Method for predicting residual life of battery based on dual drive of machine learning and physical model

Abstract

The invention requests to protect a method for predicting the residual life of a battery based on dual drive of machine learning and a physical model, and the method comprises the following steps: 1, carrying out the preprocessing of collected data, including data cleaning, outlier processing, data normalization and the like; 2, extracting characteristics related to the service life of the battery from the preprocessed data, and screening out related characteristics which have the greatest influence on the prediction of the service life of the battery through a principal component analysis characteristic selection method; 3, based on the physical equation and the internal parameters of the battery, establishing a mathematical model of the battery, and extracting parameters for predicting the RUL of the battery from the charging capacity curve by using a least square algorithm; and 4, selecting a deep learning algorithm for predicting the discharge capacity of the battery from a battery material and a working environment. Meanwhile, the features extracted from the discharge capacity in combination with the physical model are used as the input of the DNN model, and the RUL of the battery is used as the output for training. The method is suitable for different types of batteries and different application scenes, and has high universality and adaptability.

Description

Technical Field

[0001] The present invention relates to the fields of machine learning algorithms and battery physical models, and particularly to a method for predicting the remaining useful life of a battery driven by both a deep learning algorithm and a physical model. Background Art

[0002] With the wide application of batteries in multiple fields such as electric vehicles, energy storage systems, and portable electronic devices, mastering the decay and remaining useful life (RUL) of batteries is crucial for ensuring the stable operation of devices, improving energy utilization efficiency, and ensuring user safety. By predicting the battery life, the battery design and usage strategies can be optimized, the service life of the battery can be extended, the replacement cost can be reduced, and the large-scale application of batteries can be further promoted.

[0003] Machine learning algorithms are important means for modeling and processing complex data, finding patterns, and applying feedback. In battery life prediction, machine learning algorithms predict the remaining useful life of a battery by analyzing the data (such as voltage, current, temperature, capacity, etc.) collected during the charging and discharging cycles of the battery. This method does not require the construction of precise mathematical and physical models, and thus has great flexibility and adaptability. In the dual-drive prediction method, machine learning algorithms are used to extract features from massive data and train prediction models, while physical models are used to describe the physical processes and parameter changes inside the battery. By combining the two, accurate prediction and in-depth understanding of battery life can be achieved. This method combines the flexibility of machine learning algorithms and the accuracy of physical models, and can more comprehensively capture the influencing factors of battery life. However, due to the current application of machine learning algorithms in batteries focusing on voltage and specific energy, and in the case of capacity, more attention is paid to the discharge capacity of the battery. The prediction of the remaining life of the battery is usually based on the equivalent circuit model of the battery, which often requires rich experimental data and professional knowledge, and has a high computational complexity. To solve the above problems, the present invention designs a method for predicting the remaining useful life (RUL) of a battery driven by both a machine learning algorithm and a physical model. In the dual-drive prediction method, machine learning algorithms are used to extract features from massive data and train prediction models, while physical models are used to describe the physical processes and parameter changes inside the battery. By combining the two, accurate prediction and in-depth understanding of battery life can be achieved.

[0004] After retrieval, the application publication number is CN114675187A, which is a method for predicting the life of a lithium-ion battery by integrating physical mechanisms and machine learning. The method includes the following steps: The lithium-ion battery undergoes the first charge and discharge cycle and electrochemical characteristic measurements are carried out; An electrochemical characteristic curve is constructed and characteristic quantity information is extracted; A machine learning model is constructed and the characteristic quantities are brought into the machine learning model to evaluate the life performance differences of the lithium-ion battery at the time of factory; A physical degradation mechanism model is constructed to evaluate the life attenuation of the lithium-ion battery caused by the charge and discharge history; Based on the factory performance differences of the lithium-ion battery and the life attenuation during the charge and discharge process, the remaining life of the lithium-ion battery is calculated. Based on the integration of the physical degradation mechanism model and the machine learning model, the present invention can accurately and reliably predict the life of the lithium-ion battery only according to the evaluation results of two main factors, namely, the initial life inconsistency at the time of factory of the lithium-ion battery and the degradation of the charge and discharge cycle history, which is simple and efficient.

[0005] Most of the applications of machine learning in battery electrode materials focus on predicting the electrochemical performance of the battery from the material composition, or predicting the remaining life of the battery from the charge and discharge curves of the battery and other battery operating characteristics. There are few research works that directly predict the remaining life of the battery starting from the battery material composition and working environment. The present invention completes the prediction of the remaining life of the battery starting from the battery material composition and working environment by training two different deep learning models respectively and embedding the relevant mechanisms of the physical model, improving the interpretability of the RUL prediction of the battery deep learning model. Summary of the Invention

[0006] The present invention aims to solve the above problems of the prior art. A method for predicting the remaining life of a battery driven by both machine learning and a physical model is proposed. The technical solution of the present invention is as follows:

[0007] A method for predicting the remaining life of a battery driven by both machine learning and a physical model, which includes the following steps:

[0008] Step 1: Collect the charging data, temperature data, and internal resistance data of the battery from the public dataset, and perform preprocessing operations on the collected data, including data cleaning, outlier processing, and data normalization, to extract the required data volume and features;

[0009] Step 2: Respectively train deep learning models DNN for the charging capacity and the remaining useful life RUL; When the charging capacity is used as the output, the battery material components and the working environment are used as the input; When RUL is used as the output, the parameters extracted from the physical model through the charging capacity curve are used as the input;

[0010] Step 3: After model training, using the battery material composition and working environment as inputs, through the DNN model for predicting the charging capacity, predict the battery charging capacity at different cycle numbers and different voltages; use the least squares method to extract parameters from the predicted charging capacity, and through the DNN model for predicting RUL, finally achieve a complete prediction of the battery RUL.

[0011] Further, in step 2, using the battery material and working environment as inputs, predict the charging capacity of the battery; based on the physical equations and internal parameters of the battery, establish a mathematical model of the battery, and use the least squares method to identify the parameters of the physical model through the charging capacity of the battery. Use the parameters as inputs to predict the remaining life of the battery.

[0012] Further, establishing a mathematical model of the battery based on the physical equations and internal parameters of the battery, and using the least squares method to identify the parameters of the physical model through the charging capacity of the battery specifically includes: through the analysis of the capacity curve of the charging capacity, extract the parameters related to aging from the data of the charging capacity Q and the current voltage V. These parameters have a physical relationship with the remaining life of the battery. In this paper, starting from the charging capacity, use the least squares method to identify the parameters related to the remaining life and use them as input features for predicting the remaining life of the battery.

[0013] Further, in step 3, after the DNN model is trained, using the material data and working environment as inputs, after using the first DNN model to predict its charging capacity, extract the physical model parameters from the predicted charging capacity curve as inputs, and use the second DNN model to predict the remaining life of the battery; completely realize predicting the remaining life of the battery starting from the battery material and the working environment of the battery.

[0014] Further, the first DNN model is a deep learning model for predicting the charging capacity, and the input features are the charging current, working temperature, voltage window, and material composition of the battery, where the material composition is four features selected after principal component analysis from the features generated by the matminer material package;

[0015] The second DNN model is a deep learning model for predicting the remaining life of the battery, and the charging capacitance Q can be obtained by the following formula: A i is the area of the i-th peak, ω i is the width at half height of the i-th peak, V i is the center voltage of symmetry of the i-th peak, V represents the collected voltage, and Q is the capacity. The values of V and Q can be directly obtained from the database, so the input features are the parameters A i 、V i 、ω i, during the process of training the model, the data set sources of the two groups of models are separated, and the intermediate variables of the model are only used in the complete RUL prediction.

[0016] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the method for predicting the remaining battery life based on the dual-drive of machine learning and physical models as described in any one of the above.

[0017] A non-transitory computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the method for predicting the remaining battery life based on the dual-drive of machine learning and physical models as described in any one of the above.

[0018] A computer program product includes a computer program. When the computer program is executed by a processor, it implements the method for predicting the remaining battery life based on the dual-drive of machine learning and physical models as described in any one of the above.

[0019] The advantages and beneficial effects of the present invention are as follows:

[0020] The beneficial effect of the present invention is that through the trained DNN model, it can realize the rapid prediction of the remaining battery life of battery materials under different working conditions, and can effectively reduce the time cost and resource cost required for traditional experimental measurements; in addition, through the modification of different battery materials, battery materials with good electrochemical performance can be further screened. The present invention has the advantages of simple strategy implementation, low cost, easy implementation, etc., and also has good scalability. Innovatively embed the physical model mechanism into the field of battery materials to more quickly judge the expected remaining battery life of battery materials. This method integrates domain knowledge into deep learning and uses battery material data to enhance the RUL prediction of traditional data-driven methods. And this method is applicable to the RUL prediction of different battery material compositions under different working environments, such as different charging rates, temperatures, material compositions, etc. Based on this work, further research on different data and knowledge integration methods will be an interesting research direction. Description of the Drawings

[0021] Figure 1 is the overall architecture diagram and implementation principle diagram of the preferred embodiment provided by the present invention;

[0022] Figure 2 are the features used in the two DNN models of the present invention;

[0023] Figure 3 is the model training effect diagram of the present invention. Detailed Embodiments

[0024] The technical solutions in the embodiments of the present invention will be clearly and detailedly described below in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention.

[0025] The technical solution for the present invention to solve the above technical problems is as follows:

[0026] A method for predicting the remaining battery life driven by both machine learning and physical models, which includes the following steps:

[0027] 1. Data collection and preprocessing. Collect charging data, temperature data, internal resistance data, etc. of the battery. These data are collected by the battery management system (BMS) or sensors and other devices. It is necessary to preprocess the collected data first, including data cleaning, outlier processing, data normalization, etc., to ensure the quality and consistency of the data;

[0028] 2. Feature extraction and selection. Extract features related to battery life from the preprocessed data, such as material composition, working environment, etc. Through the principal component analysis (PCA) feature selection method, screen out the relevant features that have the greatest impact on battery life prediction;

[0029] 3. Physical model establishment. Based on the physical equations and internal parameters of the battery, establish a mathematical model of the battery, and use the least squares algorithm to extract the parameters for predicting the battery RUL from the charge capacity curve;

[0030] 4. Machine learning model training. Select the deep learning (DNN) algorithm to predict the battery discharge capacity starting from the battery material and working environment. At the same time, use the features extracted from the discharge capacity combined with the physical model as the input of the DNN model, and the RUL of the battery as the output for training.

[0031] Figure 1 is the overall architecture diagram and implementation principle diagram of the present invention. As Figure 1 shown, it shows the structure of the work and the corresponding processing process. First is data support, save the required data from the database locally, and screen out the working environment, material composition, charging voltage, charging capacity (intermediate variable), and the remaining battery life extracted each cycle (final output). Secondly is DNN model training. Select the DNN algorithm as the model algorithm of the present invention, use the working environment and material composition as the input to predict the charging capacity, and use the parameters identified from the charging capacity curve by the least squares method as the input to predict the remaining battery life. Finally is the complete RUL prediction. In the present invention, the data in the previous training process comes from different sources, and the model output for predicting the charging capacity will not be used as the input for predicting the RUL. Instead, in the complete RUL prediction, the charging capacity predicted starting from the working conditions and materials will be directly used to predict the RUL to achieve the complete RUL prediction.

[0032] Figure 2 are the features used in the two DNN models of the present invention. As shown in the figure, on the left side of the figure is the deep learning model for predicting the charging capacity, and the input features are the charging current, operating temperature, voltage window, and material composition of the battery. Among them, the material composition is the features generated using the matminer material package, and four features are selected after principal component analysis. On the right side of the figure is the deep learning model for predicting the remaining life of the battery, and the charging capacitance Q can be obtained by the following formula: Therefore, the input features are the parameters A i , V i , ω i identified by the least squares method in the charging capacity curve. During the process of training the model, due to the need for the effectiveness and robustness of the model, the data set sources of the two groups of models are separated, and only in the complete RUL prediction, the intermediate variable (charging capacity) of the model will be used.

[0033] Figure 3 is the model training effect diagram of the present invention. The left figure is the prediction result of the DNN model for the charging capacity. It can be seen that most of the data can be predicted well, and only the prediction efficiency is poor at the high point of the charging capacity. This may be because most of the data volume is in the middle section, and the data volume at the capacity high point is not sufficient, but it can still prove that the DNN model has accuracy in predicting the charging capacity. The right figure is the prediction result of the DNN model for the remaining life. It can be seen that most of the data points are on the baseline. The capacity at 80% is set as the battery scrapping standard. Similarly, the predicted value at the high position of the remaining life is lower than the true value. This may be because there are fewer battery materials with high cycle numbers (the higher the RUL), and more battery cycle numbers are between 300 and 500 times. Therefore, the prediction efficiency of the DNN model at the high position is not good, but the data is normally distributed. The DNN models for predicting the charging capacity and the remaining life can both well fit the true values, which shows the accuracy of the DNN model used.

[0034] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0035] A computer-readable medium includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0036] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0037] The above embodiments should be understood as being only for illustrative purposes of the present invention and not for limiting the scope of protection of the present invention. After reading the content described in the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A method for predicting the remaining battery life based on dual drive of machine learning and physical model, characterized in that: The following steps are involved: Step 1: Collect battery charging data, temperature data, and internal resistance data from public data sets, perform preprocessing operations including data cleaning, outlier processing, and data normalization on the collected data, and extract the required data volume and features; Step 2: train a deep learning model DNN for the charging capacity and the remaining service life RUL respectively; when the charging capacity is output, the battery material composition and the working environment are used as input; when RUL is output, the parameters extracted from the physical model through the charging capacity curve are used as input; Step 3. After model training, the battery material composition and working environment are used as input, and the DNN model for predicting charging capacity is used to predict the battery charging capacity under different cycle numbers and different voltages. The least squares method is used to extract parameters from the predicted charging capacity, and the DNN model for predicting RUL is used to finally achieve a complete prediction of the battery RUL.

2. The method for predicting the remaining battery life based on dual drive of machine learning and physical model according to claim 1, characterized in that: In step 2, the battery material and working environment are used as input to predict the battery's charging capacity; based on the battery's physical equations and internal parameters, a mathematical model of the battery is established, and the parameters of the physical model are identified using the least squares method through the battery's charging capacity. The parameters are used as input to predict the remaining life of the battery.

3. The method for predicting the remaining battery life based on dual drive of machine learning and physical model according to claim 2, characterized in that: The mathematical model of the battery is established based on the physical equations and internal parameters of the battery, and the parameters of the physical model are identified by the least squares method through the battery charging capacity, including: the voltage platform on the charging capacity curve can be amplified by IC analysis, the peak of the IC curve is highly sensitive to battery degradation, and the symmetry center, width, height and area information obtained from the IC peak are related to the real phase change behavior of the material. Therefore, it is reasonable to model the charging curve from the perspective of IC transformation. These parameters have a physical relationship with the remaining life of the battery. Starting from the charging capacity, this paper uses the least squares method to identify the parameters related to the remaining life, and uses it as the input feature for predicting the remaining life of the battery.

4. The method for predicting the remaining battery life based on dual drive of machine learning and physical model according to claim 2, characterized in that: In step 3, after the DNN model is trained, the material data and working environment are used as input. After the charging capacity is predicted using the first DNN model, the physical model parameters are extracted from the predicted charging capacity curve as input, and the remaining life of the battery is predicted using the second DNN model. This fully realizes the prediction of the remaining life of the battery based on the battery material and the working environment of the battery.

5. The method for predicting the remaining battery life based on dual drive of machine learning and physical model according to claim 4, characterized in that: The first DNN model is a deep learning model for predicting charging capacity. The input features are the charging current, operating temperature, voltage window and material composition of the battery. The material composition is a feature generated by the matminer material package, and four features are selected after principal component analysis. The second DNN model is a deep learning model for predicting the remaining battery life. The charging capacitor Q can be obtained by the following formula: Where n is the number of peaks, A i is the area of ​​the ith peak, ω i is the width of the ith peak at half height, V i is the central voltage of the symmetrical i-th peak, V represents the collected voltage, Q is the capacity, and the values ​​of V and Q can be directly obtained in the database, so the input feature is the parameter A identified in the charge capacity curve using the least squares method. i 、V i ,ω i ,During the model training process, the data sets of the two groups of models come from separate sources, and the intermediate ,variables of the model are used only in the complete RUL prediction.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method for predicting the remaining battery life based on dual drive based on machine learning and physical model as described in any one of claims 1 to 5 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the method for predicting the remaining life of a battery based on dual drive based on machine learning and physical model as described in any one of claims 1 to 5.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the method for predicting the remaining life of a battery based on dual drive based on machine learning and physical model as described in any one of claims 1 to 5.

Images