Method for predicting capacity fading trend of low-temperature lithium battery

Abstract

The invention provides a capacity attenuation trend prediction method of a low-temperature lithium battery, and relates to the technical field of battery health management and prediction. The capacity fading trend prediction method of the low-temperature lithium battery specifically comprises the following steps of S1, initial data acquisition, S2, feature extraction, S3, hybrid prediction model construction, S4, model training and offline verification, S5, online deployment and dynamic correction, and S6, SOH evaluation and early warning output. A hybrid prediction model is constructed by fusing an electrochemical mechanism and deep learning, so that high-precision prediction and full-life-cycle intelligent health management of lithium battery capacity fading in a low-temperature environment are realized, and the problems of accelerated battery aging and prominent safety risk in an alpine region are effectively solved; early warning and intelligent management of the health state of the battery can be realized, and the service life of the battery under low-temperature application can be prolonged.

Description

Technical Field

[0001] This invention relates to the field of battery health management and prediction technology, specifically a method for predicting the capacity decay trend of low-temperature lithium batteries. Background Technology

[0002] Lithium-ion battery capacity decay refers to the gradual reduction in the amount of electricity a battery can store and release during repeated charge-discharge cycles or long-term storage. This is caused by irreversible chemical reactions and physical changes, such as changes in electrode material structure, electrolyte decomposition, lithium dendrite growth, SEI film thickening, and loss of active lithium. This process is affected by factors such as temperature, charge / discharge rate, cutoff voltage, and usage habits, and is a core issue limiting battery life and performance.

[0003] The existing patent (publication number: CN120870899A) discloses that "this invention relates to the field of battery health management and prediction technology, specifically a method and system for predicting the capacity decay trend of lithium batteries, including: generating a deeply fused feature by combining data-driven features and electrochemical features through a dual calibration mechanism; inputting the feature sequence into an electrochemical process sensing model and encoding it into a potential state vector representing personalized degradation; then inputting the vector as an initial condition into a neural differential equation model, learning the degradation dynamics and solving it by integration, and finally generating a continuous health state decay trajectory and obtaining the prediction result. This invention overcomes the deficiency of weak generalization ability of traditional 'black box' models by deeply integrating physical mechanisms with data-driven models, and can achieve long-term, high-precision, and continuous prediction of the degradation trend throughout the entire battery life cycle using only weak signals in the early stages of the battery, with high reliability and practical value."

[0004] In the process of realizing this application, the inventors discovered that the existing technology has the following problems: most existing methods for predicting the capacity decay trend of lithium batteries only focus on the capacity decay problem of batteries at room temperature. In some cold regions, low temperature will significantly aggravate the capacity decay of lithium batteries. On the one hand, low temperature increases the viscosity of electrolyte and decreases the ionic conductivity, resulting in an increase in battery internal resistance and a temporary decrease in discharge capacity. On the other hand, when charging at low temperature, the insertion speed of lithium ions into the negative electrode is slower, which can easily cause the precipitation of metallic lithium, resulting in irreversible loss of active lithium and safety hazards. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for predicting the capacity decay trend of low-temperature lithium batteries, which helps to achieve early warning and intelligent management of battery health status and extend the service life of batteries in low-temperature applications.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting the capacity decay trend of a low-temperature lithium battery, specifically comprising the following steps: S1. Initial Data Acquisition Multiple temperature gradients were set in the range of -80℃ to 10℃ to conduct accelerated cycle charge and discharge tests to simulate actual low-temperature usage scenarios. During the test, capacity decay, internal resistance change, voltage curve, temperature distribution and cycle number were collected to form a complete aging dataset under multiple temperature conditions. S2. Feature Extraction Based on electrochemical principles, the main aging mechanisms at low temperatures are analyzed, and core input variables are screened out through correlation analysis to build a model and construct a feature set. S3. Constructing a hybrid prediction model A hybrid prediction model is constructed by integrating an electrochemical empirical model and an LSTM deep learning model. The electrochemical empirical model uses the Arrhenius equation to describe the effect of temperature on the aging rate, providing physical constraints and prior knowledge. The LSTM deep learning model processes multi-dimensional time series features, capturing long-term time-series dependencies and nonlinear decay patterns. The two models complement each other through weighted fusion or residual correction. S4. Model Training and Offline Validation The experimental data were divided into training, validation and test sets. The LSTM network was trained using early recurrent data and the parameters of the Arrhenius equation were fitted. The prediction error was evaluated on the test set. By adjusting the number of network layers, learning rate and time window, the mean absolute percentage error was controlled within 5% to ensure that the model has reliable prediction performance. S5. Online Deployment and Dynamic Correction The trained model is connected to the battery management system. Rolling predictions are made based on real-time collected temperature, SOC, and current data. An online correction mechanism is established. When the deviation between the measured capacity and the predicted value exceeds the threshold, the model parameters are fine-tuned using new data. S6. SOH Assessment and Early Warning Output The system calculates the state of health (SOH) in real time and visualizes the capacity decay trend and remaining lifespan range. It also establishes a tiered early warning system: an SOH above 80% indicates normal operation; a range of 60% to 80% suggests monitoring and optimizing usage strategies; and a replacement reminder is triggered when the SOH is below 60% or short-term failure is predicted.

[0007] Preferably, in S1, the multiple temperature gradients include -80℃, -70℃, -60℃, -50℃, -40℃, -30℃, -20℃, -10℃, 0℃, and 10℃.

[0008] Preferably, in S2, the main aging mechanisms include restricted lithium-ion diffusion, electrolyte performance degradation, SEI film thickening, and lithium deposition.

[0009] Preferably, in S2, the core input variables include real-time temperature, state of charge, current stress, cumulative cycle count, and historical capacity retention rate.

[0010] Preferably, in S3, the formula for the Arrhenius equation is as follows:

[0011] in, The aging rate constant is For activation energy, The gas constant is... This refers to absolute temperature.

[0012] Preferably, in S3, the LSTM deep learning model includes an input layer, a hidden layer, and an output layer. The input layer is used to input multi-dimensional time series features, including temperature, SOC, current, and historical capacity. The hidden layer is used by the LSTM network to capture long-term temporal dependencies and handle nonlinear decay patterns. The output layer is used to output the future capacity retention curve and remaining cycle life.

[0013] Preferably, in step S4, the ratio of the training set, validation set, and test set is 70% for the training set, 15% for the validation set, and 15% for the test set.

[0014] Preferably, in step S6, the formula for calculating the state of health (SOH) is as follows:

[0015] in, Current available capacity This is the rated capacity.

[0016] This invention provides a method for predicting the capacity decay trend of low-temperature lithium batteries. It has the following beneficial effects: 1. This invention provides a method for predicting the capacity decay trend of low-temperature lithium batteries. By constructing a hybrid prediction framework that integrates an electrochemical empirical model and LSTM deep learning, it effectively solves the key problem of existing technologies that only focus on room temperature scenarios and ignore the aggravated capacity decay in low-temperature environments. This method uses the Arrhenius equation to provide physical constraints to ensure extrapolation stability, and at the same time, it uses an LSTM network to capture the nonlinear temporal characteristics of complex aging mechanisms such as limited lithium-ion diffusion and electrolyte degradation at low temperatures. It achieves high-precision prediction with an average absolute percentage error of less than 5% in a wide temperature range of -80℃ to 10℃. Combined with an online dynamic correction mechanism and a graded SOH early warning system, it overcomes the loss of active lithium and safety hazards caused by low-temperature charging and discharging, and provides intelligent management support for the safe, efficient and economical operation of lithium batteries throughout their entire life cycle in cold regions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the workflow of the present invention; Figure 2This is a graph showing the maximum capacity degradation curve of the lithium battery under low-temperature conditions according to the present invention. Figure 3 This is a temperature aging rate fitting graph for the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for predicting the capacity decay trend of a low-temperature lithium battery, specifically including the following steps: S1. Initial Data Acquisition Multiple temperature gradients were set within the range of -80℃ to 10℃ to conduct accelerated charge-discharge tests to simulate real-world low-temperature usage scenarios. These temperature gradients included -80℃, -70℃, -60℃, -50℃, -40℃, -30℃, -20℃, -10℃, 0℃, and 10℃. During the tests, capacity decay, internal resistance changes, voltage curves, temperature distribution, and cycle counts were collected to form a complete aging dataset under multiple temperature conditions. Battery decay data at each temperature are shown below. Figure 2 As shown; Specifically, through the above steps, a multi-dimensional aging dataset covering extreme low temperatures to room temperature is constructed. This system reveals the capacity decay law, internal resistance evolution mechanism, and thermo-electrical coupling failure mode of lithium-ion batteries under low-temperature conditions. This provides crucial data support and theoretical basis for establishing low-temperature lifetime prediction models, optimizing battery thermal management system design, and formulating cold-region usage strategies. It effectively bridges the data gap between current standard testing conditions and actual low-temperature application scenarios. Key experimental data are shown in the table below:

[0020] S2. Feature Extraction Based on electrochemical principles, the main aging mechanisms at low temperatures are analyzed. These mechanisms include restricted lithium-ion diffusion, electrolyte performance degradation, SEI film thickening, and lithium deposition. Core input variables are selected through correlation analysis, including real-time temperature, state of charge, current stress, cumulative cycle count, and historical capacity retention. A model is then established to construct a feature set. Specifically, through the above steps, the abstract battery aging process is transformed into a quantifiable and modelable feature parameter system. By identifying lithium-ion diffusion limitation, electrolyte performance degradation, SEI film thickening and lithium deposition, and other key failure mechanisms, the model input variables are ensured to have clear physicochemical meaning and accurately reflect changes in the battery's internal state. This improves the interpretability and generalization ability of the prediction model and avoids the black box problem that may occur with purely data-driven methods.

[0021] S3. Constructing a hybrid prediction model A hybrid prediction model is constructed by fusing an electrochemical empirical model and an LSTM deep learning model. The electrochemical empirical model uses the Arrhenius equation to describe the effect of temperature on the aging rate, providing physical constraints and prior knowledge. The LSTM deep learning model processes multi-dimensional time series features, capturing long-term time-series dependencies and nonlinear decay patterns. The two models are complementary through weighted fusion or residual correction. The fitting results are shown below. Figure 3 As shown; The formula for the Arrhenius equation is as follows:

[0022] in, The aging rate constant is For activation energy, The gas constant is... Absolute temperature; The LSTM deep learning model includes an input layer, a hidden layer, and an output layer. The input layer is used to input multi-dimensional time series features, including temperature, SOC, current, and historical capacity. The hidden layer is used by the LSTM network to capture long-term temporal dependencies and handle nonlinear decay patterns. The output layer is used to output the future capacity retention curve and remaining cycle life. Specifically, through the above steps, this method achieves the complementary advantages of physical mechanisms and data intelligence. The Arrhenius equation provides physical constraints on temperature-aging rate based on activation energy theory, ensuring the stability and rationality of the model in data sparse or extrapolation scenarios. Meanwhile, the LSTM network makes full use of its gating mechanism to capture long-term time-series dependencies and nonlinear dynamic features in capacity decay. The two are combined through weighted fusion or residual correction, which not only overcomes the limitation of pure empirical models in handling complex nonlinearities, but also enhances the physical consistency and extrapolation reliability of pure neural network models.

[0023] S4. Model Training and Offline Validation The experimental data was divided into training, validation, and test sets, with a ratio of 70% for training, 15% for validation, and 15% for test. An LSTM network was trained using early recurrent data and the parameters of the Arrhenius equation were fitted. The prediction error was evaluated on the test set. By adjusting the number of network layers, learning rate, and time window, the mean absolute percentage error was kept below 5%, ensuring reliable predictive performance. Comparison data with other models is shown in the table below.

[0024] Specifically, through scientific data partitioning strategies and systematic hyperparameter optimization processes, the model is ensured to have good generalization ability and prediction accuracy. The training set is used to learn aging patterns, the validation set is used to prevent overfitting and guide network structure tuning, and the independent test set is used to objectively evaluate the model's performance on unseen data. Ultimately, the MAPE is controlled within 5%, laying a reliable accuracy foundation for subsequent online deployment.

[0025] S5. Online Deployment and Dynamic Correction The trained model is connected to the battery management system. Rolling predictions are made based on real-time collected temperature, SOC, and current data. An online correction mechanism is established. When the deviation between the measured capacity and the predicted value exceeds the threshold, the model parameters are fine-tuned using new data. Specifically, by continuously updating prediction results through real-time collection of operational data and establishing an adaptive correction mechanism based on prediction bias, the model can automatically adapt to individual differences, changes in usage habits, and the influence of unmodeled dynamic factors, ensuring that the prediction results remain accurate and timely throughout the entire life cycle, thus achieving truly intelligent health management.

[0026] S6. SOH Assessment and Early Warning Output The system calculates the state of health (SOH) in real time and visualizes the capacity decay trend and remaining lifespan range. It also establishes a graded early warning system: an SOH above 80% indicates normal operation; a range of 60% to 80% suggests attention and optimization of usage strategies; and a replacement reminder is triggered when the SOH is below 60% or short-term failure is predicted. The formula for calculating the state of health (SOH) is as follows:

[0027] in, Current available capacity This is the rated capacity.

[0028] Specifically, by quantifying the ratio of current available capacity to rated capacity and combining it with a threshold-based tiered early warning strategy, the system provides battery users with end-to-end management support, from normal operation and optimization suggestions to replacement reminders. This avoids resource waste caused by over-maintenance and prevents safety risks and economic losses caused by battery failure, ultimately achieving safe, efficient, and economical operation of low-temperature lithium batteries throughout their entire lifecycle.

[0029] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the capacity decay trend of a low-temperature lithium battery, characterized in that, Specifically, the following steps are included: S1. Initial Data Acquisition Multiple temperature gradients were set in the range of -80℃ to 10℃ to conduct accelerated cycle charge and discharge tests to simulate actual low-temperature usage scenarios. During the test, capacity decay, internal resistance change, voltage curve, temperature distribution and cycle number were collected to form a complete aging dataset under multiple temperature conditions. S2. Feature Extraction Based on electrochemical principles, the main aging mechanisms at low temperatures are analyzed, and core input variables are screened out through correlation analysis to build a model and construct a feature set. S3. Constructing a hybrid prediction model A hybrid prediction model is constructed by integrating an electrochemical empirical model and an LSTM deep learning model. The electrochemical empirical model uses the Arrhenius equation to describe the effect of temperature on the aging rate, providing physical constraints and prior knowledge. The LSTM deep learning model processes multi-dimensional time series features, capturing long-term time-series dependencies and nonlinear decay patterns. The two models complement each other through weighted fusion or residual correction. S4. Model Training and Offline Validation The experimental data were divided into training, validation and test sets. The LSTM network was trained using early recurrent data and the parameters of the Arrhenius equation were fitted. The prediction error was evaluated on the test set. By adjusting the number of network layers, learning rate and time window, the mean absolute percentage error was controlled within 5% to ensure that the model has reliable prediction performance. S5. Online Deployment and Dynamic Correction The trained model is connected to the battery management system. Rolling predictions are made based on real-time collected temperature, SOC, and current data. An online correction mechanism is established. When the deviation between the measured capacity and the predicted value exceeds the threshold, the model parameters are fine-tuned using new data. S6. SOH Assessment and Early Warning Output The system calculates the state of health (SOH) in real time and visualizes the capacity decay trend and remaining lifespan range. It also establishes a tiered early warning system: an SOH above 80% indicates normal operation; a range of 60% to 80% suggests monitoring and optimizing usage strategies; and a replacement reminder is triggered when the SOH is below 60% or short-term failure is predicted.

2. The method for predicting the capacity decay trend of a low-temperature lithium battery according to claim 1, characterized in that: In S1, the multiple temperature gradients include -80℃, -70℃, -60℃, -50℃, -40℃, -30℃, -20℃, -10℃, 0℃, and 10℃.

3. The method for predicting the capacity decay trend of a low-temperature lithium battery according to claim 1, characterized in that: In S2, the main aging mechanisms include restricted lithium-ion diffusion, electrolyte performance degradation, SEI film thickening, and lithium deposition.

4. The method for predicting the capacity decay trend of a low-temperature lithium battery according to claim 1, characterized in that: In S2, the core input variables include real-time temperature, state of charge, current stress, cumulative cycle count, and historical capacity retention rate.

5. The method for predicting the capacity decay trend of a low-temperature lithium battery according to claim 1, characterized in that: In S3, the formula for the Arrhenius equation is: ,in, The aging rate constant is For activation energy, The gas constant is... This refers to absolute temperature.

6. The method for predicting the capacity decay trend of a low-temperature lithium battery according to claim 1, characterized in that: In S3, the LSTM deep learning model includes an input layer, a hidden layer, and an output layer. The input layer is used to input multi-dimensional time series features, including temperature, SOC, current, and historical capacity. The hidden layer is used by the LSTM network to capture long-term temporal dependencies and handle nonlinear decay patterns. The output layer is used to output the future capacity retention curve and remaining cycle life.

7. The method for predicting the capacity decay trend of a low-temperature lithium battery according to claim 1, characterized in that: In S4, the ratio of training set, validation set and test set is 70% training set, 15% validation set and 15% test set.

8. The method for predicting the capacity decay trend of a low-temperature lithium battery according to claim 1, characterized in that: In step S6, the formula for calculating the healthy state SOH is as follows: ,in, Current available capacity This is the rated capacity.

Images