A battery pack life prediction digital twin system

Abstract

The application relates to the technical field of battery pack detection, and discloses a battery pack life prediction digital twin system, which comprises a data acquisition module, a storage and a processor.The data acquisition module is used for acquiring operation data of a battery pack; the storage is used for storing instructions; and the processor is used for executing the instructions to realize the following: a model system composed of a multi-physical field coupling model solving module, a weak degradation feature extraction module and a model parameter correction module is established; the multi-physical field coupling model solving module is used for solving a coupling equation set representing the internal electrochemical, thermal and mechanical states of the battery pack; and the weak degradation feature extraction module is used for analyzing weak feature signals representing early aging of the battery from the operation data.Through the construction of a heterogeneous algebra fusion engine, the defects of independent work and information fragmentation of various models in the traditional method are overcome, so that the analysis of the internal state of the battery is more comprehensive and accurate, and the overall accuracy of life prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of battery pack testing technology, specifically a digital twin system for predicting battery pack lifespan. Background Technology

[0002] Digital twin technology, by constructing a virtual model that maps in real time to the physical battery pack, provides important technical support for monitoring battery health status, predicting remaining lifespan, diagnosing faults, and managing safety in fields such as new energy vehicles, electrochemical energy storage power stations, and consumer electronics.

[0003] Existing technologies for constructing battery digital twin systems typically face multiple technical bottlenecks. Physical mechanism models, data-driven models, and parameter identification methods are often fragmented. They operate within independent frameworks, preventing effective bidirectional information flow. This separation leads to incomplete descriptions of battery states. Lifetime predictions are also largely limited to macroscopic electrical data such as voltage and current. These data are not sensitive to early microscopic degradation of internal battery materials. Therefore, existing early warning mechanisms are usually lagging, only responding when significant performance degradation has already occurred. Furthermore, manufacturing differences and varying usage histories between individual batteries cause severe model parameter drift. Models calibrated for specific batteries experience a significant drop in accuracy when directly applied to other batteries, fundamentally limiting generalization ability. In addition, high-precision multiphysics simulations are computationally expensive. Existing technologies often employ fixed simplification strategies to improve efficiency. This static allocation of computational resources cannot adaptively adjust to the complexity of battery states, making it difficult to balance accuracy and efficiency. Finally, once an existing model is established, its internal structure is often fixed. Faced with new aging patterns or unforeseen operating conditions, the model's adaptability is limited, and its long-term effectiveness is difficult to guarantee. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a digital twin system for predicting battery pack lifespan. This system solves the problems in existing technologies where different models based on physical mechanisms, data-driven approaches, and parameter identification are fragmented, leading to insensitivity to early micro-degradation characteristics, delayed early warnings, difficulty in overcoming individual differences, insufficient generalization ability, rigid model structure, and inflexible allocation of computing resources, making it difficult to balance the accuracy, efficiency, and adaptability of long-term predictions.

[0005] To achieve the above objectives, a first aspect of the present invention provides a digital twin system for predicting battery pack lifespan, comprising: The data acquisition module is used to acquire the operating data of the battery pack; Memory, used to store program instructions; A processor, connected to the memory, configured to execute the program instructions to perform: A model system is established, consisting of a multiphysics coupled model solution module, a weak degradation feature extraction module, and a model parameter correction module; in, The multiphysics coupling model solving module is used to solve the set of equations characterizing the coupling state of the electrochemical field, thermodynamic field and mechanical field inside the battery pack; The weak degradation feature extraction module is used to extract weak feature signals characterizing the early aging process of the battery from the operating data. The model parameter correction module is used to correct the parameters of the multiphysics coupling model based on the historical operating data of the battery pack. as well as, A heterogeneous algebraic fusion engine is constructed. The heterogeneous algebraic fusion engine is used to coordinate and control the solution process of the multiphysics coupling model solution module based on the weak feature signal and the parameter space topology determined by the model parameter correction module, and to generate the battery pack lifetime prediction information based on the output of the process.

[0006] In an optional implementation, the multiphysics coupling model solving module discretizes the coupling equations into higher-order state tensors and solves the higher-order state tensors using tensor network decomposition. Specifically, the heterogeneous algebraic fusion engine dynamically adjusts the tensor rank in the tensor network decomposition operation based on changes in the weak feature signal, adaptively allocating computational resources.

[0007] In an optional implementation, the weak degradation feature extraction module extracts the weak feature signal by constructing an equivalent Hamiltonian characterizing the internal electrochemical environment of the battery and by analyzing the eigenenergy spectrum of the equivalent Hamiltonian. The operational data includes microphysical field data acquired from an acoustic emission sensor, an electrochemical impedance spectroscopy analyzer, or a stress sensor, which is used to construct the perturbation term in the equivalent Hamiltonian.

[0008] In an optional implementation, the model parameter correction module constructs a parametric manifold by analyzing the historical operating data of the battery pack, and uses the topological invariants of the parametric manifold to correct the parameters of the multiphysics coupling model to eliminate non-essential differences between individual batteries.

[0009] In an optional implementation, the model parameter correction module specifically employs a continuous homology algorithm to analyze the historical operating data of the battery pack in order to construct the parametric manifold and extract the topological invariants.

[0010] In an optional implementation, the heterogeneous algebraic fusion engine achieves coordinated control of the multiphysics coupling model solution module by constructing a unified mathematical framework that integrates the symmetry of the solution module's computational structure, the spectral structure of the weak feature signal, and the topological structure of the parameter space at the algebraic level.

[0011] In an optional implementation, when the processor executes the program instructions, it also calculates the health status and remaining service life of the battery pack based on the battery internal state variables output by the multiphysics coupling model solving module.

[0012] In an optional implementation, the processor, when executing the program instructions, further performs the following: comparing the calculated value of the health status with the measured value from an external measurement to obtain a prediction error; and, based on the prediction error, reverse-optimizing the internal parameters of the heterogeneous algebraic fusion engine to achieve adaptive model updates.

[0013] A second aspect of the present invention provides a digital twin method for predicting battery pack lifespan, the method comprising the following steps: Acquire the operating data of the battery pack and preprocess the operating data; Build a heterogeneous algebra fusion engine; Under the control of the heterogeneous algebra fusion engine, the following operations are performed collaboratively: Solve the coupled equations characterizing the internal electrochemical, thermal, and mechanical states of the battery pack; Weak characteristic signals representing early battery aging were extracted from the operational data. The model parameters of the coupled equations are corrected based on the historical operating data of the battery pack. Based on the results of the collaborative execution operation, lifespan prediction information for the battery pack is generated.

[0014] In an optional implementation, the method further includes: Based on the lifespan prediction information, the virtual model of the battery pack digital twin is updated in real time.

[0015] This invention provides a digital twin system for predicting battery pack lifespan. It offers the following advantages: 1. This invention constructs a heterogeneous algebraic fusion engine that collaboratively controls three modules: multi-physics coupled model solving, weak degradation feature extraction, and model parameter correction. This achieves a deep fusion of physical mechanisms, microscopic features, and statistical laws. The engine ensures real-time interaction and mutual constraints between different modules, overcoming the shortcomings of traditional methods where models operate independently and information is fragmented. This results in a more comprehensive and accurate analysis of the battery's internal state, thereby improving the overall accuracy of lifetime prediction.

[0016] 2. This invention introduces a quantum-derived feature extraction module to construct an equivalent Hamiltonian from the microscopic physical field data inside the battery, and analyzes its energy spectrum changes to identify early degradation characteristics. This method can capture weak signals occurring at the material's microstructure level that indicate the onset of irreversible aging, signals that are not yet apparent in macroscopic electrical data. Therefore, it endows the system with the ability to issue early warnings before significant battery performance degradation occurs.

[0017] 3. This invention employs a model parameter correction module based on topological data analysis to construct a standardized parametric manifold from historical data of the battery fleet, and uses its topological invariants to correct the model parameters of individual batteries. This effectively filters out non-essential individual differences caused by manufacturing process fluctuations or different usage histories, enabling the model to focus on universal aging patterns and significantly enhancing the model's generalization ability and applicability across different individuals and batches of batteries.

[0018] 4. This invention constructs the solution process of the multiphysics coupling model as a dynamic tensor network, and the heterogeneous algebra fusion engine dynamically adjusts its tensor rank according to the real-time analytical system state. When the battery state is stable, the system uses a lower rank for fast calculation; while when complex conditions such as accelerated aging are detected, the tensor rank is adaptively increased for refined solution. This mechanism of allocating computing resources on demand reduces the overall computational complexity and improves the algorithm's running efficiency while ensuring the computational accuracy in key stages.

[0019] 5. This invention establishes a system self-optimization closed loop, using the error between model predictions and external measured values ​​as a loss function to inversely optimize the core generation parameters of the heterogeneous algebraic fusion engine. This mechanism does not adjust the parameters of a specific model, but rather optimizes the meta-rules themselves that coordinate the collaborative work of various modules. This enables the entire digital twin system to possess the ability to continuously learn and adaptively evolve, thereby ensuring its long-term effectiveness and reliability throughout the battery's entire lifespan. Attached Figure Description

[0020] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a schematic diagram of the data acquisition module of the present invention; Figure 3 This is a schematic diagram of the multiphysics coupling model solving module of the present invention; Figure 4 This is a schematic diagram of the weak degradation feature extraction module of the present invention; Figure 5 This is a schematic diagram of the model parameter correction module of the present invention; Figure 6This is a schematic diagram of the heterogeneous algebra fusion engine of the present invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example: Please see the appendix Figure 1 -Appendix Figure 6 This invention provides a digital twin system for predicting battery pack lifespan, comprising: Multi-source heterogeneous data acquisition and spatiotemporal alignment: This step aims to provide a comprehensive, accurate, and consistent data foundation for the subsequent construction and solution of the digital twin model. Its core task is to collect various physical quantities that reflect the battery pack's state from macroscopic operating conditions to the microscopic material level, and integrate them into a unified data structure.

[0023] At the data acquisition level, the system collects information in parallel from two different dimensions through its data acquisition module.

[0024] The first type is macroscopic operating condition data, which is usually provided by the battery management system (BMS) built into the battery pack. This data mainly describes the external operating characteristics of the battery pack as a whole or a system composed of multiple sub-modules. Specifically, it may include: the total voltage and total current of the battery pack, the state of charge (SOC) calculated based on the energy integral or open-circuit voltage method, the voltage sequence of each individual cell that makes up the battery pack, and the temperature sequence of multiple temperature probes deployed in different locations inside the battery pack.

[0025] Secondly, there is the microscopic physical field data. To capture early degradation information that is difficult to reflect in traditional BMS data and is directly related to the aging mechanism of battery internal materials and interfaces, this system preferably deploys enhanced sensing units. These sensing units are used to acquire physical signals at the microscopic level, and may specifically include: Acoustic emission (AE) sensors are used to monitor the volume changes of electrode materials caused by lithium ion insertion and extraction during charge-discharge cycles, thereby capturing the acoustic signals released by the generation and propagation of microcracks inside the material.

[0026] Electrochemical impedance spectroscopy (EIS) analyzers periodically apply a small AC excitation signal to the battery and measure its response to probe the kinetic processes at the electrode-electrolyte interface, thereby quantifying key aging indicators such as the thickening of the solid electrolyte interphase (SEI) film or the increase in charge transfer resistance.

[0027] By using stress or strain sensors, the mechanical stress and strain generated in the electrode material during lithium insertion and extraction are directly measured, providing direct experimental input for the mechanical field analysis in the subsequent multiphysics coupling model.

[0028] After acquiring the aforementioned multi-source heterogeneous raw data streams, the system will preprocess them. This preprocessing includes necessary denoising and filtering operations on the signals to eliminate interference from environmental noise and measurement errors. Subsequently, the data of different physical quantities are normalized to map them into a unified numerical range, facilitating subsequent multi-source data fusion calculations and model solving.

[0029] After preprocessing, the crucial step is spatiotemporal alignment. Because the sampling frequencies of the data from these different sources vary significantly, and the physical sensors are deployed in different locations, it is essential to unify them into a common reference frame.

[0030] In the time dimension, the system adopts a unified clock synchronization protocol, mapping all data streams to a common, discretized time axis. For low-frequency data, interpolation can be used to generate data at the corresponding time points; for high-frequency data, downsampling or feature extraction can be used for alignment.

[0031] In terms of spatial dimension, the spatial location information of each sensor is also accurately recorded and correlated with its corresponding data stream. This spatial location information provides a basis for subsequent steps such as constructing the finite element mesh and applying boundary conditions or loads at specific spatial nodes.

[0032] Finally, after the above processing, all the data is integrated into a single synchronized time-series dataset, denoted as . At any discrete timestamp A data point in this dataset can be represented by a high-dimensional state vector, with the following specific form: ; In this expression, This is the current timestamp; It is a vector that includes the voltages of all individual battery cells; The scalar represents the total current. It is a vector containing the readings from all temperature measurement points; This represents the acoustic emission signal or its characteristics acquired at that moment; Electrochemical impedance spectroscopy data representing that moment; The stress tensor or its components represent the critical location.

[0033] This synchronized time series dataset As a complete and consistent data foundation, it constitutes a high-dimensional state snapshot of the battery pack at a specific moment, and is passed to the subsequent heterogeneous algebra fusion engine construction and multiphysics collaborative solution steps, providing them with the necessary input information.

[0034] Building a heterogeneous algebra fusion engine: The core objective of this step is to construct an abstract mathematical object that serves as the "meta-controller" for subsequent computational modules—a heterogeneous algebraic fusion engine. This engine is not a module that directly performs numerical calculations, but rather a foundational framework that plays a coordinating and controlling role. Its construction aims to address the problems of poor information flow and weak internal correlation between different model modules (such as physical simulation, feature extraction, and parameter calibration) in traditional methods. By establishing a unified mathematical structure, it provides globally consistent constraints and guidance for subsequent multi-task collaborative computation.

[0035] The construction of this heterogeneous algebraic fusion engine is not a simple assembly of modules, but rather a deep fusion of the intrinsic structural information from three different computational modules in subsequent steps at an abstract algebraic level. Specifically, the engine's construction process integrates the following three aspects of information: First, the structural information comes from the multiphysics coupling model solver itself in subsequent step S3. In a preferred embodiment, the solver takes the form of a tensor network. Therefore, this step first extracts the topology of the tensor network solver, denoted as […]. More specifically, to describe all possible structural transformations (symmetric permutations of internal indices) that do not alter the physical nature of the solver during computation, this step introduces the topology. The automorphism group of is denoted as This algebraic group It accurately characterizes the inherent symmetry of the solver's computational path and all possible equivalent transformations, forming the first structural pillar of the fusion engine.

[0036] Furthermore, to incorporate the actual physical degradation state inside the battery into the engine, this step introduces physical feature information derived from the weak degradation feature extraction module in subsequent step S3. In a preferred embodiment, this feature extraction module constructs an equivalent Hamiltonian. To simulate the microscopic electrochemical environment of a battery. The intrinsic energy spectrum of this Hamiltonian is denoted as... Its structure (position of energy levels, degeneracy, and splitting) directly reflects the subtle degradation characteristics inside the battery. Therefore, this step involves analyzing the intrinsic energy spectrum. As a "fingerprint" information that characterizes the physical state of the system, it constitutes the second structural pillar of the fusion engine.

[0037] Meanwhile, to address manufacturing variations and parameter drift issues among individual batteries, this step introduces parameter space geometric information derived from the model parameter correction module in subsequent step S3. In a preferred embodiment, this correction module constructs a standardized parameter manifold by analyzing historical data from the entire battery fleet. The topological structure of a manifold is its intrinsic geometric property that does not change with coordinate transformations. This step uses homology groups from algebraic topology, denoted as […]. This is used to quantify these topological invariants (the number and dimension of "loops" or "cavities" existing in the parameter space). The homology group... It captures the universal parameter evolution laws that all batteries follow, forming the third structural pillar of the fusion engine.

[0038] Having obtained the three algebraic objects that respectively describe computational structure, physical characteristics, and statistical laws, this step integrates them into a unified mathematical framework—the heterogeneous algebraic fusion engine—through specific algebraic operations. Its mathematical construction is as follows: ; In this expression, the symbol Represents the tensor product operation, which constructs a higher-dimensional algebraic space capable of simultaneously carrying information about the solver's symmetry and the physical characteristic spectrum structure. (Symbol) The intersection operation, which preserves the structure, is used to find substructures that are compatible with the topological invariants of the parameter space and can coexist in the aforementioned fusion space.

[0039] The resulting heterogeneous algebra fusion engine It is not a numerical output in itself, but rather a structured set of constraints or a set of generation rules. In subsequent steps, this engine... This will generate a set of globally consistent collaborative regulation parameters. These parameters will be used to configure the computational accuracy of the multiphysics solver in real time and collaboratively (e.g., dynamically setting the tensor rank), guide the search direction of the weak feature extraction module, and provide prior geometric constraints for the parameter correction module. In this way, the fusion engine built in this step ensures that subsequent computational tasks are performed within a unified framework of mutual awareness and constraint, rather than a series of isolated operations.

[0040] Engine-driven multiphysics collaborative solution and state analysis: This step is the core computational phase of the system. It receives the synchronous dataset provided in the preceding steps and integrates it with the constructed heterogeneous algebraic fusion engine. Under unified control, three interrelated computational tasks are executed in parallel in a coordinated manner: solving dynamic tensor networks, extracting quantum-derived features, and topologically correcting model parameters. These three tasks are not executed independently, either sequentially or in parallel, but rather within the fusion engine. Under the generated global constraints, information exchange and collaborative evolution take place.

[0041] First, regarding the solution of the multiphysics coupling model, this embodiment employs a dynamic tensor network solver. This process begins by establishing a set of partial differential equations (PDEs) describing the internal physicochemical processes of the battery, which couple at least the electrochemical, thermodynamic, and mechanical fields. Using numerical discretization techniques such as the finite element method (FEM) or the finite difference method (FDM), this continuous PDE system is transformed into a large-scale, nonlinear, high-dimensional algebraic equation system. The state variables of this equation system at all spatially discrete nodes and time steps are uniformly organized into a high-order state tensor, denoted as [equation missing]. .

[0042] To overcome the "curse of dimensionality" problem encountered when directly solving this high-order tensor, this embodiment employs the Tensor Train (TT) decomposition method to decompose the high-order tensor. It can be approximated as the product of a series of low-order core tensors. Its mathematical form is as follows: ; In this expression, It is a third- or fourth-order core tensor. The virtual index that connects different core tensors, its dimension This is referred to as the tensor rank or TT rank. A key aspect of this embodiment is the TT rank. The value is not a pre-set fixed value, but rather a value determined by the heterogeneous algebra fusion engine. Dynamic adjustment is implemented. Engine Through a mapping function Generate a real-time rank configuration scheme: When subsequent feature extraction modules detect abnormal signals, this information will be updated to the engine. The engine then instructs the solver to locally increase the rank of TT in the physical region or state dimension related to the anomalous signal, thereby solving the problem with higher accuracy and achieving adaptive and on-demand allocation of computing resources.

[0043] Secondly, regarding the extraction of weak degradation features, this embodiment uses a quantum-derived feature extractor.

[0044] This task aims to extract weak signals characterizing early irreversible aging of batteries from the microscopic physical field data (AE and EIS signals) obtained in the aforementioned steps. These signals are difficult to identify using traditional signal processing methods. The approach involves analogizing the electrochemical environment inside the battery to a quantum system. An ideal battery in a healthy state is considered the ground state of this quantum system, possessing a known and stable equivalent Hamiltonian. .

[0045] When a battery undergoes minor aging, such as atomic-level thickening of the SEI film or the appearance of microcracks in the active material, these changes are considered as a small perturbation potential to an ideal system, denoted as . This perturbation potential is constructed in real time based on the microscopic physical field data obtained from the aforementioned steps. Therefore, the true state of the battery at any given moment is determined by a time-varying equivalent Hamiltonian. To describe: ; By solving the time-varying Schrödinger equation associated with this Hamiltonian Its intrinsic energy spectrum can be analyzed. The changes. Compared to the ground state energy spectrum. The phenomena of energy level shifts, degeneracy cancellation, or band splitting observed in the actual energy spectrum are identified as weak characteristic signals of early battery degradation. These extracted spectral features serve as "fingerprints" of the physical state and are fed back into the aforementioned engine to update the heterogeneous algebraic fusion engine. The state.

[0046] Secondly, regarding the correction of model parameters, this embodiment uses a topology data analysis corrector to achieve this.

[0047] This task aims to eliminate parameter drift between individual batteries caused by differences in manufacturing processes or varying usage histories, thereby improving the model's generalization ability. This is achieved by first collecting historical operational data from a large number of different batteries (i.e., a battery fleet), particularly key aging-related parameters such as internal resistance, capacity, and EIS characteristic parameters. These parameters constitute a point cloud dataset in a high-dimensional parameter space. .

[0048] This embodiment uses the Persistent Homology algorithm to analyze the point cloud. The algorithm identifies the inherent geometric structure of the point cloud data. It can recognize stable topological features at different scales, including connected branches, loops, and cavities. These stable topological features, unaffected by specific numerical perturbations, are used to construct a standardized, universal parametric manifold. Its topological invariants are determined by its homology group. To describe.

[0049] For a new individual battery that needs to be predicted, its initial measurement parameter vector This will be achieved through a projection mapping. Mapped to this standard parametric manifold This yields a corrected parameter vector. Through this operation, the parameters of individual batteries are "pulled back" to the common evolutionary path defined by the data of the entire battery fleet, thereby eliminating non-essential individual differences. The topological information of this manifold is also fed back to the aforementioned steps to update the heterogeneous algebraic fusion engine. .

[0050] In summary, the three computational tasks in this step are not performed in isolation, but rather through a heterogeneous algebra fusion engine. They are tightly coupled together to form a closed loop of bidirectional information flow: the results of feature extraction and parameter correction are used to update the state of the engine, and the engine, in turn, dynamically and collaboratively regulates the solution process of the physical model based on the updated state, thereby achieving a comprehensive, accurate and efficient analysis of the internal state of the battery.

[0051] Health status assessment and remaining lifespan prediction: This step is the output stage of the entire algorithm process. Its core task is to transform the high-dimensional, multi-physics field variables that characterize the microscopic physical and chemical state inside the battery, obtained from the previous steps, into macroscopic performance indicators that have direct application value for battery managers and users, namely State of Health (SOH) and Remaining Useful Life (RUL).

[0052] During State of Health (SOH) assessment, the system first utilizes the output results of the engine-driven multiphysics collaborative solution and state analysis described above. The multiphysics coupled model solver in the engine-driven multiphysics collaborative solution and state analysis, through the coordinated control of the heterogeneous algebraic fusion engine, ultimately outputs a series of key state variables describing the internal aging mechanism of the battery. These variables are not traditional externally measurable physical quantities, but rather internal parameters directly related to the electrochemical aging process, which may specifically include: Thickness of solid electrolyte interphase (SEI) film Its growth is one of the main reasons for capacity decay and increased internal resistance; The loss of active lithium ions due to SEI film growth, lithium dendrite precipitation, or dead lithium formation ; The amount of active material lost due to phase change, particle breakage, or detachment from the current collector in positive or negative electrode active materials.

[0053] After obtaining the current values ​​of these internal state variables, this step substitutes them into an explicit, electrochemical mechanism-based aging model to calculate the current... In a preferred embodiment, the health state is defined by capacity decay. It can be calculated using the following formula: ; In this expression, This represents the battery's initial full-charge capacity at the time of manufacture and is a known calibration parameter. It is the result obtained in real time from step S3 up to the current moment. The cumulative loss of active lithium. Similarly, the health state is defined by an increase in internal resistance. Alternatively, it can be obtained by comparing the increments of the calculated resistances (SEI film resistance, charge transfer resistance) with the initial internal resistance. In this way, The assessment is based directly on the quantification of internal physical aging mechanisms, rather than merely on statistical fitting of external data.

[0054] After completing the current state of health (SOH) assessment, this step will further predict the remaining useful life (RUL).

[0055] The core of RUL prediction lies in using a high-fidelity digital twin model, which has been calibrated and optimized with actual data, to perform forward extrapolation simulations of the battery's future aging trajectory. The process is as follows: The system first requires one or more possible future load profiles, denoted as... This operating condition profile can be a typical operating condition generated based on the user's historical usage habits, or it can be a profile of a specific future task specified by the user.

[0056] Subsequently, the system uses the current battery state as an initial condition in the operating load profile. Driven by this, the multiphysics coupling model described in the above steps is continuously run to simulate the aging process of the battery in the future.

[0057] During the simulation, the system continuously calculates the State of Health (SOH) value at every future moment. The simulation terminates when the SOH value first drops to a pre-set End-of-Life (EOL) threshold. This threshold is typically defined by the specific application scenario; for example, it might be defined when the battery capacity decays to 80% of its initial capacity. .

[0058] RUL is defined as starting from the current time. Time span until the simulation ends Its mathematical form can be rigorously expressed as: ; Here The symbol indicates the removal of the supremacy, ensuring that the shortest time for SOH to first reach the threshold is found.

[0059] Through the above methods, this step not only provides the battery's current health status but also, based on in-depth simulations of its internal physical mechanisms, offers information on its expected lifespan under specific future usage scenarios. These outputs will serve as the final decision-making basis for the user and will also be passed to subsequent steps for model self-optimization.

[0060] System self-optimization closed loop: This step aims to endow the entire digital twin system with the ability to continuously learn and adaptively evolve, ensuring that it can maintain the accuracy of predictions throughout the entire lifecycle of the battery pack and adapt to new aging patterns or changes in operating conditions that were not fully reflected in the initial model. This step establishes a top-level feedback and optimization loop.

[0061] This self-optimization process is typically triggered when the system acquires new, verifiable battery pack health status data, which can serve as a "ground truth." This measured data is denoted as... It can be obtained through standard laboratory capacity calibration tests, or measured using specialized diagnostic equipment during vehicle maintenance.

[0062] After obtaining new measured data Next, this step first initiates the error calculation process. The system compares the measured value with the model prediction value generated in step S4 at the same time point. A comparison is made to quantify the prediction bias of the current digital twin model. In a preferred embodiment, this prediction error... It is obtained by calculating the square of the Euclidean distance between the two, and its specific mathematical expression is as follows: ; This error value It objectively reflects the degree of difference between the current model state and the actual physical state.

[0063] The system then uses the prediction error As the objective loss function, a feedback optimization process for the model parameters is initiated. A core technical feature of this embodiment is that the goal of this optimization process is not to directly adjust the physical parameters of a single model (electrochemical model) in the above steps, but rather to optimize a higher-level heterogeneous algebraic fusion engine responsible for the coordinated regulation of the entire model system. Optimize.

[0064] Specifically, the optimization targets the steps described above used to generate the heterogeneous algebraic fusion engine. A set of core generation parameters, denoted as These parameters define the engine's structure and determine how it coordinates and controls lower-level modules based on input information (physical feature spectrum, parameter space topology).

[0065] The system employs a gradient-based optimization algorithm for this set of parameters. To update, first, calculate the loss function using the backpropagation algorithm. Regarding engine parameters gradient Given that the entire system involves complex dynamic processes described by partial differential equations, in a preferred implementation, the gradient can be efficiently solved using the adjoint method or automatic differentiation techniques.

[0066] After obtaining the gradient, the engine parameters are iteratively updated using gradient descent or its variants (Adam, RMSprop). The single-step update rule can be expressed as: In this expression, and These represent the engine parameters before and after the update, respectively. It is a positive scalar value called the learning rate, which is used to control the step size for each update.

[0067] By performing this update process, a meta-learning of the entire system is essentially being conducted. The optimization is not of a single physical parameter, but rather of the "rules" and "strategies" that coordinate multiple heterogeneous modules to work together. This optimization adjusts the engine's sensitivity to certain weak feature signals, or changes how it controls the accuracy (i.e., tensor rank) of the physics solver when it perceives a specific parameter space structure.

[0068] Ultimately, this self-optimizing closed loop makes the entire digital twin system a dynamic system that can continuously learn and evolve from a continuous stream of data. Whenever new measured data is input, the system will perform a self-examination and adjustment, thereby gradually reducing the deviation between the model and the real world, so that it can more accurately reflect the true state of the battery, even if the battery is undergoing a new and unexpected aging path.

[0069] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital twin system for predicting battery pack lifespan, characterized in that, include: The data acquisition module is used to acquire the operating data of the battery pack; Memory, used to store instructions; The processor, when executing the instructions, implements: A model system is established, consisting of a multiphysics coupled model solution module, a weak degradation feature extraction module, and a model parameter correction module; in, The multiphysics coupling model solving module is used to solve the set of coupled equations characterizing the internal electrochemical, thermal and mechanical states of the battery pack; The weak degradation feature extraction module is used to extract weak feature signals characterizing early battery aging from the operating data. The model parameter correction module is used to correct the parameters of the multiphysics coupling model based on the historical data of the battery pack. as well as, A heterogeneous algebraic fusion engine is constructed. The heterogeneous algebraic fusion engine is used to coordinate and control the solution process of the multiphysics coupling model solution module based on the weak feature signal and the parameter space topology determined by the model parameter correction module, and output the life prediction result of the battery pack.

2. The digital twin system for predicting battery pack lifespan according to claim 1, characterized in that, The multiphysics coupling model solving module is used to construct the coupling equation set into a higher-order state tensor and solve the higher-order state tensor using tensor network decomposition; the heterogeneous algebra fusion engine is specifically used to dynamically adjust the tensor rank of the tensor network decomposition according to the weak feature signal.

3. The digital twin system for predicting battery pack lifespan according to claim 1, characterized in that, The weak degradation feature extraction module is used to construct an equivalent Hamiltonian that characterizes the internal electrochemical environment of the battery, and to extract the weak feature signal by analyzing the eigenenergy spectrum of the equivalent Hamiltonian.

4. The digital twin system for predicting battery pack lifespan according to claim 1, characterized in that, The model parameter correction module is used to construct a parametric manifold from the historical data of the battery pack and use the topological invariants of the parametric manifold to correct the parameters of the multiphysics coupling model.

5. A digital twin system for predicting battery pack lifespan according to claim 4, characterized in that, The model parameter correction module is specifically used to analyze the historical data of the battery pack using a continuous homology algorithm to construct the parametric manifold and extract the topological invariants.

6. The digital twin system for predicting battery pack lifespan according to claim 4, characterized in that, The heterogeneous algebraic fusion engine achieves the coordinated control by constructing a unified algebraic framework that integrates the symmetry of the computational structure of the solution module, the spectral structure of the weak feature signal, and the topological structure of the parameter space.

7. A digital twin system for predicting battery pack lifespan according to claim 1, characterized in that, When the processor executes the instruction, it also implements: Based on the internal state variables output by the multiphysics coupling model solution module, the health status and remaining service life of the battery pack are calculated.

8. A digital twin system for predicting battery pack lifespan according to claim 7, characterized in that, When the processor executes the instruction, it also implements: The calculated value of the health status is compared with the measured value to obtain the prediction error; Based on the prediction error, the internal parameters of the heterogeneous algebra fusion engine are optimized in reverse.

9. A digital twin system for predicting battery pack lifespan according to claim 3, characterized in that, The operational data includes microscopic physical field data acquired from acoustic emission sensors, electrochemical impedance spectroscopy analyzers, or stress sensors, which is used to construct the equivalent Hamiltonian.

10. A digital twin system for predicting battery pack lifespan according to claim 1, characterized in that, The system also includes a status update module, which is used to update the virtual model of the battery pack digital twin in real time based on the life prediction results.

Images