A battery life prediction method based on a double-branch neural network

Abstract

This invention belongs to the field of battery life assessment technology, specifically relating to a battery life prediction method based on a dual-branch neural network. The method includes: acquiring the physical parameters of the battery under test during charging and discharging, including at least voltage, current, temperature, state of charge, and capacity; selecting a reference battery with the highest similarity to the battery under test from a battery database; and inputting all the physical parameters of the battery under test and the corresponding reference battery into a trained dual-branch neural network to predict the remaining life of the battery under test. This invention innovatively designs a dual-branch neural network as the battery life prediction model. One branch explicitly embeds the physical equations of battery electrode degradation, while the other branch is responsible for learning the time-dependent features of battery capacity decay. Knowledge transfer is achieved through a cross-battery feature fusion mechanism, thereby achieving high-precision and highly robust prediction of the remaining battery life under hybrid physical constraints.

Description

Technical Field

[0001] This invention belongs to the field of battery life assessment technology, specifically relating to a battery life prediction method based on a dual-branch neural network. Background Technology

[0002] Batteries (including but not limited to lithium-ion batteries, sodium-ion batteries, quasi-solid-state batteries, all-solid-state batteries, etc.) are widely used as core energy storage components in electrified vehicles (such as electric vehicles, rail transit, electric ships, electric aircraft, etc.) and energy storage systems (such as energy storage power stations, data centers, intelligent computing centers, etc.).

[0003] However, batteries inevitably age during long-term charge-discharge cycles, and the decline in their performance and health directly affects the safety and economy of electrified vehicles or energy storage systems. Therefore, accurate prediction of battery health and lifespan is of great significance for ensuring the safe and stable operation of electrified vehicles and energy storage systems and for improving battery utilization efficiency.

[0004] With the rapid development of artificial intelligence algorithms, data-driven battery life prediction models are currently mainly divided into the following two categories: The first category is purely data-driven models, such as prediction methods based on recurrent neural networks and Transformers. Although these models can learn battery degradation patterns from massive amounts of data, they lack interpretability. Their internal logic lacks an intrinsic connection with the actual electrochemical degradation mechanisms of batteries (such as solid electrolyte interfacial film growth, active material precipitation and loss, etc.), which may lead to prediction results that deviate from physical laws in long-term predictions or when dealing with untreated batteries, resulting in insufficient robustness.

[0005] The second category is data-driven models that incorporate physical information constraints. These models consider the battery degradation mechanism, embedding or inputting the electrochemical reactions inside the battery as constraints or features into the neural network, thereby improving the physical consistency and generalization ability of the model's predictions. However, existing research still suffers from problems such as incomplete integration of physical information and focusing only on a single degradation mechanism while neglecting the coupling effects of multiple degradation mechanisms.

[0006] Therefore, in order to overcome the limitations of the aforementioned data-driven methods and single-cell learning methods, there is an urgent need for a life prediction technology that can deeply integrate multiple battery degradation mechanisms and achieve efficient cross-cell knowledge transfer. Summary of the Invention

[0007] In view of the shortcomings of the prior art described above, this invention designs a dual-branch neural network that integrates multiple battery degradation mechanisms as a battery life prediction model to solve the problems of poor adaptability, lack of interpretability and insufficient fusion of physical information in traditional methods.

[0008] To achieve the above and other related objectives, this invention provides a battery life prediction method based on a dual-branch neural network, comprising: acquiring physical parameters of the battery under test during charging and discharging, including at least voltage, current, temperature, state of charge, and capacity; selecting a reference battery with the highest similarity to the battery under test from a battery database, and inputting all physical parameters of the battery under test and all physical parameters of the corresponding reference battery into a trained dual-branch neural network to predict the remaining life of the battery under test; wherein the dual-branch neural network comprises: an electrode degradation physical branch, used to calculate feature parameters characterizing the degree of battery electrode degradation based on the voltage sequence, current sequence, temperature sequence, and state of charge sequence of the battery during charging and discharging, including at least degradation parameters of the solid electrolyte interface film, loss parameters of electrochemical active materials, and electrode state parameters; a capacity sequence encoding branch, used to calculate feature parameters characterizing the battery capacity degradation trend based on the capacity sequence of the battery during charging and discharging; and a cross-battery feature fusion module, used to complementaryly fuse the feature parameters characterizing the degree of battery electrode degradation and the feature parameters characterizing the battery capacity degradation trend between the battery under test and its corresponding reference battery, and predict the remaining life of the battery under test based on the fused features.

[0009] According to a specific embodiment of the present invention, the electrode degradation physical branch includes: a first fully connected layer, used to determine the growth rate of the solid electrolyte interface film inside the battery based on the current sequence, temperature sequence, and state of charge sequence during the charging and discharging process, as a degradation parameter of the solid electrolyte interface film; a second fully connected layer, used to determine the cumulative deposition and inventory loss of electrochemical active materials inside the battery based on the current sequence, temperature sequence, and state of charge sequence during the charging and discharging process; a third fully connected layer, used to determine the loss of positive electrode material based on the current sequence, temperature sequence, state of charge sequence, and inventory loss of electrochemical active materials during the charging and discharging process; a fourth fully connected layer, used to determine the loss of negative electrode material based on the current sequence, temperature sequence, state of charge sequence, and inventory loss of electrochemical active materials during the charging and discharging process; and an encoder, used to determine the electrode consistency characteristic quantity of the battery based on the voltage sequence during the charging and discharging process, as an electrode state parameter; wherein, the cumulative deposition and inventory loss of electrochemical active materials inside the battery, the loss of positive electrode material, and the loss of negative electrode material together constitute the loss parameter of electrochemical active material.

[0010] According to a specific embodiment of the present invention, the growth rate of the solid electrolyte interface film inside the battery is calculated using the Arrhenius formula for the first fully connected layer.

[0011] According to a specific embodiment of the present invention, the second fully connected layer uses the Butler-Wolmer equation to calculate the cumulative deposition and inventory loss of electrochemical active materials inside the battery.

[0012] According to a specific embodiment of the present invention, the capacity sequence encoding branch includes: a bidirectional long short-term memory network, used to identify the long-term time dependency relationship in the capacity sequence of the battery during the charging and discharging process, and output the corresponding hidden state matrix; and a multi-head attention function, used to calculate the dependency weight of each time step on other time steps in the hidden state matrix, to obtain the battery capacity decay time series feature vector, which is used as a feature parameter to characterize the battery capacity decay trend.

[0013] According to a specific embodiment of the present invention, the cross-battery feature fusion module includes: a homogeneous feature fusion unit, used to concatenate feature parameters of the battery under test and its corresponding reference battery, which characterize the degree of battery electrode degradation, to obtain a first concatenation vector, and to concatenate feature parameters of the battery under test and its corresponding reference battery, which characterize the battery capacity degradation trend, to obtain a second concatenation vector; a heterogeneous feature fusion unit, used to concatenate the first concatenation vector and the second concatenation vector and input them into a preset attention function to generate corresponding fused features; and a remaining lifespan regression model, used to predict the remaining lifespan of the battery under test based on the fused features.

[0014] According to a specific embodiment of the present invention, the isomorphic feature fusion unit includes two fully connected layers, which are respectively used to splice the feature parameters of the battery under test and its corresponding reference battery for characterizing the degree of battery electrode degradation, and to splice the feature parameters of the battery under test and its corresponding reference battery for characterizing the battery capacity degradation trend.

[0015] According to a specific embodiment of the present invention, the training steps of the dual-branch neural network include: acquiring charge and discharge data of batteries with different remaining lifespans to construct training sample sets for short-life battery packs, medium-life battery packs, and long-life battery packs; randomly selecting two training samples from the short-life battery pack to form cross-battery sample pairs, inputting all cross-battery sample pairs into the dual-branch neural network, and calculating the total loss between the classification label output by the dual-branch neural network and the true label based on a hybrid loss function, thereby optimizing the dual-branch neural network in reverse; randomly selecting two training samples from the medium-life battery pack to form cross-battery sample pairs, inputting all cross-battery sample pairs into the dual-branch neural network, and calculating the total loss between the classification label output by the dual-branch neural network and the true label based on a hybrid loss function, thereby optimizing the dual-branch neural network in reverse; and randomly selecting two training samples from the long-life battery pack to form cross-battery sample pairs, inputting all cross-battery sample pairs into the dual-branch neural network, and calculating the total loss between the classification label output by the dual-branch neural network and the true label based on a hybrid loss function, thereby optimizing the dual-branch neural network in reverse.

[0016] According to a specific embodiment of the present invention, the calculation formula of the hybrid loss function is as follows: , , in, This represents the hybrid loss function. Represents the regression loss term. Represents the physical consistency loss term. , These are the preset coefficients. This indicates the number of samples in the current training batch. This represents the classification label of the i-th training sample. This represents the true label of the i-th training sample. This represents the trainable parameters of the remaining life regression model of the cross-battery feature fusion module in the dual-branch neural network.

[0017] According to a specific embodiment of the present invention, the physical consistency loss term includes at least a physical fusion loss term and a physical monotonicity loss term.

[0018] The beneficial effects of this invention are as follows: This invention innovatively designs a dual-branch neural network as a battery life prediction model. One branch explicitly embeds the physical equation of battery electrode degradation, while the other branch is responsible for learning the time-dependent features of battery capacity decay. Knowledge transfer is achieved through a cross-battery fusion mechanism. Finally, under hybrid physical constraints, high-precision and robust remaining life prediction is achieved, thus solving the problems of poor adaptability, lack of interpretability, and insufficient fusion of physical information in traditional battery life prediction methods. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0020] Figure 1 This is a flowchart illustrating a battery life prediction method based on a dual-branch neural network provided in one embodiment of the present invention. Figure 2 This is a schematic diagram of a dual-branch neural network model architecture provided in one embodiment of the present invention; Figure 3 This is a schematic diagram of a battery life prediction system based on a dual-branch neural network provided in one embodiment of the present invention; Figure 4 This is a structural block diagram of an electronic device provided in one embodiment of the present invention. Detailed Implementation

[0021] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0022] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0023] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, publicly known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0024] Example 1 Please see Figure 1 The battery life prediction method based on a dual-branch neural network, as shown, includes: Step S100: Obtain the physical parameters of the battery under test during the charging and discharging process, including at least voltage, current, temperature, state of charge and capacity.

[0025] Step S200: Select a reference battery from the battery database that has the highest similarity to the battery under test, and input all the physical parameters of the battery under test and all the physical parameters of the corresponding reference battery into the trained dual-branch neural network to predict the remaining life of the battery under test.

[0026] It should be noted that this application innovatively designs a dual-branch neural network architecture, comprising two parallel branches (electrode degradation physics branch and capacity sequence encoding branch) and a cross-cell feature fusion module. The electrode degradation physics branch explicitly embeds two main degradation mechanisms of the battery under charge-discharge conditions, including physical equations concerning solid electrolyte interphase (SEI) degradation and loss of electrochemically active materials. The capacity sequence encoding branch is responsible for learning the time-dependent characteristics of battery capacity degradation. Finally, the cross-cell feature fusion module integrates the complementary information of the test battery and the corresponding reference battery in the electrode degradation physics branch and capacity sequence encoding branch to predict the remaining life of the test battery.

[0027] Specifically, such as Figure 2 As shown, the electrode degradation physics branch consists of four parallel fully connected layers and an encoder. The first fully connected layer... It is mainly used to determine the growth rate of the SEI film inside the battery based on the current sequence, temperature sequence, and state of charge sequence during the charging and discharging process. Preferably, the growth rate of the SEI film can be calculated using the Arrhenius formula. This serves as a degradation parameter for the SEI membrane.

[0028] Second fully connected layer It is mainly used to determine the cumulative deposition amount of electrochemical active materials inside the battery (such as lithium deposition amount) based on the current sequence, temperature sequence, and state of charge sequence during the charging and discharging process. The amount of lithium deposited, along with the inventory loss of electrochemically active materials (such as the loss of lithium inventory, LLI), can preferably be calculated using the Butler-Wolmer equation. Lithium inventory loss (LLI).

[0029] The third fully connected layer The loss of the positive electrode material is then calculated based on the current sequence, temperature sequence, and state of charge sequence during the charging and discharging process, as well as the inventory loss of the electrochemically active materials determined above. .

[0030] Similarly, the fourth fully connected layer The loss of the battery negative electrode material is then calculated based on the current sequence, temperature sequence, and state of charge sequence during the battery's charging and discharging process, as well as the inventory loss of the electrochemically active materials determined above. .

[0031] Correspondingly, the loss of active materials in the battery can be determined based on the loss of positive electrode material and the loss of negative electrode material. Furthermore, the cumulative deposition of electrochemical active materials, the inventory loss of electrochemical active materials, and the total loss of active materials together constitute the loss parameters of electrochemical active materials.

[0032] The encoder, as an electrode state formation network, can be composed of convolutional neural networks and multilayer perceptrons. It infers the implicit electrode parameters from the voltage sequence of the battery during the charging and discharging process, and then estimates the electrode state inside the battery. Finally, it determines the electrode consistency characteristics of the battery as electrode state parameters.

[0033] It should be added here that the electrode state parameters include, but are not limited to, electrode consistency characteristics. The lithium intercalation state of the positive and negative electrodes at the beginning and end of charging and discharging, the relative capacity of the positive and negative electrodes, etc. can also be calculated simultaneously. Together with the electrode consistency characteristics, they constitute the electrode state parameters. No further restrictions are imposed on this. Modifications and refinements made by those skilled in the art to the embodiments of the present invention without departing from the spirit of the present invention still fall within the scope of the invention application patent of the present invention.

[0034] Based on the above, the growth rate of the SEI film output by the electrode degradation physics branch Cumulative deposition of electrochemical active materials, inventory loss of electrochemical active materials, and loss of battery cathode materials. Battery negative electrode material loss Together with the electrode consistency feature, they form a six-dimensional feature vector, which serves as a feature parameter to characterize the degree of battery electrode degradation.

[0035] The capacity sequence encoding branch can adopt a sequence-to-sequence (Seq2Seq) architecture, consisting of a bidirectional long short-term memory (BiLSTM) network and a multi-head attention function. The BiLSTM network captures the long-term temporal dependencies in the battery's capacity sequence during charging and discharging, outputting the corresponding hidden state matrix. The multi-head attention function calculates the dependency weights of each time step on other time steps based on the hidden state matrix, strengthening the information representation of key degradation stages, thus obtaining a battery capacity decay time-series feature vector, which serves as a feature parameter characterizing the battery capacity degradation trend.

[0036] The input to the cross-battery feature fusion module comes from the aforementioned electrode degradation physical branch and capacity sequence encoding branch. A fully connected layer and attention mechanism are used to concatenate the feature parameters used to characterize the degree of battery electrode degradation and the feature parameters used to characterize the battery capacity degradation trend, thereby obtaining the corresponding fused features. The fused features are then fed into the Remaining Useful Life (RUL) model to predict the remaining life of the battery.

[0037] In one specific embodiment, charging data of the battery under fast charging conditions (including at least voltage, current, temperature, state of charge, and capacity) can be acquired to train the aforementioned dual-branch neural network. First, all batteries can be divided into three groups based on their total lifespan (cycle life), and a training sample set can be constructed accordingly. For example, they can be divided into short-life battery groups (0-800 cycles), medium-life battery groups (800-1600 cycles), and long-life battery groups (1600-2400 cycles). Second, 30% of the training samples from each group are randomly selected as a validation sample set for cross-validation.

[0038] Furthermore, two training samples are randomly selected from short-life battery packs to form cross-battery sample pairs, and the paired samples have consistent degradation characteristics, thus laying the foundation for physical consistency knowledge transfer. Accordingly, the sample data of the sample pairs, that is, the physical parameters of the batteries represented by the two training samples during the charging process, are input together into a dual-branch neural network to train the model.

[0039] Specifically, after the physical parameters of the batteries represented by the two training samples during the charging process are input into the dual-branch neural network, two sets of feature parameters for characterizing the degree of battery electrode degradation and feature parameters for characterizing the battery capacity degradation trend will be obtained. Here, we will use the feature parameters for characterizing the degree of battery electrode degradation and the feature parameters for characterizing the battery capacity degradation trend of the first battery, and the feature parameters for characterizing the degree of battery electrode degradation and the feature parameters for characterizing the battery capacity degradation trend of the second battery as examples.

[0040] It should also be noted that the cross-battery feature fusion module includes a homogeneous feature fusion unit and a heterogeneous feature fusion unit. The homogeneous feature fusion comprises two stages: the first stage uses a fully connected layer to concatenate the feature parameters representing the degree of electrode degradation from the first battery and the second battery, resulting in a first concatenated vector; the second stage uses another fully connected layer to concatenate the feature parameters representing the battery capacity degradation trend from both batteries, resulting in a second concatenated vector. This homogeneous feature fusion allows for the alignment and integration of information from two batteries within the same modality. The heterogeneous feature fusion, after concatenating the first and second concatenated vectors, feeds them into a learnable attention function, calculating adaptive balancing weights to obtain the corresponding fused features. Finally, the fused features are fed into a RUL regression model for prediction, and the RUL regression model is jointly trained with the dual-branch neural network, enabling the dual-branch neural network to simultaneously maintain physical interpretability and prediction accuracy in cross-battery scenarios.

[0041] Furthermore, in order to ensure the prediction accuracy of the RUL regression model while making the two-branch neural network conform to the physical laws of battery degradation, the following hybrid loss function can be used. To perform reverse optimization.

[0042] , , in, Represents the regression loss term. Represents the physical consistency loss term. , These are preset coefficients, which can be set to 0.7 and 0.3 respectively. Furthermore, This indicates the number of samples in the current training batch, specifically the number of samples from short-life battery packs. This represents the classification label of the i-th training sample. This represents the true label of the i-th training sample. This represents the trainable parameters of the RUL regression model. It includes a physical fusion loss term to constrain the relative order of physical features of the two training samples before and after fusion, and a physical monotonicity loss term to force the RUL regression model to maintain monotonic consistency with the growth rate of the battery SEI film.

[0043] Similarly, the same method is used to train the dual-branch neural network for medium-life and long-life battery packs. That is, two training samples are randomly selected from the medium-life battery pack to form cross-battery sample pairs. All cross-battery sample pairs are input into the dual-branch neural network, and the total loss between the classification label output by the dual-branch neural network and the true label is calculated based on the hybrid loss function. The dual-branch neural network is then optimized in reverse. Similarly, two training samples are randomly selected from the long-life battery pack to form cross-battery sample pairs. All cross-battery sample pairs are input into the dual-branch neural network, and the total loss between the classification label output by the dual-branch neural network and the true label is calculated based on the hybrid loss function. The dual-branch neural network is then optimized in reverse. This process will not be described in detail.

[0044] Based on the above, once the dual-branch neural network is trained, it can be put into practical application, namely, predicting the remaining life of the battery under test. Specifically, for any battery under test, charging and discharging data during the early operation phase can be collected in real time using a sliding window, including voltage curves, current curves, temperature curves, state of charge curves, and capacity curves for each cycle. Further, a reference battery with the highest similarity to the battery under test is selected from the battery database. For example, this can be determined by calculating the cosine similarity or Euclidean distance between the capacity sequences of the battery under test and the batteries in the database, thus completing the pairing. Finally, the physical parameters of the paired battery under test during charging and discharging, along with the physical parameters of the corresponding reference battery, are simultaneously input into the trained dual-branch neural network. The cross-battery feature fusion module concatenates the feature parameters characterizing the degree of battery electrode degradation and the feature parameters characterizing the battery capacity degradation trend of the battery under test and its corresponding reference battery, respectively, to obtain the corresponding first and second concatenation vectors. These are then fused through complementary fusion to obtain the corresponding fused features. The fused features are then processed by a RUL regression model to output the predicted remaining life value of the battery under test corresponding to the current window.

[0045] It should be noted that the steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they contain the same logical relationship, they are all within the scope of protection of this patent. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this patent.

[0046] Example 2 Please see Figure 3 As shown, this application also provides a battery life prediction system based on a dual-branch neural network, comprising: The battery data monitoring module 10 is used to acquire the physical parameters of the battery under test during the charging and discharging process, including at least voltage, current, temperature, state of charge, and capacity. The battery life prediction module 20 is used to select a reference battery with the highest similarity to the battery under test from the battery database, and input all the physical parameters of the battery under test and all the physical parameters of the corresponding reference battery into the trained dual-branch neural network to predict the remaining life of the battery under test.

[0047] The dual-branch neural network includes: an electrode degradation physics branch, used to calculate characteristic parameters characterizing the degree of battery electrode degradation based on the voltage, current, temperature, and state of charge sequences during the charge and discharge process, including at least degradation parameters of the solid electrolyte interface film, loss parameters of electrochemical active materials, and electrode state parameters; a capacity sequence encoding branch, used to calculate characteristic parameters characterizing the battery capacity degradation trend based on the battery capacity sequence during the charge and discharge process; and a cross-battery feature fusion module, used to complementarily fuse the characteristic parameters characterizing the degree of battery electrode degradation and the characteristic parameters characterizing the battery capacity degradation trend between the battery under test and its corresponding reference battery, and predict the remaining life of the battery under test based on the fused features.

[0048] It should be noted that the battery life prediction system based on a dual-branch neural network provided in the above embodiments belongs to the same concept as the battery life prediction method based on a dual-branch neural network provided in Embodiment 1 above. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the battery life prediction method based on a dual-branch neural network provided in Embodiment 1 above can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above, and no limitation is imposed here.

[0049] Example 3 Please see Figure 4 As shown, this application also provides an electronic device, including a memory 2, a processor 1, and a program stored in the memory and executable on the processor, wherein the processor executes the steps of any of the methods described above.

[0050] The memory includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory, magnetic storage, disk, optical disk, etc. In some embodiments, the memory can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory can be an external storage device of the electronic device, such as a plug-in portable hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, Flash Card, etc. Furthermore, the memory can include both internal and external storage units of the electronic device. The memory can be used not only to store application software and various types of data installed on the electronic device, but also to temporarily store data that has been output or will be output.

[0051] In some embodiments, the processor may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory and calls data stored in the memory to perform various functions and process data of the electronic device. The processor executes the operating system and various installed application programs of the electronic device. The processor executes the application programs to implement the steps in the above method embodiments.

[0052] For example, the program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of program instruction segments capable of performing a specific function, which describe the execution process of the program in the electronic device.

[0053] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute some of the functions of the various embodiments of the present invention.

[0054] In summary, this invention innovatively designs a dual-branch neural network as a battery life prediction model. One branch explicitly embeds the physical equation of battery electrode degradation, while the other branch is responsible for learning the time-dependent features of battery capacity decay. Knowledge transfer is achieved through a cross-battery fusion mechanism, and finally, high-precision and robust remaining life prediction is completed under hybrid physical constraints. This solves the problems of poor adaptability, lack of interpretability, and insufficient fusion of physical information in traditional battery life prediction methods.

[0055] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A battery life prediction method based on a dual-branch neural network, characterized in that, include: Obtain the physical parameters of the battery under test during the charging and discharging process, including at least voltage, current, temperature, state of charge, and capacity; Select a reference battery from the battery database that has the highest similarity to the battery under test, and input all the physical parameters of the battery under test and all the physical parameters of the corresponding reference battery into the trained dual-branch neural network to predict the remaining life of the battery under test. The dual-branch neural network includes: The Electrode Degradation Physics branch is used to calculate characteristic parameters that characterize the degree of battery electrode degradation based on the voltage sequence, current sequence, temperature sequence and state of charge sequence during the charging and discharging process. These parameters include at least the degradation parameters of the solid electrolyte interface film, the loss parameters of electrochemical active materials and the electrode state parameters. The capacity sequence encoding branch is used to calculate characteristic parameters characterizing the battery capacity degradation trend based on the battery capacity sequence during the charging and discharging process. A cross-battery feature fusion module is used to complementarily fuse the feature parameters used to characterize the degree of battery electrode degradation and the feature parameters used to characterize the battery capacity degradation trend between the battery under test and its corresponding reference battery, and to predict the remaining life of the battery under test based on the fused features. 2.The battery life prediction method based on a two-branch neural network according to claim 1, wherein, The physical branch of electrode degradation includes: The first fully connected layer is used to determine the growth rate of the solid electrolyte interface film inside the battery based on the current sequence, temperature sequence and state of charge sequence during the charging and discharging process, and to serve as a degradation parameter of the solid electrolyte interface film. The second fully connected layer is used to determine the cumulative amount of electrochemical active materials deposited and the amount of inventory loss inside the battery based on the current sequence, temperature sequence and state of charge sequence during the charging and discharging process. The third fully connected layer is used to determine the loss of the positive electrode material of the battery based on the current sequence, temperature sequence, state of charge sequence and inventory loss of electrochemical active materials during the charging and discharging process. The fourth fully connected layer is used to determine the loss of the battery negative electrode material based on the current sequence, temperature sequence, state of charge sequence and inventory loss of electrochemical active materials during the charging and discharging process. The encoder is used to determine the electrode consistency characteristics of the battery based on the voltage sequence during the charging and discharging process, as electrode state parameters. Among them, the cumulative precipitation and inventory loss of electrochemical active materials inside the battery, the loss of positive electrode materials, and the loss of negative electrode materials together constitute the loss parameters of electrochemical active materials. 3.The battery life prediction method based on a two-branch neural network according to claim 2, characterized in that, The growth rate of the solid electrolyte interface film inside the battery is calculated using the Arrhenius formula for the first fully connected layer. 4.The battery life prediction method based on a two-branch neural network according to claim 2, wherein, The second fully connected layer uses the Butler-Wolmer equation to calculate the cumulative deposition and inventory loss of electrochemical active materials inside the battery.

5. The battery life prediction method based on a dual-branch neural network according to claim 1, characterized in that, The capacity sequence coding branch includes: A bidirectional long short-term memory network is used to identify long-term time dependencies in the capacity sequence of a battery during charging and discharging, and output the corresponding hidden state matrix. A multi-head attention function is used to calculate the dependency weight of each time step on other time steps based on the hidden state matrix, so as to obtain the battery capacity decay time series feature vector, which is used as a feature parameter to characterize the battery capacity decay trend.

6. The battery life prediction method based on a dual-branch neural network according to claim 1, characterized in that, The cross-battery feature fusion module includes: The isomorphic feature fusion unit is used to concatenate the feature parameters of the battery under test and its corresponding reference battery, which are used to characterize the degree of battery electrode degradation, to obtain a first concatenation vector, and to concatenate the feature parameters of the battery under test and its corresponding reference battery, which are used to characterize the battery capacity degradation trend, to obtain a second concatenation vector. The heterogeneous feature fusion unit is used to concatenate the first concatenation vector and the second concatenation vector and input them into a preset attention function to generate corresponding fused features; A remaining life regression model is used to predict the remaining life of the battery under test based on the fused features.

7. The battery life prediction method based on a two-branch neural network according to claim 6, characterized in that, The isomorphic feature fusion unit includes two fully connected layers, which are used to splice the feature parameters of the battery under test and its corresponding reference battery to characterize the degree of battery electrode degradation, and to splice the feature parameters of the battery under test and its corresponding reference battery to characterize the battery capacity degradation trend.

8. The battery life prediction method based on a dual-branch neural network according to claim 1, characterized in that, The training steps of the dual-branch neural network include: Obtain charge and discharge data of batteries with different remaining lifespans to construct training sample sets for short-life battery packs, medium-life battery packs, and long-life battery packs. Two training samples are randomly selected from the short-life battery pack to form a cross-battery sample pair. All cross-battery sample pairs are input into the dual-branch neural network. The total loss between the classification label output by the dual-branch neural network and the true label is calculated based on the hybrid loss function. The dual-branch neural network is then optimized in reverse. Two training samples are randomly selected from the medium-life battery pack to form a cross-battery sample pair. All cross-battery sample pairs are input into the dual-branch neural network. The total loss between the classification label output by the dual-branch neural network and the true label is calculated based on the hybrid loss function. The dual-branch neural network is then optimized in reverse. Two training samples are randomly selected from the long-life battery pack to form a cross-battery sample pair. All cross-battery sample pairs are input into the dual-branch neural network. The total loss between the classification label output by the dual-branch neural network and the true label is calculated based on the hybrid loss function. The dual-branch neural network is then optimized in reverse. 9.The battery life prediction method based on the dual-branch neural network according to claim 8, characterized in that, The formula for calculating the hybrid loss function is as follows: ， ， in, This represents the hybrid loss function. Represents the regression loss term. Represents the physical consistency loss term. , These are the preset coefficients. This indicates the number of samples in the current training batch. This represents the classification label of the i-th training sample. This represents the true label of the i-th training sample. This represents the trainable parameters of the remaining life regression model of the cross-battery feature fusion module in the dual-branch neural network.

10. The battery life prediction method based on a dual-branch neural network according to claim 8, characterized in that, The physical consistency loss term includes at least a physical fusion loss term and a physical monotonicity loss term.

Images