Battery temperature field prediction model training method and battery temperature field prediction method

Abstract

The embodiment of the invention provides a training method of a battery temperature field prediction model and a battery temperature field prediction method. The method comprises the following steps: generating a thermal runaway example based on a reference working condition parameter and a boundary condition temperature of a battery pack; determining characteristic parameters respectively corresponding to a plurality of modeling nodes of the battery pack in the thermal runaway example; according to the characteristic parameters corresponding to the plurality of modeling nodes, screening out modeling nodes meeting a predetermined condition from the plurality of modeling nodes as characteristic points; spatial distances and temperature variation amplitudes between the feature points and other modeling nodes meet preset conditions; generating a training set according to the reference working condition parameter and the feature parameter corresponding to each feature point; and training the battery temperature field prediction model based on the training set. According to the method, the technical problem that the model training efficiency is not ideal due to the contradiction between the battery temperature distribution gradient and the model modeling efficiency in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the technical field of battery thermal management, and in particular to a training method for a battery temperature field prediction model and a battery temperature field prediction method. Background Art

[0002] Lithium-ion batteries face some problems during the charge and discharge cycle, such as overcharge conditions that can easily lead to rapid battery temperature rise and even thermal runaway, posing a major threat to battery safety. Under complex conditions, electrochemical parameters such as battery capacity and internal resistance will change significantly, leading to more obvious nonlinear polarization phenomena, which in turn will increase internal resistance and heat generation. As heat accumulates, the temperature changes inside the battery intensify. If the temperature changes inside the battery are not accurately predicted and controlled, it may endanger the safe operation of the battery. Therefore, accurately predicting and controlling the temperature changes of lithium-ion batteries has become a key challenge to improve battery performance, ensure the safety of electric vehicles, and extend battery life.

[0003] At present, the related technologies for studying battery thermal behavior and predicting battery temperature models use physical models based on internal chemical reactions to calculate the relationship between heat generation and heat transfer for prediction. Since a large number of calculation parameters need to be processed under complex conditions, the calculation speed is slow, making it difficult to achieve real-time monitoring of battery abnormalities. Another data-driven approach to temperature prediction provided in related technologies relies heavily on the quality of the data set for its accuracy and generalization ability of deep learning models. Since it is difficult to achieve a balance between the data set's ability to represent temperature gradients and the impact of a large data set on training efficiency, there is a problem of unsatisfactory training efficiency for the battery temperature field prediction model, which in turn affects the ability to predict changes in the battery pack temperature field.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide a method for training a battery temperature field prediction model and a method for predicting a battery temperature field, so as to alleviate or solve the technical problem that it is difficult to balance the battery temperature distribution gradient and the contradiction between the modeling efficiency and the unsatisfactory model training efficiency in the related technology.

[0006] In a first aspect, an embodiment of the present application provides a method for training a battery temperature field prediction model, comprising: Generate thermal runaway examples based on the reference operating parameters and boundary condition temperatures of the battery pack; Determine characteristic parameters corresponding to multiple modeling nodes of a battery pack in a thermal runaway example, where the multiple modeling nodes are used to represent different spatial positions on the battery pack; According to the characteristic parameters respectively corresponding to the multiple modeling nodes, the modeling nodes meeting the predetermined conditions are selected from the multiple modeling nodes as the characteristic points; the spatial distance and the temperature variation amplitude between the characteristic points and other modeling nodes meet the predetermined conditions; Generate a training set according to the reference working condition parameters and the characteristic parameters corresponding to each characteristic point; Based on the training set, the battery temperature field prediction model is trained to obtain a trained battery temperature field prediction model.

[0007] In a second aspect, an embodiment of the present application provides a battery temperature field prediction method, comprising: Obtain the current operating parameters of the battery pack to be tested; The current operating condition parameters are input into the battery temperature field prediction model for processing to obtain the battery temperature field distribution of the battery pack at a future time, wherein the battery temperature field prediction model is obtained by applying any battery temperature field prediction model training method.

[0008] In a third aspect, an embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, and the program is executed by a processor to implement any of the methods described above.

[0009] In a fourth aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement any of the methods described above.

[0010] Based on the training method of the battery temperature field prediction model of the first aspect above, the present application has at least the following beneficial effects or advantages: By generating a thermal runaway example based on reference operating condition parameters and boundary condition temperatures, determining the characteristic parameters of the modeling nodes, and screening out characteristic points that meet predetermined conditions, this method can accurately locate key positions inside the battery pack, and at the same time generate a training set in combination with reference operating condition parameters, generate training data with representative operating conditions and high-confidence temperature field values, and then train a high-precision battery temperature field prediction model. The above technical means enable the model to more accurately reflect the temperature distribution of the battery pack under complex operating conditions, and for extreme situations such as thermal runaway, it can provide early warning and provide a more accurate decision-making basis for the battery management system. It significantly improves the accuracy of battery temperature field prediction, enhances the safety and reliability of the battery system, and provides strong support for efficient management and safe operation of batteries.

[0011] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments according to the present application and should not be regarded as limiting the scope of the present application.

[0013] Figure 1 A flow chart showing a method for training a battery temperature field prediction model according to an embodiment of the present application is shown; Figure 2 A battery modeling schematic diagram showing a battery temperature field prediction model training method according to an embodiment of the present application; Figure 3 A configuration curve diagram of a training method for a battery temperature field prediction model according to an embodiment of the present application is shown; Figure 4 A schematic diagram of temperature variation of a single node in a training method of a battery temperature field prediction model according to an embodiment of the present application is shown; Figure 5 A schematic diagram of a data preprocessing process of a training method for a battery temperature field prediction model according to an embodiment of the present application is shown; Figure 6 A schematic diagram of the Bi-LSTM network structure of a battery temperature field prediction model training method according to an embodiment of the present application is shown; Figure 7 A schematic diagram of the timing prediction process of the CNN-Bi-LSTM-AM network of the battery temperature field prediction model training method according to an embodiment of the present application is shown; Figure 8 A comparison diagram of the RMSE errors of the training method of the battery temperature field prediction model according to the embodiment of the present application is shown; Fig. 9 The MSE error comparison diagram of the training method of the battery temperature field prediction model of the embodiment of the present application is shown; Fig.10 A first comparison diagram showing the temperature field prediction effect for normal charging provided by an embodiment of the present application is shown; Fig.11 A second comparison diagram showing the temperature field prediction effect for normal charging provided by an embodiment of the present application; Fig.12 A third comparison diagram showing the prediction effect of the temperature field in the thermal runaway stage provided by an embodiment of the present application; Fig.13 A fourth comparison diagram showing the effect of predicting the temperature field in the thermal runaway stage provided by an embodiment of the present application; Fig.14 A comparison chart of the execution time of different algorithm inferences provided in the embodiments of the present application is shown; Fig.15 A flow chart showing a method for predicting a battery temperature field according to an embodiment of the present application is shown; Fig.16 A schematic diagram showing a flow chart of a battery temperature field prediction method according to an embodiment of the present application is shown; Fig.17 A schematic diagram showing the structure of a training device for a battery temperature field prediction model according to an embodiment of the present application; Fig.18 A schematic diagram showing the structure of a battery temperature field prediction device according to an embodiment of the present application; Fig.19 A block diagram of an electronic device provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0014] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the concept or scope of the present application. Therefore, the drawings and descriptions are considered to be exemplary in nature and not restrictive.

[0015] To facilitate understanding of the technical solutions of the embodiments of the present application, the following describes the related technologies of the embodiments of the present application. The following related technologies can be combined with the technical solutions of the embodiments of the present application as optional solutions, and they all belong to the protection scope of the embodiments of the present application.

[0016] In recent years, with the rapid development of the new energy industry, energy-saving and low-power electric vehicles (EVs) have become the focus of many researchers, and lithium-ion batteries have become one of the core technologies of these vehicles. However, lithium-ion batteries face some problems during the charge and discharge cycle. For example, as the number of cycles increases, the charge and discharge current will fluctuate significantly, and the charge and discharge time will be significantly shortened. In addition, overcharging conditions can easily lead to rapid battery temperature rise and even thermal runaway, posing a major threat to battery safety. Therefore, how to design accurate and efficient battery thermal management methods has become a key issue that needs to be solved urgently.

[0017] In battery temperature prediction, the physical model based on chemical reactions provided by related technologies is complex to calculate and difficult to monitor in real time, while the data-driven model has high requirements on data quality and limited generalization ability under complex working conditions. The above problems limit the generation efficiency and model performance of the battery temperature prediction model in practical applications.

[0018] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The several specific embodiments listed can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0019] Figure 1 A flow chart of a method for training a battery temperature field prediction model according to an embodiment of the present application is shown. Figure 1 As shown, the method may include steps S101 to S105.

[0020] Step S101: generating a thermal runaway calculation example based on reference operating condition parameters and boundary condition temperature of the battery pack; Step S102: determining characteristic parameters corresponding to a plurality of modeling nodes of a battery pack in a thermal runaway calculation example, wherein the plurality of modeling nodes are used to represent different spatial positions on the battery pack; Step S103: based on the characteristic parameters respectively corresponding to the multiple modeling nodes, a modeling node that meets the predetermined conditions is selected from the multiple modeling nodes as a characteristic point; the spatial distance and the temperature variation amplitude between the characteristic point and other modeling nodes meet the predetermined conditions; Step S104: generating a training set according to the reference working condition parameters and the characteristic parameters corresponding to each characteristic point; Step S105: Based on the training set, the battery temperature field prediction model is trained to obtain a trained battery temperature field prediction model.

[0021] In an embodiment of the present application, the thermal runaway process of the battery pack under different working conditions can be simulated. For example, the initial temperature, charge and discharge rate, ambient temperature, coolant temperature and other parameters of the battery pack are set to simulate the thermal runaway phenomenon of the battery under overcharge, short circuit or high temperature environment, and generate a thermal runaway example. In the three-dimensional model of the battery pack, multiple modeling nodes can be divided according to the spatial layout and structure of the battery, and the multiple modeling nodes are used to represent different spatial positions on the battery pack, so as to facilitate the subsequent calculation of the temperature change of each modeling node through the electrochemical-thermal coupling model. By setting and screening to meet predetermined conditions, such as the balance between the spatial distance between the feature point and other nodes and the temperature change amplitude. Effectively reduce the computational complexity, while retaining the information most sensitive to the temperature field change, and improve the computational efficiency and prediction accuracy of the model. Combining the reference operating condition parameters with the characteristic parameters of the characteristic points to generate a training set helps to fully reflect the temperature field change law of the battery pack under different working conditions, provide high-quality data support for model training, and improve the generalization ability and prediction accuracy of the model. The battery temperature field prediction model obtained through training can accurately predict the temperature distribution of the battery pack under different operating conditions, warn of thermal runaway risks in advance, provide decision support for battery thermal management, and improve the safety and reliability of the battery system.

[0022] Exemplarily, the battery pack may include an unlimited number of battery components, which may be composed of one or more single batteries (or battery cells). Multiple single batteries may be electrically connected in series or in parallel in any manner, and the packaging may be in various forms, such as cylindrical batteries, square batteries, soft-pack batteries, etc. The battery pack may also be referred to as a battery pack or a battery module in different application scenarios. The battery pack is a rechargeable secondary battery, such as a lithium-ion battery, a sodium-ion battery, a lead-acid battery, a nickel-cadmium battery, and a nickel-metal hydride battery. When the battery pack is a lithium-ion battery, it may include a liquid lithium-ion battery and a solid lithium-ion battery. The type of material used for the positive and negative electrodes may be set as required, and all are within the scope of application of the embodiments of the present application.

[0023] The electrochemical parameters of each single cell included in the above-mentioned battery pack can be set to have certain differences, which are used to fit the inconsistency between different cells in the same battery pack under actual conditions. The inconsistency exists because factors such as process parameter control in the manufacturing process will cause initial differences in parameters such as capacity and internal resistance of the single cell. During use, the environmental factors of each single cell in the battery pack and the non-uniformity of the single cell voltage will further aggravate this inconsistency through charging and discharging. The differences between the above-mentioned single cells can be characterized by the difference values ​​in a predetermined distribution form, such as a Gaussian distribution form, an empirical distribution, and a Weibull distribution. Among them, the Gaussian distribution is used to describe the degree of discreteness of parameters such as internal resistance, the empirical distribution is used for distribution forms based on fitting of actual battery test data, and the Weibull distribution is used to describe the difference distribution of battery capacity attenuation.

[0024] In step S101, by obtaining the reference operating parameters and boundary condition temperatures of the battery pack, it is helpful to fully understand the possible data combinations of the battery pack, and generate thermal runaway calculation examples based on this, which lays a data foundation for the selection of subsequent model training sets and helps to obtain a training set with strong temperature rise characterization capabilities. According to some embodiments of the present application, the reference operating condition parameters may include multiple charge and discharge rate configurations, wherein each charge and discharge rate configuration includes at least one of a constant current curve, a linear current curve, and a secondary current curve. The boundary condition temperature may also be set to multiple, and the multiple boundary condition temperatures include different coolant temperatures in the battery pack.

[0025] In step S101, a thermal runaway calculation example is generated based on the reference operating condition parameters and boundary condition temperature of the battery pack, which may specifically include the following steps: Generate multiple configuration combinations based on multiple charge and discharge rate configurations and multiple boundary condition temperatures; Generate multiple thermal runaway studies for various combinations of configurations and material properties and electrochemical parameters of the battery pack.

[0026] According to the embodiments provided in the present application, the charge and discharge rate (C-rate) is an indicator to measure the charging and discharging speed of the battery, which indicates the ability of the battery to complete charging and discharging in a unit time. It is defined based on the rated capacity of the battery and is usually represented by C. For example, 1C means that the battery completes charging and discharging within 1 hour. If the battery capacity is 1000mAh (milliampere-hour), the charging current of 1C is 1000mA. The charge and discharge rate configuration refers to the selection of different charge and discharge rates according to different application scenarios during battery testing or use. For example, the low rate (such as 0.2C, 0.5C) and high rate (such as 2C, 4C) shown in the examples, preferably, the charge and discharge rate can be set to range from 1C to 4C in this embodiment.

[0027] The above-mentioned reference operating condition parameters include a variety of charge and discharge rate configurations, and each charge and discharge rate configuration includes at least one of the configuration modes of a constant current curve, a linear current curve, and a secondary current curve. The current curve refers to the relationship between the change of current and time during the charging and discharging process of the battery. Constant current charging or discharging means that the current remains constant throughout the charging or discharging process. The linear current curve refers to the linear change of current over time, that is, the current increases or decreases at a constant slope. For example, linear charging can be expressed as the current gradually increasing from 0 to 1C, and then maintaining 1C charging until the voltage reaches the upper limit. Linear discharge can be expressed as the current gradually decreasing from 1C to 0 until the battery voltage drops to the lower limit. The secondary current curve refers to the relationship between the change of current and time in the form of a quadratic function, that is, the rate of change of current changes with time.

[0028] The optional charge and discharge rate configuration also includes combining at least two of the above constant current curve, linear current curve, and secondary current curve according to actual needs to obtain a combined curve. The combined curve can be multiple, for example: constant current-linear combined curve, constant current-secondary combined curve, constant current-linear-secondary combined curve, etc., and there is no specific limitation on the combination method.

[0029] According to some embodiments of the present application, the battery pack can set at least 4 types of charging current curves for different intervals, including constant current curves, linear current curves, secondary current curves and combination curves, etc., and the charge and discharge rates range from 1C to 4C. The diversified charging current curves can enhance the generalization ability of the battery temperature field prediction model for different current forms. The above-mentioned different intervals can be obtained by dividing the battery charge and discharge curves according to battery characteristics, such as the final stage of charge and discharge, the plateau period, etc. The final stage of charge and discharge refers to the stage when the battery is close to being full or empty. At this time, the battery voltage changes greatly. The plateau period refers to the stage when the battery voltage is relatively stable during the charge and discharge process.

[0030] In some embodiments of the present application, the above-mentioned boundary condition temperature is multiple, such as the coolant temperature of the water cooling system used by the battery pack, the battery surface temperature collected by the sensor of the battery pack, or the surface temperature of the battery pack connected to the outside, or the ambient temperature of the battery pack. Taking the coolant temperature of the battery pack as an example, multiple boundary condition temperatures can cover different coolant temperatures in the battery pack.

[0031] Exemplarily, the actual heat dissipation is simulated by the coolant flowing through the battery pack, and the preferred temperature range is from 253.15K (Kelvin) to 323.15K. The temperature of the coolant changes in a gradient from the inlet to the outlet. The temperature gradient of the coolant refers to the change in temperature of the coolant when it flows through the battery pack. The above-mentioned gradient change can be in a linear change mode. The coolant inlet temperature refers to the initial temperature of the coolant entering the battery pack cooling system (such as a liquid cold plate). The coolant inlet temperature has a significant effect on the heat dissipation effect of the battery pack. A lower coolant inlet temperature can more effectively reduce the temperature of the battery pack. The coolant outlet temperature refers to the temperature of the coolant flowing out of the battery pack cooling system. By detecting the coolant outlet temperature, the heat dissipation efficiency of the cooling system can be evaluated, and the coolant flow rate or inlet temperature can be adjusted as needed. By setting different boundary condition temperatures, the battery temperature field prediction model can adapt to different temperature environments.

[0032] In some embodiments of the present application, the specific steps of generating a thermal runaway calculation example based on the reference operating parameters and boundary condition temperature of the battery pack are as follows: Generate multiple configuration combinations based on multiple charge and discharge rate configurations and multiple boundary condition temperatures; Generate multiple thermal runaway studies for various combinations of configurations and material properties and electrochemical parameters of the battery pack.

[0033] According to some embodiments provided in this application, a variety of charge and discharge rate configurations are combined with a variety of boundary condition temperatures to generate a variety of configuration combinations. These configuration combinations cover different charge and discharge conditions and environmental conditions, providing a basis for the subsequent generation of thermal runaway examples. Select different charge and discharge rates (such as 1C, 2C, 3C, 4C, etc.), and combine different current curve forms (such as constant current, linear current, secondary current, combination curve, etc.). A variety of boundary condition temperatures can be set, including different coolant temperatures. The charge and discharge rate configuration is combined with the boundary condition temperature to generate a variety of different configuration combinations, and then generate a number of thermal runaway examples. Through the combination of a variety of charge and discharge rate configurations and boundary condition temperatures, the thermal runaway behavior of the battery under different working conditions can be reflected. Different charge and discharge rates and boundary condition temperatures have different effects on battery thermal runaway. These data differences are captured through thermal runaway examples to provide data support for subsequent model training and verification.

[0034] Exemplarily, the above thermal runaway calculation examples can also be generated based on other parameters of the battery pack, such as: the number of single cells contained in the preset battery pack, the model structure, the material properties, etc. According to the above multiple configuration combinations, as well as the design of the operating parameters such as the simulation duration, the number of time steps, the initial state of charge SOC (State of Charge) of each single cell, the initial temperature, and the electrochemical parameters, all thermal runaway calculation examples are obtained.

[0035] In step S102, the battery pack is divided into a plurality of modeling nodes according to the spatial layout of the battery pack, and the spatial position of each modeling node is defined by its geometric position in the battery pack. The above characteristic parameters are used to characterize the thermal behavior of the battery pack at different spatial positions, thereby providing a basis for subsequent model analysis and prediction. By dividing the battery pack into a plurality of modeling nodes and determining the characteristic parameters of each node, the temperature distribution and thermal behavior of the battery pack during thermal runaway can be simulated more accurately. It can effectively capture the thermal heterogeneity inside the battery pack and provide more accurate data support for the optimization of the battery thermal management system and thermal runaway warning.

[0036] Exemplarily, the lithium-ion battery pack model is established by simplifying the full-size three-dimensional model and then meshing it. The above modeling methods can be multiple, for example: Finite Element Method (FEM), by dividing the battery pack model into finite element meshes to generate modeling nodes. Or a simplified model and common node modeling method can be used to simplify the battery structure, for example, the internal structure of the battery is regarded as a whole, and hexahedral meshing is used, which requires low computational overhead and fast speed. Or an encrypted meshing method, for key areas of the battery (such as the pole area with a large temperature rise), an encrypted meshing method can be used to improve the temperature rise characterization accuracy of the model in the above area. After modeling and characterizing the battery pack using any of the above modeling methods, the above multiple modeling nodes can be obtained.

[0037] According to some embodiments of the present application, the state of lithium-ion batteries under the corresponding working conditions of each thermal runaway example is calculated by an electrochemical model and a thermal runaway model, thereby obtaining characteristic parameters corresponding to different modeling nodes. In step S102: determining the characteristic parameters corresponding to multiple modeling nodes of a battery pack in a thermal runaway example, wherein the multiple modeling nodes are used to represent different spatial positions on the battery pack, and specifically may include the following steps: Based on the simulation of polarization effect of battery pack, the ohmic overpotential, polarization overpotential and concentration overpotential in the battery pack are obtained; Determine, according to the ohmic overpotential, the polarization overpotential and the concentration overpotential, the first unit volume heat generation power corresponding to the plurality of modeling nodes respectively under the condition of normal temperature rise; Based on the first unit volume heat powers respectively corresponding to the plurality of modeling nodes, characteristic parameters respectively corresponding to the plurality of modeling nodes are determined.

[0038] According to some embodiments provided by the present application, by simulating the polarization effect of the battery pack, the internal thermal behavior of the battery during the charging and discharging process can be more accurately reflected. Ohmic overpotential, polarization overpotential and concentration overpotential are key factors affecting battery heating. By analyzing the above types of overpotentials, the first unit volume heating power of the modeling node is determined. This method can more accurately reflect the thermal characteristics of each modeling node under normal temperature rise conditions, thereby providing more reliable characteristic parameters for subsequent thermal runaway predictions.

[0039] The above-mentioned normal temperature rise can also be referred to as normal temperature rise, which refers to the reasonable increase in temperature of the battery under the design allowed charge and discharge rates, environmental conditions and operating conditions. It is usually within a safe range and will not cause battery performance degradation or thermal runaway.

[0040] Specifically, in the embodiment of the present application, a lumped model can be used to simulate the constant temperature rise in the temperature field data of the lithium-ion battery. The lumped model can be used to simulate the three main effects in the battery operation process: Ohm effect, polarization effect and concentration gradient effect, so as to reflect the state of the battery in real time.

[0041] The lumped model is a simplified thermal model that treats the battery as a system with concentrated heat capacity and ignores the internal temperature gradient. The simplicity of the lumped model makes it suitable for rapid thermal analysis of large battery packs. In order to further describe the heat generation inside the battery, the Bernardi heat generation model can be used to characterize the thermal characteristics of the battery under normal temperature rise conditions. The Bernardi heat generation model is a model used to describe the heat generation inside the battery. The heat generation rate is calculated based on parameters such as the battery's volume, current, voltage, internal resistance, temperature, and temperature influence coefficient. The lumped model and the Bernardi heat generation model can be used in combination to improve the accuracy of the battery thermal management model. In the electro-thermal coupling model, the Bernardi heat generation model can be used to calculate the heat generation rate of the battery, while the lumped model is used to describe the overall thermal behavior of the battery.

[0042] In order to obtain the above-mentioned ohmic overpotential, polarization overpotential, and concentration overpotential, they can be obtained by experimental measurement or simulation. For the ohmic overpotential, the experimental measurement method can be calculated by the current interrupt method (CIM) by interrupting the current and measuring the instantaneous change of the voltage, or the electrochemical impedance spectroscopy (EIS) by measuring the low-frequency impedance of the battery to extract the ohmic resistance. Take this as an example to illustrate:

[0043]

[0044] in, represents the activation energy of the ohmic resistor, represents the ohmic resistance at the reference temperature, represents the ohmic resistance of the battery pack, represents the reference temperature, represents the battery load current, is the battery temperature.

[0045] For polarization overpotential, the experimental measurement method can be to use the polarization curve method to measure the polarization curve (voltage-current curve) of the battery, fit the Butler-Volmer equation to extract the polarization overpotential, or the cyclic voltammetry (CV) method to analyze the shape and slope of the cyclic voltammetry curve. The simulation method can be used to obtain the polarization overpotential. For example, it is derived from the inverse form of the Butler-Volmer (BV) equation. In electrochemical reactions, the BV equation describes the relationship between overpotential and current density. Its inverse form reverses this relationship, so that the overpotential can be solved based on the current density. The calculation process is as follows:

[0046]

[0047] in, Represents the battery load current, and are the ideal gas constant and Faraday's constant, (·) represents the inverse hyperbolic sine function. is the exchange current density, represents the activation energy of the activation overpotential, represents the exchange current density at the reference temperature.

[0048] For concentration overpotential, the experimental measurement method can be to use the polarization curve method to measure the polarization curve at high current density, or the cyclic voltammetry (CV) method to measure the cyclic voltammetry curve at different scan rates to analyze the influence of concentration polarization. At low current density, the concentration overpotential can be ignored. At high current density, the concentration overpotential can be calculated by empirical formula or fitting method. The concentration overpotential is obtained by simulation. Take this as an example to illustrate:

[0049]

[0050]

[0051] Concentration overpotential The cell is modeled as a sphere with radius 1. represents the concentration distribution along the radial direction of the sphere, Indicates the battery capacity. represents the characteristic time, which is similar to the diffusion coefficient in the diffusion equation, Represents the position coordinates inside the battery, used to describe the concentration distribution The spatial changes inside the battery. In the battery model, It can be one-dimensional, two-dimensional or three-dimensional coordinates. The diffusion equation describes the change of concentration distribution S inside the battery with time t. Indicates the load current of the battery.

[0052] In the above example, Used to indicate the position along the radial direction of the battery, where Indicates the center of the battery, Represents the boundary of the battery. The boundary conditions are used to describe the concentration gradient on the battery boundary. At , the concentration gradient ∇S is zero, indicating no inflow or outflow. At the point where the concentration gradient ∇S is proportional to the current and battery capacity related to, representing the diffusion of species across the cell boundaries.

[0053] The concentration distribution can be obtained by the above method , and then based on the concentration distribution The concentrations corresponding to the positive and negative electrodes of the battery are obtained. According to the concentrations corresponding to the positive and negative electrodes, the concentration overpotential can be obtained. .

[0054] Under normal temperature rise conditions, according to the Bernardi heating model, the first unit volume heat generation power of the battery pack is , can be obtained by:

[0055]

[0056] in, represents the battery load current, is the battery volume, is the derivative of the open circuit voltage with respect to temperature, Indicates the battery temperature. is the equivalent resistance. represents the open circuit voltage, is the terminal voltage, represents the ohmic overpotential, is the polarization overpotential (also called activation overpotential), is the concentration overpotential. The lumped model focuses on solving the above three overpotentials internally to analyze the thermal and electrochemical behavior of the battery.

[0057] According to some embodiments of the present application, in step S102: determining characteristic parameters corresponding to a plurality of modeling nodes of a battery pack in a thermal runaway calculation example, wherein the plurality of modeling nodes are respectively used to represent different spatial positions on the battery pack, the following steps may also be included: Determine the solid electrolyte interface decomposition rate, negative solvent reaction rate, positive solvent reaction rate, and electrolyte decomposition reaction rate of the battery pack during thermal runaway; Determine, based on the solid electrolyte interface decomposition rate, the negative solvent reaction rate, the positive solvent reaction rate, and the electrolyte decomposition reaction rate, a second unit volume heat generation power corresponding to a plurality of modeling nodes respectively under thermal runaway conditions; Based on the second unit volume heat powers respectively corresponding to the plurality of modeling nodes, characteristic parameters respectively corresponding to the plurality of modeling nodes are determined.

[0058] Thermal runaway is a typical chain reaction process that often occurs in batteries or other chemical systems. In this process, the decomposition of the solid electrolyte interphase (SEI) serves as the initial trigger, releasing a large amount of heat and triggering subsequent chemical reactions. As heat continues to be generated, the reaction rate accelerates, and each reaction strengthens other reactions, forming a self-amplifying positive feedback loop. This cycle causes the reaction to persist and the heat accumulation to grow exponentially, eventually leading to temperature runaway and then thermal runaway.

[0059] According to some embodiments provided in the present application, during the thermal runaway process, a variety of chemical reactions will occur inside the battery, which interact and accelerate, causing a sharp increase in temperature, which may eventually cause combustion or explosion. The decomposition of the SEI film is an early stage of thermal runaway, releasing heat and leaving the negative electrode unprotected. The reaction of the negative electrode material (such as graphite) with the electrolyte occurs after the decomposition of the SEI film, and a short circuit may occur inside the battery, releasing a large amount of heat. The positive electrode material reacts with the electrolyte under high temperature conditions, which releases oxygen and further intensifies the combustion. The decomposition reaction of the electrolyte occurs at a higher temperature, releasing flammable gases. Based on the above chemical reaction rate and the predetermined heat release coefficient, the second unit volume heat power of each modeling node is calculated to reflect the heat generation at different locations inside the battery in the case of thermal runaway.

[0060] There are many ways to simulate the thermal runaway state of batteries, such as the Arrhenius equation or a simulation method based on a neural network. The Arrhenius equation is an empirical formula in chemical kinetics that describes the relationship between the reaction rate constant and temperature. In the simulation of battery thermal runaway, the Arrhenius equation can be used to describe the change of the chemical reaction rate inside the battery with temperature.

[0061] In the embodiments of the present application, the Arrhenius equation can be used to formulate mathematical models for different reactions, including SEI (solid electrolyte interface) decomposition, negative solvent reaction, positive solvent reaction, and electrolyte decomposition reaction, which can be expressed by the following mathematical formula:

[0062]

[0063]

[0064]

[0065] in, represents the solid electrolyte interface decomposition rate, represents the dimensionless amount of lithium compounds in SEI (solid electrolyte interface), represents the frequency factor of SEI decomposition reaction, represents the activation energy of SEI decomposition, R represents the ideal gas constant, Indicates the battery temperature. represents the negative solvent reaction rate, represents the dimensionless amount of lithium embedded in the carbon negative electrode, represents the frequency factor of the negative solvent reaction, represents the time of SEI layer decomposition, represents the reference time of SEI layer decomposition, represents the activation energy of the negative solvent reaction. is the positive solvent reaction rate, represents the conversion rate of the cathode solvent reaction, represents the frequency factor of the positive solvent reaction, represents the activation energy of the positive solvent reaction. represents the electrolyte decomposition reaction rate, represents the dimensionless concentration of the electrolyte, represents the frequency factor of the electrolyte decomposition reaction, It represents the activation energy of the electrolyte decomposition reaction.

[0066] Based on the above mathematical expression, we can further obtain:

[0067] in, represents the second unit volume heat generation power, It represents the heat released by decomposition of unit mass of SEI, It represents the heat released by the reaction of unit mass of negative solvent. It represents the heat released by the reaction of unit mass of positive solvent. It represents the heat released by decomposition of unit mass of electrolyte. Indicates the specific content of SEI in the negative electrode winding body, Indicates the specific content of negative electrode material in the negative electrode winding body, Indicates the specific content of positive electrode material in the positive electrode winding body, It indicates the specific content of electrolyte in the negative electrode wound body.

[0068] According to some embodiments of the present application, in step S102: determining characteristic parameters corresponding to a plurality of modeling nodes of a battery pack in a thermal runaway calculation example, wherein the plurality of modeling nodes are respectively used to represent different spatial positions on the battery pack, the following steps may also be included: Determining a conditional short-circuit current of the battery pack according to a predetermined short-circuit temperature; Determine the activation energy, state of charge and frequency factor of the battery pack during the internal short circuit of the battery; Determine, according to the conditional short-circuit current, the activation energy, the state of charge and the frequency factor, the third unit volume heat generation power corresponding to the plurality of modeling nodes respectively under the internal short-circuit condition; Based on the third unit volume heat powers respectively corresponding to the multiple modeling nodes, characteristic parameters respectively corresponding to the multiple modeling nodes are determined.

[0069] In the embodiments provided in the present application, the conditional short-circuit current of the battery pack under internal short-circuit conditions is calculated according to a predetermined short-circuit temperature. Short-circuit current is the main source of Joule heat during thermal runaway, which directly affects the temperature rise inside the battery. During the internal short-circuit process of the battery, parameters such as activation energy, state of charge, and frequency factor are determined, among which the activation energy of the internal reaction of the battery (such as SEI film decomposition, electrolyte reaction, etc.) reflects the sensitivity of the reaction rate to temperature. The state of charge of the battery affects the rate and heat release of its internal reaction. The higher the state of charge SOC, the greater the reaction rate and heat release may be. The frequency factor is a related parameter in the Arrhenius equation, which is used to characterize the maximum possible value representing the reaction rate.

[0070] According to the conditional short-circuit current, activation energy, state of charge and frequency factor, the third unit volume heat power corresponding to each modeling node is calculated, and the heat release of each modeling node under the condition of internal short circuit is quantified, providing a basis for subsequent thermal runaway simulation.

[0071] Thermal runaway reactions are usually triggered by internal short circuits, which convert electrical energy into heat. In the present application, the internal short circuit situation is simulated. Once the single cell reaches a certain temperature, the internal short circuit will be triggered. The formula is as follows:

[0072]

[0073] in, Indicates conditional short-circuit current. The temperature threshold of 57°C can be set as needed. is the frequency factor. is the charge state, is the activation energy.

[0074] Based on the above formula, the third unit volume heat power can be further obtained: :

[0075] in, It refers to electrical energy that can be converted into heat energy.

[0076] Based on the above optional embodiments, there may be multiple ways to obtain the characteristic parameters corresponding to each modeling node, for example, based on at least any one of the first unit volume heat power, the second unit volume heat power, and the third unit volume heat power corresponding to each modeling node, to obtain the characteristic parameters corresponding to each modeling node.

[0077] According to some embodiments of the present application, in step S103: based on the characteristic parameters corresponding to the multiple modeling nodes, a modeling node that meets the predetermined conditions is selected from the multiple modeling nodes as a feature point; the spatial distance and the temperature variation amplitude between the feature point and other modeling nodes meet the predetermined conditions, specifically including the following steps: Determine the average value of the distance between each candidate node and other modeling nodes in the candidate point set as the spatial uniformity parameter of the candidate point set; the candidate point set is a point set obtained by selecting some modeling nodes from multiple modeling nodes as candidate nodes; Based on the characteristic parameters corresponding to each candidate node, a characteristic matrix representing the change of temperature over time is generated; Perform dimension reduction processing on the feature matrix to obtain the temperature representative parameters of the candidate point set; When the spatial uniformity parameter and the temperature representative parameter meet predetermined conditions, each candidate node is taken as a feature point.

[0078] In an embodiment of the present application, some nodes are selected from multiple modeling nodes as candidate nodes to form a candidate point set. The average value of the distance between each candidate node and other modeling nodes is calculated, and the average value is used as the spatial uniformity parameter of the candidate point node. Based on the characteristic parameters of each candidate node, a characteristic matrix representing the temperature change over time is generated. The characteristic matrix is ​​a multidimensional data structure for storing the temperature information of each candidate node at different time points. The characteristic matrix is ​​subjected to dimensionality reduction processing to extract key parameters that can represent the temperature change of the candidate point set, which are called temperature representative parameters. The purpose of dimensionality reduction processing is to reduce the data dimension while retaining the most important information and improving the computational efficiency. When the spatial uniformity parameter and the temperature representative parameter meet the predetermined conditions, each candidate node is used as a feature point. The feature point is a point that can effectively represent the balance between the spatial distance and the temperature change amplitude in the battery pack. By selecting the feature point, the number of nodes to be processed is reduced, thereby significantly reducing the computational complexity and improving the efficiency of the simulation. The selection of the feature point is based on spatial uniformity and temperature representativeness, which can more accurately reflect the thermal runaway process inside the battery, avoid reducing the model generation efficiency due to too many nodes, or too few nodes leading to poor performance of the temperature gradient, thereby affecting the model performance.

[0079] Exemplarily, there are multiple ways to determine the distance between each candidate node and other modeling nodes, for example, Euclidean distance, Manhattan distance, Minkowski distance, etc. can be used. Euclidean distance is used to calculate the straight-line distance between two points, reflecting the geometric shortest path in space. Manhattan distance calculates the distance between two points on a grid-like path, which is suitable for scenarios where nodes are distributed in an orthogonal grid. Minkowski distance is a generalized form of Euclidean distance and Manhattan distance. The distance calculation method is adjusted by predetermined parameters. Through different parameter settings, it is retained as Manhattan distance or Euclidean distance, which can adapt to different data distributions and computing requirements.

[0080] Exemplarily, there are many ways to obtain representative parameters of temperature changes, such as principal component analysis (PCA), clustering algorithms, and wavelet transforms. Principal component analysis maps high-dimensional data to low-dimensional space through linear transformation to extract the main change patterns in temperature data. By extracting the principal components of the temperature values ​​corresponding to each modeling node in all thermal runaway examples, time series characteristics can be obtained to reflect the heat generation and heat transfer capabilities of each modeling node under different working conditions. The clustering algorithm can divide the modeling nodes in the battery pack into several categories, and the center point or representative node of each category can be used as a representative parameter of temperature change to automatically identify the spatial pattern of temperature distribution. Wavelet transform extracts the characteristics of temperature field data at different scales and times as a representative parameter, which can simultaneously capture the time domain and frequency domain characteristics of temperature changes.

[0081] Exemplarily, the above embodiment calculates the weighted characteristic value of each point in the battery pack from two aspects of spatial uniformity and temperature variation representativeness, and the weight is set as needed without specific limitation.

[0082] According to some embodiments of the present application, the method of selecting feature points can adopt a multi-objective optimization algorithm, such as a population algorithm, which includes various types, such as a particle swarm algorithm, an ant colony algorithm, and a genetic algorithm. The particle swarm algorithm simulates the social behavior of a bird group, and particles adjust their positions through individual experience and group experience. Particles move in the solution space and update their speed and position according to their own experience and group experience. The fitness function is used to evaluate the pros and cons of each particle and gradually approach the optimal solution. The ant colony algorithm optimizes the path through the deposition and volatilization of pheromones. Pheromones are deposited on the path. The higher the pheromone concentration, the better the path. The path is selected according to the pheromone concentration, and the path is gradually optimized through a positive feedback mechanism. The genetic algorithm simulates natural selection and genetic mechanisms, and optimizes the solution through selection, crossover and mutation operations. The fitness function is used to evaluate the pros and cons of each individual, the selection operation retains excellent individuals, the crossover operation generates new solutions, and the mutation operation introduces new diversity.

[0083] Specifically, according to the embodiment of the present application, the particle swarm algorithm is used as an example, there are multiple candidate point sets, and the multiple candidate point sets have different ways of selecting candidate nodes from multiple modeling nodes. Selecting feature points may include the following steps: Based on the spatial uniformity parameters and temperature representative parameters respectively corresponding to the multiple candidate point sets, fitness function values ​​respectively corresponding to the multiple candidate point sets are generated; The fitness function values ​​corresponding to the multiple candidate point sets are respectively processed to obtain the minimum value, and the candidate point set with the minimum fitness function value among the multiple candidate point sets is determined as the target point set; The candidate nodes included in the target point set are taken as feature points.

[0084] In an embodiment of the present application, for each candidate point set, a fitness function value is generated in combination with its spatial uniformity parameter and temperature representativeness parameter. The fitness function value is a comprehensive indicator used to evaluate the pros and cons of the candidate point set in terms of spatial distribution and temperature representativeness. The fitness function values ​​of all candidate point sets are processed to find the minimum value. It is regarded as the best point set, indicating that this candidate point set has achieved a balance in terms of spatial uniformity and temperature representativeness. The candidate point set with the smallest fitness function value is determined as the target point set, and then the candidate nodes in the target point set are finally selected as feature points for subsequent battery thermal runaway simulation analysis. Through the above processing, the selection of feature points is based on the comprehensive optimization of spatial position and temperature gradient, which reduces the model training efficiency caused by improper selection of feature points, or the unsatisfactory model prediction performance.

[0085] Exemplarily, for ease of understanding, according to the above-mentioned embodiment, the feature points are screened, the characteristic parameters of the lithium-ion battery state calculated by each thermal runaway example are extracted, the spatial uniformity and temperature variation representativeness of each modeling node of the battery pack are calculated, and the particle swarm optimization algorithm is used to select the feature points.

[0086] The weighted eigenvalues ​​of each point in the battery pack are calculated from the two aspects of spatial uniformity and representativeness of temperature changes. The above spatial uniformity is obtained by calculating the Euclidean distance between other modeling nodes of the battery pack through the distance matrix. The time series characteristics are composed by extracting the principal components of the temperature values ​​of each modeling node in all thermal runaway examples, reflecting the heat generation and heat transfer capabilities of each modeling node under different working conditions. Particle swarm optimization is a group intelligence optimization algorithm that simulates the foraging behavior of bird flocks for global search and has strong optimization capabilities and fast convergence characteristics. The combination of the two to obtain the spatiotemporal eigenvalues ​​for particle swarm optimization can maximize the spatial uniformity and representativeness of temperature changes, effectively improve the accuracy of feature point selection, ensure that the selected feature points can better represent the overall state of the battery pack, and at the same time enhance the sparsity of the feature subset (i.e., the selected candidate subset) in space, reduce the cost of model training, and provide a reliable basis for rapid response in the model reasoning stage.

[0087] According to some embodiments of the present application, in step S104: generate a training set according to the reference operating condition parameters and the characteristic parameters corresponding to each characteristic point. The training set can be generated by dividing it according to a predetermined ratio, and the predetermined ratio is set as needed, for example, dividing it into a training set, a validation set and a test set in a ratio of 70%: 10%: 20%. The training set is used to train the model so that the battery temperature field prediction model learns the mapping relationship between the input data and the output temperature field. The validation set is used to tune the model hyperparameters (such as learning rate, number of network layers, etc.) to prevent the model from overfitting. The test set is used to evaluate the generalization performance of the battery temperature field prediction model and verify the generalization ability of the battery temperature field prediction model.

[0088] Exemplarily, the training set includes at least five measurable battery signals: current, voltage, boundary condition temperature, SOC, and battery temperature.

[0089] According to some embodiments of the present application, the network architecture of the battery temperature field prediction model can be multiple, including a convolutional neural network layer (CNN), a bidirectional long short-term memory layer (Bi-LSTM) or a self-attention module (AM). The Bi-LSTM network is good at autonomously learning complex nonlinear time series patterns hidden in the data, while the CNN is good at integrating information of different dimensions. The above-mentioned convolutional neural network layer can be of a predetermined dimension, such as a one-dimensional convolutional neural network layer (1D-CNN).

[0090] The Bi-LSTM network layer consists of two main parts: one LSTM model is used for forward processing of past input features, and the other LSTM model is used for reverse processing of future input features. This feature of Bi-LSTM enables it to perform well in long sequence data prediction tasks, capturing past and future dependencies at each time step and learning the complex relationship between battery parameters and temperature fields on a time scale.

[0091] The self-attention mechanism is a specific type of attention mechanism that enables the battery temperature field prediction model to selectively focus on different parts of the input sequence when estimating the interdependencies between input elements.

[0092] For example, to solve the gradient vanishing and exploding problems, the Bi-LSTM network combines three gates: input gate , Forget Gate and output gate For the input sequence , can be expressed as follows:

[0093]

[0094]

[0095]

[0096]

[0097] in, represents the Hadamard product (i.e., element-wise product), yes The neuron input at this moment, is the output of the hidden layer at the corresponding moment, is the neuron state, and are the input gate, forget gate, and output gate activation vectors respectively. , and It is the weight matrix and bias parameters that need to be learned during the training process, and the subscripts t, i, and o are used to distinguish which of the input gate, forget gate, and output gate is used. is the sigmoid activation function, and Hyperbolic tangent function. The last hidden layer in the Bi-LSTM network By concatenating the forward hidden state and the backward hidden state generate.

[0098] Exemplarily, the self-attention library in the Keras framework (an open source neural network library) is used to implement the mechanism, combining self-attention with the Bi-LSTM network to effectively capture the output sequence of the Bi-LSTM and train independent layers through weight distribution to make it pay more attention to the most important output elements. The main mathematical formula of the self-attention mechanism for processing time series data and considering the context of each time step is as follows:

[0099]

[0100]

[0101]

[0102] in, is the output of the hidden state in the Bi-LSTM layer, is the sigmoid activation output of the attention network, indicating the attention score. is the element-wise sigmoid activation function, is the attention weight normalized by softmax. and They are hidden states and The weight matrix of . is the weight matrix of the attention network. Indicates that at a specific time step The attention-focused hidden states on are used to indicate the attention or importance that should be given to different neighboring elements. Captures the current time step The corresponding element-related information is combined with the hidden state in the input sequence and the corresponding attention weights get.

[0103] Data-driven deep learning models are more advantageous in terms of computational efficiency. Using battery pack history or real-time data, self-learning models can capture potential correlations in operating conditions to predict the surface temperature of lithium-ion batteries, thereby avoiding complex electrochemical process modeling. Sequential neural networks are applied to battery temperature prediction tasks and perform well in temperature field prediction of various types of batteries. Long Short-Term Memory (LSTM) networks and bidirectional LSTM (Bi-LSTM) have good capabilities in capturing time dependencies. Combining convolutional neural networks (CNNs) with LSTMs significantly improves prediction accuracy. The hybrid model with the introduction of an attention mechanism further enhances the prediction capability of dynamic temperature field changes, and improves the learning efficiency and accuracy of the model when processing complex input sequences. Models with attention mechanisms are significantly better than traditional models in terms of temperature prediction accuracy, highlighting their advantages in complex tasks.

[0104] The accuracy and generalization ability of the data-driven deep learning model largely depends on the quality of the data set. Therefore, through the processing of step S104 above, training data with representative working conditions and high-confidence temperature field values ​​are obtained, which becomes an important guarantee for improving model efficiency and model performance.

[0105] In the embodiment, a preferred model network structure setting method is provided, and a CNN-Bi-LSTM model based on the self-attention mechanism is proposed, which combines the advantages of CNN and Bi-LSTM to predict the temperature field of lithium-ion batteries. The above architecture makes full use of the spatial characteristics, temporal characteristics and dynamic weight allocation capabilities of the battery temperature field data at key time steps, thereby achieving high-precision temperature field prediction.

[0106] In step S105: based on the training set, the battery temperature field prediction model is trained to obtain a trained battery temperature field prediction model, which may include the following steps: The training set is input into the convolutional neural network layer for processing to obtain the local spatial features in the battery pack; The local spatial features are input into the bidirectional long short-term memory network layer for processing to obtain the time series features of the battery pack at different time steps; The time series features are input into the self-attention module for processing to obtain the test temperature field distribution of the battery pack at the predetermined time; When the test temperature field distribution meets the predetermined iteration conditions, a trained battery temperature field prediction model is obtained.

[0107] In the embodiment provided in the present application, the training set data is first input into the convolutional neural network layer (CNN), and the local spatial features of different nodes in the battery pack are extracted through the convolution operation, such as temperature gradient, voltage distribution, etc. The above local spatial features reflect the temperature variation pattern of the battery pack in space. It can effectively capture the spatial correlation between different nodes in the battery pack, such as the temperature variation trend of adjacent nodes.

[0108] The local spatial features are input into the bidirectional long short-term memory network layer (Bi-LSTM) for processing. Through the time series processing in both the forward and reverse directions, the long-term dependencies between battery packs at different time steps are captured. The forward LSTM processes the data from the past to the current time step, and the reverse LSTM processes the data from the future to the current time step. The combination of the two can more comprehensively reflect the changing trend of the temperature field. Bi-LSTM can effectively capture the dynamic changes of the battery temperature field over time, such as the trend of temperature rise, fall or fluctuation, thereby enhancing the prediction ability of the battery temperature field prediction model for the battery time series pattern.

[0109] The self-attention mechanism calculates the weights between features at different time steps and dynamically allocates attention resources, which can enhance the model's ability to identify key time steps. For example, when the temperature changes rapidly or thermal runaway is triggered, the model can more accurately predict the temperature field distribution.

[0110] Through the above processing, the battery temperature field prediction model outputs the test temperature field distribution of the battery pack at the predetermined time. During the training process, the model is optimized through multiple batches of iterations, and the parameters are adjusted on the validation set, and the model prediction performance is verified on the test set. When the test temperature field distribution meets the predetermined iteration conditions, the training is determined to be completed and the final battery temperature field prediction model is obtained.

[0111] Specifically, in order to prevent the model from overfitting, according to some embodiments of the present application, automatic parameter optimization and early stopping mechanism are used in the training process. When the number of times the validation set loss decreases reaches a threshold, the training is terminated in advance, making the training process more automated, effectively improving the training efficiency and reducing model overfitting.

[0112] Based on the above embodiments, the present application also provides an optional implementation method to facilitate understanding of the implementation process, which is described below.

[0113] Step S1, designing a variety of lithium-ion battery operating parameters to form a large number of thermal runaway calculation examples. Figure 2 A battery modeling schematic diagram of a battery temperature field prediction model training method according to an embodiment of the present application is shown, Figure 2 As shown in the figure, the modeling model of a battery pack consisting of 18 square batteries connected in series, each single battery, is set to have a nominal capacity of 228Ah (ampere-hour) and a volume of 2.04×10⁻³ m³. Using the finite element method, 7806 nodes were constructed on the battery pack. The simulation time of each thermal runaway case is 1800 seconds, divided into 120 time steps, each time step is 15 seconds. The initial state of charge SOC of all single batteries in the battery pack is 0, and the initial temperature is 20℃ (degrees Celsius), which can be set to the same as the ambient temperature.

[0114] Figure 3 The configuration curve diagram of the training method of the battery temperature field prediction model of the embodiment of the present application is shown as follows: Figure 3 As shown, as an exemplary configuration, 25 charging current curves are designed, including 3 constant current curves (red), 8 linear current curves (yellow), 3 secondary current curves (blue) and 11 combination curves (green), and the charging rate is set to range from 1C to 4C. Each of the above curves is selected and marked as an example in the figure. This embodiment designs 13 boundary condition temperatures ranging from 293.15K to 313.15K, and the temperature of the coolant changes linearly from the inlet to the outlet. The embodiment designs Gaussian distribution difference values ​​for the exchange current density and internal resistance in the electrochemical parameters of each single cell. Through different combinations of the above 25 charging current curves and 13 coolant temperatures, this embodiment designs a total of 325 thermal runaway calculation examples.

[0115] Step S2, calculate the lithium-ion battery state under the corresponding working conditions of each thermal runaway example through the electrochemical model and the thermal runaway model. In the embodiment, a lumped model can be used to simulate the normal temperature rise in the temperature field data of the lithium-ion battery, and the thermal runaway model can be used to simulate the short circuit situation in the battery. Table 1 is a list of thermal runaway parameters. As shown in Table 1, the parameters used in calculating thermal runaway through the Arrhenius equation in this embodiment are schematically listed, and the specific values ​​are only for illustration and are not specifically limited.

[0116] Table 1 Thermal runaway parameter list

[0117] Figure 4 A schematic diagram of single-node temperature variation of a battery temperature field prediction model training method according to an embodiment of the present application is shown. Figure 4 As shown in the figure, the simulation results of the temperature change of a modeling node in the battery pack are shown when the charging rate is 2C constant (456A). The modeling node is located on the battery cell, starting from the initial temperature of 293.15K, and the temperature continues to rise as the charging process progresses. When the temperature reaches 330K, thermal runaway occurs, causing the temperature to rise rapidly.

[0118] Step S3, extracting the battery pack node temperature, voltage, and state of charge (SOC) parameters at each time step in the calculation results of each thermal runaway example in step S2. The embodiment exemplarily uses the pyswarms particle swarm optimization library (a Python library for implementing particle swarm optimization) to evaluate the spatial uniformity and temperature variation representativeness of each modeling node, and finally obtains a feature point set, and the spatial uniformity parameter The calculation method can be expressed as:

[0119]

[0120] The Euclidean distance of each point is calculated by the distance matrix D and expressed as , i and j are used to identify different modeling nodes, N represents the total number of all modeling nodes, and (x, y, z) is used to represent the three-dimensional spatial coordinates of the modeling nodes. The sparseness of the selected point subset in space is characterized by calculating the average distance between all different points in the distance matrix. Representative parameters of temperature change The specific steps of calculation through principal component analysis are as follows: Flatten the temperature data of the selected points to obtain a feature matrix of shape (M, P), where M is the number of samples (the product of the number of examples and the number of time steps). is the number of selected points, that is, the number of nodes selected from the candidate point set.

[0121] Use PCA to reduce the dimension to The principal components are set as follows in this embodiment. .

[0122] The representative parameters of temperature variation can be expressed by calculating the variance of these principal components:

[0123] in It is Principal components. Representative parameters of temperature change It is calculated by summing up the variances of the first K principal components. The summing process actually calculates the total variability explained by the first K principal components. Since the principal components are sorted by variance, the first K principal components can explain most of the variability in the data. The number of K principal components is usually selected based on the cumulative explained variance ratio. For example, the first K principal components can be selected to explain at least 95% of the total variance, ensuring that the representative parameters of temperature change Able to fully represent the main changing trends in the data.

[0124] Fitness function of particle swarm optimization The calculation can be expressed as:

[0125] In the embodiment of battery temperature field prediction, particle swarm optimization is used to screen feature points. Fitness function Aims to balance the spatial uniformity parameters of feature points and temperature variation representative parameters , ensuring that the selected feature points are evenly distributed in the battery pack, avoiding the feature points being too concentrated or sparse, and at the same time being able to reflect the temperature change trend of the battery pack under different working conditions, thereby optimizing the selection of feature points. Find the candidate point set that minimizes the fitness function F, that is, maximizes ,maximize Ensure that feature points are evenly distributed in the battery pack to avoid oversampling or undersampling in local areas. Ensure that the characteristic points can fully reflect the temperature change trend of the battery pack.

[0126] In the particle swarm algorithm provided in this embodiment, each particle represents a selection scheme for a candidate point set. For each particle in the particle swarm algorithm, the fitness function value corresponding to the particle is calculated, and the position and speed of the particle are updated according to the particle's historical optimal solution and the global optimal solution, gradually approaching the candidate point set that best meets the conditions, and the points in the candidate point set determined after iteration are used as the selected feature points.

[0127] Step S4, dividing the example working condition parameters in step S1 and the characteristic data obtained in step S3 into a training set, a validation set and a test set in a ratio of 70%:10%:20%.

[0128] Step S5, using a sliding window of size m (for example, m=5 in this embodiment) to extract the lithium-ion battery operating parameter current, boundary condition temperature and the corresponding calculated lithium-ion battery state voltage, state of charge (SOC) and battery internal temperature field in the examples contained in the training set.

[0129] Figure 5 The data preprocessing flow diagram of the training method of the battery temperature field prediction model of the embodiment of the present application is shown. In order to achieve accurate model establishment and verification, as shown in FIG. Figure 5 As shown, the data was preprocessed, including zero filling, outlier correction and normalization operations. The rules are as follows: (1) Null value filling: If the data is null, let ,in is the parameter at time step t.

[0130] (2) Over-range correction: If If it exceeds the normal range, The normal ranges of temperature, battery voltage, SOC and battery current are , , and . Indicates the maximum value of the parameter. Indicates the minimum value of the parameter.

[0131] (3) Normalization: The minimum and maximum normalization method is used to eliminate the dimensional influence between different physical quantities. The formula is as follows:

[0132] The preprocessed training set data is input into the CNN-Bi-LSTM-AM model, and the model training is completed through multiple batch iterations and automatic parameter optimization to predict the battery temperature field distribution at future times.

[0133] In this embodiment, the 1D-CNN layer receives an input matrix in the form of (window size m=5, feature quantity n=5), and the feature quantity contains five measurable battery signals: current, voltage, boundary condition temperature, SOC, and battery temperature. The local features are captured by convolution operations, and then the feature dimension is reduced through the maximum pooling layer to reduce the computational complexity, and the Dropout layer is used to prevent overfitting. Dropout forces the network to learn more robust feature representations by randomly discarding some neurons in the network during training.

[0134] The Bi-LSTM layer processes the input feature sequence bidirectionally, captures the long-term dependencies in the sequence through the memory and forgetting mechanism, and updates the hidden layer state. Figure 6 A schematic diagram of the Bi-LSTM network structure of the battery temperature field prediction model training method of the embodiment of the present application is shown, as shown in FIG. Figure 6 As shown, W1 to W6 represent different weights, which are used to control the flow of information. Figure 6The circle in the figure represents the feature unit, and the arrows from left to right and from right to left respectively represent the direction of information flow in time series. Among them, W1 is the weight matrix of the input gate, which is used to determine the amount of new input information. W2 is the weight matrix of the forget gate, which is used to determine how much previous memory needs to be discarded. W3 is the weight matrix of the candidate memory unit, which is used to generate new candidate memory content. W4 is the weight matrix of the output gate, which is used to determine how much current memory to output. W5 is the weight matrix used to combine the candidate memory unit with the output of the forget gate to update the memory unit state. W6 is the weight matrix used to combine the output of the input gate with the candidate memory unit to generate a new hidden state.

[0135] The above weight matrix is ​​learned iteratively during the training process so that the BI-LSTM unit can accurately capture the dependencies in the time series.

[0136] The self-attention mechanism (AM) combines global and local information by calculating weighted feature representations, thereby enhancing the model's attention to key features. Figure 7 A schematic diagram of the time series prediction process of the CNN-Bi-LSTM-AM network of the battery temperature field prediction model training method of the embodiment of the present application is shown, and the input data consists of multiple time steps, each time step contains five features: (Battery temperature), , (current), (voltage) and (Boundary condition temperature). The input data is in the form of a sequence of multiple time steps, and the characteristics of each time step are represented as (T i , SOC i , I i , U i , Te i ), i is used to identify groups of different features, and n represents the total data of feature quantities. The input data passes through multiple convolutional neural network layers (ID-CNN) to extract local features and patterns in the input data. ID-CNN may represent a one-dimensional convolutional neural network, which is suitable for processing time series data. The output of the convolutional layer is passed to the bidirectional long short-term memory network layer. Bi-LSTM is able to capture long-term dependencies in time series data, and through its bidirectional structure, it is able to consider both past and future information at the same time. The self-attention mechanism is used to further capture the relationship between different time steps in the sequence. It can help the model focus on the most important part of the sequence, thereby improving the performance of the model. The output of the self-attention layer is passed to the fully connected layer. The fully connected layer is used to combine and transform the extracted features for final prediction or classification. The output of the fully connected layer is passed to the output layer to generate the final prediction result.

[0137] like Figure 7 The network structure shown combines convolutional neural networks, recurrent neural networks and self-attention mechanisms, which are suitable for processing complex time series data and can effectively capture the spatiotemporal features and long-term dependencies in the data. This embodiment introduces an early stopping mechanism of 3 times during the training process. When the performance on the validation set drops three times in a row, the training process is terminated in advance to prevent overfitting.

[0138] The experiment of this embodiment is performed on the Windows 10 operating system. The Python 3.10 programming language is used to train the neural network on the UCI data set using the TensorFlow framework. In order to improve computing efficiency, NVIDIA CUDA 11.2 technology is used to achieve GPU acceleration. The hardware configuration used in the experiment includes an RTX 3070 Ti graphics card, an i5-13400F 4.3 GHz central processing unit, and 16GB of memory. The specific experimental environment configuration is shown in Table 2.

[0139] Table 2 Experimental environment

[0140] To evaluate the CNN-Bi-LSTM-AM network provided in this embodiment, it is compared with several baseline methods, ranging from simple benchmarks to state-of-the-art algorithms. The methods considered include: 1. ARIMA (Autoregressive Integrated Moving Average); 2. SVR (Support Vector Regression); 3. LSTM (Long Short-Term Memory Neural Network); 4. CNN (Convolutional Neural Network).

[0141] Figure 8 The RMSE error comparison diagram of the training method of the battery temperature field prediction model of the embodiment of the present application is shown. Fig. 9 The MSE error comparison diagram of the training method of the battery temperature field prediction model of the embodiment of the present application is shown as follows: Figure 8 , Fig. 9 As shown in the figure, except for the RMSE (Root Mean Square Error) and MSE (Mean Squared Error) of ARIMA, which remain stable as the temperature increases, the RMSE and MSE of all other methods show an increasing trend with the increase of the initial ambient temperature. In the data set, when the ambient temperature ranges from 243.15K to 293.15K, the method provided in this embodiment is significantly better than other methods, especially at medium and low temperatures, showing excellent reasoning ability and robustness. This advantage highlights the adaptability and reliability of the model in processing tasks in the medium and low temperature range.

[0142] Fig.10A first comparison diagram showing the temperature field prediction effect of normal charging provided by an embodiment of the present application is shown. Fig.10 As shown, the good follow-up performance of the CNN-Bi-LSTM-AM provided in this embodiment to the actual temperature change is compared. Fig.11 A second comparison diagram showing the temperature field prediction effect of normal charging provided by an embodiment of the present application is shown. Fig.11 As shown in the figure, the ability of CNN, LSTM, SVR, ARIMA and other baseline models to follow the actual temperature changes is compared. Fig.10 and Fig.11 By comparison, it can be seen that the difference between the CNN-Bi-LSTM-AM provided in this embodiment and the actual temperature change is smaller than that of other models.

[0143] In order to illustrate the superiority of the CNN-BI-LSTM-AM network structure, an ablation experiment was conducted based on the data set used in this embodiment. In order to ensure the effectiveness and scientificity of the comparison, three evaluation indicators were selected, including root mean square error (RMSE), mean absolute error (MAE), and efficiency. At the same time, the impact of temperature factors on the model and the actual effect of each model in thermal runaway prediction will be deeply explored, so as to fully and deeply understand the characteristics, advantages and disadvantages of different models.

[0144] Table 3 studies the specific impact of each module on the evaluation metrics by systematically adding or removing network layers from the model. To evaluate the effectiveness of each module, Figure 7 A series of ablation experiments were conducted on key modules in the model. In the experiments, network layers such as CNN, Bi-LSTM and attention mechanism (AM) were successively deleted from the model. In addition, a substitution experiment was conducted by replacing Bi-LSTM with LSTM to further analyze the architectural differences. Under the same temperature field prediction conditions, the proposed method is the best among the 7 model structures, and the results are as follows: Table 3 An example of calculated performance (mean ± standard)

[0145] In addition, this embodiment focuses on a specific thermal runaway scenario that occurs during a certain (e.g., the 8th) charging cycle, and adopts a rolling sliding time window method to predict the temperature rise process in the test data set. Table 4 compares the average RMSE and MSE indicators of different models for thermal runaway prediction of the entire battery pack.

[0146] Table 4 Another example of calculating performance (mean ± standard)

[0147] Fig.12A third comparison diagram showing the prediction effect of the temperature field in the thermal runaway stage provided by the embodiment of the present application is shown. Fig.12 As shown in the figure, the good follow-up performance of CNN-Bi-LSTM-AM to the real temperature change during the thermal runaway stage is compared. Fig.13 The fourth comparison chart of the temperature field prediction effect of the thermal runaway stage provided by the embodiment of the present application is shown, comparing the ability of CNN, LSTM, SVR, ARIMA and other baseline models to follow the actual temperature change in the thermal runaway stage. Fig.12 and Fig.13 As shown, the experimental results show that the provided CNN-Bi-LSTM-AM model exhibits excellent prediction performance in the thermal runaway heating stage, verifying its reliability and applicability.

[0148] To demonstrate the efficiency advantage of the proposed CNN-Bi-LSTM-AM method, it is compared with other baseline methods. Each model is evaluated using 5-fold cross validation, and the experimental results are averaged over 10 experiments. Fig.14 A comparison chart of the execution time of different algorithm inferences provided in the embodiments of the present application is shown. Fig.14 As shown, the actual inference time of all models is very small, and the model provided by this embodiment has achieved significant improvement in prediction accuracy at the cost of 100 seconds of training time. In practical applications, inference performance is usually more critical. Since it provides a substantial improvement in accuracy, its practical value is improved.

[0149] Fig.15 A flow chart of a battery temperature field prediction method according to an embodiment of the present application is shown. Fig.15 As shown, the method may include step S201 and step S202.

[0150] Step S201: obtaining the current operating parameters of the battery pack to be tested; Step S202: Input the current operating condition parameters into the battery temperature field prediction model for processing to obtain the battery temperature field distribution of the battery pack at a future time, wherein the battery temperature field prediction model is obtained by applying any of the battery temperature field prediction model training methods provided in the above embodiments.

[0151] The above-mentioned execution entity can be a battery management system BMS (Battery Management System), an energy management system EMS (Energy Management System), an edge device or a remote cloud platform.

[0152] The above current operating condition parameters are parameters collected in real time, and may include the current, voltage, battery temperature, charge and discharge conditions, coolant temperature, etc. of the battery pack.

[0153] Exemplarily, a variety of sensors may be provided for the battery pack, such as current, voltage, and temperature sensors, etc., and no specific limitation is imposed on the data collection method.

[0154] In the embodiment of the present application, the acquired current operating condition parameters are input into the battery temperature field prediction model for processing. The model is obtained using the training method of any battery temperature field prediction model provided in the aforementioned embodiment. By predicting the distribution of the battery temperature field, potential thermal runaway risks can be discovered in time, preventive measures can be taken, and the safety of battery use can be improved. If an abnormal temperature distribution is detected, the system can trigger an alarm, notify the operator, or automatically take measures.

[0155] For example, the battery temperature field distribution output by the model can be used to detect abnormalities and alarm the battery pack. If there is a temperature sensor (such as a temperature sensing resistor) in the battery management system BMS of the battery pack, the predicted temperature field distribution can be used to calibrate the temperature data collected by the BMS. The above calibration method can improve the accuracy of temperature measurement, thereby enhancing the BMS's ability to monitor the battery status. Accurate temperature prediction can help optimize the battery's charging and discharging strategies, avoid operating at extreme temperatures, and extend battery life.

[0156] According to the above embodiment, the present application also provides another optional embodiment: Fig.16 A schematic diagram of the process flow of the battery temperature field prediction method according to an embodiment of the present application is shown. Fig.16 As shown, the complete process of a battery temperature field prediction and management system is demonstrated, which is divided into a model training phase and an actual use phase.

[0157] In the model training stage, preprocessing involves the selection of battery feature points, data set division, and data normalization. Feature extraction and preprocessing are performed on the collected data to provide high-quality data input for model training and real-time prediction. By constructing a thermal model, the thermal model takes into account material properties such as thermal conductivity, specific heat capacity, current, voltage, and ambient temperature, as well as charging and discharging conditions as input parameters, and after processing, it is combined with the electrochemical model to obtain the state of the lithium-ion battery through simulation calculation. The CNN-Bi-LSTM-AM model provided in some embodiments is used for training. After the model training is completed, the accuracy of the prediction is improved by automatically optimizing the model parameters.

[0158] The actual use stage includes data collection, data preprocessing and real-time model calibration. In the data collection stage, the system collects signals such as current and temperature, and records the charge and discharge current and time, and obtains the real-time ambient temperature of the battery. After the above-mentioned actual data is normalized, the position in the battery pack expected to be predicted is input, that is, the three-dimensional coordinates are input. The processed data is input into the trained CNN-Bi-LSTM-AM model for solution to obtain the prediction results of the battery temperature field. These prediction results are used for anomaly detection and alarm, and real-time calibration with the battery management system (BMS) is performed.

[0159] Through the above-mentioned simulation calculation of the electrochemical-thermal model, the battery pack status data with high confidence under typical working conditions is obtained, thus providing a basis for model training. The CNN-Bi-LSTM-AM deep learning architecture is adopted, the convolutional neural network (CNN) is used to extract local spatial features, the bidirectional long short-term memory network (Bi-LSTM) is combined to capture the timing information, and the attention mechanism (AM) is introduced to dynamically adjust the importance of the input timing features. Through the complex nonlinear relationship between the working condition parameters and the battery temperature field, the real-time and accurate monitoring of the battery temperature is finally achieved. When the method predicts abnormal temperature conditions, it can provide guidance for the battery management system to achieve effective thermal management. This method helps to optimize the life and safety of the battery and provides an innovative solution for the development of thermal management technology for lithium-ion batteries. The entire process combines physical models and data-driven machine learning models to achieve accurate prediction and management of the battery temperature field, thereby improving the safety and reliability of battery use.

[0160] Fig.17 A schematic diagram showing the structure of a training device for a battery temperature field prediction model according to an embodiment of the present application is shown. Fig.17 As shown, corresponding to the application scenario and method of the method provided in the embodiment of the present application, the embodiment of the present application also provides a training device for a battery temperature field prediction model, including: A calculation example generation module 301 is used to generate a thermal runaway calculation example based on reference operating condition parameters and boundary condition temperature of the battery pack; A parameter determination module 302 is used to determine characteristic parameters corresponding to a plurality of modeling nodes of a battery pack in a thermal runaway calculation example, wherein the plurality of modeling nodes are used to represent different spatial positions on the battery pack; The screening module 303 is used to screen out modeling nodes that meet predetermined conditions from the multiple modeling nodes as feature points according to the feature parameters corresponding to the multiple modeling nodes; the spatial distance and temperature variation range between the feature point and other modeling nodes meet the predetermined conditions; A training set generation module 304 is used to generate a training set according to the reference operating condition parameters and the characteristic parameters corresponding to each characteristic point; The training module 305 is used to train the battery temperature field prediction model based on the training set to obtain a trained battery temperature field prediction model.

[0161] Fig.18 A schematic diagram showing the structure of a battery temperature field prediction device according to an embodiment of the present application is shown in FIG. Fig.18 As shown, corresponding to the application scenario and method of the method provided in the embodiment of the present application, the embodiment of the present application also provides a battery temperature field prediction device, including: The operating condition acquisition module 401 is used to obtain the current operating condition parameters of the battery pack to be tested; The model prediction module 402 is used to input the current operating condition parameters into the battery temperature field prediction model for processing to obtain the battery temperature field distribution of the battery pack at a future moment, wherein the battery temperature field prediction model is obtained by applying any of the battery temperature field prediction model training methods provided in the above embodiments.

[0162] The functions of each module in each device in the embodiments of the present application can be found in the corresponding description in the above method, and have corresponding beneficial effects, which will not be repeated here.

[0163] Fig.19 FIG. 1 is a block diagram of an electronic device used to implement an embodiment of the present application. Fig.19 As shown, the electronic device includes: a memory 501 and a processor 502, and the memory 501 stores a computer program that can be run on the processor 502. When the processor 502 executes the computer program, the method in the above embodiment is implemented. The number of the memory 501 and the processor 502 can be one or more. In a specific implementation, the electronic device may also include a communication interface 503 for communicating with external devices and performing data exchange transmission.

[0164] In specific implementation, if the memory 501, the processor 502 and the communication interface 503 are implemented independently, the memory 501, the processor 502 and the communication interface 503 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.19 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0165] Optionally, in a specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can communicate with each other through an internal interface.

[0166] An embodiment of the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a training method for a battery temperature field prediction model and a battery temperature field prediction method provided in an embodiment of the present application.

[0167] An embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements a training method for a battery temperature field prediction model and a battery temperature field prediction method provided in an embodiment of the present application.

[0168] An embodiment of the present application also provides a chip, which includes a processor for calling and executing instructions stored in the memory from the memory, so that a communication device equipped with the chip executes the training method of the battery temperature field prediction model and the battery temperature field prediction method provided in the embodiment of the present application.

[0169] An embodiment of the present application also provides a chip, including: an input interface, an output interface, a processor and a memory, wherein the input interface, the output interface, the processor and the memory are connected via an internal connection path, and the processor is used to execute the code in the memory. When the code is executed, the processor is used to execute the training method of the battery temperature field prediction model and the battery temperature field prediction method provided in the embodiment of the application.

[0170] It should be understood that the above processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. It is worth noting that the processor may be a processor that supports the Advanced RISC Machines (ARM) architecture.

[0171] Further, optionally, the above-mentioned memory may include a read-only memory and a random access memory. The memory may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. Among them, the non-volatile memory may include a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may include a random access memory (RAM), which is used as an external cache. By way of exemplary but not limiting description, many forms of RAM are available. For example, static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct memory bus random access memory (DR RAM).

[0172] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of 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, the process or function according to the present application is generated in whole or in part. 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 from one computer-readable storage medium to another computer-readable storage medium.

[0173] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0174] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0175] Any process or method described in the flow chart or otherwise described herein can be understood as a module, fragment or portion of a code representing one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0176] The logic and / or steps described in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0177] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0178] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0179] The above are only exemplary embodiments of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope recorded in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A training method for a battery temperature field prediction model, characterized in that: include: Generate thermal runaway examples based on the reference operating parameters and boundary condition temperatures of the battery pack; Determine characteristic parameters corresponding to a plurality of modeling nodes of the battery pack in the thermal runaway calculation example, wherein the plurality of modeling nodes are respectively used to represent different spatial positions on the battery pack; According to the characteristic parameters respectively corresponding to the multiple modeling nodes, a modeling node that meets a predetermined condition is selected from the multiple modeling nodes as a characteristic point; the spatial distance and the temperature variation amplitude between the characteristic point and other modeling nodes meet the predetermined condition; Generate a training set according to the reference operating condition parameters and the characteristic parameters corresponding to each characteristic point; Based on the training set, the battery temperature field prediction model is trained to obtain the trained battery temperature field prediction model.

2. The method according to claim 1, characterized in that The step of selecting modeling nodes that meet predetermined conditions from the plurality of modeling nodes as feature points based on the feature parameters respectively corresponding to the plurality of modeling nodes includes: Determine an average value of the distance between each candidate node and other modeling nodes in the candidate point set as a spatial uniformity parameter of the candidate point set; the candidate point set is a point set obtained by selecting some modeling nodes from the multiple modeling nodes as candidate nodes; Based on the characteristic parameters corresponding to each candidate node, a characteristic matrix representing the change of temperature over time is generated; Performing dimensionality reduction processing on the feature matrix to obtain temperature representative parameters of the candidate point set; When the spatial uniformity parameter and the temperature representative parameter satisfy the predetermined condition, each candidate node is used as the feature point.

3. The method according to claim 1, characterized in that The determining of characteristic parameters respectively corresponding to a plurality of modeling nodes of the battery pack in the thermal runaway calculation example comprises: Based on the polarization effect simulation of the battery pack, an ohmic overpotential, a polarization overpotential and a concentration overpotential in the battery pack are obtained; Determine, according to the ohmic overpotential, the polarization overpotential and the concentration overpotential, a first unit volume heat generation power corresponding to each of the plurality of modeling nodes under a normal temperature rise condition; Based on the first unit volume heat powers respectively corresponding to the plurality of modeling nodes, characteristic parameters respectively corresponding to the plurality of modeling nodes are determined.

4. The method according to claim 1, characterized in that: The determining of characteristic parameters respectively corresponding to a plurality of modeling nodes of the battery pack in the thermal runaway calculation example comprises: Determining a solid electrolyte interface decomposition rate, a negative solvent reaction rate, a positive solvent reaction rate, and an electrolyte decomposition reaction rate of the battery pack during thermal runaway; Determine, based on the solid electrolyte interface decomposition rate, the negative solvent reaction rate, the positive solvent reaction rate, and the electrolyte decomposition reaction rate, a second unit volume heat generation power corresponding to each of the plurality of modeling nodes under thermal runaway conditions; Based on the second unit volume heat powers respectively corresponding to the plurality of modeling nodes, characteristic parameters respectively corresponding to the plurality of modeling nodes are determined.

5. The method according to claim 1, characterized in that The determining of characteristic parameters respectively corresponding to a plurality of modeling nodes of the battery pack in the thermal runaway calculation example comprises: Determining a conditional short-circuit current of the battery pack according to a predetermined short-circuit temperature; Determining the activation energy, state of charge and frequency factor of the battery pack during an internal short circuit of the battery; Determine, according to the conditional short-circuit current, the activation energy, the state of charge, and the frequency factor, a third unit volume heat generation power corresponding to each of the plurality of modeling nodes under an internal short-circuit condition; Based on the third unit volume heat powers respectively corresponding to the plurality of modeling nodes, characteristic parameters respectively corresponding to the plurality of modeling nodes are determined.

6. The method according to claim 1, characterized in that The battery temperature field prediction model includes a convolutional neural network layer, a bidirectional long short-term memory network layer and a self-attention module. The battery temperature field prediction model is trained based on the training set to obtain the trained battery temperature field prediction model, including: Inputting the training set into the convolutional neural network layer for processing to obtain local spatial features in the battery pack; Inputting the local spatial features into the bidirectional long short-term memory network layer for processing to obtain the time series features of the battery pack between different time steps; Inputting the time series features into the self-attention module for processing to obtain the test temperature field distribution of the battery pack at a predetermined time; When the test temperature field distribution meets a predetermined iteration condition, the trained battery temperature field prediction model is obtained.

7. The method according to any one of claims 1 to 6, characterized in that: The reference operating condition parameters include multiple charge and discharge rate configurations, wherein each charge and discharge rate configuration includes at least one of a constant current curve, a linear current curve, and a secondary current curve. The boundary condition temperatures are multiple, and the multiple boundary condition temperatures include different coolant temperatures in the battery pack. The generating of a thermal runaway calculation example based on reference operating condition parameters and boundary condition temperature of the battery pack includes: Generate multiple configuration combinations based on the multiple charge and discharge rate configurations and the multiple boundary condition temperatures; A plurality of thermal runaway simulation examples are generated according to the plurality of configuration combinations and the material properties and electrochemical parameters of the battery pack.

8. A battery temperature field prediction method, characterized in that: include: Obtain the current operating parameters of the battery pack to be tested; The current operating condition parameters are input into a battery temperature field prediction model for processing to obtain the battery temperature field distribution of the battery pack at a future moment, wherein the battery temperature field prediction model is obtained by applying the training method of the battery temperature field prediction model described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method according to any one of claims 1 to 8.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes the program to implement the method according to any one of claims 1 to 8.

Images