A lithium battery thermal runaway identification method and device, electronic equipment and storage medium

Abstract

The application belongs to the technical field of lithium ion battery safety modeling and simulation, and specifically discloses a lithium battery thermal runaway identification method and device, an electronic device and a storage medium. The method comprises the following steps: a battery temperature prediction neural network model is constructed by using the Arrhenius equation and a lithium battery temperature updating equation; the battery temperature prediction neural network model is trained by using thermal runaway experimental data samples of multiple groups of lithium batteries, and a trained battery temperature prediction neural network model is obtained; current reactant concentration logarithmic information and current battery temperature data of a lithium battery to be measured are input into the battery temperature prediction neural network model, and next-time battery temperature data of the lithium battery to be measured is obtained; thermal runaway analysis is performed according to the next-time battery temperature data and historical battery temperature data before the next-time battery temperature data, and a thermal runaway inversion result of the lithium battery to be measured is identified. Through the application, the inversion accuracy can be greatly improved, and the identification accuracy of the thermal runaway process of the lithium battery can be improved.

Description

Technical Field

[0001] This application belongs to the field of lithium-ion battery safety modeling and simulation technology, and more specifically, relates to a lithium battery thermal runaway identification method, device, electronic device and storage medium. Background Technology

[0002] Lithium-ion batteries, with their high energy density, low self-discharge rate, and long cycle life, are currently the primary energy source for electric vehicles. However, with the large-scale application of lithium-ion batteries in electric vehicles, safety incidents, particularly thermal runaway, occur frequently. Thermal runaway incidents typically manifest as a sudden rise in battery temperature, smoke, and fire. Therefore, accurate identification of lithium-ion battery thermal runaway is crucial for ensuring the safety of lithium-ion battery use.

[0003] Currently, relevant technologies for modeling lithium battery thermal runaway mainly fall into two categories: mathematical models based on physical mechanisms and pure data-driven deep learning models. However, the aforementioned mathematical models are essentially biased towards mathematical fitting rather than actual physical inversion, resulting in significant fitting errors and low inversion accuracy, thus failing to accurately identify the thermal runaway process of lithium batteries. On the other hand, pure data-driven deep learning models are prone to overfitting during training and lack explicit constraints from physical mechanisms, leading to poor model interpretability and low inversion accuracy, also failing to accurately identify the thermal runaway process of lithium batteries.

[0004] Therefore, how to better identify the thermal runaway process of lithium batteries remains a technical problem that the industry urgently needs to solve. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this application is to better identify the thermal runaway process of lithium batteries. It aims to solve the problem that the existing technology has low inversion accuracy and cannot accurately identify the thermal runaway process of lithium batteries due to the large fitting error of the mathematical model, the tendency of deep learning model training to overfit, and the poor interpretability of the model.

[0006] To achieve the above objectives, in a first aspect, this application provides a method for identifying thermal runaway in lithium batteries, comprising: Step S1: Construct a neural network model for predicting battery temperature using the Arrhenius equation and the lithium battery temperature update equation. Step S2: The battery temperature prediction neural network model is trained using multiple sets of thermal runaway experimental data samples of lithium batteries to obtain a trained battery temperature prediction neural network model. Step S3: Input the logarithmic information of the current reactant concentration and the current battery temperature data of the lithium battery under test into the battery temperature prediction neural network model to obtain the battery temperature data of the lithium battery under test at the next moment. Step S4: Perform thermal runaway analysis based on the battery temperature data at the next moment and its previous historical battery temperature data to identify the thermal runaway inversion result of the lithium battery under test.

[0007] In some embodiments, step S1 includes: A thermo-gas coupling model for thermal runaway of lithium-ion batteries was established using the Arrhenius equation. The thermal runaway thermo-gas coupling model is incorporated into the neuron structure of a neural network to construct a reaction dynamics neural network model; A battery temperature update model is constructed based on the lithium battery temperature update equation, and a battery temperature prediction neural network model is obtained by combining the reaction kinetics neural network model and the battery temperature update model; the network weight parameters of the reaction kinetics neural network model are determined based on the kinetic parameters of the thermal runaway thermo-gas coupling model.

[0008] In some embodiments, the reaction kinetic neural network model is a single hidden layer neural network, including a heat production rate neuron, a gas production rate neuron, an input layer for receiving the logarithm of reactant concentration and temperature parameters, and an output layer for outputting the rate of change of each component concentration, the total heat production rate and the total gas production rate. The weight parameters and bias parameters of the single hidden layer neural network are constrained to the physical parameters of the Arrhenius equation, which include the activation energy of the reaction, the pre-exponential factor, the reaction order, the enthalpy of the heat-producing reaction, and the enthalpy of the gas-producing reaction; the forward propagation process of the single hidden layer neural network is equivalent to the calculation process of the Arrhenius equation.

[0009] In some embodiments, the thermal runaway experimental data sample includes thermal runaway initial state data samples and corresponding full-time temperature and pressure data labels; step S2 includes: Step S21: For any of the thermal runaway experimental data samples, based on the independent variable substitution and two-stage solution strategy, the thermal runaway initial state data sample and the corresponding full-time temperature and pressure data labels are input into the battery temperature prediction neural network model to obtain the full-time battery temperature prediction information and gas production rate prediction information corresponding to the thermal runaway experimental data sample. Step S22: Using a preset loss function, the loss value is calculated based on the battery temperature prediction information and gas production rate prediction information for the entire time period, as well as the temperature and pressure data labels for the entire time period, to obtain the loss value corresponding to the thermal runaway experimental data sample. Step S23: Based on the loss value, the weight parameters of the reaction dynamics neural network model are iteratively updated using the gradient backpropagation algorithm until the preset loss function converges or the maximum number of iterations is reached, thereby obtaining the trained battery temperature prediction neural network model.

[0010] In some embodiments, the two-stage solution strategy includes a first stage and a second stage; step S21 includes: Step S211: In the first stage, the current first battery state data is input into the battery temperature prediction neural network model to obtain the battery temperature prediction information for the next moment, so as to determine the first battery state data for the next moment; the current first battery state data includes the battery state data carried by the thermal runaway initial state data sample. Step S212: When the battery temperature prediction information at the next moment indicates that the battery temperature change rate is not greater than zero, proceed to the second stage; Step S213: In the second stage, the current second battery state data is input into the battery temperature prediction neural network model to obtain the battery temperature prediction information for the next moment. Step S214: Upon determining the end of the second stage, the battery temperature prediction information and gas production rate prediction information obtained at different times in the first stage, as well as the battery temperature prediction information and gas production rate prediction information obtained at different times in the second stage, are combined in a time sequence to obtain the full-time battery temperature prediction information and gas production rate prediction information corresponding to the thermal runaway experimental data sample.

[0011] In some embodiments, step S211 includes: In the first stage, the current first battery state data is input into the reaction kinetics neural network model, and the heat generation rate prediction information corresponding to the current first battery state data is output. The heat generation rate prediction information corresponding to the current first battery state data is input into the battery temperature update model, and the independent variable of the battery temperature update model is replaced by temperature and numerical integration is performed to obtain the battery temperature prediction information at the next moment, so as to determine the first battery state data at the next moment.

[0012] In some embodiments, the preset loss function used by the battery temperature prediction neural network model in each round of model training includes a temperature loss function and a pressure loss function; The temperature loss function is determined by a weighted calculation based on the root mean square error of temperature reversal time and the maximum temperature loss error; the root mean square error of temperature reversal time is determined by calculating the root mean square of the difference between the time required for model simulation and the experimental recording time at the same temperature test time. The pressure loss function is determined by a weighted calculation based on the root mean square error of the pressure reversal time and the pressure relief time error; the root mean square error of the pressure reversal time is determined by the root mean square of the difference between the model simulation time and the experimental recording time at the same pressure test time.

[0013] Secondly, this application provides a lithium battery thermal runaway identification device, comprising: The model building module is used to build a neural network model for predicting battery temperature using the Arrhenius equation and the lithium battery temperature update equation. The model training module is used to train the battery temperature prediction neural network model using multiple sets of thermal runaway experimental data samples of lithium batteries, so as to obtain a trained battery temperature prediction neural network model. The temperature prediction module is used to input the logarithmic information of the current reactant concentration and the current battery temperature data of the lithium battery under test into the battery temperature prediction neural network model to obtain the battery temperature data of the lithium battery under test at the next moment. The thermal runaway identification module is used to perform thermal runaway analysis based on the battery temperature data at the next moment and its previous historical battery temperature data, and to identify the thermal runaway inversion result of the lithium battery under test.

[0014] Thirdly, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.

[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0016] Fifthly, this application provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0017] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0018] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: This application provides a lithium battery thermal runaway identification method, device, electronic device, and storage medium. By deeply exploring the mapping relationship between the unified paradigm of the Arrhenius equation and the topology of neural network units, the multi-step Arrhenius equation reaction mechanism is explicitly mapped to the topology of neural network units. Furthermore, a battery temperature prediction neural network model is constructed in conjunction with the lithium battery temperature update equation. This achieves a deep structural fusion of the neural network weight parameters and the physical mechanism of thermal runaway, allowing the trainable parameters of the neural network to directly correspond to the thermal runaway kinetic parameters. This transforms the model training process using thermal runaway experimental data samples into a physical optimization process for solving thermal runaway kinetic parameters, rather than a traditional data fitting process. This avoids data fitting errors and enables accurate identification of thermal runaway kinetic parameters, significantly improving inversion accuracy. Consequently, it allows for real-time and accurate inversion of the temperature change trend during lithium battery thermal runaway, enhancing the identification accuracy of lithium battery thermal runaway processes. Attached Figure Description

[0019] Figure 1 This is a schematic flowchart of the lithium battery thermal runaway identification method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the heat generation rate neuron in the reaction kinetics neural network model provided in this application embodiment; Figure 3 This is a schematic diagram of the gas production rate neuron in the reaction kinetics neural network model provided in this application embodiment; Figure 4 This is a schematic diagram of the overall network structure of the reaction dynamics neural network model provided in the embodiments of this application; Figure 5 This is a schematic diagram of the model training process of the battery temperature prediction neural network model provided in the embodiments of this application; Figure 6 This is a schematic diagram showing the comparison between the predicted data of the lithium battery thermal runaway temperature curve and the experimental data provided in the embodiments of this application; Figure 7 This is a schematic diagram of the normalized concentration change curves of four reactions in the thermal runaway of a lithium battery provided in the embodiments of this application; Figure 8 This is a schematic flowchart of the lithium battery thermal runaway identification device provided in the embodiments of this application; Figure 9 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0021] The terms "first" and "second," etc., used in the specification and claims of this application are used to distinguish different objects, not to describe a specific order of objects. For example, "first battery state data" and "second battery state data," etc., are used to distinguish different battery state data, not to describe a specific order of battery state data.

[0022] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0023] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0024] The embodiments of this application are described below with reference to the accompanying drawings.

[0025] Figure 1 This is a flowchart illustrating the lithium battery thermal runaway identification method provided in the embodiments of this application, as shown below. Figure 1 As shown, it includes: Step S1: Construct a neural network model for predicting battery temperature using the Arrhenius equation and the lithium battery temperature update equation. Step S2: Use multiple sets of thermal runaway experimental data samples of lithium batteries to train the battery temperature prediction neural network model to obtain the trained battery temperature prediction neural network model. Step S3: Input the logarithmic information of the current reactant concentration and the current battery temperature data of the lithium battery under test into the battery temperature prediction neural network model to obtain the battery temperature data of the lithium battery under test at the next moment. Step S4: Perform thermal runaway analysis based on the battery temperature data at the next moment and its previous historical battery temperature data to identify the thermal runaway inversion result of the lithium battery under test.

[0026] Specifically, the logarithmic information of reactant concentration described in the embodiments of this application refers to the data obtained by taking the logarithm of the reactant concentration generated during the internal reaction process of thermal runaway of lithium battery.

[0027] Here, the internal reactions of lithium battery thermal runaway typically include four types of reactions: solid electrolyte interface (SEI) decomposition, negative electrode reaction with electrolyte, positive electrode reaction with electrolyte, and electrolyte decomposition reaction.

[0028] The battery temperature prediction neural network model described in this application embodiment is obtained by training a large number of lithium battery thermal runaway experimental data samples. It may include a reaction kinetics neural network model and a battery temperature update model connected in sequence. The reaction kinetics neural network model is obtained by mapping the Arrhenius equation describing the rate of chemical reactions inside the battery to neurons of the neural network; the battery temperature update model can be determined according to the existing battery temperature update equation for the thermal runaway process of lithium batteries.

[0029] In the embodiments of this application, in step S1, a battery temperature prediction neural network model is constructed using the Arrhenius equation and the lithium battery temperature update equation. First, a mathematical model describing the battery thermal runaway mechanism can be constructed using the Arrhenius equation, which describes the rate of chemical reactions inside the battery, and the lithium battery temperature update equation. Then, by summarizing the correspondence between the unified paradigm of the Arrhenius equation and the topology of the neural network units, the multi-step Arrhenius reaction mechanism equation is transformed and explicitly mapped to the topology of the neural network units, thereby constructing a sequentially connected reaction kinetics neural network model and a battery temperature update model, resulting in the battery temperature prediction neural network model.

[0030] The Arrhenius equation is used to characterize four main exothermic side reaction processes inside the battery: the solid electrolyte interface (SEI) decomposition reaction, the reaction between the negative electrode and the electrolyte, the reaction between the positive electrode and the electrolyte, and the decomposition reaction of the electrolyte itself.

[0031] The rate equation for the SEI membrane decomposition reaction is as follows: ; In the formula, This represents the proportion of unstable lithium in the SEI film. This is the pre-exponential factor for the reaction; This is the activation energy of the reaction; The molar gas constant; This refers to the battery temperature.

[0032] The reaction rate equation between the negative electrode and the electrolyte is as follows: ; In the formula, This represents the proportion of lithium intercalated in the carbon anode that participates in the reaction; This is the ratio of SEI film thickness to the characteristic size of the active material; This is the activation energy of the reaction. A ne This is the pre-exponential factor for the reaction.

[0033] The reaction rate equation between the positive electrode and the electrolyte is as follows: ; In the formula, This represents the ratio of the reacted cathode material to the total cathode material. A pe This is the pre-exponential factor for the reaction. E pe This is the activation energy of the reaction.

[0034] The rate equation for electrolyte decomposition reaction is as follows: ; In the formula, This represents the ratio of remaining electrolyte to total electrolyte. A el This is the pre-exponential factor for the reaction. E el This is the activation energy of the reaction.

[0035] In this embodiment, in step S2, after constructing the battery temperature prediction neural network model, it is necessary to pre-train the model using multiple sets of lithium battery thermal runaway experimental data samples to obtain a trained model. Specifically, firstly, thermal runaway experimental data samples of lithium batteries are acquired. Here, an adiabatic accelerated calorimeter can be used to conduct thermal runaway tests on different types of lithium batteries. The battery is placed in an adiabatic environment and heated, and the HWS (Heat-Wait-Seek) mode is used to trigger thermal runaway of the lithium battery. The collected data sequence includes a time series. and the corresponding temperature sequence and non-condensable gas pressure sequence .

[0036] The thermal runaway experimental data samples can specifically include Thermal runaway initial state data samples at time points and including temperature sequences and non-condensable gas pressure sequence Data - Full-time temperature and pressure data labels. Here, thermal runaway initial state data samples may include... Battery temperature at any time Logarithmic concentration parameters of reactants and .in, This represents the reactant concentration parameter.

[0037] In this embodiment, in step S3, the logarithmic information of the current reactant concentration and the current battery temperature data of the lithium battery under test are input into the battery temperature prediction neural network model. After data recognition and prediction processing by the internal reaction kinetics neural network model and the battery temperature update model, the battery temperature data of the lithium battery under test at the next moment can be obtained. In this way, the battery temperature prediction neural network model can be used to invert the changes in the battery temperature data of the lithium battery under test in real time.

[0038] Furthermore, in this embodiment of the application, in step S4, the battery temperature data at the next moment is compared with the previously predicted historical battery temperature data to perform thermal runaway analysis, thereby obtaining the battery temperature change data of the lithium battery under test during the thermal runaway stage, thereby identifying and obtaining high-precision inversion results of the lithium battery thermal runaway process.

[0039] The lithium battery thermal runaway identification method of this application deeply explores the mapping relationship between the unified paradigm of the Arrhenius equation and the topology of neural network units. It explicitly maps the multi-step Arrhenius equation reaction mechanism into the topology of neural network units and constructs a battery temperature prediction neural network model in conjunction with the lithium battery temperature update equation. This achieves a deep structural integration of the neural network weight parameters and the physical mechanism of thermal runaway, so that the trainable parameters of the neural network directly correspond to the thermal runaway dynamic parameters. This makes the model training process using thermal runaway experimental data samples a physical solution optimization process for thermal runaway dynamic parameters, rather than a traditional data fitting process. This avoids data fitting errors, achieves accurate identification of thermal runaway dynamic parameters, and greatly improves the inversion accuracy. As a result, it can realize real-time and accurate inversion of the temperature change trend of the lithium battery thermal runaway process, and improve the identification accuracy of the lithium battery thermal runaway process.

[0040] While existing technologies have proposed Physics-informed Neural Networks (PINN) as a new paradigm for integrating data and physical knowledge, directly applying conventional PINN to lithium-ion battery thermal runaway modeling still faces key challenges. Current PINNs often use physical equations merely as external soft constraints, with no connection between network weights and physical mechanisms. Essentially, they are still a neural network-based fitting method, where fitting accuracy and generalization depend on a large amount of training data, resulting in low training efficiency and unavoidable data fitting errors. This leads to low accuracy in retrieving lithium-ion battery thermal runaway processes, making it impossible to achieve direct and accurate inversion from experimental data to thermal runaway dynamic parameters.

[0041] Based on the content of the above embodiments, as an optional embodiment, step S1 includes: A thermo-gas coupling model for thermal runaway of lithium-ion batteries was established using the Arrhenius equation. A thermal runaway thermo-gas coupling model is incorporated into the neuronal structure of a neural network to construct a reaction dynamics neural network model; A battery temperature update model is constructed based on the lithium battery temperature update equation. The battery temperature prediction neural network model is obtained by combining the reaction kinetics neural network model and the battery temperature update model. The network weight parameters of the reaction kinetics neural network model are determined based on the kinetic parameters of the thermal runaway thermo-gas coupling model.

[0042] Specifically, in the embodiments of this application, the thermal runaway thermo-gas coupling model of lithium-ion batteries can be established using the Arrhenius equation, which describes the rate of chemical reactions inside the battery, to construct a mathematical model describing the thermal runaway mechanism of the battery. The multi-step Arrhenius reaction mechanism can be explicitly mapped to a neural network unit topology through transformation to obtain a reaction kinetic neural network model.

[0043] Among them, the thermal runaway thermo-gas coupling model assumes that the battery interior is homogeneous, ignores the spatial temperature gradient, and focuses on the chemical reaction kinetics that evolve over time.

[0044] In the embodiments of this application, a ternary lithium battery of model NMC 18650 is used as an example to identify the parameters of the thermal runaway thermo-gas coupling model. This method is not limited to ternary lithium batteries of model NMC 18650, but is also applicable to thermal runaway modeling of other types of lithium-ion batteries.

[0045] It should be noted that when the method of this embodiment is applied to other types of batteries, only the parameters of the Antoine equation used to calculate the saturated vapor pressure need to be changed accordingly based on the basic parameters of cell production; the remaining steps are consistent with this embodiment.

[0046] Furthermore, in the embodiments of this application, the proposed lithium battery thermal runaway thermo-gas coupling model includes four reactions: SEI decomposition, negative electrode-electrolyte reaction, positive electrode-electrolyte reaction, and electrolyte decomposition reaction. The reaction rates of each part can be calculated based on the above four reaction equations, and are uniformly expressed as follows: ; In the formula, Here, enthalpy represents the reaction enthalpy corresponding to each part of the reaction, and heat generated per kilogram of substance. This represents the unit carbon content corresponding to each part of the reaction; The reaction rates of each part; i This represents one of the four reactions mentioned above.

[0047] Total heat generation rate during battery thermal runaway as follows: ; The rate of heat exchange between the battery and the surrounding environment as follows: ; In the formula, The overall heat transfer coefficient includes both convective and radiative heat. This represents the effective heat dissipation area of ​​the battery. The ambient temperature.

[0048] Furthermore, in the embodiments of this application, the heat generation rate of the above four reactions in thermal runaway is... According to Refactoring, as follows: ; In the formula, , , These represent the pre-exponential factor, enthalpy of reaction, and carbon content per unit for the four reactions, respectively. and Representing the logarithmic concentration term and The reaction order; This represents the activation energy of the four reactions.

[0049] Furthermore, the above equation can be mathematically transformed to incorporate the thermal runaway thermo-gas coupling model into the neuronal structure of the neural network, thus constructing a reaction dynamics neural network model, namely: ; In the formula, This can be represented as the activation function of a neural network. This is the weight matrix of the neural network. For the input matrix, This is a bias term.

[0050] Furthermore, in the embodiments of this application, the gas generated inside the battery is converted into a pressure characterization, and the total internal pressure is... Decoupling is the saturated vapor pressure of the electrolyte. Partial pressure of non-condensable gases produced by chemical decomposition ,as follows: ; The saturated vapor pressure can be calculated using the Antoine equation, as follows: ; In practical applications, the above parameters may vary depending on the battery type and model. A , B , C The calibration needs to be performed by combining the electrolyte composition data provided by the battery manufacturer with thermal analysis experiments.

[0051] Partial pressure of non-condensable gases This pressure is mainly generated by the reaction between the electrodes and the electrolyte, and the decomposition of the electrolyte during thermal runaway. Based on the ideal gas law, this pressure is as follows: ; In the formula, The mass of gas produced by the decomposition reaction depends on the specific reaction kinetics neural network model. It is the ideal gas constant; This refers to the volume of the internal gaps within the battery. denoted as , where is the molar mass of the gas mixture.

[0052] The gas production rate model adopts the same framework as the reaction heat power model, which yields: ; In the formula, Let represent the enthalpy change of gas production for each reaction, indicating the mass of gas produced per unit reaction process.

[0053] Furthermore, for each of the above reactions Its reaction rate All follow Arrhenius's law, defined as follows: ; in, Pre-exponential factor, For activation energy, Let be the ideal gas constant. For battery temperature, It is a function of reactant concentration.

[0054] Based on the above reactions, the system energy conservation equation is constructed as follows: ; in, For battery quality, For specific heat capacity, The reaction enthalpy of each reaction, The convective heat transfer coefficient is... This represents the heat dissipation area.

[0055] Furthermore, for the NMC 18650 type ternary lithium battery in this example, its electrolyte consists of 70% dimethyl phosphate (DMC) and 30% ethylene carbonate (EC) by mass. The gas production pressure equation is constructed, and its saturated vapor pressure is as follows: ; ; The partial pressures of non-condensable gases are as follows: ; ; Based on the above embodiments, as an optional embodiment, the reaction kinetic neural network model is a single hidden layer neural network, including a heat production rate neuron, a gas production rate neuron, an input layer for receiving the logarithm of reactant concentration and temperature parameters, and an output layer for outputting the concentration change rate of each component, the total heat production rate and the total gas production rate. The weight and bias parameters of a single hidden layer neural network are constrained to the physical parameters of the Arrhenius equation, which include the activation energy, pre-exponential factor, reaction order, heat production enthalpy, and gas production enthalpy. The forward propagation process of a single hidden layer neural network is equivalent to the calculation process of the Arrhenius equation.

[0056] Specifically, in the embodiments of this application, the reaction kinetics neural network model includes heat production rate neurons and gas production rate neurons. The structure of the heat production rate neuron is as follows: Figure 2 As shown, the solid lines represent the fixed weight parameters of the neural network, which are initialized by the network and remain unchanged during training; the dashed lines represent the trainable weight parameters of the neural network, corresponding to the parameters to be identified in the Arrhenius equation, including... , , The input to this neuron includes the logarithm of reactant concentration. molar gas constant ,temperature Other parameters. Here, for the reaction between the negative electrode and the electrolyte, the input also includes the ratio of the SEI film thickness to the characteristic size of the active material. After nonlinear transformation, the input signal is finally output as the instantaneous heat generation rate and reactant concentration change rate of the reaction path.

[0057] In the embodiments of this application, the structure of the gas production rate neuron is as follows: Figure 3 As shown, the gas production rate neuron directly incorporates the output of the heat production rate neuron as its input, and its input comes from the reactant concentration change rate of each reaction pathway. Gas enthalpy change parameters These are trainable weight parameters, and the output is the gas production rate. The output of the heat production rate neuron serves as the input of the gas production rate neuron, thus coupling two independent physical processes into a single, cooperating neuronal unit. Figure 2 and Figure 3 The neurons representing heat production rate and gas production rate together constitute the neurons of the reaction kinetics neural network model.

[0058] In the embodiments of this application, a reaction dynamics neural network structure with physical constraints is constructed. The complete architecture of the reaction dynamics neural network model is as follows: Figure 4 As shown, its core is a feedforward layer containing four customized neurons. Each neuron integrates a specific reaction dynamics equation, enabling simultaneous calculation of four main reaction pathways. The neurons in the reaction dynamics neural network model are... Figure 2 and Figure 3 The combination of these reactions indicates that thermal runaway involves four main reactions. Figure 4 It contains four neurons, each corresponding to a rate equation for the decomposition reaction at the solid electrolyte interface (SEI). Rate equation for the reaction between the negative electrode and the electrolyte Rate equation for the reaction between the positive electrode and the electrolyte and the rate equation for electrolyte decomposition reaction .

[0059] More specifically, such as Figure 4 As shown, the input layer of the reaction kinetics neural network model receives the initial state of each reactant and the environmental parameter vector, including the normalized logarithmic concentration term of each reactant component. Current battery temperature T (corresponding to the aforementioned parameters) and the ideal gas constant The hidden layer undergoes a nonlinear physical transformation, and its weight parameters and bias terms are not randomly initialized but directly correspond to the physical parameter set. Its network output layer is used to output the state derivative of the system, i.e., the rate of change of the concentration of each component. Total gas production rate Total heat production rate Q (corresponding to the parameters above) ), and the rate of temperature rise .

[0060] Here, the network weight parameters of the reaction kinetics neural network model are determined based on the kinetic parameters of the thermal runaway thermo-gas coupling model. Specifically, the network weight parameters (including the weight matrix and bias vector) of the reaction kinetics neural network model are constrained to the kinetic physical parameters of the Arrhenius equation in the thermal runaway thermo-gas coupling model, including the reaction activation energy. Pre-exponential factors Logarithmic concentration term reaction order Logarithmic concentration term (1- reaction order enthalpy of heat production reaction Enthalpy of gas production reaction In the reaction dynamics neural network model, the forward propagation process of a single hidden layer neural network can be directly equivalent to the calculation process of the Arrhenius equation.

[0061] The method in this application constructs a shared reaction kinetics kernel-driven thermal system by expressing heat generation power and gas generation rate as functions of the same set of reaction kinetic parameters. The gas-air coupling model explicitly maps the multi-step Arrhenius reaction mechanism into a neural network unit topology. This allows the network weight parameters of the reaction kinetics neural network model to directly correspond to the physical parameters of the Arrhenius equation in the thermal-gas coupling model of thermal runaway. Furthermore, the trainable parameters of the network directly correspond to physical parameters such as activation energy and enthalpy of reaction, enabling direct solution of the equations for the thermal runaway physical process. Compared to traditional data fitting methods, this approach can effectively achieve accurate end-to-end identification of kinetic parameters, improving the inversion and identification accuracy of lithium battery thermal runaway processes. Simultaneously, the model training process does not require a large amount of training data, resulting in high training efficiency.

[0062] Furthermore, in the embodiments of this application, a battery temperature update model is constructed based on the lithium battery temperature update equation, and a battery temperature prediction neural network model is obtained by combining the reaction kinetics neural network model and the battery temperature update model. Here, the expression of the battery temperature update equation is as follows: ; In the formula, For battery quality; This refers to the specific heat capacity of the battery. This refers to the battery volume.

[0063] Therefore, the expression for the battery temperature update model can be obtained as follows: .

[0064] In the embodiments of this application, the reaction kinetics neural network model and the battery temperature update model are connected sequentially to form a battery temperature prediction neural network model.

[0065] The method in this application constructs a reaction kinetic neural network model by mapping the Arrhenius equation describing the rate of chemical reactions inside the battery to neurons in a neural network, and combines it with a battery temperature update model to construct a battery temperature prediction neural network model. By training the model with experimental data samples of thermal runaway of lithium batteries, the thermal runaway kinetic parameters can be optimized and updated, improving the accuracy and convenience of identifying thermal runaway parameters.

[0066] In existing technologies, directly applying conventional PINN to model thermal runaway in lithium batteries still faces key challenges. First, the exponential abrupt change in reaction rate during thermal runaway in lithium batteries causes the governing equations to exhibit extreme numerical rigidity. This makes it difficult for conventional networks to accurately satisfy the differential equation constraints under discrete collocation, resulting in a geometric increase in computational load. This is one of the key reasons why it is impossible to achieve a direct and accurate inversion from experimental data to kinetic parameters.

[0067] To address the numerical stiffness problem near the thermal runaway point, this application also proposes a model training method based on independent variable substitution and a two-stage solution strategy.

[0068] Based on the above embodiments, as an optional embodiment, the thermal runaway experimental data sample includes thermal runaway initial state data samples and corresponding full-time temperature and pressure data labels; step S2 includes: Step S21: For any thermal runaway experimental data sample, based on the independent variable substitution and two-stage solution strategy, input the thermal runaway initial state data sample and the corresponding full-time temperature and pressure data labels into the battery temperature prediction neural network model to obtain the full-time battery temperature prediction information and gas production rate prediction information corresponding to the thermal runaway experimental data sample. Step S22: Using a preset loss function, the loss value is calculated based on the battery temperature prediction information and gas production rate prediction information for the entire time period, as well as the temperature and pressure data labels for the entire time period, to obtain the loss value corresponding to the thermal runaway experimental data sample. Step S23: Based on the loss value, iteratively update the weight parameters of the reaction dynamics neural network model using the gradient backpropagation algorithm until the preset loss function converges or the maximum number of iterations is reached, thus obtaining the trained battery temperature prediction neural network model.

[0069] Specifically, in the embodiments of this application, during the process of training the neural network model for predicting battery temperature using multiple sets of thermal runaway experimental data samples of lithium batteries, a network training based on independent variable substitution and a two-stage strategy is adopted.

[0070] More specifically, in the embodiments of this application, in step S21, firstly, based on the independent variable substitution and two-stage solution strategy, the thermal runaway initial state data sample and the corresponding full-time temperature and pressure data label carried by any sample in multiple sets of thermal runaway experimental data samples are input into the battery temperature prediction neural network model. After processing by the battery temperature prediction neural network model, the full-time battery temperature prediction information and gas production rate prediction information corresponding to the thermal runaway experimental data sample can be obtained.

[0071] The independent variable substitution and two-stage strategy used in this embodiment is as follows: First, in the main reaction stage of thermal runaway, the integral variable is replaced by temperature. That is, the control equations are transformed using the chain rule, as follows: ; in, These are state variables other than temperature.

[0072] In this embodiment, through this transformation, the function curve, which was originally extremely steep on the time axis, becomes smooth on the temperature axis, eliminating the root cause of gradient explosion. The numerical stiffness problem in solving the thermal runaway equation can be solved through the above-described temperature domain integral transformation.

[0073] In the numerical simulation of battery thermal runaway, the governing equations can be expressed as a set of ordinary differential equations (ODEs). To distinguish them from subsequent new ODE systems, the governing equations here are expressed as the old ODE system, as follows: ; In this embodiment, a method of independent variable substitution is used. Within the range of monotonically increasing system temperature, the integral independent variable is changed from time... Replace with temperature Define a new state variable in the new coordinate system. ,as follows: ; By the chain rule described above, for any component have: ; Based on the above discussion, this application embodiment constructs a new ODE system that is equivalent to the original system and integrates in the temperature domain, as follows: ; Based on the above derivation, the new integral formula for the ODE system is: ; In the formula, Phase 1 The battery temperature at that moment, i.e., the initial battery temperature.

[0074] The old ODE system integral formula was: ; In the formula, This is the initial time for the second phase.

[0075] Next, as Figure 5As shown, a two-stage solution strategy is employed for training. During the monotonically increasing temperature phase, integration is performed with temperature as the independent variable, employing the new ODE system described above. Upon completion of the reaction and entering the cooling phase, the final state calculated in the first stage is used as the initial condition, switching to a conventional differential equation with time as the independent variable, employing the old ODE system. Ultimately, the model outputs full-time battery temperature and gas production rate predictions corresponding to the thermal runaway experimental data samples.

[0076] Further, in the embodiments of this application, in step S22, a loss function and parameter inversion are constructed. During the training process, the Adam optimizer is used to minimize the composite loss function. Based on the gas production rate prediction information for the entire time period, the mass of gas produced by the decomposition reaction can be calculated using the predicted gas production rate data. Based on the aforementioned formula for the partial pressure of non-condensable gases, and combined with the battery temperature at the current moment, the predicted information for non-condensable gas pressure can be further calculated. Thus, the predicted information for non-condensable gas pressure for the entire time period can be calculated. Then, the predicted information for battery temperature and non-condensable gas pressure for the entire time period can be substituted into the preset loss function to calculate the loss value. Through loss analysis with the corresponding temperature and pressure data labels for the entire time period, the loss value corresponding to the thermal runaway experimental data sample is obtained.

[0077] Furthermore, in an embodiment of this application, in step S23, based on the loss value, the weight parameters of the reaction dynamics neural network model are iteratively updated using a gradient backpropagation algorithm. Following the above method, using thermal runaway experimental data samples from each group of lithium batteries, round by round ( Epoch = Epoch + 1) Iteratively update the weight parameters of the reaction kinetics neural network model until the model training reaches the maximum number of iterations or the loss value converges, and then stop training to obtain a trained battery temperature prediction neural network model.

[0078] Understandably, the full-time temperature and pressure data tags can specifically include full-time temperature data tags and full-time non-condensable gas pressure data tags collected in lithium battery thermal runaway experiments.

[0079] At this point, the network weight parameters of the reaction kinetics neural network model are the optimal kinetic parameters of the battery, including... .

[0080] The method in this application embodiment uses a large number of thermal runaway experimental data samples to repeatedly train the reaction dynamics neural network model in the battery temperature prediction neural network model, thereby controlling the model loss value within the convergence range, which helps to improve the accuracy of the lithium battery thermal runaway inversion results output by the model.

[0081] Based on the above embodiments, as an optional embodiment, the two-stage solution strategy includes a first stage of monotonically increasing battery thermal runaway temperature, and a second stage of battery cooling or non-monotonic temperature rise; step S21 includes: Step S211: In the first stage, the current first battery state data is input into the battery temperature prediction neural network model to obtain the battery temperature prediction information for the next moment, so as to determine the first battery state data for the next moment; the current first battery state data includes the battery state data carried by the thermal runaway initial state data sample. Step S212: When the battery temperature prediction information for the next moment indicates that the rate of change of battery temperature is not greater than zero, proceed to the second stage. Step S213: In the second stage, the current second battery state data is input into the battery temperature prediction neural network model to obtain the battery temperature prediction information for the next moment. Step S214: Upon determining the end of the second stage, combine the battery temperature prediction information and gas production rate prediction information obtained at different times in the first stage with the battery temperature prediction information and gas production rate prediction information obtained at different times in the second stage to obtain the full-time battery temperature prediction information and gas production rate prediction information corresponding to the thermal runaway experimental data sample.

[0082] Specifically, in the embodiments of this application, in step S211, in the first stage, the current first battery state data is input into the battery temperature prediction neural network model. Through the processing of the battery temperature prediction neural network model, the battery temperature prediction information at the next moment can be obtained, thereby determining the first battery state data at the next moment.

[0083] Specifically, the first battery state data described in the embodiments of this application refers to the battery state data input to the reaction kinetics neural network model at each moment in the first stage, which specifically includes the logarithm of the reactant concentrations of each reaction. ,temperature and molar gas constant .

[0084] It is understood that the second battery state data described in the embodiments of this application refers to the battery state data input to the reaction kinetics neural network model at each moment in the second stage, which specifically includes the logarithmic concentrations of reactants for each reaction. ,temperature and molar gas constant .

[0085] Here, the reaction dynamics neural network model determines the input for the current moment based on the output of the previous moment at each time step, until the first stage of the solution is completed.

[0086] In some embodiments, step S211 includes: In the first stage, the current state data of the first battery is input into the reaction kinetics neural network model, and the heat generation rate prediction information corresponding to the current state data of the first battery is output. The heat generation rate prediction information corresponding to the current first battery state data is input into the battery temperature update model, and the independent variable of the battery temperature update model is replaced by temperature to perform numerical integration and solve to obtain the battery temperature prediction information at the next moment, so as to determine the first battery state data at the next moment.

[0087] Specifically, in the embodiments of this application, during the iterative training of the battery temperature prediction neural network model, the numerical integration process of the differential equations involved in the reaction kinetics neural network model is divided into two consecutive stages, such as... Figure 5 As shown, the two-stage solution strategy includes a first stage in which the battery thermal runaway temperature rises monotonically, and a second stage in which the battery cools or experiences a non-monotonic temperature rise.

[0088] In the embodiments of this application, during the first stage of the monotonically rising battery thermal runaway temperature, the current first battery state data is input into the reaction kinetic neural network model in the battery temperature prediction neural network model, and the predicted information corresponding to the current first battery state data is output. This predicted information includes reactant concentration change rate prediction information, heat generation rate prediction information, and gas generation rate prediction information. The current first battery state data includes battery state data carried by the initial thermal runaway state data sample, i.e., including… Battery temperature at any time Logarithmic concentration parameters of reactants and .

[0089] Specifically, continue to refer to Figure 5 The first stage is based on the initial state of the battery (including...) Battery temperature at any time Logarithmic concentration of reactants and molar gas constant As the initial input, the governing equations are numerically calculated along the temperature dimension through forward propagation of the reaction kinetics neural network. The governing equations corresponding to this stage are the aforementioned new ODE system. The output of the reaction kinetics neural network model is the rate of change of the concentrations of the four reactants. The rate of change in the quality of gas production (i.e., gas production rate) and heat production rate The corresponding prediction data are: reactant concentration change rate prediction information, heat production rate prediction information, and gas production rate prediction information.

[0090] Next, based on the aforementioned expression of the battery temperature update model, the heat generation rate prediction information corresponding to the current first battery state data is input into the battery temperature update model in the battery temperature prediction neural network model. Then, using the new ODE system integral formula, the independent variable of the battery temperature update model is replaced by temperature instead of time for numerical integration to obtain the next time step. The battery temperature prediction information is used to determine the first battery state data at the next moment, including... Predicted battery temperature and reactant logarithmic concentration at any given time and Corresponding predicted data and molar gas constant Among them, the logarithmic concentration of reactants and The corresponding prediction data can be calculated based on the reactant concentration change rate prediction information; the gas production rate prediction information can be used to determine... The pressure prediction information of non-condensable gases obtained from real-time simulation.

[0091] In the embodiments of this application, according to the above calculation method, the input of the reaction kinetics neural network model at each time step is determined by the output of the previous time step, until the first stage of solution is completed. It can be understood that the first battery state data input at each time step includes the battery temperature determined at the previous time step. Predictive information, logarithmic concentration of reactants and Predictive information, and molar gas constant .

[0092] The method in this application embodiment trains the reaction dynamics neural network model in the first stage by using independent variable substitution, transforming the rigid differential equation in the time domain into a smooth integral form in the temperature domain. This effectively mitigates the numerical rigidity problem during the period of drastic reaction change and avoids the risk of gradient explosion or disappearance during the identification of thermal runaway parameters.

[0093] Further, in an embodiment of this application, in step S212, when it is determined that the battery temperature prediction information for the next time step indicates that the battery temperature change rate is not greater than zero, the process proceeds to the second stage. Here, when it is determined that the battery temperature prediction information for the next time step indicates that the battery temperature change rate is not greater than zero, the process proceeds to the second stage of battery cooling or non-monotonic temperature rise. Specifically, when it is determined... When the time is reached, it indicates the end of the first stage of the solution. The output of this stage is defined as the final state of the first stage, and the final state data includes the battery temperature at that moment. and the concentration of each reactant Ultimately, in the first stage, we can obtain [the following]: Battery temperature prediction and gas production rate prediction information at different starting points. Here, at the end of the first stage... , This indicates the duration of the first phase.

[0094] In the embodiments of this application, in step S213, the current second battery state data is input into the battery temperature prediction neural network model to obtain the battery temperature prediction information for the next moment. Specifically, in the second stage, the current second battery state data is first input into the reaction kinetics neural network model, which can output the heat generation rate prediction information corresponding to the current second battery state data; then, the heat generation rate prediction information corresponding to the current second battery state data is input into the battery temperature update model to obtain the battery temperature prediction information for the next moment. Furthermore, when the current moment reaches the last moment in the full-time temperature and pressure data tag data, the end of the second stage is determined, and the battery temperature prediction information and gas generation rate prediction information obtained in the second stage at different moments can be acquired. Here, the end time of the second stage... , Indicates the duration of the second phase. .

[0095] Here, the current second battery state data includes the first battery state data at the next moment. The first battery state data at the next moment is the data from the end of stage 1, i.e., the final state data of the first stage. In other words, the initial input data for the second stage is the final state data of the first stage.

[0096] At the end of the second stage, the battery temperature prediction information and gas production rate prediction information obtained at different times in the first stage, as well as the battery temperature prediction information and gas production rate prediction information obtained at different times in the second stage, are combined in time series to obtain the full-time battery temperature prediction information and gas production rate prediction information corresponding to the thermal runaway experimental data sample.

[0097] It should be noted that the final state variables of the first stage are used as the initial conditions for the second stage. The input and output terms of the reaction kinetics neural network model are the same as those in the first stage. In this stage, solving the old ODE system integral formula can yield the battery temperature at different times. Forecast information.

[0098] Furthermore, in an embodiment of this application, in step S214, upon determining the end of the second stage, the battery temperature prediction information and gas production rate prediction information obtained at different times in the first stage, and the battery temperature prediction information and gas production rate prediction information obtained at different times in the second stage, are combined in a time sequence to obtain the thermal runaway experimental data sample corresponding to the data from... The battery temperature prediction information and gas production rate prediction information for the entire time period from the start time to the end time.

[0099] The method in this application, by considering the temporal variation characteristics of lithium battery thermal runaway, divides the model training into a first stage of monotonically increasing battery thermal runaway temperature and a second stage of battery cooling or non-monotonic temperature rise, and combines the independent variable substitution method used in the first stage to solve the battery temperature microequation. This achieves network training based on independent variable substitution and a two-stage strategy, which can effectively solve the numerical rigidity problem near the thermal runaway outbreak point, significantly improve computational efficiency and numerical stability, and enhance the robustness of lithium battery thermal runaway inversion and identification.

[0100] Based on the above embodiments, as an optional embodiment, the preset loss function used by the battery temperature prediction neural network model in each round of model training includes a temperature loss function and a pressure loss function. The temperature loss function is determined by a weighted calculation based on the root mean square error of temperature reversal time and the maximum temperature loss error; the root mean square error of temperature reversal time is determined by calculating the root mean square of the difference between the time required for model simulation and the experimental recording time at the same temperature test time. The pressure loss function is determined by a weighted calculation based on the root mean square error of the pressure reversal time and the pressure relief time error; the root mean square error of the pressure reversal time is determined by the root mean square of the difference between the model simulation time and the experimental recording time at the same pressure test time.

[0101] Specifically, in the embodiments of this application, during the training process of the battery temperature prediction neural network model, the loss function is divided into two parts: temperature loss and pressure loss. The preset loss function used in each round of model training includes the temperature loss function and the pressure loss function.

[0102] The temperature loss function is based on the root mean square error of the temperature reverse time. Error due to maximum temperature loss The root mean square error of temperature reversal time is determined by weighted calculation. It is determined by calculating the root mean square of the difference between the time required for model simulation and the experimental recording time at the same temperature test time.

[0103] Based on the above embodiments, as an optional embodiment, the temperature loss function is calculated as follows: ; In the formula, Represents the temperature loss function. This represents the weighting coefficient for the maximum temperature error; The root mean square error of temperature reversal time is calculated as follows: ; In the formula, This represents the root mean square error of the temperature reversal time. Indicates the first round of model training The simulation requires the time to measure the battery temperature at each test moment. This indicates the first [number]th [item] in the corresponding thermal runaway experimental data sample. Experimental recording time at each battery temperature test moment; The calculation method for the maximum temperature loss error is as follows: ; In the formula, This represents the root mean square error of the temperature reversal time. This indicates the error due to the highest temperature loss. This represents the highest temperature simulated during a single round of model training. This represents the highest temperature measured in the corresponding thermal runaway experimental data sample.

[0104] Here, it is understandable that the parameters and All of these can be identified and obtained from the full-time temperature data tags carried by the thermal runaway experimental data samples.

[0105] Furthermore, in the embodiments of this application, the pressure loss function Based on the root mean square error of pressure reverse time and pressure relief time error The root mean square error of the pressure reverse time is determined by weighted calculation. It is determined by the root mean square of the difference between the model simulation time and the experimental recording time at the same stress test time.

[0106] More specifically, based on the above embodiments, as an optional embodiment, the pressure loss function is calculated as follows: ; In the formula, Represents the pressure loss function. This represents the weighting coefficient for the pressure relief time error; The root mean square error of the pressure reversal time is calculated as follows: ; In the formula, This indicates the root mean square error of the pressure reversal time. Indicates the first round of model training The simulated pressure of non-condensable gas at a specific pressure test moment. This indicates the first [number]th [item] in the corresponding thermal runaway experimental data sample. The pressure of the non-condensable gas measured in the experiment at each pressure test moment; The calculation method for the pressure relief time error is as follows: ; In the formula, This indicates the error in the pressure relief time. This represents the total decompression time during a single round of model training. This represents the total decompression time measured in the corresponding thermal runaway experimental data sample.

[0107] Similarly, it is understandable that the parameters and All of these can be identified and obtained from the non-condensable gas pressure data tags carried in the thermal runaway experimental data samples throughout the entire time period.

[0108] The method in this application embodiment constructs a preset loss function by introducing the root mean square error of temperature reverse time and the root mean square error of pressure reverse time. This can effectively prevent drastic temperature changes caused by a small deviation in time, further avoid affecting the gradient stability during model training, and improve the generalization ability and robustness of the reaction dynamics neural network model.

[0109] Figure 6 This paper presents a comparison between the temperature change curves calculated by the thermal runaway model based on the Arrhenius equation and the experimental measurements. From... Figure 6 As can be seen, the model calculations and experimental data show a high degree of consistency throughout the thermal runaway process. Specifically, the process can be divided into three stages: (1) the initial slow self-heating stage (approximately 700s to 1250s), in which the system temperature slowly climbs from approximately 90℃ to 120℃. The model curve and the experimental curve almost perfectly overlap, indicating that the model can accurately capture the self-heating process dominated by the initial exothermic reaction such as SEI film decomposition. (2) the accelerated heating stage (approximately 1250s to 1340s), in which the rate of temperature rise increases significantly. The model accurately predicted this accelerated trend. Although the predicted temperature is slightly higher than the experimental value, the overall slope change maintains good consistency with the experimental results. (3) the thermal runaway burst stage (after approximately 1340s), in which the system temperature soars from approximately 150℃ to over 700℃ in a very short time, exhibiting typical thermal runaway characteristics.

[0110] Figure 7The diagram clearly reveals the chain reaction characteristics of thermal runaway: when the system temperature reaches a certain threshold, a series of violent chemical reactions with high activation energy and high exothermic rates are simultaneously triggered. These reactions release enormous energy in a short period, leading to system temperature runaway, which macroscopically manifests as an explosive increase in temperature. Therefore, this diagram, from the perspective of microscopic reactant consumption, shows the reaction parameters... , , , The numerical changes powerfully explain the intrinsic dynamic mechanism of thermal runaway triggering. The identified activation energy parameters strictly follow the sequential reaction law of SEI film decomposition, negative electrode reaction with electrolyte, positive electrode reaction with electrolyte, and electrolyte decomposition reaction, effectively decoupling the complex multi-step chain reaction mechanism inside the battery.

[0111] The lithium battery thermal runaway identification device provided in this application is described below. The lithium battery thermal runaway identification device described below can be referred to in correspondence with the lithium battery thermal runaway identification method described above.

[0112] Figure 8 This is a schematic diagram of the structure of the lithium battery thermal runaway identification device provided in the embodiments of this application, as shown below. Figure 8 As shown, it includes: Model building module 10 is used to build a neural network model for predicting battery temperature using the Arrhenius equation and the lithium battery temperature update equation. The model training module 20 is used to train the battery temperature prediction neural network model using multiple sets of thermal runaway experimental data samples of lithium batteries, so as to obtain the trained battery temperature prediction neural network model. The temperature prediction module 30 is used to input the logarithmic information of the current reactant concentration and the current battery temperature data of the lithium battery under test into the battery temperature prediction neural network model to obtain the battery temperature data of the lithium battery under test at the next moment. The thermal runaway identification module 40 is used to perform thermal runaway analysis based on the battery temperature data at the next moment and its previous historical battery temperature data, and to identify the thermal runaway inversion result of the lithium battery under test.

[0113] It is understood that the detailed functional implementation of each of the above units / modules can be found in the description in the aforementioned method embodiments, and will not be repeated here.

[0114] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0115] The lithium battery thermal runaway identification device of this application deeply explores the mapping relationship between the unified paradigm of the Arrhenius equation and the topology of neural network units. It explicitly maps the multi-step Arrhenius equation reaction mechanism into the topology of neural network units and constructs a battery temperature prediction neural network model in conjunction with the lithium battery temperature update equation. This achieves a deep structural integration of the neural network weight parameters and the physical mechanism of thermal runaway, so that the trainable parameters of the neural network directly correspond to the thermal runaway dynamic parameters. This makes the model training process using thermal runaway experimental data samples a physical solution optimization process for thermal runaway dynamic parameters, rather than a traditional data fitting process. This avoids data fitting errors and achieves accurate identification of thermal runaway dynamic parameters, greatly improving the inversion accuracy. As a result, it can realize real-time accurate inversion of the temperature change trend of the lithium battery thermal runaway process, and improve the identification accuracy of the lithium battery thermal runaway process.

[0116] Based on the methods in the above embodiments, this application provides an electronic device, such as... Figure 9 As shown, the electronic device may include a processor 910, a communications interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communications interface 920, and the memory 930 communicate with each other via the communication bus 940. The processor 910 can call logical instructions in the memory 930 to execute the methods in the above embodiments.

[0117] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0118] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0119] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0120] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0121] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0122] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0123] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0124] It should be understood that expressions such as “comprising” and “may include” used in this application indicate the existence of the disclosed functions, operations, or constituent elements, and do not limit one or more additional functions, operations, and constituent elements. In this application, terms such as “comprising” and / or “having” are to be interpreted as indicating a particular characteristic, number, operation, constituent element, component, or combination thereof, but not to exclude the existence or possibility of adding one or more other characteristics, numbers, operations, constituent elements, components, or combinations thereof.

[0125] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying thermal runaway in lithium batteries, characterized in that, include: Step S1: Construct a neural network model for predicting battery temperature using the Arrhenius equation and the lithium battery temperature update equation. Step S2: The battery temperature prediction neural network model is trained using multiple sets of thermal runaway experimental data samples of lithium batteries to obtain a trained battery temperature prediction neural network model. Step S3: Input the logarithmic information of the current reactant concentration and the current battery temperature data of the lithium battery under test into the battery temperature prediction neural network model to obtain the battery temperature data of the lithium battery under test at the next moment. Step S4: Perform thermal runaway analysis based on the battery temperature data at the next moment and its previous historical battery temperature data to identify the thermal runaway inversion result of the lithium battery under test.

2. The lithium battery thermal runaway identification method according to claim 1, characterized in that, Step S1 includes: A thermo-gas coupling model for thermal runaway of lithium-ion batteries was established using the Arrhenius equation. The thermal runaway thermo-gas coupling model is incorporated into the neuron structure of a neural network to construct a reaction dynamics neural network model; A battery temperature update model is constructed based on the lithium battery temperature update equation, and a battery temperature prediction neural network model is obtained by combining the reaction kinetics neural network model and the battery temperature update model; the network weight parameters of the reaction kinetics neural network model are determined based on the kinetic parameters of the thermal runaway thermo-gas coupling model.

3. The lithium battery thermal runaway identification method according to claim 2, characterized in that, The reaction kinetics neural network model is a single hidden layer neural network, including a heat production rate neuron, a gas production rate neuron, an input layer for receiving the logarithm of reactant concentration and temperature parameters, and an output layer for outputting the concentration change rate of each component, the total heat production rate, and the total gas production rate. The weight parameters and bias parameters of the single hidden layer neural network are constrained to the physical parameters of the Arrhenius equation, which include the activation energy of the reaction, the pre-exponential factor, the reaction order, the enthalpy of the heat-producing reaction, and the enthalpy of the gas-producing reaction; the forward propagation process of the single hidden layer neural network is equivalent to the calculation process of the Arrhenius equation.

4. The lithium battery thermal runaway identification method according to claim 2, characterized in that, The thermal runaway experimental data sample includes thermal runaway initial state data samples and corresponding full-time temperature and pressure data labels; step S2 includes: Step S21: For any of the thermal runaway experimental data samples, based on the independent variable substitution and two-stage solution strategy, the thermal runaway initial state data sample and the corresponding full-time temperature and pressure data labels are input into the battery temperature prediction neural network model to obtain the full-time battery temperature prediction information and gas production rate prediction information corresponding to the thermal runaway experimental data sample. Step S22: Using a preset loss function, the loss value is calculated based on the battery temperature prediction information and gas production rate prediction information for the entire time period, as well as the temperature and pressure data labels for the entire time period, to obtain the loss value corresponding to the thermal runaway experimental data sample. Step S23: Based on the loss value, the weight parameters of the reaction dynamics neural network model are iteratively updated using the gradient backpropagation algorithm until the preset loss function converges or the maximum number of iterations is reached, thereby obtaining the trained battery temperature prediction neural network model.

5. The lithium battery thermal runaway identification method according to claim 4, characterized in that, The two-stage solution strategy includes a first stage and a second stage; step S21 includes: Step S211: In the first stage, the current first battery state data is input into the battery temperature prediction neural network model to obtain the battery temperature prediction information for the next moment, so as to determine the first battery state data for the next moment; the current first battery state data includes the battery state data carried by the thermal runaway initial state data sample. Step S212: When the battery temperature prediction information at the next moment indicates that the battery temperature change rate is not greater than zero, proceed to the second stage; Step S213: In the second stage, the current second battery state data is input into the battery temperature prediction neural network model to obtain the battery temperature prediction information for the next moment. Step S214: Upon determining the end of the second stage, the battery temperature prediction information and gas production rate prediction information obtained at different times in the first stage, as well as the battery temperature prediction information and gas production rate prediction information obtained at different times in the second stage, are combined in a time sequence to obtain the full-time battery temperature prediction information and gas production rate prediction information corresponding to the thermal runaway experimental data sample.

6. The lithium battery thermal runaway identification method according to claim 5, characterized in that, Step S211 includes: In the first stage, the current first battery state data is input into the reaction kinetics neural network model, and the heat generation rate prediction information corresponding to the current first battery state data is output. The heat generation rate prediction information corresponding to the current first battery state data is input into the battery temperature update model, and the independent variable of the battery temperature update model is replaced by temperature and numerical integration is performed to obtain the battery temperature prediction information at the next moment, so as to determine the first battery state data at the next moment.

7. The lithium battery thermal runaway identification method according to any one of claims 1-6, characterized in that, The preset loss functions used by the battery temperature prediction neural network model in each round of model training include a temperature loss function and a pressure loss function. The temperature loss function is determined by a weighted calculation based on the root mean square error of temperature reversal time and the maximum temperature loss error; the root mean square error of temperature reversal time is determined by calculating the root mean square of the difference between the time required for model simulation and the experimental recording time at the same temperature test time. The pressure loss function is determined by a weighted calculation based on the root mean square error of the pressure reversal time and the pressure relief time error; the root mean square error of the pressure reversal time is determined by the root mean square of the difference between the model simulation time and the experimental recording time at the same pressure test time.

8. A lithium battery thermal runaway identification device, characterized in that, include: The model building module is used to build a neural network model for predicting battery temperature using the Arrhenius equation and the lithium battery temperature update equation. The model training module is used to train the battery temperature prediction neural network model using multiple sets of thermal runaway experimental data samples of lithium batteries, so as to obtain a trained battery temperature prediction neural network model. The temperature prediction module is used to input the logarithmic information of the current reactant concentration and the current battery temperature data of the lithium battery under test into the battery temperature prediction neural network model to obtain the battery temperature data of the lithium battery under test at the next moment. The thermal runaway identification module is used to perform thermal runaway analysis based on the battery temperature data at the next moment and its previous historical battery temperature data, and to identify the thermal runaway inversion result of the lithium battery under test.

9. An electronic device, characterized in that, Includes memory and one or more processors; The memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions; The one or more processors invoke the computer instructions to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium comprising instructions, characterized in that: When the instructions are executed on an electronic device, the electronic device causes the electronic device to perform the method as described in any one of claims 1-7.

Images