating the difference in basic statistical characteristics between generated data and original data, using dynamic time warping algorithm to measure time series similarity, applying enhanced data sets to preliminary training tests of multiple machine learning models and comparing the performance of models trained with original data sets, and using cross-validation technology to evaluate the impact of data enhancement on model generalization ability.

[0042] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for quantitatively predicting the life decline of lithium batteries and intelligently evaluating their health status, characterized in that: The following steps are involved: S1. Multi-source data acquisition: strain gauge sensors are installed at the battery electrode sheet and the shell to obtain stress-strain data, gas sensors are integrated in the battery module to collect gas data, and battery data and environmental data are collected and processed and stored through a high-precision synchronous acquisition system with nanosecond clock signals, while monitoring data quality; S2, data preprocessing, expanding the data set through data transformation and combination and GAN-based data generation, and processing abnormal data using statistical and machine learning-based methods; S3, model construction and training, select appropriate machine learning or deep learning models, divide the data set for training and adjust the parameters until convergence; S4. Model evaluation and application: After evaluation with the test set, it is applied to the actual system and the lithium battery is managed based on the results.

2. The method for quantitatively predicting the life decline and intelligently evaluating the health status of lithium batteries according to claim 1 is characterized by: In step S1, the strain gauge sensor is based on the piezoresistive effect, and the conversion relationship of the Wheatstone bridge circuit is: and Gas sensors such as electrochemical hydrogen sensors rely on To work back on the concentration, the system collects data and checks its characteristics.

3. The method for quantitatively predicting the life decline and intelligently evaluating the health status of a lithium battery according to claim 1 is characterized in that: In step S2, data enhancement uses data transformation and combination and GAN-based data generation. GAN-based data generation includes constructing a specific structure generator network G and a discriminator network D. The generator minimizes As the goal, the discriminator maximizes The target is the Wasserstein distance metric, and the discriminator loss function in WGAN is Generator loss function And the discriminator weight gradient is clipped; data transformation and combination include translation, scaling, flipping, time series stretching and compression of the original data, and local perturbation operation and combination, and PCA is used to extract and filter features, PCA calculates the original data matrix X covariance matrix After decomposition, the principal component Y = XU (m is the number of samples, U is the eigenvector matrix), and the vector corresponding to the large eigenvalue is selected to form U new Get the reduced dimension principal component data Y new =XU new .

4. The method for quantitatively predicting the life decline and intelligently evaluating the health status of a lithium battery according to claim 1, characterized in that: In step S2, the abnormal data processing uses an abnormality detection method based on statistical models and machine learning. The abnormality detection based on the statistical model uses a multivariate Gaussian mixture model, and its initial parameters are determined according to historical normal data. Substitute real-time data into the calculated probability density value According to the probability threshold ε, the abnormality (α j , μ j ,∑ j are the weight, mean, and covariance matrix of the jth Gaussian component, k is the number of Gaussian components, is the Gaussian distribution probability density function); anomaly detection based on machine learning adopts an ensemble learning method, combining random forest and isolation forest algorithms. Random forest consists of multiple decision trees T i The isolation forest is based on the degree of isolation of data points and also uses the autoencoder network to calculate the reconstruction error. E is the encoder, D is the decoder) judges the anomaly, corrects the correctable abnormal data, and removes the uncorrectable marks.

5. The method for quantitative prediction of lithium battery life decline and intelligent health status assessment according to claim 1 is characterized by: In step S3, a neural network and a support vector machine model or a convolutional neural network, a recurrent neural network model are selected, the preprocessed data set is divided into a training set, a validation set and a test set, the model is trained using the training set, the model hyperparameters are adjusted according to the validation set performance, and the stochastic gradient descent and Adam optimization algorithms are used for training until the model converges.

6. The method for quantitative prediction of lithium battery life decline and intelligent health status assessment according to claim 1 is characterized by: In step S4, the test set is used to evaluate the model performance, and the evaluation indicators include accuracy, recall, F1 value, mean square error, and mean absolute error.

7. The method for quantitatively predicting life decline and intelligently evaluating health status of lithium batteries according to claim 6, characterized in that: In step S1, the battery data includes but is not limited to current and voltage, and the environmental data includes but is not limited to temperature, humidity, vibration, and air pressure.

8. The method for quantitative prediction of lithium battery life decline and intelligent health status assessment according to claim 1 is characterized by: In step S2, data enhancement also includes verifying data validity, calculating the difference in basic statistical characteristics between generated data and original data, using a dynamic time warping algorithm to measure time series similarity, applying the enhanced data set to preliminary training tests of multiple machine learning models and performing comparative analysis on the performance of the models trained with the original data set, and using cross-validation technology to evaluate the impact of data enhancement on the generalization ability of the model.

Images