Battery electro-thermal digital twin model construction method and device based on hybrid regularization

Abstract

The application provides a battery electro-thermal digital twin model construction method and device based on hybrid regularization. The method comprises: determining a training data set based on a virtual battery test platform; constructing an initial double-branch neural network model according to the training data set, the initial double-branch neural network model comprising an electrochemical branch and a thermodynamic branch; constructing a composite loss function comprising a partial differential equation residual loss and a thermal ordinary differential equation regularization loss, training the initial double-branch neural network model based on the composite loss function to obtain a target double-branch neural network model; constructing a hybrid regularization loss function, training the target double-branch neural network model based on the hybrid regularization loss function to obtain a battery electro-thermal digital twin model. In this way, through the combination of the double-branch neural network model of the decoupling architecture and the hybrid physical regularization collaborative design, the fidelity, calculation efficiency and physical generalization ability of the battery electro-thermal digital twin model are improved.

Description

Technical Field

[0001] This invention relates to the field of battery state assessment technology, and more specifically to a method and apparatus for constructing a battery electrothermal digital twin model based on hybrid regularization. Background Technology

[0002] With the rapid development of new energy vehicles and energy storage systems, battery management systems (BMS) are facing higher demands for accurate estimation of the internal state of batteries. In particular, the prediction of unmeasurable physical quantities such as solid phase concentration distribution, internal temperature, and overpotential is of great significance for improving safety and extending battery life.

[0003] Existing technologies mainly include: (1) High-fidelity models based on electrochemical mechanisms, such as pseudo-two-dimensional models (P2D), have high physical accuracy but high computational complexity, making them difficult to deploy in real time in embedded systems; (2) Simplified models, such as the Single Particle Model (SPM), reduce the computational load, but the prediction accuracy is insufficient under high magnification and wide temperature range conditions; (3) Pure data-driven models have the advantage of fast computation speed, but lack physical consistency, have poor generalization ability in unseen working conditions, and are prone to parameter drift in long-term prediction.

[0004] Furthermore, battery systems exhibit significant multi-timescale coupling characteristics, with electrochemical reactions operating at millisecond-level dynamics and thermal conduction processes operating at minute-level dynamics. The rigidity differences between the two are substantial. Under a unified training framework, gradient propagation between the rapid electrochemical dynamics and the slow thermal dynamics is prone to instability, leading to difficulties in model training or physical parameter drift, making it difficult for traditional models to achieve stable training and convergence.

[0005] Therefore, a battery electrothermal coupling modeling method that combines physical consistency, computational efficiency, and generalization ability is needed. Summary of the Invention

[0006] This application aims to provide a method and apparatus for constructing a battery electrothermal digital twin model based on hybrid regularization, so that the battery electrothermal digital twin model has high fidelity, high computational efficiency and strong physical generalization ability.

[0007] Firstly, a method for constructing a battery electrothermal digital twin model based on hybrid regularization is provided, the method comprising: A training dataset is determined based on a pre-built virtual battery testing platform, wherein the training dataset includes external observation data of the battery and internal unmeasurable physical state data; An initial dual-branch neural network model is constructed based on the training dataset, the initial dual-branch neural network model including an electrochemical branch and a thermodynamic branch; A composite loss function is constructed, which includes the residual loss of partial differential equations and the regularization loss of thermal ordinary differential equations. The initial bi-branch neural network model is trained based on the composite loss function to obtain the target bi-branch neural network model. A hybrid regularized loss function is constructed, and the target dual-branch neural network model is trained based on the hybrid regularized loss function to obtain a battery electrothermal digital twin model. The battery electrothermal digital twin model is used to predict the battery state of charge, battery core temperature, and battery terminal voltage.

[0008] Optionally, the training dataset determined based on the pre-built virtual battery testing platform includes: Simulations of lithium-ion batteries were conducted based on electrochemical mechanism models, generating multiple sets of battery operation data by setting different temperatures, different rates, and different dynamic load conditions. The training dataset is obtained by combining battery operation data generated under various temperatures, rates, and dynamic load conditions.

[0009] Optionally, constructing the initial two-branch neural network model based on the training dataset includes: Based on the training dataset, features are constructed to obtain input features, which include spatiotemporal structure features, load dynamic features, thermal environment features, and multi-scale frequency coding features. Based on the input features, an initial bi-branch neural network model containing electrochemical and thermodynamic branches is constructed.

[0010] Optionally, the initial dual-branch neural network model further includes: a multi-scale Fourier encoder and a non-trained Arrhenius physical layer, wherein the Arrhenius physical layer is represented as: ; ; ; in, Represents the solid-phase diffusion coefficient. Represents the exchange current density. Represents the reaction rate constant. , as well as Indicates reference temperature The physical parameters below, , , These are the corresponding apparent activation energies, where R represents the gas constant and T represents the thermodynamic temperature.

[0011] Optionally, the composite loss function is expressed as:

[0012] in, This represents the total loss value of the composite loss function. , , These are the weight parameters, This represents the residual of the constrained solid-phase diffusion equation. Key features representing the energy conservation relationship between the time derivative of the constraint temperature and the heat generation / dissipation power. This indicates the conservation of the volume integral of the concentration within the constrained particles.

[0013] Optionally, constructing the hybrid regularization loss function includes: A hybrid regularization loss function is constructed, which includes voltage L1 loss, capacity parameter anchoring loss, and battery state-of-charge mapping parameter regularization loss. The hybrid regularization loss function is expressed as follows:

[0014] in, Describe the overall objective function. Indicates voltage loss L1, This represents the battery state of charge mapping parameters. This represents the anchoring loss of the capacity parameter. , These are the battery state-of-charge mapping parameter anchoring strength coefficient and the capacity parameter anchoring strength coefficient, respectively.

[0015] Secondly, a battery electrothermal digital twin model construction device based on hybrid regularization is provided, the device comprising: The determination module is used to determine a training dataset based on a pre-built virtual battery testing platform, wherein the training dataset includes external observation data and internal unmeasurable physical state data of the battery; The first construction module is used to construct an initial dual-branch neural network model based on the training dataset, wherein the initial dual-branch neural network model includes an electrochemical branch and a thermodynamic branch; The second construction module is used to construct a composite loss function that includes the residual loss of partial differential equations and the regularization loss of thermal ordinary differential equations, and to train the initial dual-branch neural network model based on the composite loss function to obtain the target dual-branch neural network model. The third construction module is used to construct a hybrid regularized loss function, and to train the target dual-branch neural network model based on the hybrid regularized loss function to obtain a battery electrothermal digital twin model. The battery electrothermal digital twin model is used to predict the battery state of charge, battery core temperature and battery terminal voltage.

[0016] Thirdly, an electronic device is provided, comprising: The memory is configured to store instructions; The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the battery electrothermal digital twin model construction method based on hybrid regularization provided in the first aspect of the embodiments of this application.

[0017] Fourthly, a machine-readable storage medium is provided, wherein instructions are stored on the machine-readable storage medium, the instructions being used to cause a machine to execute the above-described method for constructing a battery electrothermal digital twin model based on hybrid regularization.

[0018] Fifthly, a computer program product is provided, wherein when the instructions in the computer program product are executed by the processor of an electronic device, the electronic device performs the battery electrothermal digital twin model construction method based on hybrid regularization as described above.

[0019] Based on the aforementioned method for constructing a battery electrothermal digital twin model, a training dataset is determined using a pre-built virtual battery testing platform. This dataset includes external observation data and internal, unmeasurable physical state data of the battery. An initial bi-branch neural network model is constructed based on this dataset, comprising electrochemical and thermodynamic branches. A composite loss function, incorporating residual loss from partial differential equations and regularization loss from thermodynamic ordinary differential equations, is then constructed. This composite loss function is used to train the initial bi-branch neural network model, yielding a target bi-branch neural network model. Finally, a hybrid regularization loss function is constructed, and this hybrid regularization loss function is used to train the target bi-branch neural network model, resulting in the battery electrothermal digital twin model. This model is used to predict the battery's state of charge, core temperature, and terminal voltage. Thus, by combining a decoupled bi-branch neural network model with a hybrid physical regularization co-design, the fidelity, computational efficiency, and physical generalization ability of the battery electrothermal digital twin model are improved. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the battery electrothermal digital twin model construction method based on hybrid regularization provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the battery electrothermal digital twin model construction device based on hybrid regularization provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0022] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0023] The following description, in conjunction with the accompanying drawings, details the battery electrothermal digital twin model construction method and apparatus based on hybrid regularization provided in this application, through specific embodiments and application scenarios.

[0024] Please see Figure 1 This is a flowchart illustrating the battery electrothermal digital twin model construction method based on hybrid regularization provided in this application embodiment. This method is applied to electronic devices. Figure 1 As shown, the method includes the following steps S100 to S400.

[0025] Step S100: Determine the training dataset based on the pre-built virtual battery testing platform, wherein the training dataset includes external observation data of the battery and internal unmeasurable physical state data.

[0026] In this embodiment, a high-fidelity virtual battery testing platform can first be constructed. Based on an electrochemical mechanism model, a training dataset containing different temperatures, rates, and dynamic operating conditions is generated. The training dataset includes external observation data and internal, unmeasurable physical state data of the battery. External observation data can be understood as external signals exhibited by the battery that can be directly acquired by sensors, such as voltage, current, and temperature.

[0027] Internal unmeasurable physical state data can be understood as microscopic physical quantities that occur inside the battery and cannot be directly measured by sensors in real experiments. These are also true values ​​that can only be generated by electrochemical mechanism model simulation, such as the solid-phase lithium ion concentration distribution inside electrode particles and the liquid-phase ion concentration field in the electrolyte.

[0028] Step S200: Construct an initial dual-branch neural network model based on the training dataset. The initial dual-branch neural network model includes an electrochemical branch and a thermodynamic branch.

[0029] In this embodiment, the initial dual-branch neural network model can be understood as a neural network model including an electrochemical branch and a thermodynamic branch. The electrochemical branch can handle fast dynamic electrochemical reactions at the millisecond level, while the thermodynamic branch can handle slow dynamic thermal evolution processes at the minute level. The electrochemical branch takes as input spatiotemporal features including radial basis function extensions and outputs a solid-phase concentration field; the thermodynamic branch takes as input current and ambient temperature features and outputs a core temperature. A non-trainable Arrhenius physics layer is embedded between the electrochemical and thermodynamic branches. The Arrhenius physics layer is a structured physics embedding module with no trainable parameters, used to dynamically correct kinetic parameters based on the real-time core temperature T output by the thermodynamic branch. The calculation results of the Arrhenius physics layer are fed back to the electrochemical branch and the overpotential compensation module in real time, achieving temperature-driven kinetic coupling. The mathematical expression of the Arrhenius physics layer is: ; ; ; in, Represents the solid-phase diffusion coefficient. Represents the exchange current density. Represents the reaction rate constant. , as well as Indicates reference temperature The physical parameters below, , , These are the corresponding apparent activation energies, where R represents the gas constant and T represents the thermodynamic temperature.

[0030] Step S300: Construct a composite loss function that includes the residual loss of partial differential equations and the regularization loss of thermal ordinary differential equations, and train the initial dual-branch neural network model based on the composite loss function to obtain the target dual-branch neural network model.

[0031] In this embodiment of the application, a composite loss function is constructed that includes the residual loss of the partial differential equation and the regularization loss of the thermal ordinary differential equation. The composite loss function can be expressed as:

[0032] in, This represents the total loss value of the composite loss function. , , These are the weight parameters, This represents the residual of the constrained solid-phase diffusion equation. Key features representing the energy conservation relationship between the time derivative of the constraint temperature and the heat generation / dissipation power. This indicates the conservation of the volume integral of the concentration within the constrained particles.

[0033] The initial bi-branch neural network model is trained using a composite loss function to obtain the target bi-branch neural network model. In this training phase, voltage output-related parameters are frozen, and only the network core is trained to block the interference of voltage error gradients on the physical field learning process, enabling the model to learn the internal physical field of the battery. Through this first-stage training, the network can reconstruct the solid phase concentration field and temperature field, thereby mastering the internal physical evolution laws of the battery.

[0034] Step S400: Construct a hybrid regularized loss function, train the target dual-branch neural network model based on the hybrid regularized loss function, and obtain a battery electrothermal digital twin model. The battery electrothermal digital twin model is used to predict the battery state of charge, battery core temperature, and battery terminal voltage.

[0035] In this embodiment, after completing the first stage of training, the second stage of training begins. This stage unlocks some network parameters and introduces an overpotential compensation module. A hybrid regularization strategy is employed to minimize the terminal voltage error while simultaneously applying soft anchoring constraints and state-of-charge (SOC) mapping regularization to the battery capacity scaling parameters, forming a hybrid regularization objective function. This prevents physical drift of the capacity parameters during long-term voltage fitting, thereby maintaining the physical consistency of the SOC estimation results. A hybrid regularization loss function is constructed, comprising voltage L1 loss, capacity parameter anchoring loss, and battery SOC mapping parameter regularization loss. The hybrid regularization loss function is expressed as follows:

[0036] in, Describe the overall objective function. Indicates voltage loss L1, This represents the battery state of charge mapping parameters. This represents the anchoring loss of the capacity parameter. , These are the anchoring strength coefficients for the battery state-of-charge mapping parameters and the capacity parameter, respectively, used to adjust the trade-off between data fitting accuracy and parameter stability. Through the second stage of training, physical parameter drift can be suppressed while maintaining voltage prediction accuracy.

[0037] After the two-stage training process described above, the final battery electrothermal digital twin model is obtained. This model can predict the battery's key states in real time, including battery SOC, core battery temperature, and battery terminal voltage, based on input operating data such as current and ambient temperature. Because the model incorporates physical constraints, it still exhibits good generalization ability under unseen operating conditions and can be deployed in real time within embedded battery management systems.

[0038] Through steps S100-S400, a training dataset is determined based on a pre-built virtual battery testing platform. This training dataset includes external battery observation data and internal, unmeasurable physical state data. An initial bi-branch neural network model is constructed based on the training dataset, comprising an electrochemical branch and a thermodynamic branch. A composite loss function is constructed, incorporating residual loss from partial differential equations and regularization loss from thermodynamic ordinary differential equations. The initial bi-branch neural network model is trained using this composite loss function to obtain a target bi-branch neural network model. A hybrid regularization loss function is then constructed, and the target bi-branch neural network model is trained using this hybrid regularization loss function to obtain a battery electrothermal digital twin model. This battery electrothermal digital twin model is used to predict the battery's state of charge, core temperature, and terminal voltage. Thus, by combining a decoupled bi-branch neural network model with a hybrid physical regularization co-design, the fidelity, computational efficiency, and physical generalization ability of the battery electrothermal digital twin model are improved.

[0039] In some implementations, the pre-built virtual battery testing platform determines the training dataset, including: Simulations of lithium-ion batteries were conducted based on electrochemical mechanism models, generating multiple sets of battery operation data by setting different temperatures, different rates, and different dynamic load conditions. The training dataset is obtained by combining battery operation data generated under various temperatures, rates, and dynamic load conditions.

[0040] Specifically, lithium-ion batteries can be simulated using an electrochemical P2D model or an equivalent electrochemical mechanism model, combined with a lumped thermal model to construct a battery electrothermal coupling simulation environment. Battery operating data is generated by setting different temperatures, rates, and dynamic load conditions. The generated data includes: external observables: terminal voltage, current, ambient temperature; internal physical states: solid phase concentration distribution, core temperature, etc. Training datasets are generated through combinations of various operating conditions, thereby ensuring the model's generalization ability under different temperature and rate conditions.

[0041] In some implementations, constructing an initial two-branch neural network model based on the training dataset includes: Based on the training dataset, features are constructed to obtain input features, which include spatiotemporal structure features, load dynamic features, thermal environment features, and multi-scale frequency coding features. Based on the input features, an initial bi-branch neural network model containing electrochemical and thermodynamic branches is constructed.

[0042] Specifically, after obtaining the training data, features are constructed from the data to obtain input features. Input features may include spatiotemporal structure features, load dynamic features, thermal environment features, and multi-scale frequency coding features.

[0043] Among them, the electrochemical branch receives spatiotemporal characteristics and load characteristics, while the thermal branch only receives load characteristics and ambient temperature characteristics, thereby achieving electrothermal decoupling in physical structure.

[0044] To enhance the model's ability to express dynamic behavior at different time scales, this embodiment performs multi-scale Fourier feature encoding on the time variables.

[0045] Specifically, the frequency domain is divided into three intervals: low frequency, mid frequency, and high frequency. Within each interval, a frequency set is generated using a logarithmic uniform sampling method, which includes low frequency features, mid frequency features, and high frequency features.

[0046] The low-frequency band is mainly used to describe the slow dynamic changes of batteries over long time scales, and its frequency range is:

[0047] Within this interval, 16 frequencies are sampled in a logarithmically uniform manner:

[0048] Low-frequency characteristics mainly reflect the slow evolution of battery SOC during a complete charge-discharge cycle.

[0049] The mid-frequency band is used to describe the voltage response behavior of a battery under dynamic load changes, and its frequency range is:

[0050] Within this interval, 20 frequencies are sampled in a logarithmic uniform manner:

[0051] The mid-frequency band is mainly used to characterize the response characteristics of battery polarization dynamics and current changes.

[0052] The high-frequency band is used to describe the rapid dynamic processes of a battery under transient operating conditions, and its frequency range is:

[0053] Within this interval, 12 frequencies are sampled in a logarithmically uniform manner:

[0054] High-frequency features are used to enhance the model's ability to express current abrupt changes and short-term dynamic processes.

[0055] Apply the above frequency set to the time variable Construct Fourier feature maps:

[0056] Where f is the frequency variable, The k-th sampling frequency in the low-frequency band. The k-th sampling frequency in the mid-frequency band. Let k be the k-th sampling frequency in the high-frequency band, where k is the frequency sampling number and t is the time variable. is the encoded vector of the time variable after Fourier feature mapping, and n is the total number of frequency samples, that is, the total number of sampling frequencies across all frequency bands.

[0057] By using the multi-scale Fourier encoding method described above, the neural network can simultaneously learn the dynamic changes of the battery system at different time scales, thereby improving the model's ability to express complex electrochemical-thermal coupling processes.

[0058] It should be noted that the division range and sampling quantity of low frequency, medium frequency and high frequency in this embodiment can be adjusted according to the specific application scenario, and are not limited to the above values.

[0059] In this embodiment, the electrochemical branch is used to learn the evolution of the solid-phase concentration field inside the battery. Its inputs include spatiotemporal structural features and current load features. After feature extraction through a multi-layer neural network, the solid-phase concentration field is output. The thermal branch is used to predict the battery core temperature. Its inputs include current features and ambient temperature; the output is the battery core temperature T. An Arrhenius physics layer is embedded between the electrochemical and thermal branches to achieve temperature-driven kinetic parameter correction. This physics layer does not contain trainable parameters to ensure that the model satisfies basic physical laws.

[0060] This application first constructs a high-fidelity virtual testing platform based on electrochemical mechanisms to generate training data containing different temperatures, magnifications, and dynamic operating conditions, while simultaneously obtaining external observable data and internal unmeasurable physical state data. Subsequently, a decoupled dual-branch computational structure containing an electrochemical dynamic processing module and a thermal dynamic processing module is constructed, and an Arrhenius physics layer with untrainable parameters is embedded between the two branches to correct the diffusion coefficient and reaction kinetic parameters in real time according to temperature, thereby realizing the physical embedding expression of the electro-thermal coupling relationship.

[0061] In terms of training strategy, a two-stage optimization mechanism is adopted: the first stage focuses on learning the evolution law of the concentration field and temperature field inside the battery through the residual constraint of the solid-phase diffusion equation and the constraint of the thermal energy conservation ordinary differential equation (ODE) to solve the rigidity problem between the millisecond-level dynamics of electrochemistry and the minute-level dynamics of thermal process; the second stage introduces an overpotential compensation module after unlocking some parameters, and combines the voltage error term, the soft anchoring constraint of the capacity scaling parameter and the state of charge (SOC) mapping regularization to form a hybrid regularization objective function to achieve voltage output calibration and suppress physical parameter drift.

[0062] The embodiments of this application, through the collaborative design of decoupled architecture and hybrid physical regularization, enhance physical consistency while ensuring computational efficiency, enabling the model to achieve high-precision prediction under unseen temperature conditions and significantly reducing error accumulation during long-term operation, making it suitable for real-time deployment of embedded battery management systems.

[0063] This application's embodiments address the multiphysics rigidity problem: through a dual-branch decoupling design, the dynamics of electricity and heat are handled separately, avoiding gradient conflicts. Simultaneously, zero-sample generalization is achieved: by embedding the Arrhenius physics layer, the model can extrapolate unseen low-temperature operating conditions based on physical laws. This application's embodiments exhibit extremely high computational efficiency, with single-step inference latency at the microsecond level, making it suitable for embedded deployment. This application's embodiments ensure physical consistency: through hybrid regularization (ODE constraints + parameter anchoring), overfitting and physical parameter drift in purely data-driven models are prevented.

[0064] Please see Figure 2 This is a schematic diagram of the structure of a battery electrothermal digital twin model construction device based on hybrid regularization provided in an embodiment of this application. A second aspect of this application provides a battery electrothermal digital twin model construction device based on hybrid regularization, the device comprising: The determination module is used to determine a training dataset based on a pre-built virtual battery testing platform, wherein the training dataset includes external observation data and internal unmeasurable physical state data of the battery; The first construction module is used to construct an initial dual-branch neural network model based on the training dataset, wherein the initial dual-branch neural network model includes an electrochemical branch and a thermodynamic branch; The second construction module is used to construct a composite loss function that includes the residual loss of partial differential equations and the regularization loss of thermal ordinary differential equations, and to train the initial dual-branch neural network model based on the composite loss function to obtain the target dual-branch neural network model. The third construction module is used to construct a hybrid regularized loss function, and to train the target dual-branch neural network model based on the hybrid regularized loss function to obtain a battery electrothermal digital twin model. The battery electrothermal digital twin model is used to predict the battery state of charge, battery core temperature and battery terminal voltage.

[0065] The battery electrothermal digital twin model construction apparatus based on hybrid regularization provided in the second aspect of the embodiments of this application can realize the various processes implemented in the above method embodiments and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0066] Please see Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. A third aspect of this application provides an electronic device 3000, including a processor 3100 and a memory 3200. The memory 3200 stores machine-executable instructions that can be executed by the processor 3100. The processor 3100 can execute the machine-executable instructions to implement the above-mentioned method for constructing a battery electrothermal digital twin model based on hybrid regularization.

[0067] In some embodiments, this application also provides a machine-readable storage medium storing instructions that, when executed by a processor, cause the processor to implement the above-described method for constructing a battery electrothermal digital twin model based on hybrid regularization.

[0068] In some embodiments, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the battery electrothermal digital twin model construction method based on hybrid regularization according to the above embodiments.

[0069] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0070] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0071] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0072] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0073] Computer-readable media include both permanent and non-permanent, removable and non-removable media, which can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0074] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0075] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0076] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

Claims

1. A method for constructing a battery electrothermal digital twin model based on hybrid regularization, characterized in that, The method includes: A training dataset is determined based on a pre-built virtual battery testing platform, wherein the training dataset includes external observation data of the battery and internal unmeasurable physical state data; An initial dual-branch neural network model is constructed based on the training dataset, the initial dual-branch neural network model including an electrochemical branch and a thermodynamic branch; A composite loss function is constructed, which includes the residual loss of partial differential equations and the regularization loss of thermal ordinary differential equations. The initial bi-branch neural network model is trained based on the composite loss function to obtain the target bi-branch neural network model. A hybrid regularized loss function is constructed, and the target dual-branch neural network model is trained based on the hybrid regularized loss function to obtain a battery electrothermal digital twin model. The battery electrothermal digital twin model is used to predict the battery state of charge, battery core temperature, and battery terminal voltage.

2. The method according to claim 1, characterized in that, The training dataset determined by the pre-built virtual battery testing platform includes: Simulations of lithium-ion batteries were conducted based on electrochemical mechanism models, generating multiple sets of battery operation data by setting different temperatures, different rates, and different dynamic load conditions. The training dataset is obtained by combining battery operation data generated under various temperatures, rates, and dynamic load conditions.

3. The method according to claim 1, characterized in that, The step of constructing an initial dual-branch neural network model based on the training dataset includes: Based on the training dataset, features are constructed to obtain input features, which include spatiotemporal structure features, load dynamic features, thermal environment features, and multi-scale frequency coding features. Based on the input features, an initial bi-branch neural network model containing electrochemical and thermodynamic branches is constructed.

4. The method according to claim 1, characterized in that, The initial dual-branch neural network model further includes: a multi-scale Fourier encoder and a non-trained Arrhenius physical layer, wherein the Arrhenius physical layer is represented as: ； ； ； in, Represents the solid-phase diffusion coefficient. Represents the exchange current density. Represents the reaction rate constant. , as well as Indicates reference temperature The physical parameters below, , , These are the corresponding apparent activation energies, where R represents the gas constant and T represents the thermodynamic temperature.

5. The method according to claim 1, characterized in that, The composite loss function is expressed as follows: in, This represents the total loss value of the composite loss function. , , These are the weight parameters, This represents the residual of the constrained solid-phase diffusion equation. Key features representing the energy conservation relationship between the time derivative of the constraint temperature and the heat generation / dissipation power. This indicates the conservation of the volume integral of the concentration within the constrained particles.

6. The method according to claim 1, characterized in that, The construction of the hybrid regularization loss function includes: A hybrid regularization loss function is constructed, which includes voltage L1 loss, capacity parameter anchoring loss, and battery state-of-charge mapping parameter regularization loss. The hybrid regularization loss function is expressed as follows: in, Describe the overall objective function. Indicates voltage loss L1, This represents the battery state of charge mapping parameters. This represents the anchoring loss of the capacity parameter. , These are the battery state-of-charge mapping parameter anchoring strength coefficient and the capacity parameter anchoring strength coefficient, respectively.

7. A device for constructing a battery electrothermal digital twin model based on hybrid regularization, characterized in that, The device includes: The determination module is used to determine a training dataset based on a pre-built virtual battery testing platform, wherein the training dataset includes external observation data and internal unmeasurable physical state data of the battery; The first construction module is used to construct an initial dual-branch neural network model based on the training dataset, wherein the initial dual-branch neural network model includes an electrochemical branch and a thermodynamic branch; The second construction module is used to construct a composite loss function that includes the residual loss of partial differential equations and the regularization loss of thermal ordinary differential equations, and to train the initial dual-branch neural network model based on the composite loss function to obtain the target dual-branch neural network model. The third construction module is used to construct a hybrid regularized loss function, and to train the target dual-branch neural network model based on the hybrid regularized loss function to obtain a battery electrothermal digital twin model. The battery electrothermal digital twin model is used to predict the battery state of charge, battery core temperature and battery terminal voltage.

8. An electronic device, characterized in that, include: The memory is configured to store instructions; A processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the battery electrothermal digital twin model construction method based on any one of claims 1 to 6.

9. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to execute the battery electrothermal digital twin model construction method based on hybrid regularization according to any one of claims 1 to 6.

10. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device performs the battery electrothermal digital twin model construction method based on hybrid regularization as described in any one of claims 1 to 6.

Images