CN116430245A - Battery thermal runaway prediction method based on gradient optimization multi-physical information neural network - Google Patents
Battery thermal runaway prediction method based on gradient optimization multi-physical information neural network Download PDFInfo
- Publication number
- CN116430245A CN116430245A CN202310699901.3A CN202310699901A CN116430245A CN 116430245 A CN116430245 A CN 116430245A CN 202310699901 A CN202310699901 A CN 202310699901A CN 116430245 A CN116430245 A CN 116430245A
- Authority
- CN
- China
- Prior art keywords
- neural network
- physical information
- thermal runaway
- lithium ion
- ion battery
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 112
- 238000000034 method Methods 0.000 title claims abstract description 76
- 238000005457 optimization Methods 0.000 title claims abstract description 23
- HBBGRARXTFLTSG-UHFFFAOYSA-N Lithium ion Chemical compound [Li+] HBBGRARXTFLTSG-UHFFFAOYSA-N 0.000 claims abstract description 74
- 229910001416 lithium ion Inorganic materials 0.000 claims abstract description 74
- 230000008569 process Effects 0.000 claims abstract description 36
- 238000005859 coupling reaction Methods 0.000 claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 30
- 230000008878 coupling Effects 0.000 claims abstract description 29
- 238000010168 coupling process Methods 0.000 claims abstract description 29
- 238000006243 chemical reaction Methods 0.000 claims description 36
- 239000004973 liquid crystal related substance Substances 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 21
- 239000003792 electrolyte Substances 0.000 claims description 12
- 238000000354 decomposition reaction Methods 0.000 claims description 10
- 230000036961 partial effect Effects 0.000 claims description 8
- 230000020169 heat generation Effects 0.000 claims description 7
- 238000002144 chemical decomposition reaction Methods 0.000 claims description 6
- 238000007725 thermal activation Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 4
- 238000001994 activation Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 239000013543 active substance Substances 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000002441 reversible effect Effects 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 abstract description 6
- 238000004364 calculation method Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 239000007784 solid electrolyte Substances 0.000 description 2
- WHXSMMKQMYFTQS-UHFFFAOYSA-N Lithium Chemical compound [Li] WHXSMMKQMYFTQS-UHFFFAOYSA-N 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000002427 irreversible effect Effects 0.000 description 1
- 229910052744 lithium Inorganic materials 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 238000005293 physical law Methods 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
- H01M10/00—Secondary cells; Manufacture thereof
- H01M10/42—Methods or arrangements for servicing or maintenance of secondary cells or secondary half-cells
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/367—Software therefor, e.g. for battery testing using modelling or look-up tables
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
- G06F17/12—Simultaneous equations, e.g. systems of linear equations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0499—Feedforward networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C10/00—Computational theoretical chemistry, i.e. ICT specially adapted for theoretical aspects of quantum chemistry, molecular mechanics, molecular dynamics or the like
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/10—Analysis or design of chemical reactions, syntheses or processes
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/70—Machine learning, data mining or chemometrics
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/10—Energy storage using batteries
Abstract
The application belongs to the technical field of batteries, and provides a battery thermal runaway prediction method based on a gradient optimization multi-physical information neural network, which comprises the following steps: establishing a multi-parameter coupling model for thermal runaway of the lithium ion battery; constructing a multi-physical information neural network, wherein the input of the multi-physical information neural network is time, ambient temperature and at least two spatial distribution variables of the lithium ion battery, the output is the temperature of the lithium ion battery and the dimensionless concentration of lithium ions contained in the lithium ion battery, and a loss function is determined based on the multi-parameter coupling model; training the multi-physical information neural network by using training data and optimizing the counter-propagation gradient in the training process; a trained multi-physical information neural network is used to predict thermal runaway processes of lithium ion batteries. According to the method, the training process of the neural network prediction model is supervised and optimized based on a multi-parameter physical mechanism of thermal runaway of the lithium ion battery, and the prediction precision and the robustness degree can be effectively improved.
Description
Technical Field
The application belongs to the technical field of batteries, relates to a battery safety management technology, and particularly provides a battery thermal runaway prediction method based on a gradient optimization multi-physical information neural network.
Background
The Lithium Ion Battery (LIB) has certain advantages in Battery schemes used by new energy vehicles such as electric automobiles, electric ships and the like due to the characteristics of high energy density, long duration, quick charge rate and the like, and the safety problem existing in the use process of the Lithium Ion Battery is not negligible. The main safety issue of lithium ion batteries is thermal runaway, which refers to an irreversible state in lithium ion batteries, where the local battery temperature rises and causes combustion due to initial internal short circuits until the internal reactive species of the LIB component are burned or even cause explosion of the lithium ion battery. Real-time prediction of thermal runaway of lithium ion batteries and estimation of thermal runaway phenomena under various operating conditions is an active area of research, as it not only predicts thermal runaway, but also provides information about the optimal design of lithium ion battery structures (e.g., optimal design of surface to volume ratio).
There are two main embodiments of predicting thermal runaway processes of lithium ion batteries currently available: numerical simulation is carried out on a thermal runaway process by using a Finite Element Method (FEM), a Finite Difference Method (FDM) and the like, and the physical mechanism of the method is clear and visual, but a large amount of calculation work is needed, and the method cannot respond in time to random and rapid changing environments and various actual operating conditions encountered by a lithium ion battery, so that the method is difficult to be applied to rapid prediction or estimation of the thermal runaway of the lithium ion battery; the method is implemented by using a deep learning neural network and predicting the thermal runaway process of the lithium ion battery, but the method adopts pure data driving as a main part, and repeatedly trains and learns from a given neural network structure and obtained training data to obtain a specific prediction model, namely, a mapping relation between input data and output data is established.
In recent years, physical information neural networks (Physics Informed Neural Networks, PINN) have been widely studied as prospective proxy models for many time-containing systems in engineering, and a major advantage of physical information neural networks is that they can quickly model time-dependent systems of interest without losing generality and accuracy, fundamentally, PINN differ from common data-driven deep learning models in that they are trained in conjunction with the laws of dominance of physics to ensure that they do not violate these laws and produce robust results against outliers.
However, there is no mature physical information neural network applied to lithium ion battery thermal runaway prediction at present, and the main reason is that: firstly, although the neural network learns and optimizes based on a physical model in the optimization process in the training process, the neural network does not necessarily ensure a perfect conformation of non-visible data points and a physical control equation; secondly, studies have found that the convergence of the PINN during training is largely dependent on the physical type involved, since the control equations are provided as soft constraints in the loss function, and are susceptible to out-of-distribution data; moreover, solving these problems using PINN is more challenging and difficult from a mathematical perspective due to the high degree of non-convexity of the loss function and the imbalance caused when optimizing multiple loss functions simultaneously.
Therefore, if the multi-physical information neural network is adopted to predict the thermal runaway of the lithium ion battery, the key point is how to build a multi-parameter coupling physical model for accurately describing the thermal runaway process and reasonably build a coupling training mechanism of the physical model to the neural network.
Disclosure of Invention
The invention aims to solve the problems in the prior art and provide a battery thermal runaway prediction method based on a gradient optimization multi-physical information neural network, which can monitor and optimize the training process of a neural network prediction model by utilizing a multi-parameter physical mechanism of thermal runaway of a lithium ion battery so as to ensure that the training of the prediction model is performed under the control of a physical law with practical significance and generate a steady prediction result conforming to an actual physical image.
The embodiment of the application can be realized through the following technical scheme:
a battery thermal runaway prediction method based on a gradient optimization multi-physical information neural network is used for predicting a thermal runaway process of a lithium ion battery and comprises the following steps:
establishing a multi-parameter coupling model for thermal runaway of the lithium ion battery;
constructing a multi-physical information neural network, wherein the input of the multi-physical information neural network is timeAmbient temperature->And at least two spatially distributed variables of the lithium ion battery +.>、/>The output is the temperature of the lithium ion battery>And lithium ions contained thereinDimensionless concentration ∈10->Loss function->Determining based on the multiparameter coupling model;
training the multi-physical information neural network by using training data, wherein the reverse propagation gradient of the multi-physical information neural network is optimized based on the multi-parameter coupling model in the training process;
and predicting the thermal runaway process of the lithium ion battery by using the trained multi-physical information neural network.
Preferably, the multiparameter coupling model comprises a first set of equations describing the thermodynamic reaction during thermal runaway of the lithium ion battery, and a second set of equations describing the chemical degradation reaction of various mediums during thermal runaway of the lithium ion battery; the first equation set and the second equation set are based on the rate of heat generation by the volumeAre coupled to each other.
Further, the first equation set is specifically:
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>、/>for the density, heat capacity and heat conductivity of lithium ion battery respectively, < >>The temperature of lithium ions in the electrolyte of a lithium ion battery, which is the distribution space of lithium ions +.>Time->Function of->Is->Is (are) calculated domain->To calculate the upper time limit +.>For convection heat transfer coefficient>、/>、/>、/>、/>、/>The reaction enthalpy, the specific active substance content per unit volume, the reaction factor, the activation energy, the molar gas constant and the reaction progression are respectively +.>、/>Initial values of temperature and dimensionless concentration respectively;
the second equation set is specifically:
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>heat generation rate of positive electrode and electrolyte decomposition, respectively, +.>、/>、/>Frequency factor, reaction coefficient and thermal activation energy of positive electrode decomposition reaction respectively, +.>、/>Respectively->、/>Is used for the reaction series of (a),、/>、/>、/>the frequency factor, the thermal activation energy, the lithium ion concentration and the reaction progression of the electrolyte decomposition reaction are respectively shown.
Preferably, the multi-physical information neural network comprises a first neural network and a second neural network which respectively comprise two independence; the first and second neural networks share the same input、/>、/>、/>The method comprises the steps of carrying out a first treatment on the surface of the The output of the first neural network is temperature +.>The output of the second neural network is the dimensionless concentration +.>Is a predicted value of (a).
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>、/>、/>is->Data fitting loss, partial differential equation loss, ordinary differential equation loss, boundary condition loss and initial condition loss, respectively, < >>、/>、/>、/>、/>Respectively, weighting coefficients thereof, ">、/>、/>、/>Respectively->、/>、/>、/>Predicted value of +.>,/>,,/>,/>,/>,/>、/>Variable space distribution variables ∈>、/>Corresponding heat conductivity->、/>、/>Respectively->、/>、/>Upper limit of the value of (2).
Preferably, the multi-physical information neural network further includes a voltage transformation layer, and the step 300 of optimizing a back propagation gradient of the multi-physical information neural network based on the multi-parameter coupling model in the training process, specifically, performing the following steps in back propagation:
Wherein, the liquid crystal display device comprises a liquid crystal display device,rateis a preset learning rate;
Preferably, optimizing the counter-propagating gradient of the multi-physical information neural network is only precededNAnd performed in the second back propagation.
Preferably, the battery thermal runaway prediction method based on the gradient optimization multi-physical information neural network further comprises the steps of inputting、/>、/>、/>Output->、/>And performing non-dimensionalization processing.
According to the battery thermal runaway prediction method based on the gradient optimization multi-physical information neural network, a multi-parameter coupling model describing various reaction mechanisms in the thermal runaway process of the lithium ion battery is established, and a loss function with an interconnection structure is generated by utilizing the multi-parameter coupling model, so that the output of the mutually independent neural network sharing input is constrained in a coupling mode, and the multi-physical information neural network trained under the driving and supervision of the coupling reaction mechanisms can be used for predicting the thermal runaway process of the lithium ion battery more accurately, and the prediction robustness is remarkably improved.
Drawings
FIG. 1 is a flow chart of a battery thermal runaway prediction method based on a gradient optimized multi-physical information neural network provided according to an embodiment of the present application;
fig. 2 is a schematic architecture diagram of a multi-physical information neural network according to an embodiment of the present application;
FIG. 3 is a schematic diagram of steps for implementing a method for predicting thermal runaway of a battery based on a gradient-optimized multi-physical information neural network according to an embodiment of the present application;
FIG. 4 is a schematic diagram of a prediction result of a multi-physical information neural network on a temperature of a lithium ion battery according to an embodiment of the present application;
fig. 5 is a partially enlarged schematic view of the prediction result in fig. 4.
Detailed Description
The present application will be further described below based on preferred embodiments with reference to the accompanying drawings.
By way of example, the present application provides a battery thermal runaway prediction method based on a gradient-optimized multi-physical information neural network, and fig. 1 shows a flowchart of implementation of the prediction method in some preferred embodiments, as shown in fig. 1, and the method includes the following steps:
The steps described above are described in detail below with reference to the drawings and the specific embodiments.
In the examples of the present application, step 100 is used to construct a multiparameter coupling model describing a thermal runaway process of a lithium ion battery, which refers to the occurrence of a continuous chain reaction of component degradation due to an increase in temperature or an increase in current and power dissipation caused by an exothermic reaction, which is a complex multiphysics-chemical reaction process including various thermodynamic reactions and chemical degradation reactions of various components.
Specifically, the thermodynamic reaction in the thermal runaway process of a lithium ion battery can be described by the following formula (1):
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>、/>for the density, heat capacity and heat conductivity of lithium ion battery respectively, < >>The temperature of lithium ions in the electrolyte of a lithium ion battery, which is the distribution space of lithium ions +.>Time->Function of->Is->Is (are) calculated domain->To calculate the upper time limit +.>Is the rate of heat generation by the volume.
Equation (1) above is an energy balance equation during thermal runaway of a lithium ion battery, describing the energy balance relationship among the rate of change of temperature, heat conduction and rate of heat generation of a volume, wherein、/>The specific forms of (2) and (3) are given by the following formulas (2) and (3), respectively:
wherein equation (2) describes convective heat exchange between the lithium ion battery and the external environment,for convective heat transfer coefficient, (3) derives from exothermic reaction caused by stimulated chemical reaction, and dimensionless concentration of lithium in electrolyte of lithium ion battery +.>Is proportional to the rate of change of (in particular,)>、/>、/>、/>、/>、/>The reaction enthalpy, the specific active substance content per unit volume, the reaction factor, the activation energy, the molar gas constant and the reaction progression are respectively shown.
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>the initial values of temperature and dimensionless concentration, respectively.
Meanwhile, in the thermal runaway process of a lithium ion battery, exothermic reaction caused by continuous chemical degradation of various components is an important cause of rapid temperature rise, and in general, chemical degradation occurs in the following order: the Solid Electrolyte Interface (SEI), the anode, the cathode and the electrolyte, the heat generated by the exothermic reaction is simply the sum of each contribution, and therefore,can be expressed in the form of the following formula (5):
wherein the method comprises the steps of、/>、/>、/>The solid electrolyte interface, the negative electrode, the positive electrode and the rate of heat generation by the electrolyte decomposition are respectively +.>、/>Is the main factor of the degradation exotherm during thermal runaway, therefore, in some preferred embodiments, only +.>、/>:
、/>May be represented by Arrhenius' law, e.gIn some specific embodiments, the information may be, in some embodiments,、/>can be represented by the following formulas (6), (7), respectively:
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>、/>frequency factor, reaction coefficient and thermal activation energy of positive electrode decomposition reaction respectively, +.>、Respectively->、/>Reaction progression of>、/>、/>、/>The frequency factor, the thermal activation energy, the lithium ion concentration and the reaction progression of the electrolyte decomposition reaction are respectively shown.
The equation describing the model reaction rate differs, and the temperature reaches a level over time, which results in a significant difference in rate, thereby explaining the fundamental difference in synthesis temperature between the positive electrode decomposition reaction and the electrolyte decomposition reaction.
The formulas (1) to (7) are multi-parameter coupling models for describing the thermal runaway process of the lithium ion battery, and the thermal runaway process of the lithium ion battery is predicted, namely, the equation of the models is solved to obtain the pair serving as a distribution spaceTime->And ambient temperature->Temperature of the function of>Dimensionless concentration->Is a prediction of (2).
After the multi-parameter coupling model is built through step 100, a multi-physical information neural network can be constructed through step 200, and training can be performed on the multi-physical information neural network through step 300.
In an embodiment of the present application, the multi-physical information neural network constructed in step 200 is input as timeAmbient temperature->And at least two spatially distributed variables of the lithium ion battery +.>(e.g. voltage, density, etc. are +.>Variable with a specific distribution on) output is +.>、/>The loss function is determined based on the multiparameter coupling model established in step 100.
The conventional form of physical information neural networks (Physics-Informed Neural Network, PINN) and the manner in which they are trained are known to those skilled in the art, and in general, the network model form of the PINN may be the same as that of a conventional deep neural network with time prediction capability, for example, may include an input layer, a plurality of intermediate layers, an output layer, and the like, unlike a conventional deep neural network, the PINN approximates the solution of a Partial Differential Equation (PDE) by adding an equation (also referred to as a control equation) describing a physical model as a penalty term or constraint term to an empirical loss function of the deep neural network, so that the data fitting and control equation after parameter optimization is satisfied, i.e., the model output thereof is equal to that of an equation or a system response of a time-varying system.
Specifically, the loss function of the multi-physical information neural network provided by the application is formed by adding the formulas (1) to (4) as penalty terms in a conventional empirical loss function, wherein the formulas (3) and (4) are boundary conditions and initial conditions, respectively, which will make the unique solution of the controlled partial differential equation of the formula (1), and such a physical-based constraint is also commonly called residual loss; while conventional empirical losses are also commonly referred to as data fitting losses, ideally, when each residual and data fitting loss is equal to zero, the optimization is complete, meaning that the dominant laws of physics, boundary conditions, and initial conditions are satisfied.
Furthermore, complex reactions of lithium ion batteries during thermal runaway are involved at the same timeThermodynamic and chemical degradation processes, each of formulas (1) to (4) and (5) to (7) are described by differential equations, respectively, and are performed byThe coupling relation between the physical correlation factors and given data is established, and the neural network is supervised together, so that the accuracy and the robustness of the PINN are improved, therefore, in the embodiment of the application, the thermal runaway prediction estimation is carried out through the PINN, each dynamic process is required to be modeled and trained respectively and simultaneously by the independent neural network, and the training of each dynamic process is required to be supervised in a coupling way by a unified penalty term, which is also the reason that the neural network provided by the application is called a multi-physical information neural network.
In particular, in some preferred embodiments of the present application, the loss functionCan be expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>、/>、/>is->Data fitting loss, partial differential equation loss (which represents the part of the loss function driven by the multiparameter coupling model described above), ordinary differential equation loss, boundary condition loss, and initial condition loss, respectively,/">、/>、/>、/>、/>Respectively, their weighting coefficients.
The specific expression of each of the above-described loss functions is given by the following formulas (9) to (13):
wherein, the liquid crystal display device comprises a liquid crystal display device,is mean square error>、/>、/>、/>Respectively->、/>、/>、/>Is a predicted value of (a).
FIG. 2 illustrates a specific architecture of the multiple physical information neural network established by step 200 in some preferred embodiments, as illustrated in FIG. 2, where the multiple physical information neural network includes two independent first (neural network T) and second (neural network C) neural networks, respectively, each of which may employ a neural network architecture with time prediction capability known to those skilled in the art that shares the same inputs, respectively learns and outputs temperaturesAnd non-dimensional concentration of lithium ions +.>After the partial derivative term is calculated by automatic differentiation of the above-mentioned outputs, the total loss function is constructed using the loss terms of the formula>:
Wherein, the liquid crystal display device comprises a liquid crystal display device,,/>,/>,/>,,/>,/>、/>variable space distribution variables ∈>、/>The thermal conductivity coefficient of the material is corresponding to that of the material,、/>、/>respectively->、/>、/>Upper limit of the value of (2).
Among the above-mentioned loss terms, what is important is the PDE loss, i.e., the energy balance equation, which, due to its coupling with the outputs of two differently trained networks, enables it to take into account the multi-physical characteristics of thermal runaway of the lithium ion battery and makes the multi-physical information neural network proposed in this application have a structure of interconnected loss functions.
The training data is input into the neural network, the loss function is calculated based on the output, and the training mode of updating the weights of each layer of the neural network through back propagation is known to those skilled in the art, and in some preferred embodiments, in order to prevent the gradient explosion phenomenon, as shown in fig. 2, a transformation layer (transformer) is arranged between the input and the first and second neural networks, so as to smooth the diffusion term of the nonlinear equilibrium equation.
Meanwhile, since gradient change speeds of different items of the loss function are different, in order to avoid severe oscillation of gradient values during model training caused by unbalanced back propagation gradient calculation of the multi-physical information neural network and further unstable prediction accuracy, as shown in fig. 2, in the training process of the multi-physical information neural network, the back propagation gradient of the multi-physical information neural network is optimized based on the multi-parameter coupling model so as to balance interaction among different items in the loss function during model training.
Specifically, for each iteration the total loss functionGradient operation (i.e. acquire +.>) Then find +.>And->Maximum value of item gradient, designated->I.e. +.>The method comprises the steps of carrying out a first treatment on the surface of the Respectively find +.>And->Is recorded as->、/>I.e. +.>、/>;
Then, the learning rate is setrateAnd calculate the optimization coefficient based on the following formulas (14), (15)、/>:
Then based on、/>Counter-propagating is performed to update the weights of the transformer layers while based on propagation loss +.>The weights of the first and second neural networks are updated by conventional back propagation.
Further, in some preferred embodimentsIn an example, can use、/>For->、/>The coefficients of (2) are updated to achieve the purpose of gradient optimization.
In some preferred embodiments, the above-mentioned utilization is used to further improve the high prediction accuracy of the physical information neural network while reducing the calculation time、/>Optimization of the transformer layer weights may be precededNIn a plurality of iterations (whereinNIs a preset positive integer), inN+In 1 and later iterations, only the weights of the first neural network and the second neural network are updated, so that the calculation time is reduced while the high prediction precision of the physical information neural network is improved.
In addition, the difference between the input and output scales is another factor causing the weight imbalance between the loss terms, in order to eliminate the problem that the physical quantity normalized neural network output does not match the partial differential equation having the physical dimensions, the variables and coefficients of the partial differential equation, that is, the input and output of the multi-physical information neural network should also be normalized and dimensionless processed, specifically, the non-dimensionalization processing is performed on each of the input and output variables using the following formula (16):
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>respectively->、/>Lower limit of the value of (2).
Accordingly, the formula (1), the formula (3) and the formula (4) become:
after the construction and training of the multi-physical information neural network model are completed through the steps 200 and 300, the trained multi-physical information neural network can be used to predict the thermal runaway process of the lithium ion battery in the step 400, and the prediction of specific parameters or variables using the trained neural network model is known to those skilled in the art and will not be described herein.
Fig. 3 shows a flow chart of the implementation of the prediction method in a specific embodiment that first generates a sufficient amount of data using a COMSOL Multiphysics 5.6.6 high-fidelity model and then provides the data for training to a multi-physical information neural network model that includes two different neural networks, where the neural network T is trained to predict the evolution of the temperature profile in space and time, and the neural network C is trained to predict the concentration of reactive species that degrade over time due to temperature rise, each neural network being assigned a different task, thus each network requires a different loss function, however, the networks are optimized simultaneously because they are coupled to each other by the overall control equation.
FIG. 4 shows the prediction results of the multi-physical information neural network on the temperature of the lithium ion battery provided by the embodiment, wherein the prediction results of the case of setting the learning rate optimization and the case of not setting the learning rate optimization are provided respectively and compared with the actually measured reference values; fig. 5 shows the predicted result of fig. 4 at 230 to 270 minutes in an enlarged manner.
From fig. 4 and fig. 5, it can be seen that, by using the battery thermal runaway prediction method based on the gradient optimization multi-physical information neural network provided by the application, the temperature and dimensionless concentration changes along with time in the thermal runaway process of the lithium ion battery can be accurately predicted, especially, the temperature and dimensionless concentration changes along with time have good prediction effects at the inflection point where the state is suddenly changed, and further, the prediction capability at the mutation point is further improved under the condition of setting the learning rate.
While the foregoing is directed to embodiments of the present application, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.
Claims (8)
1. A battery thermal runaway prediction method based on a gradient optimization multi-physical information neural network is used for predicting a thermal runaway process of a lithium ion battery and is characterized by comprising the following steps of:
establishing a multi-parameter coupling model for thermal runaway of the lithium ion battery;
constructing a multi-physical information neural network, wherein the input of the multi-physical information neural network is timeAmbient temperatureAnd at least two spatially distributed variables of the lithium ion battery +.>、/>The output is the temperature of the lithium ion battery>And the dimensionless concentration of lithium ions contained therein +.>Loss function->Determining based on the multiparameter coupling model;
training the multi-physical information neural network by using training data, wherein the reverse propagation gradient of the multi-physical information neural network is optimized based on the multi-parameter coupling model in the training process;
and predicting the thermal runaway process of the lithium ion battery by using the trained multi-physical information neural network.
2. The battery thermal runaway prediction method based on the gradient optimization multi-physical information neural network according to claim 1, wherein the method is characterized by comprising the following steps of:
the multi-parameter coupling model comprises a first equation set for describing a thermodynamic reaction in the thermal runaway process of the lithium ion battery and a second equation set for describing chemical degradation reactions of various mediums in the thermal runaway process of the lithium ion battery;
3. The battery thermal runaway prediction method based on the gradient optimization multi-physical information neural network according to claim 2, wherein the first equation set specifically comprises:
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>、/>for the density, heat capacity and heat conductivity of lithium ion battery respectively, < >>The temperature of lithium ions in the electrolyte of a lithium ion battery, which is the distribution space of lithium ions +.>Time->Function of->Is->Is (are) calculated domain->To calculate the upper time limit +.>For convection heat transfer coefficient>、/>、/>、/>、/>、/>The reaction enthalpy, the specific active substance content per unit volume, the reaction factor, the activation energy, the molar gas constant and the reaction progression are respectively +.>、/>Initial values of temperature and dimensionless concentration respectively;
the second equation set is specifically:
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>heat generation rate of positive electrode and electrolyte decomposition, respectively, +.>、/>、/>Respectively is positive electrodeFrequency factor, reaction coefficient and thermal activation energy of the solution reaction,/->、/>Respectively->、/>Reaction progression of>、、/>、/>The frequency factor, the thermal activation energy, the lithium ion concentration and the reaction progression of the electrolyte decomposition reaction are respectively shown.
4. The battery thermal runaway prediction method based on the gradient optimization multi-physical information neural network according to claim 1, wherein the method is characterized by comprising the following steps of:
the multi-physical information neural network comprises a first neural network and a second neural network which are independent;
5. The gradient-optimized multi-physical information neural network-based battery thermal runaway prediction method of claim 3, wherein the loss functionThe method comprises the following steps:
wherein, the liquid crystal display device comprises a liquid crystal display device,、/>、/>、/>is->Data fitting loss, partial differential equation loss, ordinary differential equation loss, boundary condition loss and initial condition loss, respectively, < >>、/>、/>、/>、/>Respectively, weighting coefficients thereof, ">、/>、/>、/>Respectively->、/>、/>、/>Predicted value of +.>,/>,/>,/>,/>,/>,/>、/>Variable space distribution variables ∈>、/>Corresponding heat conductivity->、/>、/>Respectively->、/>、/>Upper limit of the value of (2).
6. The method for predicting thermal runaway of a battery based on a gradient-optimized multi-physical information neural network of claim 5,
the multi-physical information neural network further includes a voltage transformation layer, and,
in step 300, the optimization of the back propagation gradient of the multi-physical information neural network based on the multi-parameter coupling model in the training process is specifically performed in the back propagation:
Wherein, the liquid crystal display device comprises a liquid crystal display device,rateis a preset learning rate;
7. The battery thermal runaway prediction method based on the gradient-optimized multi-physical information neural network according to claim 6, wherein the method is characterized by:
optimizing the counter-propagating gradient of the multi-physical information neural network is only precededNAnd performed in the second back propagation.
8. The battery thermal runaway prediction method based on the gradient optimization multi-physical information neural network according to claim 1, wherein the method is characterized by comprising the following steps of:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310699901.3A CN116430245A (en) | 2023-06-14 | 2023-06-14 | Battery thermal runaway prediction method based on gradient optimization multi-physical information neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310699901.3A CN116430245A (en) | 2023-06-14 | 2023-06-14 | Battery thermal runaway prediction method based on gradient optimization multi-physical information neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116430245A true CN116430245A (en) | 2023-07-14 |
Family
ID=87081928
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310699901.3A Withdrawn CN116430245A (en) | 2023-06-14 | 2023-06-14 | Battery thermal runaway prediction method based on gradient optimization multi-physical information neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116430245A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116736141A (en) * | 2023-08-10 | 2023-09-12 | 锦浪科技股份有限公司 | Lithium battery energy storage safety management system and method |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104849675A (en) * | 2015-06-17 | 2015-08-19 | 哈尔滨工业大学 | Method for obtaining electrochemical and thermal coupling models of lithium ion battery |
CN110161423A (en) * | 2019-06-26 | 2019-08-23 | 重庆大学 | A kind of dynamic lithium battery state joint estimation method based on various dimensions coupling model |
CN113344024A (en) * | 2021-04-22 | 2021-09-03 | 华中科技大学 | Lithium ion battery thermal runaway grading early warning method and early warning system |
CN114355199A (en) * | 2021-12-30 | 2022-04-15 | 重庆长安新能源汽车科技有限公司 | Battery thermal runaway risk prediction method and system based on recurrent neural network |
CN114966451A (en) * | 2022-06-06 | 2022-08-30 | 北京航空航天大学 | Power supply product residual service life online prediction method based on physical information neural network |
CN115438588A (en) * | 2022-09-29 | 2022-12-06 | 中南大学 | Temperature prediction method, system, equipment and storage medium of lithium battery |
CN115993551A (en) * | 2023-03-01 | 2023-04-21 | 北京航空航天大学 | Lithium ion battery diagnosis method integrating physical information parameters and electrochemical impedance spectrum |
-
2023
- 2023-06-14 CN CN202310699901.3A patent/CN116430245A/en not_active Withdrawn
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104849675A (en) * | 2015-06-17 | 2015-08-19 | 哈尔滨工业大学 | Method for obtaining electrochemical and thermal coupling models of lithium ion battery |
CN110161423A (en) * | 2019-06-26 | 2019-08-23 | 重庆大学 | A kind of dynamic lithium battery state joint estimation method based on various dimensions coupling model |
CN113344024A (en) * | 2021-04-22 | 2021-09-03 | 华中科技大学 | Lithium ion battery thermal runaway grading early warning method and early warning system |
CN114355199A (en) * | 2021-12-30 | 2022-04-15 | 重庆长安新能源汽车科技有限公司 | Battery thermal runaway risk prediction method and system based on recurrent neural network |
CN114966451A (en) * | 2022-06-06 | 2022-08-30 | 北京航空航天大学 | Power supply product residual service life online prediction method based on physical information neural network |
CN115438588A (en) * | 2022-09-29 | 2022-12-06 | 中南大学 | Temperature prediction method, system, equipment and storage medium of lithium battery |
CN115993551A (en) * | 2023-03-01 | 2023-04-21 | 北京航空航天大学 | Lithium ion battery diagnosis method integrating physical information parameters and electrochemical impedance spectrum |
Non-Patent Citations (1)
Title |
---|
SUNG WOOK KIM 等: "Modeling and prediction of lithium-ion battery thermal runaway via multiphysics-informed neural network", 《JOURNAL OF ENERGY STORAGE》, vol. 60, pages 3 - 12 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116736141A (en) * | 2023-08-10 | 2023-09-12 | 锦浪科技股份有限公司 | Lithium battery energy storage safety management system and method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Geng et al. | Bridging physics-based and equivalent circuit models for lithium-ion batteries | |
Garg et al. | Design and analysis of capacity models for Lithium-ion battery | |
Fang et al. | A prediction model based on artificial neural network for surface temperature simulation of nickel–metal hydride battery during charging | |
Ansari et al. | Discharge, rest and charge simulation of lead-acid batteries using an efficient reduced order model based on proper orthogonal decomposition | |
CN116430245A (en) | Battery thermal runaway prediction method based on gradient optimization multi-physical information neural network | |
CN112964992B (en) | Method, device and medium for processing temperature information in battery based on AUKF | |
CN110336057B (en) | Construction method and application of two-dimensional temperature distribution observer of cross-flow galvanic pile | |
Zhang et al. | Internal temperature prediction model of the cylindrical lithium-ion battery under different cooling modes | |
Li et al. | A novel state of charge estimation method of lithium‐ion batteries based on the IWOA‐AdaBoost‐Elman algorithm | |
Lai et al. | New data optimization framework for parameter estimation under uncertainties with application to lithium-ion battery | |
Yu et al. | Parallel battery pack charging strategy under various ambient temperatures based on minimum lithium plating overpotential control | |
CN115587512A (en) | ANSYS TwinBuilder-based lithium battery thermoelectric coupling digital twin model construction method | |
Parhizi et al. | An iterative analytical model for aging analysis of Li-ion cells | |
Abdullaeva et al. | Optimal variable estimation of a Li-ion battery model by fractional calculus and bio-inspired algorithms | |
Singh et al. | A two-layer framework for optimal control of battery temperature and microgrid operation | |
Namdari et al. | A multiscale entropy-based long short term memory model for lithium-ion battery prognostics | |
Deng et al. | Physics-dominated neural network for spatiotemporal modeling of battery thermal process | |
Zhou et al. | Data-driven real-time prediction of pouch cell temperature field under minimal sensing | |
Guo et al. | Nonlinear state-variable method for solving physics-based li-ion cell model with high-frequency inputs | |
Harb et al. | Mathematical Model of the Discharge Behavior of a Spirally Wound Lead‐Acid Cell | |
Nateghi et al. | A thermo-chemo-mechanically coupled model for cathode particles in lithium–ion batteries | |
CN110059342B (en) | Parameter estimation method for P2D model of lithium ion battery | |
Arpino et al. | A robust model and numerical approach for solving solid oxide fuel cell (SOFC) problems | |
WO2024016500A1 (en) | Real-time temperature measurement method for traction battery pack | |
Hao et al. | Adaptive Model-Based Reinforcement Learning for Fast Charging Optimization of Lithium-Ion Batteries |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20230714 |