A lithium battery negative electrode potential prediction method based on mechanism data fusion

Abstract

The present application relates to a kind of lithium battery negative electrode potential prediction method based on mechanism data fusion, including the construction of negative electrode multiple reaction coupling electrochemical model, the construction of multiple physical field coupling model, output final identification parameter, construct lithium precipitation pressure residual feature, output negative electrode potential prediction value.The lithium battery negative electrode potential prediction method based on mechanism data fusion, the accurate characterization of the evolution process of lithium precipitation is realized, provides accurate interface potential information for lithium precipitation critical state identification, simultaneously provides multiple source constraints for parameter identification and pressure feature extraction by multiple physical field coupling, overcome the defects of insufficient information fusion of existing model, effectively avoid parameter compensation and local optimal problem, by constructing lithium precipitation pressure residual feature provides input with clear physical meaning for prediction model, without a large number of measured data online prediction, give consideration to local dynamic feature capture and long-time dependence learning ability, realize the unification of physical interpretability and online prediction ability.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery state prediction technology, specifically a method for predicting the negative electrode potential of lithium batteries based on mechanism data fusion. Background Technology

[0002] Lithium-ion batteries have been widely used in electric vehicles, energy storage systems, and consumer electronics due to their advantages such as high energy density, long cycle life, and low self-discharge rate. When batteries are operated under extreme charging conditions such as low temperature, high rate, or high state of charge, the polarization of the negative electrode is significantly enhanced, and the negative electrode potential continues to decrease. When the negative electrode potential is lower than the lithium metal / lithium ion equilibrium potential, a lithium metal deposition reaction will be triggered. Lithium metal deposition not only directly leads to the loss of active lithium and the decay of battery capacity, but also further induces a series of side reactions such as dead lithium generation, SEI film thickening, and increased interface impedance. In severe cases, it may cause safety accidents such as internal short circuits and thermal runaway.

[0003] Currently, there are three main methods for predicting the state of lithium-ion batteries: The first method is based on experimental detection, which uses disassembly and observation, reference electrode or offline impedance analysis to detect lithium plating, and can provide more accurate information on the negative electrode interface. The second approach is a modeling method based on electrochemical mechanisms. This method simulates the internal processes of the battery by constructing sub-models such as negative electrode lithium intercalation kinetics and SEI film growth, and uses a multi-physics coupling model to simulate and analyze the battery state. The third approach is a data-driven method, which builds a data-driven prediction model by collecting external measurable signals such as current, terminal voltage, and temperature, and combines machine learning algorithms such as neural networks to predict battery status. In practical applications, lithium plating risk assessment relies heavily on externally measurable signals such as terminal voltage, state of charge, and temperature, lacking direct characterization of the negative electrode interface potential. This makes it difficult to identify critical lithium plating states in a timely manner and fails to provide accurate interface information for lithium plating early warning. Secondly, existing electrochemical models do not adequately consider the coupling relationship and feedback mechanism between lithium metal deposition, reversible lithium stripping, dead lithium generation, and SEI film growth. They lack closed-loop feedback descriptions of the impact of by-reaction products on interfacial film resistance, effective reaction area, and negative electrode polarization, resulting in limited calculation accuracy of negative electrode polarization and interfacial impedance, making it difficult to accurately reflect the lithium plating evolution process. Furthermore, parameter identification methods often employ a single target of terminal voltage or a local optimization strategy, failing to establish a multi-objective constraint mechanism for terminal voltage, negative electrode potential, and electrochemical impedance. They also lack a two-stage identification framework of global search and local optimization, which easily leads to problems such as parameter compensation, non-unique results, and getting trapped in local optima, making it difficult to simultaneously guarantee global search capability and local identification accuracy. Therefore, a lithium battery negative electrode potential prediction method based on mechanism data fusion is proposed to solve the above problems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a lithium battery negative electrode potential prediction method based on mechanism data fusion. It has the advantages of integrating mechanism constraints and data-driven approaches, possessing both physical interpretability and online prediction capabilities. It solves the problems in existing technologies, such as the difficulty in identifying the critical state of lithium plating, the lack of side reaction coupling and feedback mechanisms leading to limited accuracy in negative electrode polarization calculation, non-unique parameter identification results that are prone to getting trapped in local optima, and the difficulty in obtaining training samples for data-driven models and insufficient online prediction accuracy.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting the negative electrode potential of a lithium battery based on mechanism data fusion, comprising the following steps; S1. Obtain the geometric parameters, electrochemical parameters, and side reaction parameters of the battery, and construct a coupled electrochemical model of multiple side reactions at the negative electrode; S2. A multi-physics coupling model is constructed based on the energy conservation equation, the total volumetric strain of the battery negative electrode active material during the charging and discharging process, and the coupled electrochemical model of multiple side reactions of the negative electrode. S3. Distinguish the parameters to be identified in the multiphysics coupling model, construct a multi-objective optimization function, and then perform global and local two-stage search and optimization on the parameters to be identified, obtain the parameter identification results at each stage, and select the parameter identification result with the smallest comprehensive error as the final identified parameter. S4. Input the final identification parameters into the multiphysics coupling model established in S2, and output the reference pressure, negative electrode potential related data, and negative electrode potential data. Construct the lithium plating pressure residual characteristics based on the difference between the measured pressure and the reference pressure. S5. Using historical side reaction parameters and historical lithium plating pressure residual characteristics as inputs, an input sequence is constructed using a sliding time window, and a time-series prediction network is built. The negative electrode potential related data and lithium plating pressure residual characteristics are input into the time series prediction network, and the predicted value of the negative electrode potential is output.

[0006] Furthermore, a method for constructing a coupled electrochemical model of multiple side reactions at the negative electrode is introduced in step S1; Model parameters, model boundary conditions, and initial conditions are set based on the pre-processed battery geometric parameters, electrochemical parameters, and side reaction parameters. Based on battery geometric parameters and electrochemical parameters, the following models were constructed: local total reaction current shunting model of negative electrode, lithium metal deposition model triggered by negative electrode interface potential, coupled evolution model of reversible lithium metal, dead lithium and SEI film, coupled model between SEI film growth and lithium metal deposition, feedback model of by-reaction products on negative electrode interface film resistance, and feedback model of by-reaction products on effective reaction area of ​​negative electrode. Based on the above six models, as well as model parameters, model boundary conditions, and initial conditions, a coupled electrochemical model of multiple side reactions at the negative electrode is constructed.

[0007] Furthermore, the above six model construction methods are further introduced in step S1; Based on battery geometric parameters and electrochemical parameters, the current density of the negative electrode main lithium intercalation reaction, the current density of the SEI film growth side reaction, the current density of lithium metal deposition, and the current density of reversible lithium metal stripping are obtained, and a local total reaction current shunt model of the negative electrode is constructed to output the total current density. Based on battery geometric parameters and electrochemical parameters, the negative electrode solid phase potential, negative electrode liquid electrolyte potential, lithium metal or lithium ion equilibrium potential, and interface film resistance are obtained. Combined with the total current density, a lithium metal deposition model triggered by the negative electrode interface potential is constructed, and the lithium metal deposition overpotential is output. Based on battery geometric parameters and electrochemical parameters, the reversible lithium capacity and the rate constant of reversible lithium to dead lithium conversion are obtained. Combined with lithium deposition current density and reversible lithium stripping current density, a coupled evolution model of reversible lithium, dead lithium and SEI film is constructed to output the reversible lithium capacity and the reversible lithium change rate. The molar amount of dead lithium is calculated based on the rate of change of reversible lithium metal and the rate constant of the conversion of reversible lithium metal to dead lithium, and is used as the cumulative amount of by-reaction products. Based on battery geometric parameters and electrochemical parameters, the exchange current density of SEI film growth side reactions, the SEI film growth side reaction transfer coefficient, the SEI film growth side reaction overpotential, the initial value of SEI film thickness, the promoting coefficient of reversible lithium metal on secondary SEI formation, the promoting coefficient of dead lithium on secondary SEI formation, and the reference SEI film thickness are obtained. Combined with the reversible lithium metal capacity and the molar amount of dead lithium, a coupling model between SEI film growth and lithium metal deposition is constructed, and the SEI film growth side reaction current density and the current SEI film thickness are output. Based on battery geometric parameters and electrochemical parameters, the equivalent ionic conductivity of the SEI film, the contribution coefficient of reversible lithium metal to the interface impedance, and the contribution coefficient of dead lithium to the interface impedance are obtained. Combined with the current SEI film thickness, reversible lithium metal capacity, and dead lithium molar amount, a feedback model of the side reaction products on the negative electrode interface film resistance is constructed, and the negative electrode interface film resistance is output. Based on battery geometric parameters and electrochemical parameters, the initial reaction specific surface area of ​​the negative electrode, the shielding coefficient of the SEI film on the active reaction area of ​​the negative electrode, the shielding coefficient of reversible lithium metal on the active reaction area of ​​the negative electrode, and the shielding coefficient of dead lithium on the active reaction area of ​​the negative electrode are obtained. Combined with the current SEI film thickness, reversible lithium metal capacity, and dead lithium molar amount, a feedback model of the side reaction products on the effective reaction area of ​​the negative electrode is constructed, and the effective reaction specific surface area of ​​the negative electrode is output. The negative electrode interface film resistance and the effective reaction surface area of ​​the negative electrode are fed back into the calculation of the lithium metal deposition overpotential to form a closed-loop feedback of the side reaction products to the polarization of the negative electrode.

[0008] Furthermore, a multiphysics coupling model construction method is introduced in step S2; An electrochemical-thermal coupling sub-model was constructed based on battery geometric parameters, electrochemical parameters, and energy conservation equations; An electrochemical-mechanical coupling sub-model was constructed based on battery geometric parameters, electrochemical parameters, and the total volumetric strain of the battery negative electrode active material during the charging and discharging process. In the multi-physics coupling model, the steady-state operating point is solved under a given operating condition by the negative electrode multi-side reaction coupled electrochemical model. A small alternating perturbation signal is applied at the steady-state operating point and the voltage response at different frequencies is obtained. The complex impedance spectrum of the battery is obtained based on the ratio of the voltage response to the perturbation signal, and an electrochemical-impedance coupling sub-model is established. A multiphysics coupling model is constructed by fusing the electrochemical-thermal coupling sub-model, the electrochemical-mechanical coupling sub-model, and the electrochemical-impedance coupling sub-model.

[0009] Furthermore, step S2 further introduces methods for constructing electrochemical-thermal coupler models and electrochemical-mechanical coupler models; Based on the battery's geometric parameters and electrochemical parameters, the battery's thermal conductivity, battery density, specific heat capacity, reaction heat generation rate, polarization heat generation rate, ohmic heat generation rate, and convective heat transfer coefficient are obtained. An electrochemical-thermal coupling sub-model is constructed by combining the multi-reaction coupling electrochemical model of the negative electrode to output the battery temperature field. The model parameters of the coupled electrochemical model of multiple side reactions of the negative electrode are corrected based on the battery temperature field, and the heat generated by the side reactions in the coupled electrochemical model of multiple side reactions of the negative electrode is fed back to the electrochemical-thermal coupling sub-model as a heat source to update the battery temperature field. Based on battery geometric parameters and electrochemical parameters, elastic strain, chemical strain caused by lithium ion insertion or extraction, thermal strain caused by temperature change, strain caused by lithium metal deposition, irreversible strain caused by dead lithium accumulation, and strain caused by SEI film growth are obtained. An electrochemical-mechanical coupled sub-model is constructed by combining the battery temperature field and the accumulation of by-reaction products to output the total volumetric strain. Then, the external pressure response of the battery is calculated based on the total volumetric strain.

[0010] Furthermore, a multiphysics coupling model parameter identification method is introduced in step S3; The sensitivity of the parameters to be identified in the multiphysics coupling model is distinguished, the sensitivity distinction results are obtained, the parameters to be identified are selected based on the sensitivity distinction results, and a multi-objective optimization function is constructed with the terminal voltage error, negative electrode potential error and electrochemical impedance error as objectives. A multi-objective genetic algorithm is used to perform a global search of the parameters to be identified: An initial parameter population is generated within the range of values ​​of each parameter to be identified. Each group of parameters is input into a multiphysics coupling model to calculate the objective function value. Candidate parameters are filtered according to Pareto non-dominated sorting and crowding distance. A new generation of population is generated through selection, crossover, and mutation. The process is repeated iteratively until the convergence condition is met, and the Pareto non-dominated solution set is obtained and used as the parameter identification result in the global search phase. The initial parameters are selected based on the Pareto non-dominated solution set, which satisfies the physical constraints and the comprehensive error meets the preset conditions. A trust region is established with the initial parameters as the center, and a Gaussian process surrogate model is constructed based on the existing parameters and the corresponding objective function values. Candidate parameters are generated within the trust region using a Gaussian process surrogate model. Then, the objective function value of the candidate parameters is calculated using a multiphysics coupling model. The trust region is adjusted according to the candidate parameters, and the iteration is repeated until the stopping condition is met to obtain the parameter identification results in the local optimization stage. Based on the parameter identification results from both the global search and local optimization phases, the parameter set that minimizes the combined terminal voltage error, negative electrode potential error, and electrochemical impedance error while satisfying physical constraints is selected as the final identification parameters.

[0011] Furthermore, a lithium plating pressure residual feature extraction method is introduced in step S4; Replace the original parameter values ​​in the multiphysics coupling model with the final identified parameters to obtain the multiphysics coupling model after parameter calibration. The baseline pressure under the condition of no lithium metal deposition was calculated based on the multiphysics coupling model after parameter calibration, and the negative electrode potential related data was generated. The residual characteristics of lithium plating pressure are constructed based on the difference between the measured pressure and the reference pressure. When the difference between the measured pressure and the reference pressure is greater than the preset tolerance threshold, the difference is used as the residual characteristic of lithium plating pressure; otherwise, the residual characteristic of lithium plating pressure is zero.

[0012] Furthermore, step S5 introduces a method for constructing and training a temporal prediction network; The negative electrode potential correlation data and lithium plating pressure residual characteristics are used as input features; The negative electrode potential data generated by the multiphysics coupling model under the corresponding operating conditions is used as training labels and aligned with the input features at the corresponding time. A sliding time window is used to construct the input time series from the input features of multiple consecutive sampling times; A temporal prediction network is constructed, which includes a temporal convolution module and a recurrent memory module. The temporal convolution module contains residual connections. The input time series is fed into the temporal prediction network. After the temporal convolution module extracts local dynamic features and multi-scale temporal features, it is fed into the recurrent memory module to learn the long-term dependencies between temporal features. Finally, the negative electrode potential prediction value is output through the fully connected layer. The predicted negative electrode potential is compared with the training label, and a loss function is constructed based on the error between the two. The time series prediction network is trained with the goal of minimizing the loss function.

[0013] Compared with existing technologies, this invention provides a method for predicting the negative electrode potential of lithium batteries based on mechanism data fusion, which has the following beneficial effects: 1. This lithium battery negative electrode potential prediction method based on mechanism data fusion constructs a coupled electrochemical model of multiple side reactions of the negative electrode and establishes a closed-loop feedback mechanism for side reaction products. It realizes the accurate characterization of the coupling evolution and interactive feedback between lithium metal deposition, reversible lithium stripping, dead lithium generation and SEI film growth, and provides accurate negative electrode interface potential information for the identification of lithium plating critical state, effectively improving the calculation accuracy of negative electrode polarization and interface impedance. 2. This lithium battery negative electrode potential prediction method based on mechanism data fusion constructs an electrochemical-thermal-mechanical-impedance multi-physics coupling model, which provides multi-source physical field constraints for parameter identification and pressure feature extraction. Among them, the thermal-electric bidirectional coupling ensures the model accuracy over a wide temperature range, the mechanical coupling enables the external pressure response to reflect the cumulative expansion effect of by-reaction products, and the impedance coupling provides frequency domain constraints for parameter identification, overcoming the shortcomings of insufficient multi-source physical information fusion and single constraint in existing models. 3. This lithium battery negative electrode potential prediction method based on mechanism data fusion effectively avoids the problems of parameter compensation, non-unique identification results, and getting trapped in local optima by constructing a multi-objective constraint mechanism of terminal voltage, negative electrode potential, and electrochemical impedance, and adopting a two-stage parameter identification method that combines global search and local optimization. At the same time, it ensures both global search capability and local identification accuracy. Furthermore, by constructing lithium plating pressure residual features, these features can directly characterize the additional expansion caused by lithium metal deposition and the accumulation of by-reaction products, providing high-value input features with clear physical interpretability for subsequent prediction models, and overcoming the defect that lithium plating features in pressure signals are submerged by normal expansion. 4. This lithium battery negative electrode potential prediction method based on mechanism data fusion uses negative electrode potential data generated by a multi-physics coupling model as training labels and negative electrode potential related data and lithium plating pressure residual features as inputs to construct a temporal prediction network combining temporal convolution and recurrent memory. This enables online prediction without the need for a large number of measured negative electrode potential labels, effectively solving the problem of difficulty in obtaining training samples for existing data-driven models. At the same time, it takes into account both local dynamic feature capture and long-term dependent learning capabilities, improving the generalization ability under operating conditions and online computing efficiency while ensuring prediction accuracy, and achieving a unity of physical interpretability and online prediction capability. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a voltage curve fitting diagram between the experimental results and the model simulation of this invention; Figure 3 This is a graph showing the fitting of the experimental results and the temperature curves from the model simulation of this invention.

[0015] Figure 4 This is a graph showing the fitting of the experimental results and the pressure curves from the model simulation of this invention. Figure 5 The impedance spectrum is the result of simulation and experimental data of the model of this invention; Figure 6 This is a graph showing the fitting of the negative electrode potential curve between the experimental results and the model simulation of this invention. Figure 7 This is a prediction diagram of the negative electrode potential of the lithium-ion battery using the TCN-LSTM of this invention; Figure 8 This is a model prediction diagram for the present invention without pressure feature input. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1: Please refer to Figure 1-8 The lithium battery negative electrode potential prediction method based on mechanism data fusion in this embodiment includes the following steps; S1. Obtain the geometric parameters, electrochemical parameters, and side reaction parameters of the battery, and construct a coupled electrochemical model of multiple side reactions at the negative electrode; S2. A multi-physics coupling model is constructed based on the energy conservation equation, the total volumetric strain of the battery negative electrode active material during the charging and discharging process, and the coupled electrochemical model of multiple side reactions of the negative electrode. S3. Distinguish the parameters to be identified in the multiphysics coupling model, construct a multi-objective optimization function, and then perform global and local two-stage search and optimization on the parameters to be identified, obtain the parameter identification results at each stage, and select the parameter identification result with the smallest comprehensive error as the final identified parameter. S4. Input the final identification parameters into the multiphysics coupling model established in S2, and output the reference pressure, negative electrode potential related data, and negative electrode potential data. Construct the lithium plating pressure residual characteristics based on the difference between the measured pressure and the reference pressure. S5. Using historical side reaction parameters and historical lithium plating pressure residual characteristics as inputs, an input sequence is constructed using a sliding time window, and a time-series prediction network is built. The negative electrode potential related data and lithium plating pressure residual characteristics are input into the time series prediction network, and the predicted value of the negative electrode potential is output.

[0018] It should be noted that a complete technical framework of "mechanism modeling → parameter identification → pressure feature extraction → time series prediction" has been constructed. The computational characterization of lithium plating evolution is achieved through a side-reaction electrochemical model, overcoming the shortcomings of traditional methods that rely on external measurable signals and cannot directly characterize the negative electrode interface. Multi-physics coupling models provide multi-source constraints, overcoming the lack of information in pure electrochemical models. Global and local two-stage optimization overcomes the tendency of single optimization to get trapped in local optima. Pressure residual characteristics enhance the observability of abnormal lithium plating expansion. Mechanistic models generate training labels to replace large amounts of measured data, overcoming the difficulty of obtaining training samples in data-driven methods. This solves the problem of existing technologies lacking a negative electrode potential prediction method that integrates mechanistic constraints and data-driven approaches, while also possessing physical interpretability and online prediction capabilities.

[0019] In particular, step S1 further introduces a method for constructing a coupled electrochemical model of multiple side reactions at the negative electrode; Model parameters, model boundary conditions, and initial conditions are set based on the pre-processed battery geometric parameters, electrochemical parameters, and side reaction parameters. Based on battery geometric parameters and electrochemical parameters, the following models were constructed: local total reaction current shunting model of negative electrode, lithium metal deposition model triggered by negative electrode interface potential, coupled evolution model of reversible lithium metal, dead lithium and SEI film, coupled model between SEI film growth and lithium metal deposition, feedback model of by-reaction products on negative electrode interface film resistance, and feedback model of by-reaction products on effective reaction area of ​​negative electrode. Based on the above six models, as well as model parameters, model boundary conditions, and initial conditions, a coupled electrochemical model of multiple side reactions at the negative electrode is constructed.

[0020] It should be noted that a closed-loop coupling structure of "deposition → product accumulation → interface deterioration → promoted deposition" is established, forming a closed-loop feedback of by-reaction products to negative electrode polarization. This solves the technical problem that existing electrochemical models have not established the coupling relationship between lithium metal deposition and SEI growth, nor have they formed a feedback mechanism for by-reaction products on interfacial film resistance and effective reaction area, which limits the accuracy of negative electrode polarization calculation.

[0021] In addition, the above six model construction methods are further introduced in step S1; Based on battery geometric parameters and electrochemical parameters, the current density of the negative electrode main lithium intercalation reaction, the current density of the SEI film growth side reaction, the current density of lithium metal deposition, and the current density of reversible lithium metal stripping are obtained, and a local total reaction current shunt model of the negative electrode is constructed to output the total current density. It should be noted that the total Faraday reaction current density at the negative electrode is... satisfy: ; In the formula, The current density for the lithium intercalation reaction at the negative electrode. This represents the current density of the side reactions during SEI film growth. This represents the lithium metal deposition current density. This represents the reversible lithium metal stripping current density. Based on battery geometric parameters and electrochemical parameters, the negative electrode solid phase potential, negative electrode liquid electrolyte potential, lithium metal or lithium ion equilibrium potential, and interface film resistance are obtained. Combined with the total current density, a lithium metal deposition model triggered by the negative electrode interface potential is constructed, and the lithium metal deposition overpotential is output. It should be noted that the lithium metal deposition model triggered by the negative electrode interface potential involves lithium metal deposition overpotential. satisfy: ; In the formula, This is the overpotential for lithium metal deposition. The negative solid-phase potential, The potential of the negative electrode liquid phase electrolyte, The equilibrium potential of lithium metal / lithium ion. For interface film resistance, This represents the total current density. when When the current density is below a preset deposition threshold, the lithium metal deposition reaction is activated, and the lithium metal deposition current density... satisfy: ; In the formula, The effective exchange current density for lithium metal deposition reaction, The transfer coefficient of the lithium metal deposition reaction. This is the lithium deposition activation function; the lithium deposition activation function is used to characterize the process by which the lithium metal deposition reaction gradually starts as the negative electrode interface potential decreases; when When the lithium deposition threshold is not lower than the preset deposition threshold, the lithium plating activation function is zero. Preferably, It can be represented as: ; In the formula, The slope coefficient of the lithium plating activation function. This is the lithium plating trigger threshold; when Below hour, The value gradually increases, thus characterizing the change process of the lithium plating reaction from weak to strong; Based on battery geometric parameters and electrochemical parameters, the reversible lithium capacity and the rate constant of reversible lithium to dead lithium conversion are obtained. Combined with lithium deposition current density and reversible lithium stripping current density, a coupled evolution model of reversible lithium, dead lithium and SEI film is constructed to output the reversible lithium capacity and the reversible lithium change rate. The molar amount of dead lithium is calculated based on the rate of change of reversible lithium metal and the rate constant of the conversion of reversible lithium metal to dead lithium, and is used as the cumulative amount of by-reaction products. It should be noted that reversible metallic lithium can be stripped off again during subsequent discharge or the recovery of the negative electrode potential; dead lithium no longer participates in the reversible electrochemical reaction, which will cause capacity loss, increased interfacial impedance and a decrease in the effective reaction area of ​​the negative electrode. Reversible metallic lithium molar amount satisfy: ; In the formula, This represents the lithium metal deposition current density. This represents the reversible lithium metal stripping current density. is the rate constant for the reversible conversion of metallic lithium to dead lithium.

[0022] Furthermore, the molar amount of dead lithium satisfies: In the formula, The molar amount of dead lithium per unit area of ​​the negative electrode; As reversible metallic lithium gradually loses its electrochemical activity and transforms into dead lithium, Over time, it accumulates and, as an irreversible byproduct, further affects the state of the negative electrode interface. Reversible lithium metal stripping current density satisfy: ; In the formula, The rate coefficient for the reversible lithium metal stripping reaction is given. The transfer coefficient of the reversible lithium metal stripping reaction. The activation function for lithium stripping; Lithium stripping activation function Used to characterize the degree of reversible lithium metal stripping after the negative electrode interface potential recovers. By establishing the transformation relationship between reversible lithium metal, dead lithium, and stripping reaction, the recoverable and unrecoverable parts after lithium metal deposition can be described, thereby improving the model's ability to characterize capacity loss, film resistance growth, and negative electrode potential evolution. Based on battery geometric parameters and electrochemical parameters, the exchange current density of SEI film growth side reactions, the SEI film growth side reaction transfer coefficient, the SEI film growth side reaction overpotential, the initial value of SEI film thickness, the promoting coefficient of reversible lithium metal on secondary SEI formation, the promoting coefficient of dead lithium on secondary SEI formation, and the reference SEI film thickness are obtained. Combined with the reversible lithium metal capacity and the molar amount of dead lithium, a coupling model between SEI film growth and lithium metal deposition is constructed, and the SEI film growth side reaction current density and the current SEI film thickness are output. It should be noted that the coupling model between SEI film growth and lithium metal deposition is established, and the side reaction current density during SEI film growth is also considered. satisfy: ; In the formula, The exchange current density for the side reactions during SEI film growth. The side reaction transport coefficient for SEI film growth is given. This is the overpotential for the side reaction during SEI film growth. For the SEI film thickness, For reference SEI film thickness, The promoting factor for secondary SEI formation by reversible lithium metal is... The promoting factor of dead lithium on secondary SEI formation SEI film thickness satisfy: ; In the formula, For the molar mass of the SEI membrane, For the SEI film density, The coefficient representing the secondary SEI coverage induced by dead lithium; Therefore, the growth of SEI film is no longer determined solely by the overpotential of traditional side reactions, but is also influenced by the combined effects of reversible metallic lithium and dead lithium accumulation. After lithium metal deposition, secondary SEI formation is induced, and dead lithium accumulation also promotes the thickening of the interface film, thus forming a competitive coupling relationship between SEI film growth, lithium plating reaction and dead lithium formation. Based on battery geometric parameters and electrochemical parameters, the equivalent ionic conductivity of the SEI film, the contribution coefficient of reversible lithium metal to the interface impedance, and the contribution coefficient of dead lithium to the interface impedance are obtained. Combined with the current SEI film thickness, reversible lithium metal capacity, and dead lithium molar amount, a feedback model of the side reaction products on the negative electrode interface film resistance is constructed, and the negative electrode interface film resistance is output. It should be noted that the feedback model for the effect of by-reaction products on the negative electrode interfacial film resistance satisfies the following: ; In the formula, The equivalent ionic conductivity of the SEI membrane. This represents the contribution coefficient of reversible lithium metal to the interface impedance. The contribution coefficient of dead lithium to the interface impedance; Based on battery geometric parameters and electrochemical parameters, the initial reaction specific surface area of ​​the negative electrode, the shielding coefficient of the SEI film on the active reaction area of ​​the negative electrode, the shielding coefficient of reversible lithium metal on the active reaction area of ​​the negative electrode, and the shielding coefficient of dead lithium on the active reaction area of ​​the negative electrode are obtained. Combined with the current SEI film thickness, reversible lithium metal capacity, and dead lithium molar amount, a feedback model of the side reaction products on the effective reaction area of ​​the negative electrode is constructed, and the effective reaction specific surface area of ​​the negative electrode is output. The negative electrode interface film resistance and the effective reaction surface area of ​​the negative electrode are fed back into the calculation of the lithium metal deposition overpotential to form a closed-loop feedback of the side reaction products to the polarization of the negative electrode.

[0023] It should be noted that a feedback model for the effect of by-reaction products on the effective reaction area of ​​the negative electrode is established, and the effective reaction surface area of ​​the negative electrode is... satisfy: ; In the formula, The initial surface area of ​​the negative electrode reaction. This represents the shielding coefficient of the SEI film on the negative electrode active reaction area. is the shielding coefficient of reversible lithium metal on the active reaction area of ​​the negative electrode. The shielding coefficient of dead lithium on the active reaction area of ​​the negative electrode; A decrease in the negative electrode interface potential triggers a lithium metal deposition reaction. After lithium metal deposition, some of the lithium metal can be reversibly stripped off, while the other part is converted into dead lithium. Reversible lithium metal and dead lithium promote the formation of a secondary SEI film. The thickening of the SEI film, the coverage of reversible lithium metal, and the accumulation of dead lithium will jointly lead to an increase in the negative electrode interface film resistance and a decrease in the effective reaction area. The increase in film resistance and the decrease in the effective reaction area will further enhance the negative electrode polarization and reduce the negative electrode interface potential.

[0024] It should be noted that the current splitting model achieves quantitative allocation of total current; the deposition model uses overpotential and threshold to quantitatively identify the critical state of lithium deposition; the evolution model divides deposited lithium into two independent state variables, reversible and dead lithium, and establishes a transformation equation; the SEI coupling model describes the quantitative promotion of SEI film growth by lithium deposition products; the film resistance model establishes a linear superposition relationship between film resistance and SEI thickness, reversible lithium content, and dead lithium content; the area model establishes an exponential shading relationship of effective area; and it solves the specific technical problem of lack of quantitative modeling of each link in the side reaction coupling and feedback relationship.

[0025] Example 2: Please refer to Figure 1-8 Based on Example 1, a lithium battery negative electrode potential prediction method based on mechanism data fusion is proposed, and a multi-physics field coupling model construction method is further introduced in step S2. An electrochemical-thermal coupling sub-model was constructed based on battery geometric parameters, electrochemical parameters, and energy conservation equations; An electrochemical-mechanical coupling sub-model was constructed based on battery geometric parameters, electrochemical parameters, and the total volumetric strain of the battery negative electrode active material during the charging and discharging process. In the multi-physics coupling model, the steady-state operating point is solved under a given operating condition by the negative electrode multi-side reaction coupled electrochemical model. A small alternating perturbation signal is applied at the steady-state operating point and the voltage response at different frequencies is obtained. The complex impedance spectrum of the battery is obtained based on the ratio of the voltage response to the perturbation signal, and an electrochemical-impedance coupling sub-model is established. It should be noted that the calculation formula is as follows: ; ; ; In the formula, and These are the current and terminal voltage at the steady-state operating point, respectively. and Angular frequency The complex amplitude of the small-signal current disturbance and the complex amplitude of the terminal voltage response; This represents the real part of the complex impedance. This represents the imaginary part of the complex impedance. A multiphysics coupling model is constructed by fusing the electrochemical-thermal coupling sub-model, the electrochemical-mechanical coupling sub-model, and the electrochemical-impedance coupling sub-model.

[0026] It should be noted that electrochemical-thermal coupling enables temperature-based correction of reaction kinetic parameters, giving the model the ability to describe low-temperature polarization; electrochemical-mechanical coupling enables characterization of volumetric strain caused by the accumulation of by-reaction products, giving the model the ability to indirectly observe lithium plating through external pressure response; and electrochemical-impedance coupling provides frequency domain constraints for the identification of multi-objective parameters. All three sub-models use the state variables of by-reaction products as coupling driving parameters to realize the state constraints of the electrochemical model by multi-dimensional information of thermo-mechanical-impedance, thus solving the technical problem of insufficient fusion of multi-source physical information in existing models.

[0027] In particular, step S2 further introduces methods for constructing electrochemical-thermal coupler models and electrochemical-mechanical coupler models; Based on the battery's geometric parameters and electrochemical parameters, the battery's thermal conductivity, battery density, specific heat capacity, reaction heat generation rate, polarization heat generation rate, ohmic heat generation rate, and convective heat transfer coefficient are obtained. An electrochemical-thermal coupling sub-model is constructed by combining the multi-reaction coupling electrochemical model of the negative electrode to output the battery temperature field. The model parameters of the coupled electrochemical model of multiple side reactions of the negative electrode are corrected based on the battery temperature field, and the heat generated by the side reactions in the coupled electrochemical model of multiple side reactions of the negative electrode is fed back to the electrochemical-thermal coupling sub-model as a heat source to update the battery temperature field. It should be noted that the internal temperature field of the battery satisfies the energy conservation equation: ; in, The thermal conductivity of the battery, , These are battery density and specific heat capacity, respectively. , , These are the heat generation rates of reaction heat, polarization heat, and ohmic heat, respectively.

[0028] When the battery surface comes into contact with air, heat exchange occurs; the convective heat transfer is as follows: ; in, Where is the battery thermal conductivity, and h is the convective heat transfer coefficient. For ambient temperature, This refers to the surface temperature of the battery. The kinetic parameters in the electrochemical model, such as the diffusion coefficients of the liquid and solid phases and the reaction rate constant, are highly dependent on temperature, meaning they are greatly affected by temperature changes. The model is usually based on the Arrhenius equation, as shown in the following formula: ; In the formula, For temperature The dynamics and transport parameters to be corrected are as follows. Reference temperature The parameter values ​​below; This is the current absolute temperature of the battery; Reference absolute temperature; The apparent activation energy is the corresponding parameter. The molar gas constant; Based on battery geometric parameters and electrochemical parameters, elastic strain, chemical strain caused by lithium ion insertion or extraction, thermal strain caused by temperature change, strain caused by lithium metal deposition, irreversible strain caused by dead lithium accumulation, and strain caused by SEI film growth are obtained. An electrochemical-mechanical coupling sub-model is constructed by combining the battery temperature field and the accumulation of by-reaction products to output the total volumetric strain. Then, the external pressure response of the battery is calculated based on the total volumetric strain. It should be noted that, in constructing an electrochemical-mechanical coupling model constrained by pressure characteristics, the total volumetric strain generated by the negative electrode active material during charging and discharging is jointly determined by lithium-ion insertion / extraction, temperature change, SEI film growth, lithium metal deposition, and dead lithium accumulation, satisfying the following: ; In the formula, For total volumetric strain, For elastic strain, Chemical strain caused by lithium-ion insertion / extraction For thermal strain, The strain is caused by lithium metal deposition. This is due to irreversible strain caused by the accumulation of dead lithium. This refers to the strain caused by the growth of the SEI film.

[0029] The battery's external pressure response satisfies: ; In the formula, Let t be the external pressure of the battery at time t. This is the initial preload. The equivalent stiffness coefficient of the battery structure; Compared to using only voltage, temperature, and SOC signals, pressure characteristics can provide observable information on negative electrode volume changes and the accumulation of by-reaction products, especially reflecting irreversible expansion caused by lithium metal deposition and dead lithium accumulation.

[0030] It should be noted that by using Arrhenius temperature correction and feedback of heat generation from side reactions to the temperature field, a two-way coupling of "temperature → electrochemical parameters → heat generation → temperature" is formed; by incorporating the strain of lithium metal deposition, dead lithium accumulation, and SEI film growth into the total volumetric strain, the external pressure response can reflect the cumulative expansion effect of side reaction products.

[0031] In addition, a multiphysics coupling model parameter identification method is further introduced in step S3; The sensitivity of the parameters to be identified in the multiphysics coupling model is distinguished, the sensitivity distinction results are obtained, the parameters to be identified are selected based on the sensitivity distinction results, and a multi-objective optimization function is constructed with the terminal voltage error, negative electrode potential error and electrochemical impedance error as objectives. A multi-objective genetic algorithm is used to perform a global search of the parameters to be identified: An initial parameter population is generated within the range of values ​​of each parameter to be identified. Each group of parameters is input into a multiphysics coupling model to calculate the objective function value. Candidate parameters are filtered according to Pareto non-dominated sorting and crowding distance. A new generation of population is generated through selection, crossover, and mutation. The process is repeated iteratively until the convergence condition is met, and the Pareto non-dominated solution set is obtained and used as the parameter identification result in the global search phase. The initial parameters are selected based on the Pareto non-dominated solution set, which satisfies the physical constraints and the comprehensive error meets the preset conditions. A trust region is established with the initial parameters as the center, and a Gaussian process surrogate model is constructed based on the existing parameters and the corresponding objective function values. Candidate parameters are generated within the trust region using a Gaussian process surrogate model. Then, the objective function value of the candidate parameters is calculated using a multiphysics coupling model. The trust region is adjusted according to the candidate parameters, and the iteration is repeated until the stopping condition is met to obtain the parameter identification results in the local optimization stage. Based on the parameter identification results from both the global search and local optimization phases, the parameter set that minimizes the combined terminal voltage error, negative electrode potential error, and electrochemical impedance error while satisfying physical constraints is selected as the final identification parameters.

[0032] It should be noted that the objective function for the terminal voltage is: ; The objective function for negative electrode potential is: ; The impedance objective function is: ; In the formula, the subscript sim represents the model simulation value, and the subscript exp represents the experimental measurement value. , and These represent the number of data points for terminal voltage, negative electrode potential, and impedance, respectively. Construct a multi-objective optimization problem: ; To select initial parameters for TuRBO local optimization, we selected a set of parameters from the Pareto non-dominated solution set obtained from NSGA-II that have small normalized synthesis error and satisfy the physical constraints of the model as the initial parameters for TuRBO local optimization. A local trust region is established centered on the initial parameters, and a Gaussian process surrogate model is constructed based on the existing parameter samples and their objective function values: ; In the formula, The mean function of a Gaussian process. The covariance kernel function; Candidate parameters are generated in the trust region using a Gaussian process surrogate model, and the terminal voltage, negative electrode potential and impedance response corresponding to the candidate parameters are calculated by a real multiphysics coupling model. When candidate parameters can reduce the overall target error, expand the trust region; when no effective improvement is achieved after multiple iterations, shrink the trust region; repeat candidate parameter generation, model calculation and proxy model update until the preset number of model calls is reached, there is no improvement after continuous iterations or the trust region size is lower than the preset threshold.

[0033] It should be noted that by using multi-objective constraints of three dimensions—terminal voltage, negative electrode potential, and electrochemical impedance—the defects of non-unique parameter compensation and identification results caused by single objectives are overcome; the NSGA-II global search ensures full exploration of the parameter space; the TuRBO local optimization uses high-quality solutions in the Pareto solution set as starting points for fine-grained search; and the synergistic mechanism of global exploration and local development solves the technical problem that existing methods cannot simultaneously guarantee global search capability and local identification accuracy.

[0034] In step S4, a method for extracting lithium plating pressure residual features is further introduced. Replace the original parameter values ​​in the multiphysics coupling model with the final identified parameters to obtain the multiphysics coupling model after parameter calibration. The baseline pressure under the condition of no lithium metal deposition was calculated based on the multiphysics coupling model after parameter calibration, and the negative electrode potential related data was generated. The residual characteristics of lithium plating pressure are constructed based on the difference between the measured pressure and the reference pressure. When the difference between the measured pressure and the reference pressure is greater than the preset tolerance threshold, the difference is used as the residual characteristic of lithium plating pressure; otherwise, the residual characteristic of lithium plating pressure is zero.

[0035] It should be noted that the reference pressure calculated based on the electrochemical-thermal-mechanical coupling model is caused by normal lithium intercalation expansion and thermal expansion when no lithium metal deposition occurs. It is represented as: Then, based on the difference between the experimentally measured pressure and the reference pressure, the residual characteristics of the lithium plating pressure are extracted, which are expressed as follows: ; In the formula, The residual characteristics of lithium plating pressure are as follows: The pressure measured in the experiment, The reference pressure before lithium plating occurs. These are the tolerance thresholds corresponding to pressure measurement errors and model errors; In this invention, the original pressure value is not directly used. Instead of using it as input to the neural network, it extracts the residual features of lithium plating pressure, which can more directly characterize the abnormal expansion of lithium plating. .

[0036] It should be noted that the reference pressure calculated by the multiphysics coupling model after parameter calibration only includes normal lithium intercalation expansion and thermal expansion. The normal expansion component is subtracted from the measured pressure to overcome the defect that the lithium plating characteristics are submerged by normal expansion. The interference of measurement error and model error is eliminated by the tolerance threshold. This feature directly characterizes the additional expansion caused by lithium metal deposition and the accumulation of by-reaction products. It solves the technical problem that the pressure signal cannot be effectively used for lithium plating state identification.

[0037] Example 3: Please refer to Figure 1-8 Based on Example 2, a lithium battery negative electrode potential prediction method based on mechanism data fusion is proposed, and a time-series prediction network construction and training method is further introduced in step S5. The negative electrode potential correlation data and lithium plating pressure residual characteristics are used as input features; The negative electrode potential data generated by the multiphysics coupling model under the corresponding operating conditions is used as training labels and aligned with the input features at the corresponding time. A sliding time window is used to construct the input time series from the input features of multiple consecutive sampling times; A temporal prediction network is constructed, which includes a temporal convolution module and a recurrent memory module. The temporal convolution module contains residual connections. The input time series is fed into the temporal prediction network. After the temporal convolution module extracts local dynamic features and multi-scale temporal features, it is fed into the recurrent memory module to learn the long-term dependencies between temporal features. Finally, the negative electrode potential prediction value is output through the fully connected layer. The predicted negative electrode potential is compared with the training label, and a loss function is constructed based on the error between the two. The time series prediction network is trained with the goal of minimizing the loss function.

[0038] It should be noted that the input features are: ; In the formula, for Battery current at any given moment for Battery terminal voltage at time, for The state of charge of the battery at any given time. for The surface temperature of the battery at any given time. The residual characteristics of lithium plating pressure obtained according to step S4; From time Select a preset window length from the data collected previously. The input vector constitutes the time step. The corresponding input time series: ; In the formula, For the first The input time series corresponding to each moment. The preset sliding time window length; time Corresponding input time series The TCN module uses causal convolution, which ensures that the output features at the current time are determined only by the input data at the current time and before it. time The corresponding TCN output features are represented as follows: ; In the formula, For a moment The input vector; , For a moment Previous historical input vectors; For a moment The corresponding causal convolution output features; This represents a mapping function consisting of a causal convolution kernel and corresponding nonlinear operations; The TCN module uses dilated convolution to expand the temporal receptive field by setting the dilation rate. Historical input data is sampled at intervals; The effective receptive field corresponding to a single-layer one-dimensional dilated convolution satisfies: ; In the formula, This represents the effective receptive field length of a single-layer one-dimensional dilated convolution on the input time series. The kernel size; The expansion rate; To improve network training stability, residual connections are configured in the TCN module, in the following form: ; In the formula, These are the input features for the residual block; The residual mapping is obtained after the input features are processed by causal dilated convolution, normalization, nonlinear activation and random deactivation; Output for residual blocks; It is a non-linear activation function; For time Corresponding input time series The feature extraction process of the TCN module is represented as follows: ; In the formula, For the TCN module, based on the input time series The extracted feature sequences.

[0039] feature sequence Input LSTM module: In the formula, This represents the hidden state output by the LSTM module at the corresponding time. Further, the hidden state output by the LSTM module By inputting the fully connected layer, we obtain the predicted negative electrode potential at the corresponding time step: ; In the formula, For a moment The corresponding predicted negative electrode potential; These are the weights of the fully connected layer; For biasing the fully connected layer; During training, the predicted negative electrode potential output by the TCN-LSTM negative electrode potential prediction model is used. Training labels with corresponding negative electrode potentials Compare the two and construct a loss function based on the error between them; .

[0040] It should be noted that training labels generated by a multi-physics coupling model are used to replace measured data, overcoming the difficulty in obtaining negative electrode potential labels; physical priors are introduced through lithium plating pressure residual features to enhance the model's sensitivity to lithium plating states; the TCN-LSTM architecture balances local dynamic feature extraction and long-term dependency learning; and a collaborative architecture of "mechanism label generation + data-driven online prediction" achieves a balance between physical interpretability and online computational efficiency. This resolves the technical contradictions of difficulty in obtaining training samples for data-driven models and the computational complexity of pure mechanistic models, making them difficult to apply online.

[0041] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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 a process, method, article, or apparatus. Without further limitations, 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 said element.

[0042] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the negative electrode potential of a lithium battery based on mechanism data fusion, characterized in that, Includes the following steps; S1. Obtain the geometric parameters, electrochemical parameters, and side reaction parameters of the battery, and construct a coupled electrochemical model of multiple side reactions at the negative electrode; S2. A multi-physics coupling model is constructed based on the energy conservation equation, the total volumetric strain of the battery negative electrode active material during the charging and discharging process, and the coupled electrochemical model of multiple side reactions of the negative electrode. S3. Distinguish the parameters to be identified in the multiphysics coupling model, construct a multi-objective optimization function, and then perform global and local two-stage search and optimization on the parameters to be identified, obtain the parameter identification results at each stage, and select the parameter identification result with the smallest comprehensive error as the final identified parameter. S4. Input the final identification parameters into the multiphysics coupling model established in S2, and output the reference pressure, negative electrode potential related data, and negative electrode potential data. Construct the lithium plating pressure residual characteristics based on the difference between the measured pressure and the reference pressure. S5. Using historical side reaction parameters and historical lithium plating pressure residual characteristics as inputs, an input sequence is constructed using a sliding time window, and a time-series prediction network is built. The negative electrode potential related data and lithium plating pressure residual characteristics are input into the time series prediction network, and the predicted value of the negative electrode potential is output.

2. The lithium battery negative electrode potential prediction method based on mechanism data fusion according to claim 1, characterized in that: The S1 step further introduces a method for constructing a coupled electrochemical model of multiple side reactions at the negative electrode. Model parameters, model boundary conditions, and initial conditions are set based on the pre-processed battery geometric parameters, electrochemical parameters, and side reaction parameters. Based on battery geometric parameters and electrochemical parameters, the following models were constructed: local total reaction current shunting model of negative electrode, lithium metal deposition model triggered by negative electrode interface potential, coupled evolution model of reversible lithium metal, dead lithium and SEI film, coupled model between SEI film growth and lithium metal deposition, feedback model of by-reaction products on negative electrode interface film resistance, and feedback model of by-reaction products on effective reaction area of ​​negative electrode. Based on the above six models, as well as model parameters, model boundary conditions, and initial conditions, a coupled electrochemical model of multiple side reactions at the negative electrode is constructed.

3. The lithium battery negative electrode potential prediction method based on mechanism data fusion according to claim 2, characterized in that: The above six model construction methods are further introduced in step S1; Based on battery geometric parameters and electrochemical parameters, the current density of the negative electrode main lithium intercalation reaction, the current density of the SEI film growth side reaction, the current density of lithium metal deposition, and the current density of reversible lithium metal stripping are obtained, and a local total reaction current shunt model of the negative electrode is constructed to output the total current density. Based on battery geometric parameters and electrochemical parameters, the negative electrode solid phase potential, negative electrode liquid electrolyte potential, lithium metal or lithium ion equilibrium potential, and interface film resistance are obtained. Combined with the total current density, a lithium metal deposition model triggered by the negative electrode interface potential is constructed, and the lithium metal deposition overpotential is output. Based on battery geometric parameters and electrochemical parameters, the reversible lithium capacity and the rate constant of reversible lithium to dead lithium conversion are obtained. Combined with lithium deposition current density and reversible lithium stripping current density, a coupled evolution model of reversible lithium, dead lithium and SEI film is constructed to output the reversible lithium capacity and the reversible lithium change rate. The molar amount of dead lithium is calculated based on the rate of change of reversible lithium metal and the rate constant of the conversion of reversible lithium metal to dead lithium, and is used as the cumulative amount of by-reaction products. Based on battery geometric parameters and electrochemical parameters, the exchange current density of SEI film growth side reactions, the SEI film growth side reaction transfer coefficient, the SEI film growth side reaction overpotential, the initial value of SEI film thickness, the promoting coefficient of reversible lithium metal on secondary SEI formation, the promoting coefficient of dead lithium on secondary SEI formation, and the reference SEI film thickness are obtained. Combined with the reversible lithium metal capacity and the molar amount of dead lithium, a coupling model between SEI film growth and lithium metal deposition is constructed, and the SEI film growth side reaction current density and the current SEI film thickness are output. Based on battery geometric parameters and electrochemical parameters, the equivalent ionic conductivity of the SEI film, the contribution coefficient of reversible lithium metal to the interface impedance, and the contribution coefficient of dead lithium to the interface impedance are obtained. Combined with the current SEI film thickness, reversible lithium metal capacity, and dead lithium molar amount, a feedback model of the side reaction products on the negative electrode interface film resistance is constructed, and the negative electrode interface film resistance is output. Based on battery geometric parameters and electrochemical parameters, the initial reaction specific surface area of ​​the negative electrode, the shielding coefficient of the SEI film on the active reaction area of ​​the negative electrode, the shielding coefficient of reversible lithium metal on the active reaction area of ​​the negative electrode, and the shielding coefficient of dead lithium on the active reaction area of ​​the negative electrode are obtained. Combined with the current SEI film thickness, reversible lithium metal capacity, and dead lithium molar amount, a feedback model of the side reaction products on the effective reaction area of ​​the negative electrode is constructed, and the effective reaction specific surface area of ​​the negative electrode is output. The negative electrode interface film resistance and the effective reaction surface area of ​​the negative electrode are fed back into the calculation of the lithium metal deposition overpotential to form a closed-loop feedback of the side reaction products to the polarization of the negative electrode.

4. The lithium battery negative electrode potential prediction method based on mechanism data fusion according to claim 1, characterized in that: The S2 step further introduces a method for constructing a multiphysics coupling model; An electrochemical-thermal coupling sub-model was constructed based on battery geometric parameters, electrochemical parameters, and energy conservation equations; An electrochemical-mechanical coupling sub-model was constructed based on battery geometric parameters, electrochemical parameters, and the total volumetric strain of the battery negative electrode active material during the charging and discharging process. In the multi-physics coupling model, the steady-state operating point is solved under a given operating condition by the negative electrode multi-side reaction coupled electrochemical model. A small alternating perturbation signal is applied at the steady-state operating point and the voltage response at different frequencies is obtained. The complex impedance spectrum of the battery is obtained based on the ratio of the voltage response to the perturbation signal, and an electrochemical-impedance coupling sub-model is established. A multiphysics coupling model is constructed by fusing the electrochemical-thermal coupling sub-model, the electrochemical-mechanical coupling sub-model, and the electrochemical-impedance coupling sub-model.

5. The lithium battery negative electrode potential prediction method based on mechanism data fusion according to claim 4, characterized in that: The S2 step further introduces methods for constructing electrochemical-thermal coupler models and electrochemical-mechanical coupler models; Based on the battery's geometric parameters and electrochemical parameters, the battery's thermal conductivity, battery density, specific heat capacity, reaction heat generation rate, polarization heat generation rate, ohmic heat generation rate, and convective heat transfer coefficient are obtained. An electrochemical-thermal coupling sub-model is constructed by combining the multi-reaction coupling electrochemical model of the negative electrode to output the battery temperature field. The model parameters of the coupled electrochemical model of multiple side reactions of the negative electrode are corrected based on the battery temperature field, and the heat generated by the side reactions in the coupled electrochemical model of multiple side reactions of the negative electrode is fed back to the electrochemical-thermal coupling sub-model as a heat source to update the battery temperature field. Based on battery geometric parameters and electrochemical parameters, elastic strain, chemical strain caused by lithium ion insertion or extraction, thermal strain caused by temperature change, strain caused by lithium metal deposition, irreversible strain caused by dead lithium accumulation, and strain caused by SEI film growth are obtained. An electrochemical-mechanical coupled sub-model is constructed by combining the battery temperature field and the accumulation of by-reaction products to output the total volumetric strain. Then, the external pressure response of the battery is calculated based on the total volumetric strain.

6. The lithium battery negative electrode potential prediction method based on mechanism data fusion according to claim 5, characterized in that: The S3 step further introduces a method for identifying parameters of a multiphysics coupling model. The sensitivity of the parameters to be identified in the multiphysics coupling model is distinguished, the sensitivity distinction results are obtained, the parameters to be identified are selected based on the sensitivity distinction results, and a multi-objective optimization function is constructed with the terminal voltage error, negative electrode potential error and electrochemical impedance error as objectives. A multi-objective genetic algorithm is used to perform a global search of the parameters to be identified: An initial parameter population is generated within the range of values ​​of each parameter to be identified. Each group of parameters is input into a multiphysics coupling model to calculate the objective function value. Candidate parameters are filtered according to Pareto non-dominated sorting and crowding distance. A new generation of population is generated through selection, crossover, and mutation. The process is repeated iteratively until the convergence condition is met, and the Pareto non-dominated solution set is obtained and used as the parameter identification result in the global search phase. The initial parameters are selected based on the Pareto non-dominated solution set, which satisfies the physical constraints and the comprehensive error meets the preset conditions. A trust region is established with the initial parameters as the center, and a Gaussian process surrogate model is constructed based on the existing parameters and the corresponding objective function values. Candidate parameters are generated within the trust region using a Gaussian process surrogate model. Then, the objective function value of the candidate parameters is calculated using a multiphysics coupling model. The trust region is adjusted according to the candidate parameters, and the iteration is repeated until the stopping condition is met to obtain the parameter identification results in the local optimization stage. Based on the parameter identification results from both the global search and local optimization phases, the parameter set that minimizes the combined terminal voltage error, negative electrode potential error, and electrochemical impedance error while satisfying physical constraints is selected as the final identification parameters.

7. The lithium battery negative electrode potential prediction method based on mechanism data fusion according to claim 6, characterized in that: The S4 step further introduces a method for extracting lithium plating pressure residual features; Replace the original parameter values ​​in the multiphysics coupling model with the final identified parameters to obtain the multiphysics coupling model after parameter calibration. The baseline pressure under the condition of no lithium metal deposition was calculated based on the multiphysics coupling model after parameter calibration, and the negative electrode potential related data was generated. The residual characteristics of lithium plating pressure are constructed based on the difference between the measured pressure and the reference pressure. When the difference between the measured pressure and the reference pressure is greater than the preset tolerance threshold, the difference is used as the residual characteristic of lithium plating pressure; otherwise, the residual characteristic of lithium plating pressure is zero.

8. The lithium battery negative electrode potential prediction method based on mechanism data fusion according to claim 7, characterized in that: The S5 step further introduces a method for constructing and training a temporal prediction network; The negative electrode potential correlation data and lithium plating pressure residual characteristics are used as input features; The negative electrode potential data generated by the multiphysics coupling model under the corresponding operating conditions is used as training labels and aligned with the input features at the corresponding time. A sliding time window is used to construct the input time series from the input features of multiple consecutive sampling times; A temporal prediction network is constructed, which includes a temporal convolution module and a recurrent memory module. The temporal convolution module contains residual connections. The input time series is fed into the temporal prediction network. After the temporal convolution module extracts local dynamic features and multi-scale temporal features, it is fed into the recurrent memory module to learn the long-term dependencies between temporal features. Finally, the negative electrode potential prediction value is output through the fully connected layer. The predicted negative electrode potential is compared with the training label, and a loss function is constructed based on the error between the two. The time series prediction network is trained with the goal of minimizing the loss function.

Images