A battery life prediction method and related devices

Abstract

The application belongs to the technical field of lithium ion batteries, and discloses a battery life prediction method and related equipment, obtains running data of a to-be-tested battery and extracts a health feature vector, then inputs the feature vector into a trained physical information neural network, and outputs a health state prediction value. The network is trained through a composite loss function, wherein a data fitting loss term constrains the prediction error of the model on the known observation points, and a physical regularization loss term is constructed based on a battery degradation kinetics equation, and the degradation rate coefficient is set as a trainable parameter. This physical constraint mechanism uses automatic differentiation technology to calculate the first derivative of the output with respect to the cycle number, and takes the absolute value of the degradation rate coefficient, forcing the prediction trajectory to strictly follow the electrochemical law of monotonically decreasing health state, completely eliminating the abnormal capacity rebound phenomenon common in pure data-driven models from a mathematical point of view, and avoiding false decisions made by the battery management system due to blind optimistic prediction.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion battery technology, specifically a battery life prediction method and related equipment. Background Technology

[0002] In the global energy transition to cleaner energy, lithium-ion batteries have become the core energy storage component for electric vehicles and smart grids. Accurately predicting the remaining battery life (RUL) is crucial for ensuring the safe operation of the battery management system (BMS) and for developing reasonable operation and maintenance strategies.

[0003] The performance of lithium-ion batteries degrades irreversibly with charge-discharge cycles. Existing RUL prediction methods struggle to balance fitting ability, physical rationality, and cross-individual adaptability, thus failing to meet the requirements of high-reliability industrial applications.

[0004] Currently, commonly used lithium-ion battery RUL prediction methods are mainly pure data-driven models, including LSTM and Transformer types. These methods, with their powerful nonlinear fitting capabilities, can predict RUL based on battery-related data and have become the mainstream technical solution in this field in recent years. They are widely used in various battery RUL prediction scenarios, and by leveraging their advantages in data processing and fitting, they have solved the problem of accurately predicting battery RUL to a certain extent.

[0005] Current pure data-driven models face significant technical bottlenecks in practical high-reliability industrial applications. Generally, they fail to balance fitting ability, physical plausibility, and cross-individual adaptability, making them unsuitable for real-world engineering needs. Specifically, firstly, they lack physical interpretability; the models are essentially "black boxes," unconstrained by objective electrochemical laws, and prone to "non-physical oscillations" that violate physical principles during long-term extrapolation, potentially leading to BMS safety decision errors and posing safety hazards. Secondly, they exhibit weak cross-individual generalization ability; influenced by the heterogeneity of individual batteries, data distribution is easily skewed, resulting in a significant drop in prediction accuracy when general models are applied to entirely new target batteries. Furthermore, traditional retraining methods require a large amount of full-lifecycle data, which is insufficient for scenarios with only limited early data during the "cold start" phase of practical engineering. Therefore, existing pure data-driven models cannot meet the core requirements of this field for RUL prediction methods. Summary of the Invention

[0006] This invention provides a battery life prediction method and related equipment, which solves the problems that pure data-driven models lack physical interpretability and have weak generalization ability, and cannot meet the actual engineering needs of battery RUL prediction.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A battery life prediction method includes: Acquire operational monitoring data of the lithium-ion battery under test, and extract health feature vectors reflecting the battery's health status from it; The health feature vector is input into the trained prediction model, and the prediction result is output. The trained prediction model is a physical information neural network. The input includes the current cycle number of the battery and the health feature vector. The output is the predicted health status value of the current cycle. The physical information neural network is trained by a composite loss function that includes a data fitting loss term and a physical regularization loss term. The physical regularization loss term is constructed based on the battery degradation dynamics equation, and the battery degradation rate coefficient is set as a trainable system parameter.

[0008] Preferably, the health feature vector is extracted from the battery's standardized constant current-constant voltage charging phase, and the health features include: constant voltage rise time within a preset voltage range, constant current charging time, constant voltage charging time, current integral during the constant current phase, current integral during the constant voltage phase, and the proportion of the constant current phase to the total charging time.

[0009] Preferably, the hidden layer of the physical information neural network uses the hyperbolic tangent function as the activation function.

[0010] Preferably, in the composite loss function, the data fitting loss term uses the mean square error to constrain the prediction error of the model at known observation points, and the physical regularization loss term uses automatic differentiation technology to calculate the first derivative of the network output with respect to the number of cycles, and constructs a physical residual equation based on a semi-empirical exponential decay model, wherein the absolute value of the decay rate coefficient is taken, and the predicted health status value is forced to decrease monotonically with time.

[0011] Preferably, the trained prediction model is obtained through two-stage adaptive training, specifically including: in the first stage, using the full lifecycle data of the source domain battery to jointly optimize the network weights, biases, and trainable physical parameters, so that the physical parameters converge to a global prior value reflecting the average degradation level of the source domain population; in the second stage, when the model is deployed on the target battery, using the early historical operating data before the prediction start point of the target battery, the network weights and trainable physical parameters are iteratively updated, so that the trainable physical parameters evolve from the global prior value to an individual-specific value reflecting the true degradation rate of the target battery.

[0012] Preferably, the early historical operating data before the target battery prediction start point is the monitoring data of 30% to 40% of the total number of cycles before the target battery reaches the preset failure threshold, or the monitoring data of the first N charge-discharge cycles, where N ranges from 105 to 140.

[0013] A battery life prediction system, comprising: Data acquisition module: used to acquire the operation monitoring data of the lithium-ion battery under test, and extract the health feature vector reflecting the health status of the battery from it; The prediction module internally stores a trained prediction model, which is used to receive the health feature vector and output the remaining life prediction result. The trained prediction model is a physical information neural network. Its input includes the current cycle number of the battery and the health feature vector, and its output is the predicted health status value of the current cycle. The physical information neural network is trained by a composite loss function that includes a data fitting loss term and a physical regularization loss term. The physical regularization loss term is constructed based on the battery degradation dynamics equation, and the battery degradation rate coefficient is set as a trainable system parameter. The output module is used to output the remaining lifetime prediction results.

[0014] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of a battery life prediction method.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a battery life prediction method.

[0016] A computer program product includes a computer program that, when executed by a processor, implements the steps of a battery life prediction method.

[0017] Compared with existing technologies, this invention has the following advantages: This invention provides a battery life prediction method. It acquires the operating data of the battery under test and extracts a health feature vector. This feature vector is then input into a trained physical information neural network, which outputs a predicted health state value. This network is trained using a composite loss function. The data fitting loss term constrains the model's prediction error at known observation points, while the physical regularization loss term is constructed based on the battery degradation kinetic equation, and the degradation rate coefficient is set as a trainable parameter. This physical constraint mechanism utilizes automatic differentiation technology to calculate the first derivative of the output with respect to the number of cycles, while simultaneously taking the absolute value of the degradation rate coefficient. This forces the predicted trajectory to strictly follow the monotonically decreasing electrochemical law of health state, mathematically eliminating the abnormal capacity recovery phenomenon commonly found in purely data-driven models and preventing the battery management system from making erroneous decisions due to blindly optimistic predictions. Furthermore, the trained model undergoes two-stage adaptive training: first, offline global pre-training is performed using full-lifetime data from the source domain batteries, causing the physical parameters to converge to global prior values ​​reflecting the average degradation level of the group; then, during deployment, online local fine-tuning of the network weights and trainable physical parameters is performed using a small amount of early historical data from the target battery, causing the degradation rate coefficient to quickly converge to a specific value reflecting individual characteristics. This mechanism effectively overcomes the heterogeneity problem between different batteries, achieving high-precision cross-domain prediction without collecting full-lifetime data from the target battery, solving the "cold start" problem in engineering applications, and completing the dynamic transfer from general knowledge of the group to individual specific characteristics at a low computational cost. The final output health status evolution trajectory is smooth and monotonically decreasing, and the remaining lifetime prediction results can be directly used to guide battery management strategy adjustments or safety warnings. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a battery life prediction method according to an embodiment of the present invention; Figure 2 This is a block diagram of a battery life prediction system according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0024] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0025] like Figure 1 As shown, the present invention provides a method for predicting the remaining battery life based on an adaptive physical information network, which can run in a computing device or battery management system (BMS) containing a processor and memory.

[0026] First, operational monitoring data of the lithium-ion battery under test is acquired, and a health feature vector reflecting the state of health (SOH) of the battery is extracted from it. The discharge process of a battery in actual operation is often affected by random load fluctuations, containing a large amount of non-stationary noise. To improve the signal-to-noise ratio of the features, this embodiment extracts health features only from the standardized constant current-constant voltage (CC-CV) charging stage of the battery. The charge and discharge data of the battery in the i-th cycle are acquired, and specific health features are extracted to form a feature vector. The health features specifically include: constant voltage rise time within a preset voltage range (e.g., preferably 3.8V-4.0V), constant current charging time, constant voltage charging time, current integral during the constant current stage, current integral during the constant voltage stage, and the proportion of the constant current stage to the total charging time. The maximum-minimum normalization method is used to map the extracted feature vector and the number of cycles to the [0,1] interval to eliminate the interference of different dimensions on the neural network weight update. This feature extraction method effectively filters out the non-stationary noise of the discharge process, enabling the subsequent model to focus on the essential laws of battery degradation.

[0027] Next, the health feature vector is input into the trained prediction model, which outputs the remaining lifespan prediction result. The trained prediction model is a Physical Information Neural Network (PINN), whose inputs include the current cycle number of the battery and the health feature vector, and whose output is the predicted health status value for the current cycle. The PINN is trained using a composite loss function that includes a data fitting loss term and a physical regularization loss term. The physical regularization loss term is constructed based on the battery degradation kinetic equation, and the battery degradation rate coefficient is set as a trainable system parameter.

[0028] Specifically, a multi-layer feedforward deep neural network is constructed as a differentiable function approximator. The input layer of this network contains the current cycle number k of the battery and the health feature vector extracted in the current cycle. The design aims to simultaneously capture macroscopic temporal decay trends and microscopic local electrochemical state changes. The hidden layer comprises M hidden layers, each containing N neurons (e.g., in this embodiment, preferably 3 fully connected layers with 64 neurons per layer). The output layer outputs the predicted health state value corresponding to the current cycle. The hidden layers of the physical information neural network use the hyperbolic tangent function (Tanh) as the activation function. This setting ensures that the higher-order gradient information of the physical equations is continuous and well-defined during backpropagation, thus avoiding the problems of incorrect calculation of physical residuals and network non-convergence caused by using conventional activation functions such as ReLU that are not differentiable or have discontinuous derivatives. This is a necessary condition for achieving effective coupling between physical mechanisms and deep networks.

[0029] This embodiment uses a semi-empirical exponential decay model to describe the battery's dynamic behavior, assuming that the battery's aging rate is proportional to its current health state. This mechanism is then transformed into the total loss function of a neural network. It consists of two parts: the data fitting loss term. and physical regularization loss term The data fitting loss term uses the mean squared error to constrain the model's prediction error at known observation points:

[0030] in This represents the number of samples in the training set.

[0031] The physical regularization loss term is used to calculate the network output using automatic differentiation techniques. For the number of loops The first derivative is used to construct the physical residual equation:

[0032] in This represents the number of configuration points. The physical degradation rate coefficient, which characterizes the rate of individual battery degradation, is set as a trainable system parameter in this invention. Here, [the following is a more detailed explanation of the parameter]. Take absolute value It is a key constraint mechanism that forces the derivative of the decay rate to always be negative, mathematically guaranteeing that... It decreases monotonically over time, completely eliminating the phenomenon of non-physical capacity recovery.

[0033] The total loss function is defined as: ,in and This is the balance coefficient.

[0034] The trained prediction model is obtained through two-stage adaptive training.

[0035] Phase 1 (Offline Global Pre-training): The network weights, biases, and trainable physical parameters are jointly optimized using the full lifecycle data of the source domain batteries. The network weights learn a general degradation feature mapping relationship, and the physical parameters converge to global prior values ​​reflecting the average degradation level of the source domain population. .

[0036] Phase 2 (Online Local Fine-tuning): When the model is deployed on the target battery, the network is initialized as follows: The network weights and trainable physical parameters are iteratively updated using early historical operating data prior to the target battery prediction start point. Specifically, this early historical operating data includes, for example, monitoring data representing 30% to 40% of the total number of cycles before the target battery reaches a preset failure threshold, or monitoring data from the first N charge-discharge cycles, where N ranges from 105 to 140. In this embodiment, data from the first 105, 120, or 140 cycles is preferred. The preferred number of iterations for online fine-tuning of network weights is 50. During this process, the trainable physical parameters... quickly from The evolution converges to reflect the true degradation rate of the target battery. Through this adaptive mechanism, the model successfully achieved dynamic transfer from "general knowledge of the group" to "individual-specific characteristics," effectively overcoming the data distribution offset problem caused by the heterogeneity of different individual batteries and solving the prediction problem in the "cold start" stage of engineering.

[0037] After completing the online fine-tuning, freeze the weights and individual physical parameters of the fully connected layers. .

[0038] Construct a sequence of future cycle counts and input it into the network. Utilize the network's extrapolation capabilities to generate a smooth and monotonically decreasing future SOH evolution trajectory.

[0039] The system automatically scans the predicted trajectory, and when the predicted value... When the battery first drops and crosses a preset failure threshold (e.g., 80% of rated capacity) set according to different battery application scenarios, the corresponding cycle number is marked as the predicted failure time. .

[0040] The final output is the remaining useful life calculation result: (in The remaining lifespan prediction result is used to determine the current number of running cycles, and this result is sent to the battery management system. The battery management system executes specific control commands based on the remaining lifespan prediction result, including but not limited to adjusting the battery thermal management strategy, reducing the charging and discharging power limit, or triggering a power outage protection safety warning.

[0041] To verify the generalization performance of the adaptive transfer learning method described in this invention during cross-domain deployment, this embodiment uses a publicly available dataset battery with a rated capacity of 2 Ah and a failure threshold of 70%, as well as a self-tested battery with a rated capacity of 2.6 Ah, subjected to high-rate charge-discharge and a failure threshold of 80%. The results show that the method of this invention can achieve high-precision lifetime prediction for target batteries with different material systems, different charge-discharge regimes, and different failure thresholds. This method completely eliminates non-physical oscillations, ensuring system safety; simultaneously, it achieves rapid domain adaptation with a small amount of early target battery data, demonstrating significant computational efficiency advantages.

[0042] Corresponding to the above method, the present invention also provides a battery life prediction system, such as... Figure 2 As shown. The system includes: The data acquisition module is used to acquire the operation monitoring data of the lithium-ion battery under test and extract the health feature vector reflecting the health status of the battery. The prediction module, which internally stores a trained prediction model, is used to receive the health feature vector and output the remaining life expectancy prediction result. The output module is used to output the remaining lifetime prediction results.

[0043] The trained prediction model in the prediction module is a physical information neural network. Its input includes the current cycle number and health feature vector of the battery, and its output is the predicted health status value of the current cycle. This network is trained using a composite loss function that includes a data fitting loss term and a physical regularization loss term. The physical regularization loss term is constructed based on the battery degradation kinetic equation, and the battery degradation rate coefficient is set as a trainable system parameter. The output module communicates with the battery management system. When the output RUL result meets preset conditions, it triggers the BMS to perform battery thermal management strategy adjustment, charge / discharge power derating, or power outage protection safety warning.

[0044] A computer device is provided according to an embodiment of the present invention. This computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0045] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention.

[0046] The computer device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device may include, but is not limited to, a processor and memory.

[0047] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0048] The memory can be used to store the computer program and / or module, and the processor implements various functions of the computer device by running or executing the computer program and / or module stored in the memory, and by calling the data stored in the memory.

[0049] If the modules / units integrated into the computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0050] Based on this understanding, the present invention can implement all or part of the processes in the methods of the above embodiments, or it can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0051] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0052] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for predicting battery life, characterized in that... ,include: Acquire operational monitoring data of the lithium-ion battery under test, and extract health feature vectors reflecting the battery's health status from it; The health feature vector is input into the trained prediction model, and the prediction result is output. The trained prediction model is a physical information neural network. The input includes the current cycle number of the battery and the health feature vector. The output is the predicted health status value of the current cycle. The physical information neural network is trained by a composite loss function that includes a data fitting loss term and a physical regularization loss term. The physical regularization loss term is constructed based on the battery degradation dynamics equation, and the battery degradation rate coefficient is set as a trainable system parameter.

2. The battery life prediction method according to claim 1, characterized in that, The health feature vector is extracted from the battery's standardized constant current-constant voltage charging phase. The health features include: constant voltage rise time within a preset voltage range, constant current charging time, constant voltage charging time, current integral during the constant current phase, current integral during the constant voltage phase, and the proportion of the constant current phase to the total charging time.

3. The battery life prediction method according to claim 1, characterized in that, The hidden layer of the physical information neural network uses the hyperbolic tangent function as the activation function.

4. The battery life prediction method according to claim 1, characterized in that, In the composite loss function, the data fitting loss term uses the mean square error to constrain the prediction error of the model at known observation points, and the physical regularization loss term uses automatic differentiation technology to calculate the first derivative of the network output with respect to the number of cycles, and constructs a physical residual equation based on a semi-empirical exponential decay model, where the absolute value of the decay rate coefficient is taken, and the predicted health status value is forced to decrease monotonically with time.

5. The battery life prediction method according to claim 1, characterized in that, The trained prediction model is obtained through two-stage adaptive training, specifically including: the first stage, using the full life cycle data of the source domain battery to jointly optimize the network weights, biases, and trainable physical parameters, so that the physical parameters converge to the global prior value reflecting the average degradation level of the source domain population; the second stage, when the model is deployed on the target battery, using the early historical operating data before the prediction start point of the target battery, to iteratively update the network weights and trainable physical parameters, so that the trainable physical parameters evolve from the global prior value to the individual-specific value reflecting the true degradation rate of the target battery.

6. The battery life prediction method according to claim 1, characterized in that, The early historical operating data before the target battery prediction start point is the monitoring data of 30% to 40% of the total number of cycles before the target battery reaches the preset failure threshold, or the monitoring data of the first N charge-discharge cycles, where N ranges from 105 to 140.

7. A battery life prediction system, characterized in that, include: Data acquisition module: used to acquire the operation monitoring data of the lithium-ion battery under test, and extract the health feature vector reflecting the health status of the battery from it; The prediction module internally stores a trained prediction model, which is used to receive the health feature vector and output the remaining life prediction result. The trained prediction model is a physical information neural network. Its input includes the current cycle number of the battery and the health feature vector, and its output is the predicted health status value of the current cycle. The physical information neural network is trained by a composite loss function that includes a data fitting loss term and a physical regularization loss term. The physical regularization loss term is constructed based on the battery degradation dynamics equation, and the battery degradation rate coefficient is set as a trainable system parameter. The output module is used to output the remaining lifetime prediction results.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the battery life prediction method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of a battery life prediction method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of a battery life prediction method according to any one of claims 1-6.

Images