Battery pack thermal runaway risk identification and warning system and method

Abstract

This application discloses a battery pack thermal runaway risk identification and early warning system and method, relating to the field of battery risk early warning. The method includes: real-time acquisition of battery pack operating parameter datasets via a multimodal sensor network to perform mixed anomaly detection on the battery pack; obtaining real-time anomaly detection results to predict the thermal state of the battery pack; constructing a thermal state evolution trend map for thermal runaway risk assessment; determining the thermal runaway risk level; triggering multi-level early warning signals based on the thermal runaway risk level; and generating a thermal runaway suppression strategy. This addresses the technical problems of high false alarm rates, delayed response, and inability to dynamically adapt to battery aging caused by the limitations of single-sensor monitoring in existing battery thermal runaway early warning systems. It achieves the technical effect of improving the detection rate of early signs of thermal runaway and shortening the early warning time through multimodal sensing.

Description

Technical Field

[0001] This application relates to the field of battery risk warning, and in particular to a battery pack thermal runaway risk identification and warning system and method. Background Technology

[0002] With the increasing energy density of power battery packs, thermal runaway has become a core threat to the safety of lithium-ion batteries. Existing technologies mostly rely on single sensors (such as temperature or voltage) for threshold judgments, but battery thermal runaway exhibits multi-physics coupling characteristics (such as gas generation, deformation, and localized temperature rise), making it difficult for traditional methods to detect early signs. Furthermore, existing early warning systems generally suffer from response lag, with an average delay exceeding 500ms from anomaly detection to triggering protection, failing to meet the real-time requirements of high-rate charging and discharging scenarios.

[0003] Existing technologies suffer from technical problems such as high false alarm rate, delayed response, and inability to dynamically adapt to battery aging due to the limitations of single sensor monitoring in battery thermal runaway early warning. Summary of the Invention

[0004] This application provides a battery pack thermal runaway risk identification and early warning system and method to solve the technical problems of high false alarm rate, slow response and inability to dynamically adapt to battery aging caused by the limitations of single sensor monitoring in existing battery thermal runaway early warning systems. It achieves the technical effect of improving the detection rate of early signs of thermal runaway and shortening the early warning time through multimodal sensing.

[0005] This application provides a battery pack thermal runaway risk identification and early warning system. The system is applied to a battery pack thermal runaway risk identification and early warning method, comprising: a data acquisition module for real-time acquisition of a battery pack operating parameter dataset through a multimodal sensor network; a detection module for performing hybrid anomaly detection on the battery pack based on the operating parameter dataset to obtain real-time anomaly detection results; a prediction module for predicting the thermal state of the battery pack based on the real-time anomaly detection results and constructing a thermal state evolution trend map; and an evaluation module for performing thermal runaway risk evaluation based on the thermal state evolution trend map, determining the thermal runaway risk level, triggering multi-level early warning signals according to the thermal runaway risk level, and generating a thermal runaway suppression strategy.

[0006] In a possible implementation, the acquisition module performs the following processing: a first sensing unit is used to deploy an array of infrared temperature sensors distributed on the surface of the battery pack and between individual cells to form a first sensing layer; a second sensing unit is used to embed a miniature voltage sensor at the positive and negative electrode connection of the battery pack to form a second sensing layer; a third sensing unit is used to install a gas composition spectrometer in the exhaust channel of the battery pack to form a third sensing layer; a fourth sensing unit is used to attach a flexible piezoelectric thin film sensor to the inside of the battery pack casing to form a fourth sensing layer; and an alignment unit is used to synchronize and align the first sensing layer, the second sensing layer, the third sensing layer, and the fourth sensing layer according to the running timestamp to generate the running parameter dataset.

[0007] In a possible implementation, the detection module performs the following processes: a feature extraction unit, used to extract spatial features based on the running parameter dataset to generate a high-dimensional feature vector; a classification unit, used to perform multi-classification based on the high-dimensional feature vector, construct a multi-classification decision, and perform local feature enhancement on the running parameter dataset according to the multi-classification decision to obtain multiple enhanced feature data; and a weighted fusion unit, used to retrieve historical detection results, perform weighted fusion of the multiple enhanced feature data with the historical detection results, and generate a real-time anomaly detection result, wherein the real-time anomaly detection result includes an anomaly type label and a confidence score.

[0008] In a possible implementation, the prediction module performs the following processes: a selection unit, used to dynamically select a subset of input features based on the anomaly detection results, and construct a multi-dimensional time-series input matrix for the target time window; a thermal runaway prediction unit, used to perform thermal runaway prediction on the multi-dimensional time-series input matrix to obtain a thermal state prediction dataset, the thermal state prediction dataset including predicted values ​​of temperature change rate, gas production rate, and deformation acceleration; a first state evolution unit, used to perform state evolution based on the battery pack's temperature change rate to obtain spatial temperature distribution data; a second state evolution unit, used to perform state evolution based on the battery pack's gas production rate to obtain gas diffusion path data; a third state evolution unit, used to perform state evolution based on the battery pack's predicted deformation acceleration to obtain mechanical deformation vector data; and an association and integration unit, used to associate and integrate the spatial temperature distribution data, the gas diffusion path data, and the mechanical deformation vector data to draw the thermal state evolution trend map.

[0009] In a possible implementation, the prediction module performs the following processes: a data filtering unit, used to filter from the running parameter dataset according to the anomaly type label to generate a first feature subset; a weighting unit, used to weight the first feature subset based on the confidence score to generate a second feature subset; a data extraction unit, used to extract the moving average, standard deviation, and kurtosis coefficient of the second feature subset within a target time window to construct a multidimensional statistical feature matrix; and a fusion unit, used to perform feature-level fusion of the multidimensional statistical feature matrix with the electrochemical mechanism simulation data to generate the multidimensional time-series input matrix.

[0010] In a possible implementation, the evaluation module performs the following processes: a multidimensional decomposition unit, used to decompose the thermal state evolution trend diagram into multidimensional data to obtain multidimensional data, which includes thermal accumulation rate data, thermal diffusion gradient data, and thermal runaway trigger probability data; a dynamic adjustment unit, used to dynamically adjust the weighting coefficients of the thermal accumulation rate data, the thermal diffusion gradient data, and the thermal runaway trigger probability data according to the battery pack's health state parameters, to generate a comprehensive risk score; and a risk determination unit, used to determine the risk level based on the comprehensive risk score to obtain the thermal runaway risk level.

[0011] In a possible implementation, the evaluation module performs the following processes: a traversal unit, used to construct a risk level quantification matrix based on the comprehensive risk score, and traverse the risk level quantification matrix to extract the thermal diffusivity factor, electrochemical instability index, and cumulative deformation energy value of the battery pack; a threshold setting unit, used to set multi-level risk thresholds, including a first risk threshold, a second risk threshold, and a third risk threshold; a first determination unit, used to determine a level I risk when the thermal diffusivity factor exceeds the first risk threshold and the electrochemical instability index is lower than the second risk threshold; a second determination unit, used to determine a level II risk when the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the cumulative deformation energy value is lower than the third risk threshold; and a third determination unit, used to determine a level III risk when the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the cumulative deformation energy value exceeds the third risk threshold.

[0012] In a possible implementation, the evaluation module performs the following processes: a dimensional analysis unit, used to analyze the matrix dimensions of the risk level quantification matrix to determine the thermodynamic, electrochemical, and mechanical dimensions; a first calculation unit, used to calculate the product of the maximum temperature difference between adjacent cells in the temperature field and the thermal conduction rate as the thermal diffusion factor based on the thermodynamic dimension and the thermal diffusion gradient data; a second calculation unit, used to calculate the electrochemical instability index based on the electrochemical dimension, using the thermal runaway trigger probability data combined with the gas concentration change curve and voltage drop rate; and a third calculation unit, used to analyze the stress concentration area of ​​the battery pack's casing through the deformation vector field combined with the thermal accumulation rate data based on the mechanical dimension, and calculate the cumulative deformation energy value.

[0013] In a possible implementation, the evaluation module performs the following processing: a fourth calculation unit, used to calculate the three-dimensional risk diffusion of the battery pack based on real-time temperature distribution data and the thermal conduction equation, to determine the thermal runaway propagation path; a first triggering unit, used to trigger a first warning signal when the thermal runaway risk level is Level I, locate the abnormal position coordinates according to the thermal runaway propagation path, and generate an early warning message; a second triggering unit, used to trigger a second warning signal when the thermal runaway risk level is Level II, dynamically adjust the liquid cooling system according to the thermal runaway propagation path, generate a liquid cooling control command, and distribute the flow of the cooling pipes of the liquid cooling system; a third triggering unit, used to trigger a third warning signal when the thermal runaway risk level is Level III, activate the directional fire extinguishing device based on the thermal runaway propagation path, and generate an emergency pressure relief channel.

[0014] This application also provides a method for identifying and warning of thermal runaway risk in battery packs, comprising: acquiring a dataset of operating parameters of the battery pack in real time through a multimodal sensor network; performing hybrid anomaly detection on the battery pack based on the operating parameter dataset to obtain real-time anomaly detection results; predicting the thermal state of the battery pack based on the real-time anomaly detection results to construct a thermal state evolution trend map; assessing the thermal runaway risk based on the thermal state evolution trend map to determine the thermal runaway risk level; triggering multi-level warning signals based on the thermal runaway risk level; and generating a thermal runaway suppression strategy.

[0015] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0016] The battery pack thermal runaway risk identification and early warning system and method provided in this application relate to the field of battery risk early warning technology. It solves the technical problems of high false alarm rate, slow response and inability to dynamically adapt to battery aging caused by the limitations of single sensor monitoring in existing battery thermal runaway early warning systems. It achieves the technical effect of improving the detection rate of early signs of thermal runaway and shortening the early warning time through multimodal sensing. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings of the embodiments of this application will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0018] Figure 1 This is a schematic diagram of the structure of the battery pack thermal runaway risk identification and early warning system provided in the embodiments of this application;

[0019] Figure 2 This is a flowchart illustrating the battery pack thermal runaway risk identification and early warning method provided in an embodiment of this application. Detailed Implementation

[0020] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] According to the battery pack thermal runaway risk identification and early warning system of the embodiments of this application, such as Figure 1 As shown, this system addresses the technical problems of high false alarm rates, delayed response, and inability to dynamically adapt to battery aging caused by the limitations of single-sensor monitoring in existing battery thermal runaway early warning systems. It achieves the technical effect of improving the detection rate of early signs of thermal runaway and shortening the warning time through multi-modal sensing. The battery pack thermal runaway risk identification and early warning system includes: a data acquisition module 10, a detection module 20, a prediction module 30, and an evaluation module 40.

[0024] The acquisition module 10 is used to acquire the battery pack's operating parameter dataset in real time through a multimodal sensor network. The specific configuration of the acquisition module 10 will be described in detail below. As mentioned above, the acquisition module 10 may further include: a first sensing unit for distributing an array of infrared temperature sensors on the surface of the battery pack and between individual cells to form a first sensing layer; a second sensing unit for embedding miniature voltage sensors at the positive and negative electrode connections of the battery pack to form a second sensing layer; a third sensing unit for installing a gas composition spectrometer in the battery pack's exhaust channel to form a third sensing layer; a fourth sensing unit for attaching a flexible piezoelectric thin film sensor to the inner side of the battery pack casing to form a fourth sensing layer; and an alignment unit for synchronizing and aligning the first, second, third, and fourth sensing layers according to their operating timestamps to generate the operating parameter dataset.

[0025] Firstly, for the first sensing layer, a high-density infrared temperature sensor array (such as FLIR A315, accuracy ±1℃) is used on the surface of the battery module and between individual cells, arranged in a matrix with a 5mm spacing. Each individual cell corresponds to 3 temperature measurement points. The signal transmission line is integrated through a flexible FPC circuit board. The sampling frequency of the sensor array is set to 10Hz, and it is fixed to the battery surface with thermally conductive adhesive to reduce contact thermal resistance. The second sensing layer uses a miniature voltage sensor (such as TI INA226, resolution 0.1mV) with an embedded mounting process. The sensor probe is embedded in a slot at the positive and negative terminals, and the surface is covered with an insulating ceramic coating (thickness ≤0.2mm). Differential measurement technology is used to eliminate the influence of contact resistance, and the sampling frequency is set to 10Hz in sync with the temperature sensing layer. The third sensing layer uses a gas analyzer (such as Ocean Insight HR2000+ spectrometer) installed in the main exhaust channel of the battery pack. In-situ detection is performed through a quartz glass window. A gas pump is used to establish a gas circulation path at a flow rate of 200mL / min to monitor the characteristic absorption peaks of characteristic gases such as CO and HF in real time. The spectral scanning interval is set to 500ms. The flexible piezoelectric thin-film sensor of the fourth sensing layer (such as Measurement Specialties LDT0-028K) is made into an irregularly shaped structure that matches the curved surface of the battery casing using laser cutting technology. It is attached to the inside of the casing with conductive silver paste and is set to a sampling rate of 100Hz to capture micro-deformation signals (resolution 0.1μm). The time synchronization module uses an AD9548 clock generator chip to provide a unified clock source for each sensing layer through the IEEE 1588 Precision Time Protocol (PTP). A timestamp marking mechanism is established in the FPGA processing module (Xilinx Zynq UltraScale+). Data frames are transmitted using a CAN FD bus and an 8-byte timestamp is attached (accuracy ±1μs). Finally, the time synchronization module performs millisecond-level alignment of the data of each sensing layer. This means that the data of the four layers of sensing are interpolated and resampled through a time alignment algorithm to generate a synchronized operating parameter dataset with a time deviation of less than 2ms. The data storage format adopts a layered HDF5 structure, which contains four sets of two-dimensional arrays (time × spatial coordinates) with timestamps for temperature, voltage, gas concentration, and deformation, thus determining the operating parameter dataset.

[0026] The detection module 20 is used to perform hybrid anomaly detection on the battery pack based on the operating parameter dataset to obtain real-time anomaly detection results. The specific configuration of the detection module 20 will be described in detail below. As mentioned above, the detection module 20 may further include: a feature extraction unit, used to extract spatial features based on the operating parameter dataset to generate a high-dimensional feature vector; a classification unit, used to perform multi-classification based on the high-dimensional feature vector, construct a multi-classification decision, and perform local feature enhancement on the operating parameter dataset according to the multi-classification decision to obtain multiple enhanced feature data; and a weighted fusion unit, used to retrieve historical detection results, perform weighted fusion of the multiple enhanced feature data with the historical detection results, and generate a real-time anomaly detection result, wherein the real-time anomaly detection result includes anomaly type labels and confidence scores.

[0027] Spatial feature extraction and anomaly detection are performed using a three-dimensional convolutional neural network (3D-CNN): First, the four layers of sensor data are constructed into a four-dimensional input tensor (time × spatial coordinates × sensor layer × physical quantity). Cross-sensor layer feature fusion is then performed using three 3×3×3 convolutional kernels. The first convolutional layer extracts local temperature-voltage correlation features (using ReLU activation function), the second convolutional layer captures gas-deformation coupling features (using SELU activation function to enhance nonlinear expression), and the third convolutional layer generates a 128-dimensional high-dimensional feature vector. Next, the feature vector is input into an improved support vector machine (SVM) classifier, employing a Mahalanobis distance-weighted multi-class decision tree to classify anomalies. The system is categorized into three types: normal, early thermal runaway (gas production-dominated, temperature rise-dominated), and severe anomaly. A sliding window mechanism (5-second window length, 1-second step) is introduced during the classification process. Residual connections are made between feature vectors from consecutive time windows, and temporal feature enhancement is achieved through a gated recurrent unit (GRU). This generates enhanced feature data containing spatiotemporal correlations, i.e., real-time anomaly detection results. Historical detection results are stored in a circular buffer database of the edge computing unit (capacity retaining data from the most recent 72 hours). An exponential decay weighted algorithm (decay coefficient λ=0.95) is used to fuse the current enhanced features with historical data. Specifically, the joint confidence is calculated using a Bayesian probabilistic framework.

[0028] ;

[0029] in, For current enhancement features, For historical features, the final anomaly type label is determined by the maximum posterior probability, and the confidence score is normalized by the Softmax function.

[0030] The prediction module 30 is used to predict the thermal state of the battery pack based on the real-time anomaly detection results and construct a thermal state evolution trend map. The specific configuration of the prediction module 30 will be described in detail below. As mentioned above, the prediction module 30 may further include: a selection unit, used to dynamically select a subset of input features based on the anomaly detection results and construct a multi-dimensional time-series input matrix for a target time window; a thermal runaway prediction unit, used to perform thermal runaway prediction on the multi-dimensional time-series input matrix to obtain a thermal state prediction dataset, which includes predicted values ​​for temperature change rate, gas production rate, and deformation acceleration; a first state evolution unit, used to perform state evolution based on the temperature change rate of the battery pack to obtain spatial temperature distribution data; a second state evolution unit, used to perform state evolution based on the gas production rate of the battery pack to obtain gas diffusion path data; a third state evolution unit, used to perform state evolution based on the predicted value of deformation acceleration of the battery pack to obtain mechanical deformation vector data; and an association and integration unit, used to associate and integrate the spatial temperature distribution data, the gas diffusion path data, and the mechanical deformation vector data to draw the thermal state evolution trend map.

[0031] Based on the anomaly type labels (such as gas production-dominant or temperature rise-dominant) and confidence scores in real-time anomaly detection results, a subset of key features is dynamically selected using Mahalanobis distance and mutual information. For example, when a gas production anomaly is detected (confidence > 0.8), three types of features—gas concentration, temperature gradient, and voltage fluctuation rate—are prioritized. A multi-dimensional time-series input matrix (dimension: time step × number of features × spatial nodes) is constructed using a sliding time window (5 minutes in length, 10 seconds in step size). Simultaneously, the theoretical heat production rate output by the electrochemical model is introduced as a physical constraint feature to achieve feature-level fusion.

[0032] The input matrix is ​​further fed into an improved Bi-LSTM prediction network, whose structure includes bidirectional LSTM layers (128 nodes per layer) and an attention mechanism layer. The temperature prediction branch uses a three-dimensional finite element heat conduction model for physical constraints, predicting the temperature field distribution for the next 2 minutes every 10 seconds (spatial resolution 1 cm³). The gas production prediction branch combines CFD simulation data to calculate the gas diffusion path through concentration gradients, outputting the main diffusion direction and velocity vector. The deformation prediction branch uses finite element analysis to calculate the stress distribution of the shell, outputting the deformation acceleration vector field.

[0033] The three types of prediction data are mapped to a unified coordinate system by a spatiotemporal alignment algorithm. The temperature field is interpolated quickly using an octree spatial index, the gas path is generated as a streamline trajectory using a particle tracking method, and the deformation vector is transformed using a Lagrange description. A visualization engine (based on the PyVista library) constructs a 3D dynamic trend chart. The process may include mapping predicted temperature data to a 3D mesh model of the battery pack, using the HSV color space (red→blue corresponding to 45℃→80℃) to represent the temperature distribution, overlaying isothermal surfaces (5℃ intervals) to display the heat accumulation area, and then driving the SmoothedParticle Hydrodynamics (SPH) fluid dynamics model based on the predicted gas production rate to generate gas clouds with varying transparency (CO is displayed as orange, HF as purple). The length of the streamline arrows is proportional to the diffusion velocity. Deformation vector arrows (scaling factor 50:1) are drawn on the surface of the battery casing, with the arrow direction indicating the deformation trend and the color depth (green→red) reflecting the magnitude of deformation acceleration (0→5mm / s²). The Marching Cubes algorithm is used to extract the intersection area of ​​temperature >60℃, gas production rate >50ppm / s, and deformation acceleration >3mm / s², and the core area of ​​thermal runaway is highlighted with pulsed red light. The system updates the trend chart every 30 seconds and supports user interaction (rotation / zoom / profile viewing). It also overlays and displays key parameters (such as maximum temperature rise rate and total gas production) in the form of digital instruments. When the deviation between the predicted data and the measured values ​​exceeds 15%, the system automatically triggers online fine-tuning of the LSTM network (learning rate 0.001, batch size 32) to obtain the thermal state evolution trend chart.

[0034] The specific configuration of the prediction module 30 will be described in detail below. As mentioned above, the prediction module 30 may further include: a data filtering unit, used to filter from the running parameter dataset according to the anomaly type label to generate a first feature subset; a weighting unit, used to weight the first feature subset based on the confidence score to generate a second feature subset; a data extraction unit, used to extract the moving average, standard deviation, and kurtosis coefficient of the second feature subset within a target time window to construct a multidimensional statistical feature matrix; and a fusion unit, used to perform feature-level fusion of the multidimensional statistical feature matrix with the electrochemical mechanism simulation data to generate the multidimensional time-series input matrix.

[0035] From the operational parameter dataset, parameters closely related to specific anomalies (i.e., sensor data with a correlation exceeding a set threshold) are extracted based on anomaly type labels, forming the first feature subset. A confidence scoring mechanism is then introduced to weight each feature in the first feature subset. The confidence score reflects the reliability of a feature in predicting thermal runaway risk. By weighting the influence of different features in risk assessment, a second feature subset is generated.

[0036] Subsequently, within the target time window, statistical analysis is performed on the second feature subset, calculating statistical indicators such as the moving average, standard deviation, and kurtosis coefficient. The moving average is obtained by setting a fixed-size window on the data sequence and calculating the arithmetic mean of the data within that window. As the window slides across the data sequence, a series of averages are obtained, thus reflecting the local trend of the data.

[0037] Standard deviation is a statistic that measures the dispersion of data. It is calculated by averaging the squares of the differences between each value in a data series and the mean, and then taking the square root. A larger standard deviation indicates that the data points are more dispersed, and vice versa.

[0038] Kurtosis is a statistical measure describing the shape of a data distribution, reflecting the sharpness of the distribution. The kurtosis coefficient is calculated by averaging the fourth power of the difference between each value in the data sequence and the mean, then dividing by the fourth power of the standard deviation. A kurtosis coefficient greater than 3 indicates that the data distribution is sharper than a normal distribution; a coefficient less than 3 indicates a flatter distribution. These statistical calculations provide important basis for data analysis.

[0039] The aforementioned indicators, from the perspectives of time series stability, dispersion, and distribution patterns, reveal the changing patterns and potential risks of the characteristics. Based on this, a multidimensional statistical feature matrix was constructed. To further improve the accuracy of risk assessment, this matrix was fused with electrochemical mechanism simulation data at the feature level. This process integrates the two at the feature level. First, the multidimensional statistical feature matrix and the electrochemical mechanism simulation data were standardized to ensure numerical comparability. Then, using specific fusion algorithms, such as concatenation, weighted averaging, or more complex machine learning algorithms, the two were fused in the feature space. This not only combines the real-time performance and accuracy of experimental data but also incorporates a deeper understanding of the mechanisms in the simulation data, thereby generating a multidimensional time-series input matrix containing rich information and deep mechanisms. This matrix serves as the input to the subsequent risk assessment model, providing a more reliable guarantee for the safe operation of the battery pack.

[0040] The evaluation module 40 is used to perform thermal runaway risk assessment based on the thermal state evolution trend diagram, determine the thermal runaway risk level, trigger multi-level early warning signals according to the thermal runaway risk level, and generate thermal runaway suppression strategies.

[0041] The specific configuration of the evaluation module 40 will be described in detail below. As mentioned above, the evaluation module 40 may further include: a multidimensional decomposition unit, used to decompose the thermal state evolution trend diagram into multiple dimensions to obtain multidimensional data, wherein the multidimensional data includes thermal accumulation rate data, thermal diffusion gradient data, and thermal runaway trigger probability data; a dynamic adjustment unit, used to dynamically adjust the weight coefficients of the thermal accumulation rate data, the thermal diffusion gradient data, and the thermal runaway trigger probability data according to the health state parameters of the battery pack, and generate a comprehensive risk score; and a risk determination unit, used to determine the risk level based on the comprehensive risk score to obtain the thermal runaway risk level.

[0042] Based on four-dimensional spatiotemporal data (three-dimensional space + time) of thermal state evolution trend map, key risk parameters are extracted using tensor decomposition technology. Local temperature change rate is calculated through spatiotemporal derivative operation. Gaussian filtering is used to eliminate noise and determine the thermal accumulation rate data. Furthermore, a three-dimensional thermal conduction tensor is constructed. The anisotropic diffusion coefficient is solved using the finite difference method to determine the thermal diffusion gradient data. Finally, the gas generation rate and deformation acceleration are fused to construct regression coefficients to determine the probability data of thermal runaway triggering.

[0043] Further, an adaptive weight matrix based on the state of health (SOH) of the battery is established. When SOH > 80%, the weights for thermal accumulation rate are w1 = 0.5, diffusion gradient w2 = 0.3, and trigger probability w3 = 0.2. When 40% ≤ SOH ≤ 80%, the weights are dynamically adjusted according to w1 = 0.3 + 0.2 × (SOH / 40), w2 = 0.4, and w3 = 0.3 − 0.1 × (SOH / 40). When SOH < 40%, the trigger probability weight is increased to w3 = 0.5, and the remaining weights are reduced proportionally. A comprehensive risk score is generated, and the risk level is determined based on the comprehensive risk score, thereby obtaining the thermal runaway risk level of the battery pack. This comprehensive, accurate, and dynamic assessment of the thermal runaway risk of the battery pack ensures the safe and stable operation of the battery pack.

[0044] The specific configuration of the evaluation module 40 will be described in detail below. As mentioned above, the evaluation module 40 may further include: a traversal unit, used to construct a risk level quantification matrix based on the comprehensive risk score, and traverse the risk level quantification matrix to extract the thermal diffusivity factor, electrochemical instability index, and deformation energy accumulation value of the battery pack; a threshold setting unit, used to set multi-level risk thresholds, the multi-level risk thresholds including a first risk threshold, a second risk threshold, and a third risk threshold; a first determination unit, used to determine a level I risk when the thermal diffusivity factor exceeds the first risk threshold and the electrochemical instability index is lower than the second risk threshold; a second determination unit, used to determine a level II risk when the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the deformation energy accumulation value is lower than the third risk threshold; and a third determination unit, used to determine a level III risk when the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the deformation energy accumulation value exceeds the third risk threshold.

[0045] Based on the three-dimensional thermal state evolution data of the battery pack, a three-dimensional quantification matrix incorporating thermodynamics, electrochemistry, and mechanics was established. Thermal diffusivity (...) The calculation uses the improved Green's function method:

[0046] ;

[0047] in, The maximum temperature difference between adjacent cells. Let k be the heat production rate of the heat source. For the distance between heat sources, The thermal conductivity attenuation coefficient of the material. For effective heat dissipation surface area.

[0048] Electrochemical instability index ( ) Calculated by coupling the gas concentration abrupt change rate with the voltage relaxation time:

[0049] ;

[0050] in, The electrochemical coupling coefficient is... is the voltage recovery time constant.

[0051] Deformation energy accumulation ( Integrating the energy density through the outer shell strain:

[0052] ;

[0053] in, For stress tensor, For strain tensor.

[0054] Furthermore, using a sliding window statistical method, the first risk threshold can be updated and determined every 5 minutes. The second risk threshold is determined based on the Arrhenius equation, and the third risk threshold is determined by relating it to the fatigue characteristics of the material. When the thermal diffusivity factor exceeds the first risk threshold and the electrochemical instability index is lower than the second risk threshold, it is determined to be a Level I risk, i.e., a local overheating risk, triggering a yellow warning (flashing warning on the HMI interface). When the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the cumulative deformation energy value is lower than the third risk threshold, it is determined to be a Level II risk, i.e., a thermo-electric coupling risk, triggering an orange warning. When the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the cumulative deformation energy value exceeds the third risk threshold, it is determined to be a Level III risk. That is, when all three indicators exceed the threshold, the system triggers a red warning within 10ms.

[0055] The specific configuration of the evaluation module 40 will be described in detail below. As mentioned above, the evaluation module 40 may further include: a dimensional analysis unit, used to analyze the matrix dimensions of the risk level quantification matrix and determine the thermodynamic dimension, electrochemical dimension, and mechanical dimension; a first calculation unit, used to calculate the product of the maximum temperature difference between adjacent cells in the temperature field and the heat conduction rate as the thermal diffusion factor based on the thermodynamic dimension and the thermal diffusion gradient data; a second calculation unit, used to calculate the electrochemical instability index based on the electrochemical dimension, the thermal runaway trigger probability data combined with the gas concentration change curve and voltage drop rate; and a third calculation unit, used to analyze the stress concentration area of ​​the battery pack's casing through the deformation vector field combined with the heat accumulation rate data based on the mechanical dimension, and calculate the deformation energy accumulation value.

[0056] In the construction and parameter calculation of the risk level quantification matrix, the system achieves three-dimensional risk assessment through multi-physics coupling analysis, including thermodynamic, electrochemical, and mechanical dimensions. Furthermore, based on three-dimensional temperature field data (spatial resolution 1 cm³) collected by an infrared temperature sensor array, a thermal network model between battery cells is constructed. The thermal diffusivity factor calculation process is as follows:

[0057] The system iterates through all adjacent cell pairs in the three-dimensional temperature field data and performs real-time temperature difference calculations on all adjacent cell pairs. It records the maximum temperature difference value and then corrects it by combining Fourier's law with the change in contact thermal resistance caused by aging to obtain the heat conduction rate. Finally, it multiplies the maximum temperature difference value with the heat conduction rate and uses it as the heat diffusion factor. The higher the heat diffusion factor, the stronger the thermal diffusion capability of the battery pack.

[0058] Furthermore, based on the electrochemical dimension, combining gas spectral data (CO / HF concentrations CgasCgas) and voltage sensor data, the electrochemical instability index is calculated: The gas concentration is monitored in real time using the CUSUM algorithm; when the rate of change in gas concentration exceeds the baseline, an event marker is triggered to obtain the gas concentration abrupt change detection result. Simultaneously, the voltage change rate is calculated within a 1-second time window, and the relaxation time constant is extracted using Hilbert transform. The exponential synthesis formula for the electrochemical instability index is as follows:

[0059] ;

[0060] in, The electrochemical instability index, For gas concentration, This is the critical value for gas concentration. Let be the relaxation time constant. This is a reference relaxation time under healthy conditions.

[0061] Furthermore, based on the mechanical dimension, the cumulative deformation energy value is constructed using strain data (sampling rate 1kHz) from a flexible piezoelectric thin-film sensor. This involves interpolating discrete strain measurements into a full-field strain tensor based on a shell finite element model, calculating the stress tensor based on the anisotropic parameters of the composite material (obtained by multiplying the full-field strain tensor by the stiffness matrix), and also incorporating the thermal accumulation rate to compensate for Young's modulus temperature. The mechanical energy integral is accelerated using a GPU (CUDA kernel function), achieving a processing speed of 1M elements / ms. All dimensional data are timestamped (error <1ms), and quaternion interpolation is used to compensate for sensor latency. This approach reduces the time required for 3D risk assessment from 120ms using traditional methods to 18ms, while controlling the error of cross-dimensional coupling analysis to within 3%.

[0062] The specific configuration of the evaluation module 40 will be described in detail below. As mentioned above, the evaluation module 40 may further include: a fourth calculation unit, used to calculate the three-dimensional risk diffusion of the battery pack based on real-time temperature distribution data and the heat conduction equation, and determine the thermal runaway propagation path; a first triggering unit, used to trigger a first warning signal when the thermal runaway risk level is Level I, locate the abnormal position coordinates according to the thermal runaway propagation path, and generate an early warning message; a second triggering unit, used to trigger a second warning signal when the thermal runaway risk level is Level II, dynamically adjust the liquid cooling system according to the thermal runaway propagation path, generate a liquid cooling control command, and distribute the flow of the cooling pipes of the liquid cooling system; and a third triggering unit, used to trigger a third warning signal when the thermal runaway risk level is Level III, activate the directional fire extinguishing device based on the thermal runaway propagation path, and generate an emergency pressure relief channel.

[0063] Based on real-time infrared temperature data (sampling rate 10Hz), a three-dimensional non-uniform thermal conduction model of the battery pack is constructed. The thermal diffusion process can be described by using a parabolic PDE with variable coefficients. The thermal conductivity of the battery pack varies with SOH and temperature. The battery pack is then discretized into 5 million tetrahedral elements. The temperature field evolution is solved in parallel using the explicit Euler method. The three-dimensional temperature distribution is updated every 50ms. At the same time, the Dijkstra algorithm is used to find the path with the maximum temperature gradient. The thermal runaway propagation direction vector is defined. Paths with a propagation rate exceeding 0.5m / s are marked as high-risk channels, and the thermal runaway propagation path is determined.

[0064] Furthermore, when the thermal runaway risk level is Level I, a first warning signal is triggered. The abnormal location coordinates are located according to the thermal runaway propagation path. This means that the hot spot coordinates are detected by the extreme value of the second derivative of the temperature field, with a positioning accuracy of ±3mm. A JSON format message is constructed to generate a warning message, thereby activating the three axial fans closest to the hot spot (wind speed v=5+10×(Score−0.4) m / s) and limiting the charging current through the BMS.

[0065] Furthermore, when the thermal runaway risk level is Level II, a second warning signal is triggered, and the liquid cooling system is dynamically adjusted according to the thermal runaway propagation path. This means that based on the propagation path topology, the liquid cooling pipeline is divided into a priority cooling zone (upstream of the path) and a secondary cooling zone (downstream of the path), and liquid cooling control commands are generated to distribute the flow of the cooling pipeline of the liquid cooling system. The flow distribution can be controlled by using a PID-Smith predictor. When the local temperature rise rate is >3℃ / s, the flow is switched to turbulent mode (Reynolds number Re>4000), which improves the heat transfer coefficient by 30%.

[0066] Furthermore, when the thermal runaway risk level is Level III, a third warning signal is triggered, and the directional fire extinguishing device is activated based on the thermal runaway propagation path. This means that the extinguishing agent spray angle is calculated according to the propagation path vector, with an angle control accuracy of ±0.5° and a response time of <50ms. Pulse width modulation (PWM) is used to control the extinguishing agent flow rate. Then, the pressure gradient field is solved inversely based on the shell deformation data to determine the optimal pressure relief point sequence. A shape memory alloy (SMA) actuator is used to open the pressure relief valve. The opening sequence satisfies the optimal pressure relief point sequence, ensuring that the pressure release wavefront and the heat propagation wavefront meet in reverse to generate an emergency pressure relief channel. The response time of the emergency pressure relief channel is reduced from the traditional 150ms to 35ms, and the extinguishing agent coverage efficiency is improved to 92%.

[0067] This application addresses the technical problems of high false alarm rate, delayed response, and inability to dynamically adapt to battery aging caused by the limitations of single sensor monitoring in existing battery thermal runaway early warning systems. It achieves the technical effect of improving the detection rate of early signs of thermal runaway and shortening the warning time through multimodal sensing.

[0068] In the above text, refer to Figure 1 A battery pack thermal runaway risk identification and early warning system according to an embodiment of this application is described in detail. Next, reference will be made to... Figure 2 This application describes a method for identifying and warning of thermal runaway risk in battery packs according to embodiments of the present application.

[0069] This application provides a method for identifying and warning of thermal runaway risk in a battery pack. The method is applied to a battery pack thermal runaway risk identification and warning system and includes:

[0070] The battery pack's operating parameter dataset is collected in real time using a multimodal sensor network; hybrid anomaly detection is performed on the battery pack based on the operating parameter dataset to obtain real-time anomaly detection results; thermal state prediction is performed on the battery pack based on the real-time anomaly detection results to construct a thermal state evolution trend map; thermal runaway risk assessment is performed based on the thermal state evolution trend map to determine the thermal runaway risk level; multi-level early warning signals are triggered based on the thermal runaway risk level, and a thermal runaway suppression strategy is generated.

[0071] The deployment process of a multimodal sensor network includes:

[0072] An array of infrared temperature sensors is distributed across the surface of the battery pack and the gaps between individual cells to form a first sensing layer; a miniature voltage sensor is embedded at the connection between the positive and negative electrodes of the battery pack to form a second sensing layer; a gas composition spectrometer is installed in the exhaust channel of the battery pack to form a third sensing layer; a flexible piezoelectric thin film sensor is attached to the inside of the battery pack casing to form a fourth sensing layer; the first, second, third, and fourth sensing layers are synchronized and aligned according to their operating timestamps to generate the operating parameter dataset.

[0073] Based on the aforementioned operational parameter dataset, hybrid anomaly detection is performed on the battery pack to obtain real-time anomaly detection results, including:

[0074] Spatial features are extracted from the runtime parameter dataset to generate a high-dimensional feature vector. Multi-classification is performed based on the high-dimensional feature vector to construct a multi-classification decision. Local feature enhancement is performed on the runtime parameter dataset according to the multi-classification decision to obtain multiple enhanced feature data. Historical detection results are retrieved, and the multiple enhanced feature data are weighted and fused with the historical detection results to generate a real-time anomaly detection result. The real-time anomaly detection result includes anomaly type labels and confidence scores.

[0075] Based on the real-time anomaly detection results, the thermal state of the battery pack is predicted, and a thermal state evolution trend diagram is constructed, including:

[0076] Based on the anomaly detection results, a subset of input features is dynamically selected to construct a multi-dimensional time-series input matrix for the target time window. Thermal runaway prediction is performed on the multi-dimensional time-series input matrix to obtain a thermal state prediction dataset, which includes predicted values ​​for temperature change rate, gas production rate, and deformation acceleration. State evolution is performed based on the battery pack's temperature change rate to obtain spatial temperature distribution data. State evolution is also performed based on the battery pack's gas production rate to obtain gas diffusion path data. Furthermore, state evolution is performed based on the battery pack's predicted deformation acceleration to obtain mechanical deformation vector data. Finally, the spatial temperature distribution data, the gas diffusion path data, and the mechanical deformation vector data are correlated and integrated to plot the thermal state evolution trend map.

[0077] Based on the anomaly detection results, a subset of input features is dynamically selected to construct a multi-dimensional time-series input matrix for the target time window, including:

[0078] The first feature subset is generated by filtering the runtime parameter dataset according to the anomaly type label; the first feature subset is weighted based on the confidence score to generate a second feature subset; the moving average, standard deviation and kurtosis coefficient of the second feature subset within the target time window are extracted to construct a multidimensional statistical feature matrix; the multidimensional statistical feature matrix is ​​fused with the electrochemical mechanism simulation data at the feature level to generate the multidimensional time series input matrix.

[0079] Based on the aforementioned thermal state evolution trend diagram, a thermal runaway risk assessment is performed to determine the thermal runaway risk level, including:

[0080] The thermal state evolution trend map is decomposed into multiple dimensions to obtain multidimensional data, which includes thermal accumulation rate data, thermal diffusion gradient data, and thermal runaway trigger probability data. The weighting coefficients of the thermal accumulation rate data, thermal diffusion gradient data, and thermal runaway trigger probability data are dynamically adjusted according to the health state parameters of the battery pack to generate a comprehensive risk score. The risk level is determined based on the comprehensive risk score to obtain the thermal runaway risk level.

[0081] The risk level is determined based on the comprehensive risk score to obtain the thermal runaway risk level, including:

[0082] Based on the comprehensive risk score, a risk level quantification matrix is ​​constructed. The matrix is ​​then traversed to extract the thermal diffusivity factor, electrochemical instability index, and cumulative deformation energy of the battery pack. Multiple risk thresholds are set, including a first risk threshold, a second risk threshold, and a third risk threshold. When the thermal diffusivity factor exceeds the first risk threshold and the electrochemical instability index is lower than the second risk threshold, the risk is classified as Level I. When the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the cumulative deformation energy is lower than the third risk threshold, the risk is classified as Level II. When the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the cumulative deformation energy exceeds the third risk threshold, the risk is classified as Level III.

[0083] The thermal diffusivity, electrochemical room temperature index, and cumulative deformation energy of the battery pack are extracted by traversing the aforementioned risk level quantification matrix, including:

[0084] The matrix dimensions of the risk level quantification matrix are analyzed to determine the thermodynamic, electrochemical, and mechanical dimensions. Based on the thermodynamic dimension, the product of the maximum temperature difference between adjacent cells in the temperature field and the thermal conduction rate is calculated using the thermal diffusion gradient data as the thermal diffusion factor. Based on the electrochemical dimension, the electrochemical instability index is obtained by combining the thermal runaway trigger probability data with the voltage drop rate of the gas concentration change curve. Based on the mechanical dimension, the stress concentration area of ​​the battery pack's casing is analyzed using the deformation vector field combined with the thermal accumulation rate data to calculate the cumulative deformation energy.

[0085] Multiple warning signals are triggered based on the aforementioned thermal runaway risk level, including:

[0086] Based on real-time temperature distribution data, a three-dimensional risk diffusion calculation of the battery pack is performed using the heat conduction equation to determine the thermal runaway propagation path. When the thermal runaway risk level is Level I, a first warning signal is triggered, and the abnormal location coordinates are located according to the thermal runaway propagation path to generate an early warning message. When the thermal runaway risk level is Level II, a second warning signal is triggered, and the liquid cooling system is dynamically adjusted according to the thermal runaway propagation path to generate a liquid cooling control command and distribute the flow of the cooling pipes of the liquid cooling system. When the thermal runaway risk level is Level III, a third warning signal is triggered, and the directional fire extinguishing device is activated based on the thermal runaway propagation path to generate an emergency pressure relief channel.

[0087] The battery pack thermal runaway risk identification and early warning system provided in this application can execute the battery pack thermal runaway risk identification and early warning method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects of the execution method.

[0088] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this application.

[0089] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A battery pack thermal runaway risk identification and early warning system, characterized in that, The system includes: The acquisition module is used to acquire the battery pack's operating parameter dataset in real time through a multimodal sensor network; The detection module is used to perform mixed anomaly detection on the battery pack based on the operating parameter dataset and obtain real-time anomaly detection results; The prediction module is used to predict the thermal state of the battery pack based on the real-time anomaly detection results and construct a thermal state evolution trend map. The thermal state evolution trend map is drawn after being associated and integrated with spatial temperature distribution data, gas diffusion path data, and mechanical deformation vector data. The assessment module is used to assess the risk of thermal runaway based on the thermal state evolution trend diagram, determine the risk level of thermal runaway, trigger multi-level early warning signals according to the risk level of thermal runaway, and generate a thermal runaway suppression strategy. The detection module includes: The feature extraction unit is used to extract spatial features based on the running parameter dataset and generate a high-dimensional feature vector. The classification unit is used to perform multi-classification based on the high-dimensional feature vector, construct multi-classification decision, and perform local feature enhancement on the running parameter dataset according to the multi-classification decision to obtain multiple enhanced feature data. The weighted fusion unit is used to retrieve historical detection results, perform weighted fusion of the multiple enhanced feature data with the historical detection results, and generate real-time anomaly detection results, which include anomaly type labels and confidence scores. The data acquisition module includes: The first sensing unit is used to deploy an array of infrared temperature sensors in a distributed manner on the surface of the battery pack and between individual cells to form the first sensing layer. The second sensing unit is used to embed a miniature voltage sensor at the connection between the positive and negative terminals of the battery pack to form a second sensing layer. The third sensing unit is used to install a gas composition spectrometer in the battery pack exhaust channel to form the third sensing layer. The fourth sensing unit is used to attach a flexible piezoelectric thin film sensor to the inside of the battery pack casing to form the fourth sensing layer; The alignment unit is used to synchronize and align the first sensing layer, the second sensing layer, the third sensing layer, and the fourth sensing layer according to the running timestamp to generate the running parameter dataset. The prediction module includes: The selection unit is used to dynamically select a subset of input features based on the anomaly detection results and construct a multi-dimensional time-series input matrix for the target time window. The thermal runaway prediction unit is used to perform thermal runaway prediction on the multi-dimensional time-series input matrix to obtain a thermal state prediction dataset, which includes predicted values ​​of temperature change rate, gas production rate, and deformation acceleration. The first state evolution unit is used to perform state evolution based on the temperature change rate of the battery pack to obtain spatial temperature distribution data. The second state evolution unit is used to perform state evolution based on the gas production rate of the battery pack to obtain gas diffusion path data. The third state evolution unit is used to perform state evolution based on the predicted deformation acceleration value of the battery pack to obtain mechanical deformation vector data; The association and integration unit is used to associate and integrate the spatial temperature distribution data, the gas diffusion path data, and the mechanical deformation vector data to draw the thermal state evolution trend diagram; The prediction module includes: A data filtering unit is used to filter the running parameter dataset according to the anomaly type label to generate a first feature subset; A weighting unit is used to weight the first feature subset based on the confidence score to generate a second feature subset; The data extraction unit is used to extract the moving average, standard deviation, and kurtosis coefficient of the second feature subset within the target time window, and construct a multidimensional statistical feature matrix. The fusion unit is used to perform feature-level fusion of the multidimensional statistical feature matrix and the electrochemical mechanism simulation data to generate the multidimensional time-series input matrix.

2. The battery pack thermal runaway risk identification and early warning system as described in claim 1, characterized in that, The evaluation module includes: A multidimensional decomposition unit is used to perform multidimensional decomposition on the thermal state evolution trend diagram to obtain multidimensional data, which includes thermal accumulation rate data, thermal diffusion gradient data, and thermal runaway trigger probability data. The dynamic adjustment unit is used to dynamically adjust the weighting coefficients of the thermal accumulation rate data, the thermal diffusion gradient data, and the thermal runaway trigger probability data according to the health status parameters of the battery pack, and generate a comprehensive risk score. The risk assessment unit is used to determine the risk level based on the comprehensive risk score and obtain the thermal runaway risk level.

3. The battery pack thermal runaway risk identification and early warning system as described in claim 2, characterized in that, The evaluation module includes: The traversal unit is used to construct a risk level quantification matrix based on the comprehensive risk score, and to traverse the risk level quantification matrix to extract the thermal diffusivity factor, electrochemical instability index and deformation energy accumulation value of the battery pack. A threshold setting unit is used to set multi-level risk thresholds, which include a first risk threshold, a second risk threshold, and a third risk threshold. The first determination unit is used to determine the risk as Level I when the thermal diffusivity factor exceeds the first risk threshold and the electrochemical instability index is lower than the second risk threshold. The second determination unit is used to determine a Level II risk when the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the cumulative deformation energy value is lower than the third risk threshold. The third determination unit is used to determine the risk level as III when the thermal diffusivity factor exceeds the first risk threshold, the electrochemical instability index exceeds the second risk threshold, and the cumulative deformation energy exceeds the third risk threshold.

4. The battery pack thermal runaway risk identification and early warning system as described in claim 3, characterized in that, The evaluation module includes: The dimension analysis unit is used to analyze the matrix dimensions of the risk level quantification matrix and determine the thermodynamic dimension, electrochemical dimension, and mechanical dimension. The first calculation unit is used to calculate the product of the maximum temperature difference between adjacent units in the temperature field and the heat conduction rate, based on the thermodynamic dimension and the heat diffusion gradient data, as the heat diffusion factor. The second calculation unit is used to calculate the electrochemical instability index based on the electrochemical dimension, the thermal runaway trigger probability data, and the voltage drop rate of the gas concentration change curve. The third calculation unit is used to analyze the stress concentration area of ​​the battery pack's outer casing based on the mechanical dimension, by combining the deformation vector field with the thermal accumulation rate data, and to calculate the cumulative deformation energy value.

5. The battery pack thermal runaway risk identification and early warning system as described in claim 3, characterized in that, The evaluation module includes: The fourth calculation unit is used to calculate the three-dimensional risk diffusion of the battery pack based on real-time temperature distribution data and the thermal conduction equation, and to determine the thermal runaway propagation path. The first triggering unit is used to trigger a first early warning signal when the thermal runaway risk level is Level I, locate the abnormal location coordinates according to the thermal runaway propagation path, and generate an early warning message. The second triggering unit is used to trigger a second early warning signal when the thermal runaway risk level is Level II, dynamically adjust the liquid cooling system according to the thermal runaway propagation path, generate liquid cooling control commands, and distribute the flow of the cooling pipes of the liquid cooling system. The third triggering unit is used to trigger a third early warning signal when the thermal runaway risk level is Level III, and to activate the directional fire extinguishing device based on the thermal runaway propagation path to generate an emergency pressure relief channel.

6. A method for identifying and warning of thermal runaway risk in battery packs, characterized in that, The method is used to implement the battery pack thermal runaway risk identification and early warning system according to any one of claims 1-5, and the method includes: Real-time acquisition of battery pack operating parameter datasets via a multimodal sensor network; Based on the aforementioned operational parameter dataset, hybrid anomaly detection is performed on the battery pack to obtain real-time anomaly detection results. Based on the real-time anomaly detection results, the thermal state of the battery pack is predicted, and a thermal state evolution trend diagram is constructed. Based on the thermal state evolution trend diagram, a thermal runaway risk assessment is performed to determine the thermal runaway risk level. Multi-level early warning signals are triggered according to the thermal runaway risk level, and a thermal runaway suppression strategy is generated.

Images