A power battery thermal runaway early warning and detection system

By dynamically integrating and controlling multi-source sensor data acquisition and multi-level early warning models, the problems of single sensor dimension, lack of graded early warning and model rigidity in power battery thermal runaway early warning are solved, realizing efficient and accurate detection and active protection against battery thermal runaway.

CN123112697APending Publication Date: 2026-11-17ZHONGCHUANG INT INSPECTION & CERTIFICATION (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611077223.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing power battery thermal runaway early warning technologies suffer from problems such as single sensing dimension, lack of hierarchical early warning mechanism, poor adaptability of fixed models, and imperfect linkage control, resulting in high false alarm rate, high false alarm rate, and inability to provide timely and effective protection.

Method used

The system employs a multi-source sensor data acquisition module to collect multi-dimensional physical quantity data in real time. A comprehensive thermal feature vector is generated through a feature extraction and fusion module. A multi-level early warning model library is used for hierarchical early warning. A hierarchical alarm and linkage control module is used to execute linkage protection measures. The system's adaptability is improved by combining a self-learning and model update module.

Benefits of technology

It achieves comprehensive and multi-dimensional perception of battery thermal runaway, significantly improves early feature capture capability, reduces false alarm and false alarm rates, provides phased early warning information, ensures safety and adaptability, and realizes a leap from passive alarm to active protection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123112697A_ABST
    Figure CN123112697A_ABST
Patent Text Reader

Abstract

This invention relates to the field of battery early warning and detection technology, and discloses a power battery thermal runaway early warning and detection system, including: a multi-source sensor data acquisition module; a feature extraction and fusion module; a multi-level early warning model library; a thermal runaway risk comprehensive assessment module; and a graded alarm and linkage control module. This invention, through the thermal runaway risk comprehensive assessment module, inputs the comprehensive thermal feature vector into multiple early warning models, calculates the risk index corresponding to each model, and generates the final thermal runaway risk level and remaining safety time estimate based on dynamic decision logic. This solves the problems of model rigidity and poor adaptability in existing technologies, achieving adaptive assessment for different battery types and operating conditions, and significantly reducing false alarm and false negative rates.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of battery early warning and detection technology, specifically to a power battery thermal runaway early warning and detection system. Background Technology

[0002] Thermal runaway of power batteries is one of the most serious safety incidents in electric vehicles and energy storage systems. It can be caused by mechanical abuse, electrical abuse, or thermal abuse, leading to violent exothermic chemical reactions inside the battery, ultimately causing fire or even explosion. Therefore, achieving early warning and accurate detection of thermal runaway is crucial for ensuring personal and property safety.

[0003] Existing battery thermal runaway early warning technologies mainly fall into the following categories: Single-parameter threshold-based methods: These methods monitor battery temperature or voltage, triggering an alarm when a preset threshold is exceeded. While simple, these methods have a high false alarm rate and cannot detect anomalies in the early stages of thermal runaway (such as the initial stage of an internal short circuit). Equivalent circuit model-based methods: These methods determine the battery state by identifying changes in battery model parameters (such as internal resistance and capacitance). These methods require high model accuracy and struggle to capture rapidly changing nonlinear characteristics before thermal runaway. Gas detection-based methods: These methods provide early warning by detecting characteristic gases such as hydrogen and CO released before battery thermal runaway. These methods have a relatively fast response time, but single gas sensors are susceptible to environmental interference and cannot distinguish between different stages of the thermal runaway process.

[0004] Existing technologies suffer from the following main problems: Limited sensing dimensions: Most systems rely solely on temperature or voltage, neglecting multi-dimensional information such as casing strain, internal pressure, and characteristic gases, resulting in short warning windows and low accuracy. Lack of tiered warning mechanisms: Existing systems often employ a binary logic of "safe unless alarmed," failing to provide phased and tiered warning information during thermal runaway development, hindering gradual intervention. Fixed models and poor adaptability: Fixed thresholds or models cannot adapt to the differences in thermal runaway characteristics under different battery chemistry systems, aging levels, and operating conditions, leading to frequent false alarms or missed alarms. Inadequate linkage control: After an alarm, there is a lack of intelligent linkage with the BMS, thermal management system, and fire suppression system, preventing automatic execution of protective measures such as current reduction, cooling, and power cut-off at critical moments.

[0005] In view of this, we propose a power battery thermal runaway early warning and detection system. Summary of the Invention

[0006] The purpose of this invention is to provide a power battery thermal runaway early warning and detection system to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A power battery thermal runaway early warning and detection system, the system comprising: The multi-source sensor data acquisition module is used to acquire multi-dimensional physical quantity data during battery operation in real time. The multi-dimensional physical quantity data includes at least: single cell voltage, charging and discharging current, electrode temperature, casing strain, internal air pressure, and characteristic gas concentration. The feature extraction and fusion module is used to extract multiple feature parameters that can characterize the precursors of battery thermal runaway from the multidimensional physical quantity data, and after performing spatiotemporal alignment and normalization on the feature parameters, generate a comprehensive thermal feature vector through an adaptive weighted fusion algorithm. A multi-level early warning model library is used to store multiple pre-trained thermal runaway early warning models, including an early warning model based on abnormal voltage drop, a mid-term early warning model based on gas production rate, and an emergency alarm model based on temperature mutation rate. The thermal runaway risk comprehensive assessment module is used to input the comprehensive thermal feature vector into the multiple early warning models respectively, calculate the risk index corresponding to each model, and generate the final thermal runaway risk level and remaining safe time estimate based on a dynamic decision logic. The graded alarm and linkage control module is used to execute graded alarms according to the thermal runaway risk level and trigger corresponding linkage control strategies. The linkage control strategies include at least: adjusting the charging and discharging parameters of the battery management system, starting the thermal management system, cutting off the high-voltage circuit, and sending alarm information to the cloud monitoring platform.

[0008] Preferably, the multi-source sensor data acquisition module includes: Voltage and current sensors are used to acquire the terminal voltage of each individual cell and the charging and discharging current of the battery pack at a sampling frequency of not less than 10Hz. A temperature sensor array is arranged on the battery terminals, the surface of the casing, and inside the module to collect temperature data at multiple points. Strain sensors are attached to the surface of the battery casing to detect casing deformation caused by internal gas generation. A pressure sensor is sealed and installed inside the cavity of the battery module to detect changes in the internal pressure of the module. A gas sensor array is used to detect characteristic gases released before thermal runaway of a battery, the characteristic gases including at least hydrogen, carbon monoxide, and methane.

[0009] Preferably, the feature parameters extracted by the feature extraction and fusion module include at least: The voltage drop rate is obtained by calculating the magnitude of voltage drop in a single cell per unit time. The rate of temperature rise is obtained by calculating the increase in temperature sensor readings per unit time. The strain rate of change is obtained by calculating the rate of change of the shell strain per unit time. The rate of increase in air pressure is obtained by calculating the increase in air pressure inside the module per unit time. The characteristic gas concentration growth rate is obtained by calculating the growth rate of the characteristic gas concentration per unit time.

[0010] Preferably, the formula for calculating the comprehensive hot feature vector generated by the adaptive weighted fusion algorithm is as follows: ; in, Let be the normalized feature vector of the i-th feature parameter. Let n be the adaptive weight corresponding to time t, and n be the total number of feature parameters; the adaptive weight It is dynamically calculated based on the signal-to-noise ratio and historical contribution of each characteristic parameter at the current moment.

[0011] Preferably, the early warning model in the multi-level early warning model library is based on the characteristics of abnormal voltage drop, and its output early risk index The calculation formula is: ; Among them, among them, For the rate of voltage drop, This represents the minimum deviation between the voltage of a single cell and the average voltage within the group. , These are the model coefficients. This is the Sigmoid normalization function.

[0012] Preferably, the formula for calculating the final thermal runaway risk level Rfinal by the thermal runaway risk comprehensive assessment module is as follows: ; in, This is an early-stage risk index. This is a medium-term risk index. This is an emergency risk index. , , The confidence weights for each model are dynamically adjusted based on the historical prediction accuracy of each model under the current operating conditions.

[0013] Preferably, the thermal runaway risk comprehensive assessment module further outputs an estimated remaining safe time based on the changing trend of the comprehensive thermal feature vector and using a pre-set thermal runaway time prediction model. The prediction model was obtained by fitting multiple sets of thermal runaway experimental data.

[0014] Preferably, the hierarchical alarm strategy executed by the hierarchical alarm and linkage control module includes: Level 1 Warning (Attention Level): Risk of thermal runaway When the data falls below the first threshold, only the data is recorded and a "Pay attention" message is displayed on the monitoring interface. Level II Warning (Alert Level): When When the current falls between the first and second thresholds, an audible and visual alarm is triggered, and the BMS charging current limit is automatically adjusted, activating the active cooling system. Level 3 alarm (emergency level): When When the second threshold is exceeded, the high-voltage circuit is immediately cut off, the fire extinguishing device is activated, and an emergency alarm message is sent to the cloud and surrounding vehicles through the vehicle communication module.

[0015] Preferably, the system further includes a self-learning and model update module, which is used to incrementally update the model parameters in the multi-level early warning model library by using complete sensor data, feature vectors, model output results and final conclusions as training samples after each thermal runaway event (including false alarms and real events), so as to improve the system's adaptability to different battery types and operating conditions.

[0016] Preferably, the system further includes a visualization diagnostic module, which is used to display in real time the voltage consistency, temperature distribution heat map, characteristic gas concentration change curve and thermal runaway risk level time axis of each individual battery in a graphical manner, and supports retrospective analysis of historical events.

[0017] By employing the above technical solution, the present invention provides a power battery thermal runaway early warning and detection system. It possesses at least the following beneficial effects: This invention uses a multi-source sensor data acquisition module to collect multi-dimensional physical quantity data such as single-cell battery voltage, charging and discharging current, electrode temperature, shell strain, internal air pressure, and characteristic gas concentration in real time. This enables comprehensive and multi-dimensional perception of the precursors of battery thermal runaway, solving the problems of single sensing dimensions and incomplete early warning information in the prior art, and significantly improving the ability to capture early thermal runaway characteristics. The feature extraction and fusion module extracts multiple feature parameters such as voltage drop rate, temperature rise rate, strain change rate, gas pressure rise rate, and characteristic gas concentration growth rate from multidimensional data. An adaptive weighted fusion algorithm is used to generate a comprehensive thermal feature vector, which realizes the effective integration and dimensionality reduction of multi-source heterogeneous features. This solves the problems of single features being easily disturbed and having poor robustness in the existing technology, and provides high-quality feature input for subsequent accurate evaluation. By storing early warning models based on abnormal voltage drops, mid-term warning models based on gas production rate, and emergency alarm models based on temperature mutation rate in a multi-level early warning model library, the system achieves phased and hierarchical detection of the entire thermal runaway development process. This solves the problem of the lack of a graded early warning mechanism and the inability to distinguish the stages of thermal runaway development in existing technologies. The system can issue attention-level warnings in the early stage of thermal runaway, warning-level warnings in the mid-term, and emergency alarms in emergency situations, thus buying valuable time for users to respond. The comprehensive thermal runaway risk assessment module inputs the comprehensive thermal feature vector into multiple early warning models, calculates the risk index corresponding to each model, and generates the final thermal runaway risk level and remaining safety time estimate based on dynamic decision logic. This solves the problems of model rigidity and poor adaptability in the existing technology, realizes adaptive assessment for different battery types and operating conditions, and significantly reduces the false alarm rate and false alarm rate. The graded alarm and linkage control module executes graded alarms based on the thermal runaway risk level and triggers corresponding linkage control strategies, which solves the problems of imperfect linkage control and inability to automatically execute protection measures in the existing technology. It realizes the leap from passive alarm to active protection, effectively curbs the further development of thermal runaway, and maximizes the protection of personnel and property safety. By using the self-learning and model update module to use the complete data of each thermal runaway event as training samples, the early warning model is incrementally updated, which solves the problem that the existing intraoperative model cannot evolve on its own and its performance degrades after long-term operation. This enables the system to continuously adapt to battery aging, environmental changes and the emergence of new material systems, and achieves intelligent improvement that becomes more accurate with use. The visualization diagnostic module displays the voltage consistency, temperature distribution heatmap, characteristic gas concentration change curve, and thermal runaway risk level time axis of each individual battery cell in a graphical and real-time manner. This solves the problem that the results are not intuitive and are not conducive to the quick judgment of operation and maintenance personnel in the existing technology. It enables operators to grasp the safety status of the battery pack at a glance and supports the retrospective analysis of historical events, providing a channel for accident investigation and system optimization. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of the invention, form part of this application: Figure 1 This is a schematic diagram of the overall structure of a power battery thermal runaway early warning and detection system according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 , Figure 1 This invention provides a power battery thermal runaway early warning and detection system; the invention provides a power battery thermal runaway early warning and detection system, comprising: The multi-source sensor data acquisition module 101 is used to acquire multi-dimensional physical quantity data during battery operation in real time. The multi-dimensional physical quantity data includes at least: single cell voltage, charging and discharging current, electrode temperature, casing strain, internal air pressure, and characteristic gas concentration. It should be noted that the comprehensiveness and real-time nature of the data acquisition by this multi-source sensor data acquisition module directly determine the accuracy and timeliness of the early warning. The multi-dimensional physical quantity data covers four dimensions: electrical, thermal, mechanical, and gaseous. Voltage and current reflect the internal electrochemical reaction state of the battery; temperature is the most direct indicator of thermal runaway; shell strain can capture the micron-level deformation caused by internal gas generation, which is a key physical signal in the early stages of thermal runaway; changes in internal gas pressure can reflect the accumulation of gas inside the sealed battery; characteristic gases (hydrogen, carbon monoxide, methane) are indicative products of electrolyte decomposition or active material reaction release inside the battery. Through the synchronous acquisition of multi-dimensional data, the system can cross-verify the precursors of thermal runaway from different physical mechanisms, effectively avoiding the problem of false alarms or missed alarms from a single sensor.

[0021] The feature extraction and fusion module 102 is used to extract multiple feature parameters that can characterize the precursors of battery thermal runaway from the multidimensional physical quantity data, and after performing spatiotemporal alignment and normalization processing on the feature parameters, generate a comprehensive thermal feature vector through an adaptive weighted fusion algorithm. It should be noted that the original multidimensional sensing data may be asynchronous on the time axis (e.g., gas sensors respond slowly, while voltage sensors respond quickly). Therefore, spatiotemporal alignment is required first to ensure that all feature parameters correspond to the battery state at the same moment. Normalization maps feature parameters with different dimensions (e.g., voltage is in volts, and gas pressure is in Pascals) to a unified 0-1 range, facilitating subsequent fusion calculations. The adaptive weighted fusion algorithm is the core innovation, and its weights... It is not fixed, but dynamically adjusted according to the signal-to-noise ratio of each feature parameter at the current moment. For example, when a vehicle accelerates rapidly, the current noise increases, and the signal-to-noise ratio of the voltage and current features may decrease, while the strain and air pressure features are less affected by vibration and their signal-to-noise ratios are relatively stable. Therefore, the weight of the latter will be automatically increased to ensure that the quality of the fused features is not affected by a single noise source.

[0022] A multi-level early warning model library 103 is used to store multiple pre-trained thermal runaway early warning models, including an early warning model based on abnormal voltage drop, a mid-term early warning model based on gas production rate, and an emergency alarm model based on temperature mutation rate. It's important to note that thermal runaway doesn't happen instantly, but rather as a gradual process from internal short circuits, gas production, and temperature rise to final thermal runaway. The early warning model targets the initial stage of thermal runaway—an internal short circuit causing an abnormal voltage drop. At this point, the temperature may not have risen significantly, but the voltage signal is already abnormal. The mid-stage warning model targets the gas production stage, where the electrolyte decomposes inside the battery, generating a large amount of gas, leading to increased gas pressure and a rise in the concentration of characteristic gases. The emergency alarm model targets the final stage before thermal runaway—a rapid rise in temperature. These three models correspond to different physical characteristics and time windows, forming a complete warning chain from early to emergency, providing users with valuable response time ranging from several minutes to several seconds.

[0023] The thermal runaway risk comprehensive assessment module 104 is used to input the comprehensive thermal feature vector into the multiple early warning models respectively, calculate the risk index corresponding to each model, and generate the final thermal runaway risk level and remaining safe time estimate based on a dynamic decision logic. It should be noted that it inputs the comprehensive thermal feature vector into three early warning models respectively to obtain three risk indices. , , The dynamic decision-making logic does not simply take the maximum or average value, but rather takes the maximum value and multiplies it by a confidence weight, where the confidence weight... , , The model is dynamically adjusted based on its historical prediction accuracy under current operating conditions. For example, in low-temperature environments, the response of gas sensors may slow down, leading to a decrease in the accuracy of the medium-term model, and its confidence weights are adjusted accordingly. It will automatically decrease, thereby reducing the model's influence on the final decision. This dynamic adjustment mechanism enables the system to adapt to different environmental conditions and battery states, significantly improving the robustness of the evaluation.

[0024] The graded alarm and linkage control module 105 is used to execute graded alarms according to the thermal runaway risk level and trigger corresponding linkage control strategies. The linkage control strategies include at least: adjusting the charging and discharging parameters of the battery management system, starting the thermal management system, cutting off the high-voltage circuit, and sending alarm information to the cloud monitoring platform. It should be noted that this module translates risk assessment results into practical action instructions, achieving a closed loop from perception to protection. The tiered alarm strategy is designed following the principle of progressive intervention: in the first-level warning stage, only data is recorded and observation is prompted to avoid unnecessary interference; in the second-level warning stage, BMS parameters are proactively adjusted and the cooling system is activated to attempt to curb the development of thermal runaway; in the third-level alarm stage, the most severe measures are taken—cutting off the high-voltage circuit and activating the fire extinguishing device—to ensure safety to the greatest extent possible. Simultaneously, alarm information is sent to the cloud and surrounding vehicles via the vehicle communication module, enabling information sharing between vehicles and between vehicles and the cloud, providing support for collaborative rescue efforts.

[0025] The multi-source sensor data acquisition module includes: Voltage and current sensors are used to acquire the terminal voltage of each individual cell and the charging and discharging current of the battery pack at a sampling frequency of not less than 10Hz. A temperature sensor array is arranged on the battery terminals, the surface of the casing, and inside the module to collect temperature data at multiple points. Strain sensors are attached to the surface of the battery casing to detect casing deformation caused by internal gas generation. A pressure sensor is sealed and installed inside the cavity of the battery module to detect changes in the internal pressure of the module. A gas sensor array for detecting characteristic gases released before thermal runaway of a battery, the characteristic gases including at least hydrogen, carbon monoxide, and methane; It is important to note that the selection and deployment of each sensor are crucial. Voltage and current sensors must have a sampling frequency of at least 10Hz to capture the rapid voltage drop; temperature sensor arrays need to be placed in multiple locations on the battery terminals, casing surface, and inside the module, as the initial hotspot of thermal runaway may appear in different locations; strain sensors need to be attached to the battery casing surface and have micron-level resolution to detect the minute deformations caused by early gas production; pressure sensors need to be sealed and installed in the internal cavity of the module to accurately sense changes in internal pressure; and gas sensor arrays need to have high sensitivity and fast response characteristics for hydrogen, carbon monoxide, and methane, as these gases are characteristic markers of battery thermal runaway.

[0026] The feature parameters extracted by the feature extraction and fusion module include at least the following: The voltage drop rate is obtained by calculating the magnitude of voltage drop in a single cell per unit time. The rate of temperature rise is obtained by calculating the increase in temperature sensor readings per unit time. The strain rate of change is obtained by calculating the rate of change of the shell strain per unit time. The rate of increase in air pressure is obtained by calculating the increase in air pressure inside the module per unit time. The characteristic gas concentration growth rate is obtained by calculating the growth rate of the characteristic gas concentration per unit time. It's important to note that all five characteristic parameters are dynamic features in the form of rates of change, not static values. This is because thermal runaway is a rapidly developing process, and its precursors are manifested by drastic changes in various physical quantities, rather than reaching a certain absolute value. For example, a battery may have a high absolute temperature in a high-temperature environment, but as long as the rate of temperature rise is stable, it may not necessarily be a precursor to thermal runaway; conversely, if the temperature rises rapidly in a short period of time, even if the absolute value is not high, it may still be a dangerous signal. Therefore, using the rate of change as a characteristic parameter allows for more sensitive capture of the dynamic characteristics of thermal runaway while eliminating the influence of static environmental factors.

[0027] The formula for calculating the comprehensive hot feature vector generated by the adaptive weighted fusion algorithm is as follows: ; in, Let be the normalized feature vector of the i-th feature parameter. Let n be the adaptive weight corresponding to time t, and n be the total number of feature parameters; the adaptive weight It is dynamically calculated based on the signal-to-noise ratio and historical contribution of each feature parameter at the current moment; It should be noted that the formula effectively integrates multi-source heterogeneous features. Among them, It is the feature vector after normalization of the i-th feature parameter, and its value ranges from 0 to 1; The weights are adaptive weights corresponding to time t, with the sum of all weights being 1. The dynamic calculation of the weights is based on the signal-to-noise ratio (SNR) of each feature parameter at the current time: the system maintains a sliding window for each feature parameter, calculating the ratio of the average power of the signal to the noise power within the window; the higher the SNR, the greater the weight. Simultaneously, historical contribution is introduced as a correction factor; that is, feature parameters with higher prediction accuracy over a past period receive a certain weight gain. This dual adjustment mechanism ensures the stability and accuracy of the fused features under complex conditions.

[0028] The early warning models in the multi-level early warning model library are based on the characteristics of abnormal voltage drops, and their output is an early risk index. The calculation formula is: ; Among them, among them, For the rate of voltage drop, This represents the minimum deviation between the voltage of a single cell and the average voltage within the group. , These are the model coefficients. It is the Sigmoid normalization function; It should be noted that, among them, It is the rate of voltage drop, reflecting the rapid voltage drop caused by an internal short circuit in the battery; This represents the minimum deviation between the voltage of a single cell and the average voltage within the group, reflecting the consistency of the individual cells—when an internal short circuit occurs in a single cell, its voltage will be significantly lower than that of other cells. These two characteristics are expressed through coefficients. and Perform a linear combination, then pass it through the Sigmoid function. Mapped to probability values ​​between 0 and 1, serving as an early risk index. Coefficient. and This is obtained through training with a large amount of normal and abnormal data. For example, if the training data shows that the voltage drop rate is more important for early warning, then... It will be greater than .

[0029] The formula used by the comprehensive thermal runaway risk assessment module to calculate the final thermal runaway risk level Rfinal is as follows: ; in, This is an early-stage risk index. This is a medium-term risk index. This is an emergency risk index. , , The confidence weights for each model are dynamically adjusted based on the historical prediction accuracy of each model under the current operating conditions. It should be noted that the formula uses a strategy of taking the maximum value to determine the final risk level, reflecting the idea of ​​the "barrel effect"—the overall safety level of a system is determined by the most severe risk. Specifically, , , The confidence weights for each model are initially set to 1 and dynamically adjusted based on the historical prediction accuracy of each model under current operating conditions. For example, if the early warning model has a high prediction accuracy in the recent period, then... It will increase; if the intermediate-term model performs poorly in low-temperature environments, then It will be appropriately reduced. This dynamic adjustment mechanism enables the system to adapt to different operating conditions, avoiding overall misjudgment due to the failure of a certain model under specific conditions.

[0030] The thermal runaway risk comprehensive assessment module also outputs an estimate of the remaining safe time based on the changing trend of the comprehensive thermal feature vector and using a pre-set thermal runaway time prediction model. The prediction model was obtained by fitting multiple sets of thermal runaway experimental data; It is worth noting that the remaining safe time The output is a significant innovation of this invention, providing clear countdown guidance for emergency response. Based on extensive experimental data on thermal runaway, the predictive model establishes a relationship between the time interval from the "first anomaly in characteristic parameters" to the "trigger of thermal runaway" and the current trend of characteristic parameter changes. The model can take the form of multiple linear regression or an exponential decay model, and the system updates once per second. The countdown is displayed on the interface. For example, when an abnormal voltage drop is detected, the prediction model outputs "Estimated remaining safe time: 180 seconds," and adjusts it in real time based on changes in characteristic parameters. If the time is less than 30 seconds, the system will automatically trigger a level 3 alarm, providing a final time window for personnel evacuation and emergency response.

[0031] The hierarchical alarm strategy executed by the hierarchical alarm and linkage control module includes: Level 1 Warning (Attention Level): Risk of thermal runaway When the data falls below the first threshold, only the data is recorded and a "Pay attention" message is displayed on the monitoring interface. Level II Warning (Alert Level): When When the current falls between the first and second thresholds, an audible and visual alarm is triggered, and the BMS charging current limit is automatically adjusted, activating the active cooling system. Level 3 alarm (emergency level): When When the second threshold is exceeded, the high-voltage circuit is immediately cut off, the fire extinguishing device is activated, and an emergency alarm message is sent to the cloud and surrounding vehicles through the vehicle communication module. It is worth noting that the tiered alarm strategy embodies the principles of precise response and minimal intervention. Level 1 alerts (attention level) only record data in the background and display a notification on the monitoring interface, without generating any disruptive alarms; suitable for low-risk but noteworthy situations. Level 2 alerts (warning level) issue audible and visual alarms and automatically take intervention measures (adjusting charging current limits, initiating cooling); suitable for medium-risk situations requiring proactive intervention. Level 3 alerts (emergency level) take the most severe measures (cutting off high voltage, initiating fire suppression, sending an alarm); suitable for extremely high-risk situations endangering life. This tiered mechanism avoids the "boy who cried wolf" effect caused by frequent alarms while ensuring decisive action can be taken in truly critical moments.

[0032] The system also includes a self-learning and model update module, which is used to incrementally update the model parameters in the multi-level early warning model library by using complete sensor data, feature vectors, model output results and final conclusions as training samples after each thermal runaway event (including false alarms and real events), so as to improve the system's adaptability to different battery types and operating conditions. It is worth noting that the self-learning and model update modules endow the system with the ability to continuously evolve. After each event, the system stores the complete "data-feature-decision-conclusion" chain as a training sample in the local database. When a certain number (e.g., 100) are accumulated, the system automatically starts incremental learning to update the coefficients of the early warning model. , Decision boundary and confidence weights of the mid-term SVM model , , For example, if multiple false alarms occur in low-temperature environments, the system will automatically reduce the confidence weight of the mid-term model in low-temperature environments. This reduces false alarms. This closed-loop learning mechanism allows the system to become increasingly accurate in matching the real characteristics of the protected battery as it runs, achieving personalized adaptation that becomes more accurate with use.

[0033] The system also includes a visualization diagnostic module, which is used to display in real time the voltage consistency, temperature distribution heat map, characteristic gas concentration change curve and thermal runaway risk level time axis of each individual battery in a graphical manner, and supports retrospective analysis of historical events. It's worth noting that the visualization diagnostic module serves as the "window" between the system and the user, directly impacting system usability and user experience. This module displays key information intuitively on the vehicle's central control screen or cloud management platform, using dashboards, heatmaps, and graphs. For example, the voltage consistency curve uses different colored lines to represent the voltage of each individual battery cell; if the voltage of a cell experiences an abnormal drop, its line immediately turns red and flashes. The temperature distribution heatmap uses a red-yellow-blue gradient to represent the temperature distribution in different areas of the module, making hotspots readily apparent. The characteristic gas concentration curve displays real-time changes in the concentration of gases such as hydrogen and carbon monoxide. The risk level timeline uses green-yellow-red three-segment color bands to represent historical risk changes. Users can click on any point in time to view detailed sensor data at that time, supporting retrospective playback analysis of historical events, providing a powerful tool for accident investigation and system optimization.

[0034] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A power battery thermal runaway early warning and detection system, characterized in that, The system includes: The multi-source sensor data acquisition module is used to acquire multi-dimensional physical quantity data during battery operation in real time. The multi-dimensional physical quantity data includes at least: single cell voltage, charging and discharging current, electrode temperature, casing strain, internal air pressure, and characteristic gas concentration. The feature extraction and fusion module is used to extract multiple feature parameters that can characterize the precursors of battery thermal runaway from the multidimensional physical quantity data, and after performing spatiotemporal alignment and normalization on the feature parameters, generate a comprehensive thermal feature vector through an adaptive weighted fusion algorithm. A multi-level early warning model library is used to store multiple pre-trained thermal runaway early warning models, including an early warning model based on abnormal voltage drop, a mid-term early warning model based on gas production rate, and an emergency alarm model based on temperature mutation rate. The thermal runaway risk comprehensive assessment module is used to input the comprehensive thermal feature vector into the multiple early warning models respectively, calculate the risk index corresponding to each model, and generate the final thermal runaway risk level and remaining safe time estimate based on a dynamic decision logic. The graded alarm and linkage control module is used to execute graded alarms according to the thermal runaway risk level and trigger corresponding linkage control strategies. The linkage control strategies include at least: adjusting the charging and discharging parameters of the battery management system, starting the thermal management system, cutting off the high-voltage circuit, and sending alarm information to the cloud monitoring platform.

2. The power battery thermal runaway early warning and detection system according to claim 1, characterized in that, The multi-source sensor data acquisition module includes: Voltage and current sensors are used to acquire the terminal voltage of each individual cell and the charging and discharging current of the battery pack at a sampling frequency of not less than 10Hz. A temperature sensor array is arranged on the battery terminals, the surface of the casing, and inside the module to collect temperature data at multiple points. Strain sensors are attached to the surface of the battery casing to detect casing deformation caused by internal gas generation. A pressure sensor is sealed and installed inside the cavity of the battery module to detect changes in the internal pressure of the module. A gas sensor array is used to detect characteristic gases released before thermal runaway of a battery, the characteristic gases including at least hydrogen, carbon monoxide, and methane.

3. The power battery thermal runaway early warning and detection system according to claim 1, characterized in that, The feature parameters extracted by the feature extraction and fusion module include at least the following: The voltage drop rate is obtained by calculating the magnitude of voltage drop in a single cell per unit time. The rate of temperature rise is obtained by calculating the increase in temperature sensor readings per unit time. The strain rate of change is obtained by calculating the rate of change of the shell strain per unit time. The rate of increase in air pressure is obtained by calculating the increase in air pressure inside the module per unit time. The characteristic gas concentration growth rate is obtained by calculating the growth rate of the characteristic gas concentration per unit time.

4. The power battery thermal runaway early warning and detection system according to claim 1, characterized in that, The formula for calculating the comprehensive hot feature vector generated by the adaptive weighted fusion algorithm is as follows: ; in, Let be the normalized feature vector of the i-th feature parameter. Let n be the adaptive weight corresponding to time t, and n be the total number of feature parameters; the adaptive weight It is dynamically calculated based on the signal-to-noise ratio and historical contribution of each characteristic parameter at the current moment.

5. The power battery thermal runaway early warning and detection system according to claim 1, characterized in that, The early warning models in the multi-level early warning model library are based on the characteristics of abnormal voltage drops, and their output is an early risk index. The calculation formula is: ; Among them, among them, For the rate of voltage drop, This represents the minimum deviation between the voltage of a single cell and the average voltage within the group. , These are the model coefficients. This is the Sigmoid normalization function.

6. The power battery thermal runaway early warning and detection system according to claim 1, characterized in that, The formula used by the comprehensive thermal runaway risk assessment module to calculate the final thermal runaway risk level Rfinal is as follows: ; in, This is an early-stage risk index. This is a medium-term risk index. This is an emergency risk index. , , The confidence weights for each model are dynamically adjusted based on the historical prediction accuracy of each model under the current operating conditions.

7. The power battery thermal runaway early warning and detection system according to claim 1, characterized in that, The thermal runaway risk comprehensive assessment module also outputs an estimate of the remaining safe time based on the changing trend of the comprehensive thermal feature vector and using a pre-set thermal runaway time prediction model. The prediction model was obtained by fitting multiple sets of thermal runaway experimental data.

8. The power battery thermal runaway early warning and detection system according to claim 1, characterized in that, The hierarchical alarm strategy executed by the hierarchical alarm and linkage control module includes: Level 1 Warning (Attention Level): Risk of thermal runaway When the data falls below the first threshold, only the data is recorded and a "Pay attention" message is displayed on the monitoring interface. Level II Warning (Alert Level): When When the current falls between the first and second thresholds, an audible and visual alarm is triggered, and the BMS charging current limit is automatically adjusted, activating the active cooling system. Level 3 alarm (emergency level): When When the second threshold is exceeded, the high-voltage circuit is immediately cut off, the fire extinguishing device is activated, and an emergency alarm message is sent to the cloud and surrounding vehicles through the vehicle communication module.

9. The power battery thermal runaway early warning and detection system according to claim 1, characterized in that, The system also includes a self-learning and model update module, which, after each thermal runaway event (including false alarms and real events), uses complete sensor data, feature vectors, model output results and final conclusions as training samples to incrementally update the model parameters in the multi-level early warning model library, so as to improve the system's adaptability to different battery types and operating conditions.

10. A power battery thermal runaway early warning and detection system according to claim 1, characterized in that, The system also includes a visualization diagnostic module, which is used to display in real time the voltage consistency, temperature distribution heat map, characteristic gas concentration change curve and thermal runaway risk level time axis of each individual cell in a graphical manner, and supports retrospective analysis of historical events.