CN-121848931-B - Lithium battery thermal runaway state identification method based on physical information neural network

CN121848931BCN 121848931 BCN121848931 BCN 121848931BCN-121848931-B

Abstract

The invention relates to the technical field of electric variable measurement and test, in particular to a lithium battery thermal runaway state identification method based on a physical information neural network. The method comprises the steps of obtaining broadband fluctuation current, voltage response tracks and surface layer temperature signals, generating multidimensional electric measurement data streams, injecting the multidimensional electric measurement data streams into a battery physical information network, stripping dynamic polarization internal resistance sequences and deep heat source tensors, constructing an electric heating cross-domain residual error model to quantify charge heating residual errors, extracting residual error gradient flow deviation correction electric characteristic references, carrying out online evaluation on the dynamic polarization internal resistance sequences, and outputting a thermal runaway critical blocking signal when step electric parameter shock waves are identified. According to the invention, the physical information neural network and the electrothermal cross-domain residual error model are utilized, the electrothermal coupling physical law is fused to conduct online calculation, and meanwhile, the reference drift caused by battery aging is continuously corrected by extracting residual error gradient flow, so that the high-precision low-delay online measurement and reliable early warning of the lithium battery thermal runaway critical state are realized.

Inventors

Ke Xiwen
WANG YONG
DU XIANZHEN
Shao changwang
ZHANG JINGPENG

Assignees

山东精工电子科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260318

Claims (8)

1. The lithium battery thermal runaway state identification method based on the physical information neural network is characterized by comprising the following steps of: The method comprises the steps of obtaining broadband fluctuation current spontaneously superimposed on a battery charging loop of a new energy automobile, synchronously collecting a voltage response track and a surface layer temperature signal of a battery end, carrying out time domain interleaving reconstruction on the broadband fluctuation current, the voltage response track and the surface layer temperature signal, and generating a multidimensional electric measurement data stream; The multi-dimensional electric measurement data stream is injected into a battery physical information network, and a dynamic polarization internal resistance sequence and a deep heat source tensor which characterize the polarization characteristics of the current battery are stripped; Forcing the charge heating residual error to collapse towards zero value as a convergence boundary, and extracting residual gradient flow; performing on-line evaluation on the dynamic polarization internal resistance sequence constrained by the deep heat source tensor according to the electric characteristic standard of the residual gradient flow continuous deviation rectification battery physical information network; When the step electric parameter shock wave is identified in the electric characteristic standard, the battery charging loop is cut off, a thermal runaway critical blocking signal is output, and the thermal runaway risk of the battery is converged within a safety threshold.
2. The method for identifying the thermal runaway state of the lithium battery based on the physical information neural network according to claim 1, wherein the specific implementation process of obtaining the broadband fluctuation current spontaneously superimposed on the charging loop of the battery of the new energy automobile and synchronously collecting the voltage response track of the battery terminal and the surface layer temperature signal comprises the following steps: The method comprises the steps of continuously monitoring a battery charging loop by using a bandwidth current transformer, capturing broadband fluctuation current which is spontaneously superimposed along with a charge transfer process, establishing a voltage sensing channel which shares a global clock with the broadband fluctuation current, continuously sampling potential difference between battery terminals and generating a voltage response track, capturing surface layer temperature signals which are diffused to the outside by heat conduction in the battery according to a temperature measuring node attached to a battery shell, synchronously introducing the broadband fluctuation current, the voltage response track and the surface layer temperature signals into a buffer queue, and carrying out baseline calibration and anti-aliasing filtering.
3. The method for identifying the thermal runaway state of the lithium battery based on the physical information neural network according to claim 2, wherein the specific implementation process for carrying out time domain interleaving reconstruction on the broadband fluctuation current, the voltage response track and the surface layer temperature signal to generate the multidimensional electrical measurement data stream comprises the following steps: The method comprises the steps of obtaining a time stamp mark of broadband fluctuation current, voltage response tracks and surface temperature signals after baseline calibration, uniformly interpolating the broadband fluctuation current, the voltage response tracks and the surface temperature signals to the same time resolution according to a global clock, splicing the broadband fluctuation current, the voltage response tracks and the surface temperature signals after time alignment into a multi-channel time sequence according to a characteristic fusion mechanism, performing time domain interleaving reconstruction operation on the multi-channel time sequence, establishing a joint state matrix representing time lag relevance among variables, and carrying out dimension normalization and tensor encapsulation on each dimension characteristic of the joint state matrix to generate a multi-dimensional electrical measurement data stream.
4. The method for identifying the thermal runaway state of the lithium battery based on the physical information neural network according to claim 1, wherein the specific implementation process of injecting the multidimensional electrical measurement data stream into the battery physical information network and stripping the dynamic polarization internal resistance sequence and the deep heat source tensor representing the polarization characteristics of the current battery comprises the following steps: The method comprises the steps of establishing a battery physical information network integrating electrothermal coupling physical priori knowledge and a depth feature extraction architecture, forward injecting the multidimensional electrical measurement data stream to a hidden layer of the battery physical information network through an input layer, driving the battery physical information network to conduct online calculation by utilizing an equivalent circuit physical rule embedded inside, decoupling a dynamic polarization internal resistance sequence containing ohmic impedance and polarization capacitance from the multidimensional electrical measurement data stream, synchronously triggering an implicit solving operator of a thermal chemistry equation set, and mapping residual time sequence features into deep heat source tensors reflecting internal heat generation rate distribution.
5. The method for identifying the thermal runaway state of the lithium battery based on the physical information neural network according to claim 1, wherein the specific implementation process for constructing an electrothermal cross-domain residual error model and carrying out boundary test according to the dynamic polarized internal resistance sequence and the deep heat source tensor to quantify the charge heating residual error reflecting the unbalanced state of thermoelectric coupling comprises the following steps: The method comprises the steps of taking the dynamic polarization internal resistance sequence and the deep heat source tensor as independent variables and dependent variables of mutual verification, constructing an electrothermal cross-domain residual error model representing mismatch degree between joule heat dissipation and actual temperature rise, setting thermoelectric conversion boundary conditions representing safe operation conditions, inputting the dynamic polarization internal resistance sequence and the deep heat source tensor into the thermoelectric conversion boundary conditions for boundary test, calculating a difference value between excessive ohmic heat generation caused by abnormal polarization and theoretical heat transfer distribution in the deep heat source tensor, converting the difference value into an error index through high-dimensional space projection, and quantifying charge heating residual errors reflecting internal critical states.
6. The method for identifying a thermal runaway state of a lithium battery based on a physical information neural network according to claim 1, wherein the specific implementation process of forcing the charge heating residual error to collapse towards a zero value as a convergence boundary and extracting a residual error gradient stream comprises the following steps: Introducing the charge heating residual into a loss function of a battery physical information network, constructing a target optimization equation taking the charge heating residual as a penalty term, starting an automatic differential engine to perform back propagation calculation on the target optimization equation, forcing the charge heating residual to collapse towards zero value as a convergence boundary, continuously collecting a direction vector representing a state evolution trend on a loss plane in an iterative optimization process of network weight updating, and extracting residual gradient flow containing a nonlinear mapping relation along a track direction with the fastest decrease of the penalty term.
7. The method for identifying the thermal runaway state of the lithium battery based on the physical information neural network according to claim 1, wherein the specific implementation process for carrying out on-line evaluation on the dynamic polarization internal resistance sequence constrained by the deep heat source tensor according to the electrical characteristic standard of the residual gradient flow continuous deviation rectification battery physical information network comprises the following steps: The method comprises the steps of feeding back an extracted residual gradient flow to a parameter updating controller of a battery physical information network, continuously rectifying electric characteristic reference drift caused by impedance increase caused by battery aging by utilizing an error compensation direction indicated by the residual gradient flow, obtaining a dynamic polarization internal resistance sequence subjected to thermodynamic cross validation constraint by a deep heat source tensor, carrying out on-line evaluation on the deviation degree of the dynamic polarization internal resistance sequence from a normal polarization track by combining the dynamic updated electric characteristic reference, and outputting an evaluation index sequence reflecting the characteristics of dynamic evolution of internal interface impedance and thermal runaway evolution.
8. The method for identifying a thermal runaway state of a lithium battery based on a physical information neural network according to claim 7, wherein when a step electric parameter shock wave is identified in the electric characteristic reference, the specific implementation process of cutting off the battery charging loop and outputting a thermal runaway critical blocking signal to converge the thermal runaway risk of the battery within a safety threshold comprises: The method comprises the steps of continuously running mutation detection logic in a monitoring window of an electrical characteristic standard, triggering a hardware protection interrupt instruction when nonlinear mutation occurs in an evaluation index sequence input to a judging node and step-electric parameter shock waves are recognized in the electrical characteristic standard, cutting off a battery charging loop through a physical switch element according to the hardware protection interrupt instruction and blocking external electric energy from being continuously injected, synchronously outputting a thermal runaway critical blocking signal through a communication interface, intervening in the charging and discharging process of a battery based on the thermal runaway critical blocking signal, and converging the thermal runaway risk of the battery within a safety threshold.

Description

Lithium battery thermal runaway state identification method based on physical information neural network Technical Field The invention relates to the technical field of electric variable measurement and test, in particular to a lithium battery thermal runaway state identification method based on a physical information neural network. Background With the wide application of lithium ion batteries in new energy automobiles, the thermal safety problem of the batteries is increasingly prominent. In a complex actual operation condition, the lithium battery may trigger an internal complex chained exothermic side reaction due to factors such as overcharge, overdischarge, internal short circuit or external high temperature, and further evolve into thermal runaway. In order to prevent thermal runaway, external electrical variables (such as terminal voltage, charge and discharge current) and surface temperature of the battery need to be measured through conventional sensors, and core thermal states inside the battery, which cannot be directly measured, are estimated and identified in real time and early warned. At present, the prior art is divided into two types, namely a method based on a pure mechanism model, wherein the method constructs a complex physical model comprising a plurality of partial differential equations according to the law of conservation of energy, heat transfer theory and electrochemical dynamics, and the thermal state inside the battery is deduced and calculated through measurement input of external electric parameters and thermal parameters. The second type is a method based on pure data driving, which directly utilizes the sensor measurement to obtain massive external electric variables and temperature characteristics to train a deep learning network, and attempts to fit a nonlinear mapping relation between the external measured variables and the internal thermal runaway state through massive data. However, the prior art has inherent drawbacks in practical applications. In purely mechanical model-based approaches, partial differential equations describing the evolution of thermal runaway of the cell and internal electrochemical reactions often have very strong non-linear and rigid characteristics. Under the extremely limited embedded calculation force of the vehicle-mounted battery system, the difficulty of real-time online numerical solution of the equations is extremely high, and the real-time performance of internal state measurement is extremely poor. Meanwhile, the pure mechanism model highly depends on accurate initial physical parameters, and the parameters can drift remarkably along with the electric characteristic attenuation caused by battery aging, so that the measurement and identification accuracy of the model in the later period of the battery life cycle is greatly reduced. On the other hand, pure data-driven based methods lack the constraints of thermodynamic and electrochemical basic laws at all, the nature of which is only probabilistic fitting in a mathematical sense. When the model faces extreme electric working conditions or complex and changeable environments which are not covered in the training data set, the model is extremely easy to output results against physical common knowledge such as energy conservation and the like, so that missed report or frequent false report is caused. Because thermal runaway belongs to a very low probability dangerous event, the cost for acquiring high-quality destructive thermal runaway label data covering the whole life cycle of a battery is extremely high and dangerous, and a conventional neural network cannot establish reliable state evaluation and test benchmarks when a limit working condition data support of a sufficient scale is lacking. In summary, in the prior art, under the condition of lacking massive destructive test data, an identification network with signal processing capability and ensuring that an estimation result strictly follows the law of battery electric-thermal coupling physics cannot be constructed, so that high-precision and low-delay online measurement and reliable identification of a lithium battery thermal runaway critical state are difficult to realize. Therefore, a lithium battery thermal runaway state identification method based on a physical information neural network is provided. Disclosure of Invention The invention aims to provide a lithium battery thermal runaway state identification method based on a physical information neural network, which is used for carrying out thermal runaway state identification on a lithium battery of a new energy automobile. In order to achieve the above purpose, the present invention provides the following technical solutions: a lithium battery thermal runaway state identification method based on a physical information neural network comprises the following steps: The method comprises the steps of obtaining broadband fluctuation current spontaneously superimposed on a battery ch