A lithium battery internal temperature prediction and reconstruction method based on multi-scale feature fusion

Abstract

This invention discloses a method for predicting and reconstructing the internal temperature of lithium batteries based on multi-scale feature fusion, relating to the field of lithium battery internal temperature prediction. The method includes constructing an electro-chemical-thermal coupled model of a lithium battery module, obtaining a temperature field dataset using finite element simulation and physical experiments, inputting the temperature field dataset into a network model based on multi-scale feature fusion to estimate the initial spatial distribution of the internal temperature of the lithium battery module, reconstructing the final spatiotemporal temperature distribution using weighted spatiotemporal fusion, constructing a state-space model based on the final spatiotemporal temperature distribution, and deriving the cumulative distribution function of the remaining thermal runaway time to predict the remaining thermal runaway time, thus achieving early warning. This invention overcomes the limitations of complex physical and chemical models, facilitating online monitoring and providing early warning decision support, thereby improving safety and efficiency.

Description

TECHNICAL FIELD

[0001] The application relates to the field of lithium battery internal temperature prediction, and in particular to a lithium battery internal temperature prediction and reconstruction method based on multi-scale feature fusion. BACKGROUND

[0002] Fast charging is an effective way to overcome the bottleneck of the development of the electric vehicle industry, such as range anxiety and long charging time. How to realize safe and healthy high-power fast charging of electric vehicles is a challenging and practically significant problem.

[0003] The internal temperature of the lithium battery monomer / module is a core parameter representing thermal runaway. High current fast charging easily leads to an increase in battery temperature, aggravation of side reactions, emission of a large amount of heat and harmful gas, and causes the battery to catch fire and explode. The lithium battery module is only equipped with a limited number of temperature sensors to measure the average value as the surface temperature, which is far from the actual surface temperature field. The existing fast charging control strategy often replaces the internal temperature of the battery with the surface temperature, and the difference between the internal and external temperatures is large, which leads to serious lag in thermal runaway diagnosis.

[0004] How to use the limited temperature sensor monitoring value to accurately depict the surface and internal temperature field of the lithium battery and realize intelligent early warning of thermal runaway has become one of the basic problems of lithium battery health management. At present, the methods for realizing temperature monitoring of the lithium battery module and effectively early warning of thermal runaway mainly include physical models and chemical models. However, the existing methods have great limitations in application range and complex process, and are difficult to implement online.

[0005] In order to solve these problems, a lithium battery internal temperature prediction and reconstruction method based on multi-scale feature fusion is urgently needed. SUMMARY

[0006] To solve the above problems, the application provides a lithium battery internal temperature prediction and reconstruction method based on multi-scale feature fusion, which constructs a surface temperature field of a lithium battery module, estimates an internal temperature field of the lithium battery module, and realizes thermal runaway early warning based on evolution of the temperature field of the lithium battery module. The surface temperature field construction provides a data basis for the internal temperature field estimation of the lithium battery module, and the internal temperature field estimation of the lithium battery module makes the thermal runaway early warning more effective and accurate.

[0007] A lithium battery internal temperature prediction and reconstruction method based on multi-scale feature fusion uses a machine learning method to infer the internal temperature, realizes spatiotemporal reconstruction of the internal temperature, and ensures that the local battery monomer temperature is disconnected before a fire / explosion is caused by sudden temperature change, and the method comprises the following steps:

[0008] S1, constructing an electro-chemical-thermal coupling model of a lithium battery module, and obtaining a temperature field data set by using finite element simulation experiments and physical experiments;

[0009] S2, a network model based on multi-scale feature fusion is constructed, and the temperature field dataset is input into the network model based on multi-scale feature fusion to perform multi-view feature fusion to estimate the initial internal temperature spatial distribution of the lithium battery module;

[0010] S3, the initial internal temperature spatial distribution of the lithium battery module is reconstructed by using weighted spatio-temporal fusion to obtain the final temperature spatio-temporal distribution;

[0011] S4, a state space model is constructed based on the final temperature spatio-temporal distribution, and the thermal runaway residual time is derived cumulative distribution function;

[0012] S5, the thermal runaway residual time is predicted based on the cumulative distribution function to realize early warning.

[0013] Preferably, the specific content of obtaining the temperature field dataset in S1 by using finite element simulation experiment and physical experiment includes:

[0014] S101, for the battery module electro-chemical-thermal coupling model, a finite element simulation software is used to divide grid elements, set boundary conditions, and calculate to obtain the temperature of the battery grid element, and further obtain the temperature field distribution inside the battery;

[0015] S102, a physical experiment of lithium battery fast charging is set, the charging voltage and current are collected, and the internal temperature of the battery is detected in real time by using an electrochemical impedance instrument frequency response analyzer;

[0016] S103, the dataset of the internal temperature and charging voltage , current , surface temperature of the module obtained by simulation experiment and physical experiment is subjected to data transformation to obtain feature quantities related to the internal temperature , ;

[0017] Among them, indicates the feature quantity attribute related to the internal temperature of each vertex at different positions of the module , , , respectively indicates the change rate of the surface temperature, charging current and charging voltage with time .

[0018] Preferably, the specific content of dividing grid elements and setting boundary conditions by using the finite element simulation software is:

[0019] Each battery grid element is taken as a node to construct a spatio-temporal graph​ wherein is a set of nodes, is a set of edges of the spatio-temporal graph, is a binary unweighted adjacency matrix, expressed as:

[0020] ;

[0021] wherein, is a node has an edge to node .

[0022] Preferably, the network model based on multi-scale feature fusion adopts a graph convolutional neural network temperature field data set to obtain temperature field spatial features;

[0023] The temperature field data set is captured in time series using a Transformer framework, and the internal temperature is inferred and estimated for online testing.

[0024] Preferably, the specific content of reconstructing the initial lithium battery module internal temperature spatial distribution to obtain the final temperature spatio-temporal distribution using weighted spatio-temporal fusion is:

[0025] The temperature changes over time, and the time weight is calculated according to the time proximity of the data points in the initial lithium battery module internal temperature spatial distribution;

[0026] The internal temperature of the battery is spatially distributed, and the temperature distribution between each grid cell is mutually influenced, and the spatial weight is calculated according to the spatial position distance in the initial lithium battery module internal temperature spatial distribution using Euclidean distance;

[0027] Based on the measured data and the internal temperature of the lithium battery in the initial lithium battery module internal temperature spatial distribution, the interpolation weight of physical consistency is calculated;

[0028] The time, space, and physical consistency weights are combined and normalized, and the final temperature spatio-temporal distribution is obtained based on the weighted fusion model.

[0029] Preferably, the expression of the time weight is:

[0030] ;

[0031] wherein, is the time, is the sensor data time, is the time decay coefficient, is the time weight, is the exponential;

[0032] The expression of the space weight is:

[0033] ;

[0034] wherein, is the spatial distance between the target position and the current position, is the spatial decay coefficient, is the spatial weight, represents different positions of the module, is the index.

[0035] The expression of the physical consistency weight is:

[0036] ;

[0037] wherein, is the temperature difference predicted by the network model based on multi-scale feature fusion, is the measured temperature difference;

[0038] Combine the time, space, and physical consistency weights and normalize them:

[0039] ;

[0040] ;

[0041] wherein, is the fusion of time, space, and physical weights, is the normalized weight.

[0042] Preferably, the expression of the final temperature spatiotemporal distribution based on the weighted fusion model is:

[0043] ;

[0044] wherein, is the initial grid cell temperature value from the network model prediction, is the normalized weight, is the temperature value of the final temperature spatiotemporal distribution, represents the numbers of different grid cells around.

[0045] Preferably, a state space model is constructed based on the final temperature spatiotemporal distribution, and the thermal runaway remaining time The specific content of the cumulative distribution function is:

[0046] Based on the final temperature spatiotemporal distribution and combined with the characteristics of the thermal runaway temperature rise stage, the monitored temperature is compared with the set threshold value to determine the current state of the battery and construct a state space model.

[0047] The lithium battery thermal runaway remaining time is defined as the temperature The first time to reach the thermal runaway threshold is the first arrival time, and the expression of the remaining time of lithium battery thermal runaway is:

[0048] ;

[0049] Wherein, is the thermal runaway temperature threshold, and inf represents the lower limit function, is the change time, and the first arrival time The corresponding cumulative probability distribution function is:

[0050] ;

[0051] Wherein, is The internal temperature at the moment, is the initial internal temperature, , , The surface temperature, charging current and charging voltage at the moment , Indicates The surface temperature, charging current and charging voltage at the moment , , State function, is the drift coefficient, which represents the degradation rate function of the internal temperature of lithium battery module affected by the implicit variables of surface temperature, current and voltage, is the diffusion coefficient, is the Brownian motion, is the probability function, is the upper limit;

[0052] Solve the first arrival time distribution of According to the current internal temperature, surface temperature, current and voltage, the probability density function of the change of surface temperature, current and voltage from the current moment to the thermal runaway threshold is obtained;

[0053] The expected value of the cumulative degradation rate function of lithium battery is obtained, and the cumulative distribution function of the remaining time of thermal runaway at the moment is derived.

[0054] Preferably, the expression of the state space model is:

[0055] ;

[0056] Wherein, , , The surface temperature, charging current and charging voltage at the moment , Indicates​ Time , , State function, is a drift coefficient, which represents the degradation rate function of the internal temperature of the lithium battery module affected by the surface temperature, current and voltage implicit variables, is a diffusion coefficient, is Brownian motion, , the new parameter , , , , which is uniformly represented as a parameter set , representing a normal distribution.

[0057] The parameters can be obtained by the state space model The parameters are input into the residual time cumulative distribution function, and the thermal runaway residual time can be obtained.

[0058] In summary, compared with the prior art, the lithium battery internal temperature prediction and reconstruction method based on multi-scale feature fusion of the present application overcomes the limitations of complex physical model and chemical model process, so as to facilitate online implementation of monitoring and provide early warning decision support, thereby improving safety and efficiency, and the specific advantages are as follows:

[0059] 1. The deep learning model is used for multi-scale spatial feature extraction and temperature prediction of data, which can automatically capture the complex time characteristics in battery temperature change without relying on traditional physical model and chemical model, has stronger self-adaptive ability and precision, and can work effectively under various complex working conditions.

[0060] 2. Through the weighted space-time fusion technology, the problem of incomplete temperature measurement caused by fixed sensor position, uneven measurement interval and the like can be solved, the demand for high-density sensor arrangement is reduced, the whole lithium battery module temperature space-time distribution can be accurately predicted, the hardware cost is reduced, and complex physical model calculation is avoided; the network model improves the intelligent level of the battery management system, can realize online monitoring, improves the real-time response ability of the system, and effectively warns the thermal runaway.

[0061] The technical method of the present application will be further described in detail below by means of the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 is a flow chart of the lithium battery internal temperature prediction method based on multi-scale feature fusion of the present application;

[0063] Figure 2A network model module diagram based on multi-scale feature fusion of the application. DETAILED DESCRIPTION

[0064] The technical method of the application is further described below by means of the drawings and examples. It should be noted that the relative arrangement, numerical expressions and values of the components and steps set forth in these examples do not limit the scope of the application unless otherwise specifically stated.

[0065] The following description of at least one exemplary embodiment is merely exemplary in nature and is in no way intended to limit the application or its application or uses.

[0066] Techniques, systems, and devices known to those of ordinary skill in the relevant art can not be discussed in detail herein. However, where appropriate, techniques, systems, and devices should be considered part of the description.

[0067] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as a limitation. Thus, other examples of the exemplary embodiments can have different values.

[0068] Unless otherwise defined, technical terms or scientific terms used in the present application shall have the ordinary meaning commonly understood by one of ordinary skill in the art to which the application pertains.

[0069] The internal temperature of a lithium battery is not only related to its physical properties, but also closely related to environmental temperature and other factors, and the influence of external factors on the internal temperature of the lithium battery needs to be fully considered. When monitoring the surface temperature field of the lithium battery, the number of sensors is limited, the data acquisition temperature space is limited and unevenly distributed, which poses great challenges to feature extraction capability and temperature space reconstruction. The present application fully considers the influence of environmental temperature on the internal temperature of the battery, combines multi-scale spatial feature fusion and time, physical consistency, and spatial weighted temperature space reconstruction, and effectively utilizes the data obtained by the limited number of surface sensors to estimate the internal temperature of the battery and warn of thermal runaway.

[0070] The present application provides a lithium battery internal temperature prediction and reconstruction method based on multi-scale feature fusion, as shown in Figure 1 The present application provides a lithium battery internal temperature prediction and reconstruction method based on multi-scale feature fusion, as shown in

[0071] Further, the specific content of S1, i.e., obtaining the temperature field data set by finite element simulation experiment and physical experiment, includes: S101, for the battery module electro-chemical-thermal coupling model, using finite element simulation software to divide grid elements, set boundary conditions, calculate, get the temperature of the battery grid element, and further get the temperature field distribution inside the battery.

[0072] S102, set up the physical experiment of lithium battery fast charging, collect the charging voltage, current, and detect the internal temperature of the battery in real time through the electrochemical impedance instrument frequency response analyzer.

[0073] S103, data transformation is performed on the data sets of the internal temperature and charging voltage , current , surface temperature of the module obtained from the simulation experiment and the physical experiment , , .

[0074] Among them, represents different positions of the module, and each vertex has an internal temperature related feature quantity attribute , , , represents the rate of change of the surface temperature, charging current and charging voltage with time .

[0075] Further, the finite element simulation software is used to divide the grid elements, and the specific content of the boundary condition is that each battery grid element is taken as a node, and a space-time graph is constructed , wherein is a node set, is an edge set of the space-time graph, is a binary unweighted adjacency matrix, and the expression is: ; wherein, is an edge from node to node .

[0076] S2, construct a network model based on multi-scale feature fusion, input the temperature field data set into the network model based on multi-scale feature fusion, perform multi-view feature fusion on the lithium battery internal temperature, and estimate the initial lithium battery module internal temperature spatial distribution.

[0077] Further, as shown in Figure 2 , the network model based on multi-scale feature fusion adopts a graph convolutional neural network temperature field data set to obtain temperature field spatial features; a Transformer framework is used to capture time series of the temperature field data set, estimate the internal temperature, and perform online testing.

[0078] The graph convolutional neural network captures neighbor information in different ranges by constructing different scale graph convolution operations, and can realize multi-scale learning through different convolution layers. Given a graph and the feature vector of each node , the graph convolution operation updates the feature of a node by .

[0079] wherein, is the feature of node at the layer, is the set of neighbor nodes of node , is the weight matrix of the layer, is the activation function, is the number of neighbors of node , i.e., the number of edges connected to node , is one neighbor node of node .

[0080] The local feature extraction updates the feature of each node by direct neighbor node information, for each node , its feature is updated by fusing the features of direct neighbors: ;

[0081] The global feature extraction can further obtain global information by stacking multiple graph convolution layers, and can capture the relationship between distant nodes in the graph, thereby capturing features at the global scale.

[0082] In the multi-scale neural network, the features extracted from different scales are spliced to obtain a higher-dimensional feature representation, expressing more complex nonlinear relationships. Assuming that the features obtained from different scales are , the final feature can be obtained by splicing operation: .

[0083] After multi-scale extraction of spatial features, the Transformer framework is used to capture time series, estimate internal temperature and perform online testing.

[0084] The online test based on the above network model obtains the temperature distribution of the lithium battery grid unit, and constructs the internal temperature spatial distribution of the lithium battery module. The number of lithium battery temperature monitoring sensors is limited, and the estimated temperature distribution is gridized, which may ignore the temperature change between grids and cause safety hazards in the thermal runaway early warning process. Fully considering the influence of time, space and model prediction consistency on the final temperature spatiotemporal distribution, weighted spatiotemporal fusion is used to comprehensively determine the final temperature spatiotemporal distribution.

[0085] S3, the initial lithium battery module internal temperature spatial distribution is reconstructed to obtain the final temperature spatiotemporal distribution by using weighted spatiotemporal fusion.

[0086] Further, the specific content of reconstructing the initial lithium battery module internal temperature spatial distribution to obtain the final temperature spatio-temporal distribution by using weighted spatio-temporal fusion is that the temperature changes with time, and the time weight is calculated according to the time proximity of the data points in the initial lithium battery module internal temperature spatial distribution.

[0087] The internal temperature of the battery is spatially distributed, and the temperature distribution between each grid cell is mutually influenced, and the spatial weight is calculated using Euclidean distance according to the spatial position distance in the initial lithium battery module internal temperature spatial distribution.

[0088] Based on the measured data and the internal temperature of the lithium battery in the initial lithium battery module internal temperature spatial distribution, the interpolation weight of physical consistency is calculated.

[0089] The time, space and physical consistency weights are combined and normalized, and the final temperature spatio-temporal distribution is obtained based on the weighted fusion model.

[0090] Further, the expression of the time weight is: .

[0091] Wherein, is the time, is the sensor data time, is the time decay coefficient, is the time weight, is the exponential.

[0092] The expression of the space weight is: .

[0093] Wherein, is the spatial distance between the target position and the current position, is the spatial decay coefficient, is the space weight, indicates different positions of the module, is the exponential.

[0094] The expression of the physical consistency weight is: .

[0095] Wherein, is the temperature difference predicted by the network model based on multi-scale feature fusion, is the measured temperature difference.

[0096] The time, space and physical consistency weights are combined and normalized:

[0097] .

[0098] .

[0099] wherein, is the time, space, physical weight fusion, is the normalized weight.

[0100] Further, through the weighted fusion model, the temperature inside the battery can be inferred and reconstructed: .

[0101] wherein, is the initial grid cell temperature value from the network model prediction, is the normalized weight, is the temperature value of the final temperature spatiotemporal distribution, represents the number of different grid cells around.

[0102] Thermal runaway is an irreversible nonlinear temperature rise process. Temperature is the most direct parameter to characterize this process. Therefore, based on the internal temperature obtained in the above research and combined with the characteristics of the thermal runaway temperature rise stage, the monitored temperature is compared with the set threshold value to judge the current state of the battery.

[0103] S4, based on the final temperature spatiotemporal distribution, a state space model is constructed, and the thermal runaway remaining time is derived cumulative distribution function.

[0104] S5, based on the thermal runaway remaining time cumulative distribution function to predict the thermal runaway remaining time, realize early warning.

[0105] Further, based on the final temperature spatiotemporal distribution, a state space model is constructed, and the thermal runaway remaining time is derived cumulative distribution function. The specific content is: based on the final temperature spatiotemporal distribution and combined with the characteristics of the thermal runaway temperature rise stage, the monitored temperature is compared with the set threshold value to judge the current state of the battery and construct a state space model.

[0106] The lithium battery thermal runaway remaining time is defined as the temperature which reaches the thermal runaway threshold for the first time, i.e. the first arrival time, and the expression of the lithium battery thermal runaway remaining time is: .

[0107] wherein, is the thermal runaway temperature threshold, inf represents the lower limit function, is the change time.

[0108] The first arrival time The corresponding cumulative probability distribution function is:

[0109] .

[0110] in, for The internal temperature at any given moment The initial internal temperature, , , They are respectively Surface temperature, charging current, and charging voltage at all times. express time , , State function, The drift coefficient characterizes the degradation rate function of the internal temperature of a lithium battery module as a function of implicit variables such as surface temperature, current, and voltage. Where is the diffusion coefficient. For Brownian motion, Let be a probability function. This is the upper bound.

[0111] Solve Given the time-varying characteristics of battery module surface temperature, current, and voltage during fast charging, we aim to determine the remaining time of thermal runaway. The cumulative distribution function is a probability density function that calculates the changes in surface temperature, current, and voltage at the current moment up to the thermal runaway threshold, based on the current internal temperature, surface temperature, current, and voltage.

[0112] To obtain the expected value of the cumulative degradation rate function within the lithium battery, the following is derived: Remaining time for thermal runaway The cumulative distribution function.

[0113] Furthermore, in order to estimate the parameters in the residual time cumulative distribution function model, a state-space model is first constructed to characterize the temperature evolution process. The expression of the state-space model is as follows:

[0114] .

[0115] in, for The internal temperature at any given moment The initial internal temperature, , , They are respectively Surface temperature, charging current, and charging voltage at all times. express time , , State function, is a drift coefficient, representing a degradation rate function of the lithium battery module internal temperature affected by the surface temperature, current, voltage implicit variables, is a diffusion coefficient, is Brownian motion, , a new parameter , , , , is uniformly expressed as a parameter set , is a normal distribution.

[0116] The particle filter, improved Kalman filter and other methods are used to estimate and update the model parameters, and the residual time accumulation model is used to predict the thermal runaway residual time, so as to realize early warning.

[0117] The parameters can be obtained through a state space model The parameters are input into the residual time cumulative distribution function, and the thermal runaway residual time can be obtained.

[0118] The lithium battery internal temperature prediction and weighted space-time reconstruction model based on multi-scale feature fusion combined with the machine learning method proposed by the application make up for the deficiency of the physical model, fully consider the environmental temperature and other factors, estimate the internal temperature field by using the surface temperature field, and better model performance results are obtained, which is convenient for online monitoring and effective thermal runaway early warning.

[0119] Finally, it should be noted that: the above examples are only used to illustrate the technical method of the application and not to limit it, although the application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: it can still modify or replace the technical method of the application, and these modifications or equivalent replacements cannot make the modified technical method deviate from the spirit and scope of the technical method of the application.​

Claims

1. A method for predicting and reconstructing the internal temperature of a lithium battery based on multi-scale feature fusion, characterized in that, Includes the following steps: S1. Construct an electro-chemical-thermal coupling model of a lithium battery module and obtain a temperature field dataset using finite element simulation experiments and physical experiments. S2. Construct a network model based on multi-scale feature fusion, input the temperature field dataset into the network model based on multi-scale feature fusion to perform multi-view feature fusion to estimate the internal temperature of the lithium battery and obtain the initial spatial distribution of the internal temperature of the lithium battery module. S3. The final temperature spatiotemporal distribution is obtained by reconstructing the internal temperature spatial distribution of the initial lithium battery module using weighted spatiotemporal fusion. S4. Construct a state-space model based on the final temperature spatiotemporal distribution and derive the remaining thermal runaway time. Cumulative distribution function; S5, Based on thermal runaway remaining time The cumulative distribution function predicts the remaining time of thermal runaway, enabling early warning. The network model based on multi-scale feature fusion uses a graph convolutional neural network to process the temperature field dataset to obtain the spatial features of the temperature field. The Transformer framework was used to capture time series data of the temperature field dataset, infer and estimate the internal temperature, and conduct online testing. The final temperature spatiotemporal distribution is obtained by reconstructing the initial internal temperature spatial distribution of the lithium battery module using weighted spatiotemporal fusion. Time weights are calculated based on the temporal proximity of data points in the initial internal temperature spatial distribution of the lithium battery module. Spatial weights are calculated using Euclidean distance based on the spatial distances between spatial locations in the initial internal temperature spatial distribution of the lithium battery module. Based on measured data and the internal temperature of the lithium battery in the initial internal temperature spatial distribution of the lithium battery module, the weight of physical consistency is calculated by interpolation between the two. By combining and normalizing the weights of temporal, spatial, and physical consistency, the final spatiotemporal temperature distribution is obtained based on a weighted fusion model.

2. The method for predicting and reconstructing the internal temperature of a lithium battery based on multi-scale feature fusion according to claim 1, characterized in that, The specific content of obtaining the temperature field dataset using finite element simulation experiments and physical experiments in S1 includes: S101. For the electro-chemical-thermal coupling model of the battery module, the finite element simulation software is used to divide the grid cells, set the boundary conditions, and perform calculations to obtain the temperature of the battery grid cells, thereby obtaining the temperature field distribution inside the battery. S102. Set up a physical experiment for fast charging of lithium batteries, collect charging voltage and current, and detect the internal temperature of the battery in real time through an electrochemical impedance spectroscopy and a frequency response analyzer. S103. The internal temperature and charging voltage of the module obtained from simulation and physical experiments. Current Surface temperature The dataset is transformed to obtain the internal temperature. Related features , ; in This represents different positions of the module, each vertex. They all have characteristic properties related to internal temperature. , , , These represent surface temperature, charging current, and charging voltage as a function of time, respectively. The rate of change.

3. The method for predicting and reconstructing the internal temperature of a lithium battery based on multi-scale feature fusion according to claim 2, characterized in that, The specific steps for mesh generation and boundary condition setting using finite element simulation software are as follows: Each battery grid cell is used as a node to construct a spatiotemporal graph. ,in For a set of nodes, For the edge set of the spacetime graph, It is a binary unweighted adjacency matrix, expressed as: ； in, For nodes To the node One of the edges.

4. The method for predicting and reconstructing the internal temperature of a lithium battery based on multi-scale feature fusion according to claim 3, characterized in that: The expression for time weighting is: ； in, It is time. It is the moment of sensor data. It is the time decay coefficient. It is time weighting. It is an index; The expression for spatial weights is: ； in, It is the spatial distance between the target location and the current location. It is the spatial attenuation coefficient. It is spatial weight. Indicates different positions of the module. It is an index; The expression for the physical consistency weight is: ； in, The temperature difference is predicted using a network model based on multi-scale feature fusion. It is the measured temperature difference. It is the physical consistency weight; Combine and normalize the temporal, spatial, and physical consistency weights: ； ； in, It is a fusion of time, space, and physical weights. These are the normalized weights.

5. The method for predicting and reconstructing the internal temperature of a lithium battery based on multi-scale feature fusion according to claim 4, characterized in that, The expression for the final spatiotemporal distribution of temperature, obtained based on the weighted fusion model, is as follows: ； in, These are the initial grid cell temperature values ​​predicted by the network model. These are the normalized weights. It is the temperature value of the final spatiotemporal temperature distribution. This represents the number of the surrounding different grid cells.

6. The method for predicting and reconstructing the internal temperature of a lithium battery based on multi-scale feature fusion according to claim 5, characterized in that, A state-space model is constructed based on the final temperature spatiotemporal distribution, and the remaining time of thermal runaway is derived. The specific contents of the cumulative distribution function are as follows: Based on the final temperature spatiotemporal distribution and combined with the characteristics of the thermal runaway temperature rise stage, the monitored temperature is compared with the set threshold to determine the current state of the battery and construct a state space model. The remaining time of thermal runaway in a lithium battery is defined as the temperature. The time when the thermal runaway threshold is first reached is called the initial time. The expression for the remaining time of thermal runaway in a lithium battery is: ； in, Let inf be the thermal runaway temperature threshold, and let inf be the infim function. For varying times, first arrival time Corresponding cumulative probability distribution function for: ； in, for The internal temperature at any given moment The initial internal temperature, , , They are respectively Surface temperature, charging current, and charging voltage at all times. express time , , State function, The drift coefficient characterizes the degradation rate function of the internal temperature of a lithium battery module as a function of implicit variables such as surface temperature, current, and voltage. The diffusion coefficient is... For Brownian motion, Let be a probability function. The supremum; Solve The first arrival time distribution is obtained by calculating the probability density function of the changes in surface temperature, current and voltage from the current time to the thermal runaway threshold based on the current internal temperature, surface temperature, current and voltage. To obtain the expected value of the cumulative degradation rate function within the lithium battery, the following is derived: Remaining time for thermal runaway The cumulative distribution function.

7. The method for predicting and reconstructing the internal temperature of a lithium battery based on multi-scale feature fusion according to claim 6, characterized in that, The expression for the state-space model is: ； in, , , They are respectively Surface temperature, charging current, and charging voltage at all times. express time , , State function, The diffusion coefficient is... For Brownian motion, Add parameters , , , , uniformly represented as parameter set , This represents a normal distribution.

8. The method for predicting and reconstructing the internal temperature of a lithium battery based on multi-scale feature fusion according to claim 7, characterized in that, The parameters can be obtained through the state-space model. By inputting the parameters into the cumulative distribution function of the remaining time, the remaining time of thermal runaway can be obtained.

Images