Battery life prediction method based on estimating decay curve parameters using lstm
By combining a parametric model with an LSTM network, early battery cycle features are extracted and a cross-cycle mapping relationship is established, which solves the problem of insufficient battery life prediction accuracy in real vehicle scenarios and achieves high accuracy and stability prediction with limited data.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA AUTOMOTIVE ENG RES INST
- Filing Date
- 2025-04-03
- Publication Date
- 2026-05-19
AI Technical Summary
Existing battery life prediction technologies struggle to obtain complete degradation cycle data in real-world vehicle scenarios, leading to insufficient model training and decreased prediction accuracy.
By combining a parametric model with an LSTM network, a cross-cycle mapping relationship of "early features → aging parameters" is established by extracting the feature set of the early cycle of the battery. The LSTM model is then used to learn the nonlinear correlation between battery performance degradation signals and long-term degradation trajectories to predict battery life.
With limited historical data, it significantly improves the accuracy of battery life prediction and the model's generalization ability, adapting to different battery types and aging conditions.
Smart Images
Figure CN120254630B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery technology, and more specifically to a battery life prediction method based on LSTM estimation of degradation curve parameters. Background Technology
[0002] Lithium-ion batteries, as high-efficiency energy storage devices, are widely used in electric vehicles, energy storage systems, and other fields. As battery usage time increases, their capacity gradually decreases. Accurately predicting the remaining battery life (RUL) is crucial for maintenance planning, reuse, and safety early warning. Current battery life prediction technologies mainly include:
[0003] Model-based methods describe battery aging mechanisms by establishing electrochemical or empirical models, such as equivalent circuit models and SEI film growth models, to characterize the relationship between battery characteristics and capacity decay. However, these methods rely on accurate modeling of the physicochemical reactions inside the battery, require extensive experimental data to support parameter calibration, and have limited model generalization capabilities, making it difficult to adapt to the complexity of different battery types and operating conditions.
[0004] Data-driven approaches, fueled by the accumulation of battery monitoring data, have led to the introduction of machine learning algorithms (such as Support Vector Regression (SVR) and Gaussian Process Regression (GPR)) and deep learning networks (such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN)) into the field of battery life prediction. These methods extract features from historical charge-discharge data to construct a mapping between capacity decay and cycle count. While deep learning algorithms possess automatic feature extraction capabilities and can directly process raw time-series data, their performance is highly dependent on large-scale labeled data (such as tagged capacity from complete charge-discharge cycles). In practical applications, especially in real-world vehicle scenarios, early battery cycle data is limited (such as data within the warranty period), and battery failure cycles are long, making it difficult to obtain complete decay cycle data. This results in insufficient model training and a significant decrease in prediction accuracy. Summary of the Invention
[0005] The purpose of this invention is to propose a battery life prediction method based on LSTM estimation of degradation curve parameters, which can improve the accuracy of battery life prediction when historical data is limited.
[0006] To achieve the above objectives, this invention provides a battery life prediction method based on LSTM estimation of degradation curve parameters, comprising:
[0007] Collect charge and discharge data for each cycle of different batteries to build a battery pack database;
[0008] Based on charging process data, charging segments are screened, and tag capacity is calculated using ampere-hour integration and OCV-SOC correction method;
[0009] Based on the capacity decay curve trajectory, a parametric model is used for fitting to obtain the aging parameter set for fitting the capacity decay curve of the battery pack.
[0010] Extract feature sets that are highly correlated with lifetime prediction under different early-cycle conditions of batteries;
[0011] The extracted feature set is used as the input of the model, and the aging parameter set is used as the output of the model. A battery aging parameter estimation model is established based on LSTM.
[0012] After obtaining the battery capacity aging parameters, the number of later battery cycles is substituted into the parameterized model to obtain the battery life prediction.
[0013] Beneficial effects of the basic solution: Traditional battery life prediction methods often require a large amount of historical charge-discharge data to train the model, which not only increases the difficulty of data collection but also limits the application of the model in real-world vehicle scenarios. In real-world scenarios, battery packs typically only have early operational data available (such as data within the warranty period of electric vehicles), and battery data is often difficult to obtain comprehensively and continuously. This method can directly use the charge-discharge segments recorded by the existing BMS to complete life prediction without waiting for the battery to completely fail. Furthermore, this technical solution significantly reduces the need for historical data by extracting early cyclic feature sets that are highly correlated with life prediction and using an LSTM model for aging parameter estimation. Even with only limited early battery data, the remaining battery life can be predicted relatively accurately.
[0014] This technical solution combines parametric models with data-driven methods. It utilizes parametric models to describe the intrinsic laws governing battery life degradation and employs an LSTM network to establish a cross-cycle mapping relationship between "early features → aging parameters." By learning the nonlinear correlation between early performance degradation signals and long-term degradation trajectory parameters end-to-end, it overcomes the limitation of traditional methods that require observation of the entire degradation cycle to model the battery. The advantage of this solution lies in its ability to flexibly select or adjust the parametric model based on different battery types and their degradation characteristics, thereby more accurately capturing the degradation rate of the battery under different aging states. Furthermore, the LSTM model has powerful sequence data processing capabilities, enabling it to learn and memorize long-term dependencies in battery charge and discharge data, further improving prediction accuracy. This not only enhances the model's understanding of the intrinsic mechanisms of battery life degradation but also improves its generalization ability across different battery types and aging states, making the prediction results more reliable and stable.
[0015] As a feasible and preferred approach, charging segments are screened based on charging process data, including the following:
[0016] The missing data is filled using interpolation.
[0017] Analyze battery cycle data, preprocess the data, and smooth the data using a moving average filter. Let the window size be... For data sequences The result after moving average filtering for:
[0018]
[0019] in, ,when When the data exceeds the range, use the boundary value;
[0020] Select the charging segments that meet the requirements; based on the battery's OCV-SOC table, correct the SOC based on OCV to reduce the error in SOC estimation and obtain an accurate SOC.
[0021] As a feasible and preferred solution, the battery's labeled capacity is calculated using the ampere-hour integration method. The formula for the ampere-hour integration method is as follows:
[0022]
[0023] In the formula, and These are the battery at time steps. and SOC; Is the battery in time step The current, It is the cumulative charge obtained by integrating the current over time.
[0024] As a feasible and preferred approach, based on the capacity decay curve trajectory, a parametric model is used for fitting to obtain the aging coefficient set of the battery pack's capacity decay curve fitting, including the following:
[0025] Based on the calculated tag capacity, plot the capacity decay diagram of the battery pack;
[0026] Observe the shape of the battery capacity decay curve, analyze the inherent law of battery life decay, and establish a parameterized model;
[0027] By fitting the capacity decay curve of the battery pack with different parameterization models, the model parameter sets under different parameterization models are obtained;
[0028] Error analysis is performed on the model parameter set to select the optimal fitting aging parameter set.
[0029] As a feasible and preferred approach, error analysis of the model parameter set is performed using the maximum absolute error (MaxAE), root mean square error (RMSE), and mean absolute error (MAE), expressed as follows:
[0030]
[0031] Compare these error indices under different parameterization models.
[0032] As a feasible and preferred approach, a feature set highly correlated with lifetime prediction under different early battery cycles is extracted, including the following:
[0033] Feature datasets are extracted based on early cycle data of the battery pack;
[0034] Correlation analysis was performed on the extracted features using the Pearson correlation coefficient to obtain a feature dataset that is highly correlated with battery capacity decay.
[0035] As a feasible and preferred approach, a battery aging parameter estimation model is established based on LSTM, which specifically includes the following:
[0036] An LSTM model was built, consisting of three hidden layers and three fully connected layers. Activation functions were set, and test and training sets were established for the extracted feature data and aging parameters. The model was then trained. The hyperparameters of the LSTM model were optimized using the Bayesian optimization method.
[0037] As a feasible preferred option, the LSTM model also includes a ReLU activation function and a linear activation function, with the ReLU activation function used for the hidden layers and the linear activation function used for the last fully connected layer.
[0038] As a feasible and preferred approach, test and training sets are established based on the extracted feature data and aging parameters, and the model is trained, including the following:
[0039] Extracted feature data ( ) and aging parameters ( Establish test and training sets, and reshape them into three-dimensional arrays of shape [number of samples, sequence length, number of features]. Then train the model, input the test data into the trained model, and estimate the aging parameters.
[0040] As a feasible and preferred approach, for the hyperparameters in the LSTM model, based on the Bayesian optimization method, the range of hyperparameters is set, the objective function is defined, and the optimal combination of hyperparameters is found. Attached Figure Description
[0041] Figure 1 This is a flowchart illustrating the overall method of the present invention.
[0042] Figure 2 This is a schematic diagram of the logic of the present invention.
[0043] Figure 3 This is a logical diagram of step S2 in an embodiment of the present invention.
[0044] Figure 4 This is a logical diagram of step S3 in an embodiment of the present invention.
[0045] Figure 5 This is a logical diagram of step S4 in an embodiment of the present invention.
[0046] Figure 6 This is a logical diagram of step S5 in an embodiment of the present invention.
[0047] Figure 7 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation
[0048] To make the technical solution and advantages of this application clearer, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. It is understood that the specific embodiments described herein are only some embodiments of the present invention, and are only used to explain this application, not to limit it. It should be noted that the technical features or combinations of technical features described in the following embodiments should not be considered isolated; they can be combined with each other to achieve better technical effects. The same reference numerals appearing in the accompanying drawings of the following embodiments represent the same features or components, and can be applied to different embodiments.
[0049] Furthermore, unless otherwise defined, the technical or scientific terms used in this invention description shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains.
[0050] The present invention will now be described in further detail with reference to the accompanying drawings:
[0051] Reference numerals: Electronic device 500, processor 501, communication interface 502, memory 503, bus 504.
[0052] Reference Figure 1 and Figure 2 This disclosure provides a battery life prediction method based on LSTM estimation of degradation curve parameters, including the following steps:
[0053] Step S1: Collect charge / discharge data for each cycle of different batteries to build a battery pack database. Charge / discharge data is typically recorded by the Battery Management System (BMS), including current, voltage, time, temperature, and SOC data. After collection, the charge / discharge data is organized and stored in the database to construct the battery pack database.
[0054] Step S2: Based on the charging process data, the charging segments are screened, and those that meet the requirements are selected. The tag capacity is calculated using ampere-hour integration and OCV-SOC correction methods, referring to... Figure 3 ,include:
[0055] Step S21: During data acquisition, some data may be missing due to sensor malfunction or communication problems. For missing data such as current, voltage, and temperature, interpolation can be used to fill in the missing data.
[0056] The data is preprocessed by smoothing it using a moving average filter. The window size is set to 1. For data sequences The result after moving average filtering for:
[0057]
[0058] in, ,when When the data exceeds the range, the boundary value is used.
[0059] Not all charging segments are suitable for calculating tag capacity. Screening criteria can be set as charging current being stable within a certain range, charging voltage being within the normal range, etc.
[0060] Step S22: Correct the SOC based on the OCV (Open Circuit Voltage) according to the OCV-SOC table (Open Circuit Voltage-State of Charge table). When the battery is in an open circuit state (i.e., not undergoing charging or discharging operations), measure its open circuit voltage and then look up the corresponding SOC value in the OCV-SOC table.
[0061] Step S23: Calculate the battery's tag capacity using the ampere-hour integration method. The calculation formula is as follows:
[0062]
[0063] in, and These are the battery pack at the time step. and SOC, It is the battery pack in time step The current; This is the maximum available capacity of the battery pack. Because... It can be obtained by integrating the current recorded by the BMS over time, so the key to obtaining accurate battery capacity is to obtain accurate... By looking up the SOC in the OCV table, the error in SOC estimation is reduced, and an accurate SOC is obtained.
[0064] Step S3: Based on the capacity decay curve trajectory, a parametric model is used for fitting to obtain the aging coefficient set of the battery pack's capacity decay curve fitting, referring to... Figure 4 ,include:
[0065] Step S31: Based on the calculated tag capacity, plot the battery pack capacity decay diagram with the number of cycles as the horizontal axis and the tag capacity as the vertical axis.
[0066] Step S32: Observe the shape of the battery capacity degradation curve, analyze the inherent law of battery life degradation, and establish a parameterized model. In this embodiment, the optional battery degradation parameterized model includes, but is not limited to, the following:
[0067]
[0068] Step S33: Fit the capacity decay curve of the battery pack using different parameterization models to obtain the model parameter set under different parameterization models.
[0069] Step S34: Perform error analysis on the model parameter set using metrics such as maximum absolute error (MaxAE), root mean square error (RMSE), and mean absolute error (MAE). The expressions for these metrics are:
[0070]
[0071] By comparing these error indices under different parameterized models, the aging coefficient set corresponding to the model with the smallest error is selected as the optimal fitting aging coefficient set.
[0072] Step S4: Based on the lifetime prediction requirements, extract feature sets highly correlated with lifetime prediction under different early-stage battery cycles, referring to... Figure 5 ,include:
[0073] Step S41: Extract a feature dataset based on the early cycling data of the battery pack. Features in the feature dataset include, but are not limited to:
[0074] Charging capacity: The amount of electricity the battery receives during each charge, calculated by integrating the charging current over the charging time, and measured in ampere-hours (Ah).
[0075] Charging time: The time taken from the start of charging to the end of charging, in minutes (min) or hours (h).
[0076] Average charging temperature: The average temperature of the battery during the charging process, which is obtained by averaging the temperature data at various time points during the charging process, and the unit is degrees Celsius (°C).
[0077] Charging voltage change rate: The amount of change in charging voltage per unit time during the charging process, reflecting the trend of voltage change during the charging process.
[0078] Step S42: Perform correlation analysis on the extracted features to determine which features are highly correlated with battery capacity degradation. The Pearson correlation coefficient can be used for correlation analysis. The expression for the Pearson correlation coefficient is:
[0079]
[0080] Based on the calculation results, feature datasets with high correlation to battery capacity decay were selected.
[0081] Step S5: Using the extracted feature set as the model input and the aging parameter set as the model output, establish a battery aging parameter estimation model based on LSTM, referring to... Figure 6 The specific steps include:
[0082] Step S51: Establish the LSTM model. In this embodiment, the LSTM estimation model consists of three hidden layers and three fully connected layers. Each layer consists of multiple units, and all units in each layer are interconnected. The number of units decreases sequentially, and the number of units output by the last fully connected layer is the number of parameterized model coefficients.
[0083] The model activation functions include ReLU and linear activation functions. Specifically, ReLU is used in the hidden layers to mitigate the vanishing gradient problem and increase the model's non-linear expressiveness. Linear activation functions are used in the last fully connected layer for regression tasks, outputting the predicted aging parameter values.
[0084] Extracted feature data ( ) and aging parameters ( Establish test and training sets, and reshape them into three-dimensional arrays of shape [number of samples, sequence length, number of features]. Then train the model, input the test data into the trained model, and estimate the aging parameters.
[0085] Step S52: Optimize the hyperparameters of the LSTM model using Bayesian optimization. For hyperparameters in the LSTM model such as the number of hidden layer units, number of layers, learning rate, learning decay rate, and batch size, the optimal hyperparameter combination is found using Bayesian optimization. This improves model accuracy and yields the best estimation result.
[0086] S6: After obtaining the battery capacity aging parameters, substitute the battery's later cycle number into the parameterized model to obtain the final battery life prediction.
[0087] This disclosure also provides a battery life prediction system based on LSTM estimation of degradation curve parameters, which utilizes the aforementioned battery life prediction method based on LSTM estimation of degradation curve parameters.
[0088] Those skilled in the art will understand that implementing all or part of the battery life prediction method based on LSTM-based estimation of degradation curve parameters can be accomplished by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium. When executed, the program can include the processes of various embodiments of the battery life prediction method based on LSTM-based estimation of degradation curve parameters. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0089] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the aforementioned battery life prediction method based on LSTM estimation of degradation curve parameters. In this application embodiment, the processor is the control center of the computer method and can be a physical machine processor or a virtual machine processor.
[0090] Reference Figure 7The electronic device 500 includes at least one processor 501, at least one communication interface 502, at least one memory 503, and at least one bus 504. The bus 504 is used for communication between these components, the communication interface 502 is used for signaling or data communication with other node devices, and the memory 503 stores machine-readable instructions executable by the processor 501. When the electronic device 500 is running, the processor 501 communicates with the memory 503 via the bus 504. When the machine-readable instructions are invoked by the processor 501, they execute the steps of the battery life prediction method based on LSTM estimation of degradation curve parameters as described above.
[0091] The above content is merely an embodiment of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can improve and implement this solution based on the guidance provided in this application and their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.
Claims
1. A battery life prediction method based on LSTM estimation of degradation curve parameters, characterized in that: include: Collect charge and discharge data for each cycle of different batteries to build a battery pack database; Based on charging process data, charging segments are screened, and tag capacity is calculated using ampere-hour integration and OCV-SOC correction method; Based on the capacity decay curve trajectory, a parametric model is used for fitting to obtain the aging parameter set for fitting the capacity decay curve of the battery pack. Extract feature sets that are highly correlated with lifetime prediction under different early-cycle conditions of batteries; The extracted feature set is used as the input of the model, and the aging parameter set is used as the output of the model. A battery aging parameter estimation model is established based on LSTM. After obtaining the battery capacity aging parameters, the number of later battery cycles is substituted into the parameterized model to obtain the battery life prediction.
2. The battery life prediction method based on LSTM estimation of degradation curve parameters according to claim 1, characterized in that: Based on charging process data, charging segments are filtered, including the following: Use interpolation to fill in the missing data; Analyze battery cycle data, preprocess the data, and smooth the data using a moving average filter. Let the window size be... For data sequences The result after moving average filtering for: in, ,when When the data exceeds the range, use the boundary value; Select the charging segments that meet the requirements; based on the battery's OCV-SOC table, correct the SOC based on OCV to reduce the error in SOC estimation and obtain an accurate SOC.
3. The battery lifetime prediction method based on LSTM estimation of degradation curve parameters according to claim 2, characterized in that: The battery's labeled capacity is calculated using the ampere-hour integration method. The formula for the ampere-hour integration method is as follows: In the formula, and These are the battery at time steps. and SOC; Is the battery in time step The current, It is the cumulative charge obtained by integrating the current over time.
4. The battery life prediction method based on LSTM estimation of degradation curve parameters according to claim 1, characterized in that: Based on the capacity decay curve trajectory, a parametric model is used for fitting to obtain the aging coefficient set of the battery pack's capacity decay curve fitting, including the following: Based on the calculated tag capacity, plot the capacity decay diagram of the battery pack; Observe the shape of the battery capacity decay curve, analyze the inherent law of battery life decay, and establish a parameterized model; By fitting the capacity decay curve of the battery pack with different parameterization models, the model parameter sets under different parameterization models are obtained; Error analysis is performed on the model parameter set to select the optimal fitting aging parameter set.
5. The battery life prediction method based on LSTM estimation of degradation curve parameters according to claim 1, characterized in that: Error analysis of the model parameter set is performed using the maximum absolute error (MaxAE), root mean square error (RMSE), and mean absolute error (MAE), expressed as follows: Compare these error indices under different parameterization models.
6. The battery life prediction method based on LSTM estimation of degradation curve parameters according to claim 1, characterized in that: We extracted feature sets highly correlated with lifetime prediction under different early-stage battery cycles, including the following: Feature datasets are extracted based on early cycle data of the battery pack; Correlation analysis was performed on the extracted features using the Pearson correlation coefficient to obtain a feature dataset that is highly correlated with battery capacity decay.
7. The battery life prediction method based on LSTM estimation of degradation curve parameters according to claim 1, characterized in that: A battery aging parameter estimation model based on LSTM is established, specifically including the following: An LSTM model was built, consisting of three hidden layers and three fully connected layers. Activation functions were set, and test and training sets were established for the extracted feature data and aging parameters. The model was then trained. The hyperparameters of the LSTM model were optimized using the Bayesian optimization method.
8. The battery life prediction method based on LSTM estimation of degradation curve parameters according to claim 7, characterized in that: The LSTM model also includes the ReLU activation function and the linear activation function. The ReLU activation function is used in the hidden layers, and the linear activation function is used in the last fully connected layer.
9. The battery life prediction method based on LSTM estimation of degradation curve parameters according to claim 7, characterized in that: Establish test and training sets for the extracted feature data and aging parameters, and train the model, including the following: Extracted feature data ( ) and aging parameters ( Establish test and training sets, and reshape them into three-dimensional arrays of shape [number of samples, sequence length, number of features]. Then train the model, input the test data into the trained model, and estimate the aging parameters.
10. The battery life prediction method based on LSTM estimation of degradation curve parameters according to claim 7, characterized in that: For the hyperparameters in the LSTM model, based on the Bayesian optimization method, we set the range of hyperparameters, define the objective function, and find the optimal combination of hyperparameters.