Lithium ion battery thermal runaway multi-level early warning system and method and storage medium

Abstract

The application discloses a lithium ion battery thermal runaway multi-level early warning system and method and a storage medium, and belongs to the field of battery safety monitoring. Apriori The system comprises a data acquisition device, a feature extraction module, a warning strategy module and a warning model module. The method comprises the following steps: collecting thermal, electrical, mechanical and gas multi-physical field parameters; screening key feature points by using mutual information method; mining the association rules between feature points and warning levels by using an algorithm to generate a four-level warning strategy; and constructing a warning model integrating level evaluation and residual time prediction. The application overcomes the defects of the prior art, such as dependence on a single parameter and lack of grading capability, by multi-parameter coupling analysis and intelligent algorithm mining, realizes early, accurate and graded warning and residual time prediction of lithium ion battery thermal runaway, and significantly improves the safety warning accuracy and emergency disposal capability of the battery system.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery safety monitoring technology, specifically to a multi-level early warning system, method, and storage medium for lithium-ion battery thermal runaway. Background Technology

[0002] Lithium-ion batteries, due to their high energy density and long cycle life, have become core components in electrochemical energy storage and electric vehicles. However, under conditions such as overcharging, overheating, and mechanical abuse, they are prone to internal chain exothermic reactions, leading to thermal runaway, which can then cause serious safety accidents such as fires and explosions, posing a significant threat to life and property. Therefore, developing efficient and reliable early warning technologies for thermal runaway is crucial to ensuring their safe large-scale application.

[0003] Currently, in engineering practice and existing technologies, battery thermal runaway early warning systems mostly rely on threshold judgments of single parameters such as voltage and temperature, or a few parameters, resulting in relatively simple warning logic. This approach struggles to comprehensively capture the complex evolution and coupling relationships of multiple physical field parameters in the early stages of thermal runaway, easily leading to untimely warnings or high rates of false alarms and missed alarms. Furthermore, the correlation between the acquisition, feature extraction, and early warning alarms of different parameters is weak, lacking methods for deep fusion and intelligent mining of multi-dimensional information from thermal, electrical, mechanical, and gas dimensions.

[0004] In addition, existing early warning schemes generally lack a scientific and quantitative effect evaluation system, making it difficult to comprehensively assess the timeliness, accuracy, system reliability, and rationality of the early warning characteristics. This results in a poor match between the early warning level and the actual development stage of thermal runaway, and fails to provide effective support for refined graded response.

[0005] In terms of technical implementation, existing methods mostly use a single modeling tool, which makes it difficult to simultaneously address the efficient construction of association rule mining of massive historical data and real-time dynamic prediction models. The actual effectiveness, prediction accuracy, and adaptability of the models under complex working conditions all need to be verified and optimized through rigorous and precise experiments.

[0006] In summary, developing a systematic technical solution that can integrate multi-dimensional information, possess intelligent feature extraction and evaluation capabilities, and ultimately achieve early accurate warning and hierarchical control has become a technical bottleneck that urgently needs to be overcome in this field. Summary of the Invention

[0007] This invention aims to overcome the problems existing in early warning of thermal runaway in lithium-ion batteries, such as single parameters, poor correlation, lack of hierarchical strategies and effective evaluation systems, and insufficient model accuracy and practicality.

[0008] To address the aforementioned problems, in a first aspect, the present invention provides a multi-level early warning system for thermal runaway in lithium-ion batteries, comprising:

[0009] The data acquisition device is used to collect multidimensional characteristic parameters of multiple physical fields, including heat, electricity, force, and gas, during the thermal runaway process of lithium-ion batteries.

[0010] The feature extraction module is communicatively connected to the data acquisition device and is used to receive the multidimensional feature parameters and use the mutual information method to select several key feature points with the highest correlation to the thermal runaway state from the feature parameters of the multiple physical fields.

[0011] The early warning strategy module is communicatively connected to the feature extraction module and is used to mine the association relationship between the key feature points and different early warning levels based on the association rule mining algorithm to generate a multi-level early warning strategy.

[0012] The early warning model module is communicatively connected to the early warning strategy module and is used to construct an early warning model based on the multi-level early warning strategy. The early warning model includes at least an early warning level assessment unit and a thermal runaway remaining time prediction unit.

[0013] The warning level assessment unit is used to determine and output the current warning level based on real-time data, and the thermal runaway remaining time prediction unit is used to output the predicted thermal runaway remaining time.

[0014] Preferably, the data acquisition device is specifically used to acquire parameters such as voltage, temperature, expansion force, and hydrogen concentration.

[0015] Preferably, the key feature points include at least one selected from the group consisting of voltage plateau period, expansion force inflection point, expansion force change rate, safety valve temperature, hydrogen concentration, battery surface temperature, voltage change rate, and battery back side temperature rise rate.

[0016] Preferably, the system further includes an early warning evaluation module; the early warning evaluation module is used to evaluate the early warning effect of the key feature points and calculate a comprehensive score based on an evaluation system including response speed, response accuracy, system reliability and hierarchical rationality using the entropy weight method.

[0017] Preferably, the association rule mining algorithm is as follows: Apriori algorithm;

[0018] The multi-level early warning strategy is a four-level early warning strategy including Level 1 to Level 4;

[0019] The Level 1 warning is associated with the occurrence of voltage plateau and expansion force inflection point, while the Level 4 warning is associated with anomalies in voltage change rate, battery surface temperature, and battery back temperature rise rate.

[0020] Secondly, the present invention provides a multi-level early warning method for thermal runaway of lithium-ion batteries using the system described in the first aspect above, comprising the following steps:

[0021] S1. Multidimensional characteristic parameters of multiple physical fields, including heat, electricity, force, and gas, are collected through multiple sensors during the thermal runaway process of lithium-ion batteries;

[0022] S2. The mutual information method is used to screen the feature points of the multidimensional feature parameters, so as to select several key feature points with the highest correlation with the thermal runaway state from the feature parameters of the multiple physical fields.

[0023] S3. Based on the association rule mining algorithm, mine the association relationship between the key feature points and different warning levels to generate a multi-level warning strategy;

[0024] S4. Based on the aforementioned multi-level early warning strategy, construct an early warning model to output the early warning level and the predicted remaining time of thermal runaway based on real-time data.

[0025] Preferably, in step S2, the feature point selection using the mutual information method includes the following sub-steps:

[0026] S21. Determine the effective time window for each feature point;

[0027] S22. The feature point data is divided into intervals using the equal-frequency binning principle;

[0028] S23. Calculate the joint probability distribution and mutual information value of each feature point and the thermal runaway state, and select the feature points with the highest mutual information value as key feature points.

[0029] Preferably, in step S3, the multi-level early warning strategy is a four-level early warning strategy; the method further includes the following after step S3:

[0030] Based on a pre-defined evaluation system, the entropy weight method is used to evaluate the early warning effect of the key feature points and calculate the comprehensive score of each feature point.

[0031] Preferably, after step S4, the method further includes:

[0032] S5. Based on the warning level output by the warning model, implement corresponding emergency response measures;

[0033] The emergency response measures include one or more of the following: stopping charging, injecting nitrogen, ventilating and exhausting, and injecting liquid carbon dioxide.

[0034] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the second aspect.

[0035] Compared with existing technologies, this invention, through multi-physics field coupling monitoring of heat, electricity, force, and gas, comprehensively and systematically captures the characteristic evolution of different stages of the entire process of thermal runaway incubation for the first time, laying the foundation for early identification. Furthermore, by using the mutual information method to intelligently screen out the key feature points with the highest correlation to thermal runaway from the multi-physics field parameters, and introducing an early warning effect evaluation system based on the entropy weight method to quantitatively evaluate them, the accuracy of feature extraction and the reliability of system decision-making are significantly improved.

[0036] Based on this, utilize Apriori The algorithm mines strong correlation rules between feature points and warning levels, thereby constructing a four-level warning strategy, realizing the leap from coarse alarm to refined hierarchical warning;

[0037] Ultimately, the hierarchical early warning model that integrates the above strategies can not only output accurate early warning levels, but also uniquely provide the function of predicting the remaining time of thermal runaway. This wins a valuable time window and provides key decision-making basis for implementing differentiated graded emergency response measures such as stopping charging, injecting inert gas, ventilation and cooling, and powerful cooling, thus realizing the transformation from passive alarm to proactive and precise prevention and control.

[0038] The entire solution was fully validated through multi-tool collaborative modeling and on a full-scale experimental platform, demonstrating its high early warning accuracy (remaining time prediction error less than 18s) and sufficient safety margin. It has strong engineering practicality and promotion value, and ultimately comprehensively improves the safety early warning and protection capabilities of lithium-ion battery systems. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0040] Figure 1 This is a schematic diagram of the battery simulation experiment structure in an embodiment of the present invention;

[0041] Figure 2 This is an exploded view of the battery simulation experiment structure in an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of the thermocouple layout on the battery surface in an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of the gas collection and detection structure layout during the battery simulation experiment in this embodiment of the invention;

[0044] Figure 5 This is a schematic diagram of the multi-layer early warning model architecture constructed in an embodiment of the present invention;

[0045] Figure 6Based on the embodiments of the present invention SPSS A network diagram showing the correlation between feature points and early warning levels established by the software;

[0046] Figure 7 Used in the embodiments of the present invention MATLAB / Simulink This enables the modular design process of a hierarchical early warning model.

[0047] Figure 8 This is a flowchart illustrating the implementation of the early warning logic for the early warning strategy in an embodiment of the present invention;

[0048] Figure 9 This is a data graph showing the safety margin of the 1C overcharge abuse verification and early warning strategy and early warning model used in this embodiment of the invention;

[0049] In the diagram: 1. Battery; 2. Heating plate; 3. Insulation cotton; 4. Screw; 5. Metal clamp; 6. Thermocouple; 7. Pressure sensor. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0051] Example 1:

[0052] This embodiment provides a multi-level early warning system for thermal runaway in lithium-ion batteries. To more clearly illustrate the data acquisition device, [the following is used]... Figures 1 to 4 The diagram shows a self-built full-size energy storage battery compartment experimental platform used to simulate the data collection process of a data acquisition device and verify the effectiveness of the entire early warning system. The platform includes a square battery 1 as the test object, with a heating plate 2 powered by 220V AC attached to the side of the battery 1 to simulate overheating and abuse conditions. The battery 1 and heating plate 2 are wrapped in insulating cotton 3 to reduce heat loss and ensure concentrated heat injection. A clamp consisting of metal plates 5 and screws 4 clamps the battery 1 from both sides to apply constant pressure and install sensing equipment. A flat diaphragm pressure sensor 7 is clamped and fixed between the two metal plates 5, with its sensing surface facing the center of the side of the battery 1, used to monitor changes in expansion force during thermal runaway.

[0053] Please see Figure 3 In terms of temperature monitoring, multiple K-type armored thermocouples 6 are used as temperature sensors and are respectively attached to the center surface of the front of the battery 1 to measure the surface temperature of the battery. Ts The center of the back is used to measure the temperature of the back of the battery. TB and the outlet of the safety valve used to measure the temperature of the safety valve. Tsv .

[0054] A voltage sensor has two probes connected to the positive and negative terminals of battery 1, respectively, to monitor its terminal voltage. V And its changes.

[0055] Please see Figure 4 In terms of gas monitoring, an all-in-one explosion-proof gas detector is installed inside the battery compartment. Its sampling port is connected to the vicinity of the safety valve of battery 1 via a gas collection pipeline to monitor the hydrogen concentration. H2 A vacuum pump operates continuously, drawing gas from the cabin into a gas detector for analysis.

[0056] All these sensors are connected via signal lines to a multi-channel data signal converter, specifically a 7018 module, which is responsible for converting the analog signals acquired by the sensors into digital signals. A computer acts as the host computer, through... RS485 The communication protocol sends polling commands to the 7018 module, receives and stores all converted digital signals, thus completing the data collection function of the data acquisition device.

[0057] The collected multidimensional feature parameters are transmitted to the feature extraction module. The core algorithm of this module is the mutual information method, which aims to filter out the key feature points with the highest correlation to the final thermal runaway state from massive amounts of data on multiple physical fields, including heat, electricity, force, and gas. The algorithm's calculation is based on the basic formula of mutual information:

[0058] ;

[0059] in X and Y They represent two random variables, and These are their marginal probability distributions. It is their joint probability distribution.

[0060] This formula measures whether the variables are known. X After setting the value, how much can be reduced regarding the variable? Y The uncertainty. In this embodiment, X Represents a certain feature parameter. Y This represents a state of thermal runaway. MI The higher the value, the stronger the correlation between this feature and thermal runaway.

[0061] Specifically, the first step is to determine the effective time window for each feature point. The calculation method involves taking the average value of the sequence of times the feature point appears in multiple experiments. and standard deviation Valid window Typically defined in an interval Inside; among them, n For the number of experiments, The time difference between the appearance of the feature point and the occurrence of thermal runaway in each experiment. For the first i The time when the feature point appears in this experiment.

[0062] Next, the data of feature points is divided into intervals using the equal-frequency binning principle, and the number of bins is determined. k Typically, the data sample size is taken. n The square root, that is .

[0063] Finally, the joint probability distribution of each feature point and the thermal runaway state label is calculated and substituted into the mutual information formula for further calculation. Through this process, eight key feature points with the highest mutual information values ​​are selected from 23 candidate parameters. These feature points comprehensively cover the four physical fields of heat, electricity, force, and gas, specifically including: voltage plateau period. Vpp Inflection point of expansion force Ftp Abnormal rate of change of expansion force Frate-abn Abnormal temperature of safety valve Tsv-abn Hydrogen concentration appeared H2-appear Abnormal battery surface temperature Ts-abn Abnormal voltage change rate Vrate-abn And abnormal temperature rise rate on the back of the battery TB-rate-abn .

[0064] To scientifically evaluate the early warning effectiveness of the aforementioned feature points, the system also integrates an early warning evaluation module. This module operates based on a preset evaluation system, which includes four core indicators:

[0065] Response speed A 1, defined as the average time difference between the appearance of a feature point and the occurrence of thermal runaway, is calculated using the following formula: ,in The moment when thermal runaway occurs. t The time when the feature point appears;

[0066] Response accuracy A 2, defined as the absolute value of the error in predicting the timing of thermal runaway at the time the feature point appears, is calculated using the following formula: ,in t pre This is the predicted time of thermal runaway.

[0067] Early warning system reliability A3. This is a comprehensive indicator calculated by weighting sensor failure rate, lifespan, and environmental tolerance. The calculation formula is as follows: A3 =0.5 × failure rate + 0.25 × lifespan + 0.25 × environmental tolerance;

[0068] and the rationality of the classification A 4. Used to evaluate the stability of the time interval between feature points, its specific calculation formula is as follows: .in, This formula represents the time intervals at which different characteristic points appear within the same warning level, or the time intervals at which the same characteristic point appears in multiple experiments. It aims to calculate the degree of dispersion of these time points.

[0069] This module uses the entropy weight method to calculate the weights of each indicator in the comprehensive evaluation. The process begins by normalizing the calculation results of each indicator, mapping them to a range of 0 to 1. The normalization formula is as follows:

[0070] ;in Representing the j The first feature point i Individual indicator values, Representing the j The first feature point i The maximum value of each indicator. Representing the j The first feature point i The minimum value of each indicator.

[0071] Then calculate the entropy value and information entropy redundancy of each indicator;

[0072] Finally, the weights of each indicator are determined based on the entropy value, and a weighted sum is calculated to obtain the comprehensive score for each feature point. The scoring criteria are divided into "good" (0.8 ≤ Score ≤1), "Good (0.7≤)" Score <0.8), "General (0.6≤ Score <0.7), "Poor ( Score The evaluation system uses four levels: <0.6). This system allows for a quantitative comparison of the quality of different feature points and their appropriate allocation to different warning levels, thus preventing a warning level from becoming ineffective due to poor feature point performance.

[0073] Based on the results of feature extraction and evaluation, the early warning strategy module begins operation. This module can... SPSSModeler Tool call AprioriThe algorithm mines strong correlation rules between key feature points and warning levels. Optionally, the minimum support is set to 0.2 and the minimum confidence to 0.8. Through mining, a four-level warning strategy coupled with thermal, electrical, mechanical, and gaseous parameters is generated. Specifically:

[0074] Level 1 is the earliest warning level, corresponding to the early stage of thermal runaway abuse. Its rules are based on the voltage plateau period. Vpp The main feature point is a duration of 20 seconds or more, and the first occurrence of the expansion force inflection point is the key feature point. Ftp As an auxiliary feature point, at this time the layered positive electrode material NaNiO2 inside the battery begins to decompose, and the separator is damaged;

[0075] Level 2 mid-term warning, corresponding to the mid-term abuse, is based on the first expansion force change rate. Frate >0.5 kg / s The characteristic point is that gas is generated inside the battery every second and the duration of this state is ≥5 seconds. At this time, the battery bulges and deforms due to the production of gas inside the battery.

[0076] Level 3 mid-to-late stage warning corresponds to the mid-to-late stage of abuse, and its rules are based on the safety valve temperature. Tsv >75℃ and hydrogen concentration H2 A characteristic point is when the pressure is >100ppm and both states last for ≥5s. At this point, the internal pressure of the battery is too high, causing the safety valve to release pressure.

[0077] Level 4 is the most urgent warning, corresponding to the later stages of abuse. Its rules are based on the rate of voltage change. Vrate Greater than 0.1 V / s Furthermore, the duration is greater than or equal to 3 seconds as the main characteristic point, and the battery surface temperature is the primary indicator. Ts Temperatures greater than 110°C and the rate of temperature rise on the back of the battery TB- rate A value greater than 1℃ / s and a duration of ≥3s for both are auxiliary characteristic points. At this point, the internal physicochemical reaction of the battery accelerates, and the separator is about to collapse.

[0078] These rules together form a Boolean logic expression, as shown below, which clearly defines the triggering conditions for each warning level:

[0079] .

[0080] Finally, the early warning model module constructed a hierarchical early warning model based on the above four-level early warning strategy. This model adopts... Matlab / Simulink The tool is implemented, and its architecture is as follows: Figure 5 As shown. The model can be divided into four layers from bottom to top:

[0081] The data processing layer is responsible for filtering and noise reduction of the acquired raw signals;

[0082] The feature point extraction layer uses a specific algorithm to extract key feature points in real time, for example:

[0083] Voltage plateau period Vpp Identification is performed using the wavelet transform modulus maxima method. MATLAB middle wavedec The function performs wavelet decomposition on the voltage signal, extracting plateau features with a duration > 20s and a plateau threshold ≤ 0.05V from the detail coefficients. It outputs the signal indicating successful extraction of this feature point, containing both numerical and temporal information. Its mathematical expression is as follows:

[0084] ;

[0085] In the formula, This is the instantaneous voltage value at the current moment. For sliding windows (window length) Δt The average voltage within the range, in volts. This is the voltage plateau fluctuation threshold, reflecting the characteristic of a gradual voltage drop in the early stages of thermal runaway. In this embodiment, it is set to 0.05V. Δt For platform duration. (Usage) DetectChange When the module detects that the condition is met, it outputs a pulse (indicating...). Vpp (Feature points appear).

[0086] Expansion force characteristic points Ftp(first) pass Derivative The module calculates the first / second derivatives of the expansion force data curve. LogicOperator The module determines the first peak condition of the first derivative and outputs the Boolean operation result (the first occurrence of the first derivative passing through 0 and the second derivative being negative).

[0087] ;

[0088] Characteristic points of expansion force change rate Frate pass Derivative The module calculates the first derivative of the expansion force data curve, and then... MovingAverage The module calculates the mean of the sliding window (sampling for 5 seconds within the window) using... RateLimiter Module determines the rate of change threshold Frate >0.5 kg / s (≥5s) Output the signal indicating successful extraction of the feature point, including numerical and time information.

[0089] H2-appear Feature points through GasSensorInterface collection H2 Data, and by Comparator The module determines the threshold when H2 When the value is >100ppm (≥5s), output a signal indicating successful extraction of the feature point, including numerical and time information.

[0090] Safety valve temperature characteristic points Tsv Temperature data access KalmanFilter Module, then by RelationalOperator determination Tsv If the temperature exceeds 75℃, output the signal indicating successful feature point extraction, including numerical and time information.

[0091] Voltage change rate Vrate Feature points are calculated using the second-order central difference method, combined with MovingAverage Sliding window mean (3s sampling within the window) filtering processing. Voltage signal is then... Derivative The module calculates the rate of change, through Comparator Module determination threshold Vrate If the value is greater than 0.1V / s (≥3s), output a signal indicating successful extraction of the feature point, including numerical and time information.

[0092] Side temperature feature points Ts Temperature data access KalmanFilter Module, also composed of RelationalOperator The module determines the threshold when Ts When the temperature is >110℃, output a signal indicating successful extraction of the feature point, including numerical and time information.

[0093] Backside temperature change rate TB-rate Feature points are used Derivative The module calculates the rate of temperature change, through Comparator Module determination threshold TB-rate If the value is greater than 1℃ / s (≥3s), output a signal indicating successful extraction of the feature point, including numerical and time information.

[0094] The early warning level assessment layer is the core of the model; it is based on... Stateflow The state machine implementation encodes the above Boolean logic expression into state transition logic, such as... Figures 6 to 8 As shown, the system determines the currently met conditions in real time and outputs the corresponding warning level; the thermal runaway remaining time prediction layer, based on the time mapping relationship between different warning levels and the occurrence of thermal runaway obtained by fitting a large amount of experimental data, then... LookupTable The lookup table module enables dynamic estimation of the remaining time of thermal runaway under the current state. The entire model works collaboratively, ultimately outputting both the warning level and the predicted remaining time through the output layer.

[0095] It adopts Stateflow The state machine implementation strictly follows the Boolean expressions of the aforementioned four-level early warning strategy. This state machine contains five states. Normal Levels 1, 2, 3, and 4 are used to implement tiered early warning systems and output signals, as detailed below:

[0096] The initial state is Normal No warning, output 0;

[0097] The system obtains the state values ​​of all feature points from the feature point extraction layer at a cycle of one second. The state transition condition is as follows:

[0098] If the current state level is less than 1 and satisfies Vpp True or Ftp If true, then enter Level 1 state and output 1;

[0099] If the current state level is less than 2 and Frate-abn If true, then enter Level 2 state and output 2;

[0100] If the current state level is less than 3 and simultaneously meets the following conditions: Tsv-abn For true and H2-appear If true, then enter Level 3 state and output 3;

[0101] If the current state level is less than 4 and satisfies Vrate-abn True or simultaneously satisfy Ts-abn For true and TB-rate- abn If true, then enter Level 4 state and output 4.

[0102] The state transition is unidirectional, meaning that once a higher-level warning is reached, it will not revert to the previous one. The output of the state machine is the current warning level of the system.

[0103] The thermal runaway remaining time prediction layer uses a time mapping relationship between different warning levels and the occurrence of thermal runaway, fitted from a large amount of historical experimental data, to predict this relationship through multiple... LookupTable The lookup module is implemented, with each warning level corresponding to an independent lookup module. Its input is the information of the main feature points that are currently triggered and the environmental parameters, and its output is the predicted remaining time under that level. This predicted value will be dynamically updated as time goes by and new feature points appear. The entire model works together and finally outputs the warning level and the predicted remaining time value simultaneously through the output layer.

[0104] In summary, the system described in this embodiment constructs a data acquisition device integrating multi-physics field sensing of heat, electricity, force, and gas, and sequentially performs intelligent feature extraction based on mutual information, scientific effect evaluation based on entropy weight method, and... AprioriThrough association rule mining and hierarchical model construction based on multi-tool collaboration, a complete early warning system for lithium-ion battery thermal runaway was finally realized, capable of outputting graded early warning signals and predicting remaining time. Its technical effectiveness was verified through the following experimental examples, demonstrating that the system can achieve early warning, with the battery temperature far below the thermal runaway critical temperature when each warning level is triggered, and the remaining time prediction error being less than 18 seconds. This provides sufficient safety margin and reliable decision-making basis for taking graded emergency response measures.

[0105] Example 2:

[0106] This embodiment provides a multi-level early warning method for thermal runaway of lithium-ion batteries. The implementation of this method depends on the experimental platform and system components described in Embodiment 1.

[0107] The method begins with step S1, data acquisition. Specifically, this involves simultaneously acquiring complete data on the changes over time of multidimensional characteristic parameters of multiple physical fields—thermal, electrical, mechanical, and gaseous—during thermal runaway triggered by the battery at different overcharge rates (e.g., 0.5C, 1C, 2C, 3C) or different heating powers (e.g., 50W, 100W, 300W, 500W). This data is converted into digital signals by the data converter module 7018 and stored by a host computer, forming a foundational database for subsequent analysis.

[0108] Based on sufficient data, step S2, feature extraction, is performed. The core of this step is to use the mutual information method to filter massive amounts of multidimensional feature parameters in order to locate several key feature points with the highest correlation to the thermal runaway state from multiple physical fields.

[0109] Its specific sub-steps include S21, determining the effective time window for each feature point. This window is based on the statistical distribution of the occurrence times of the feature point in historical experiments, and calculates the average value. and standard deviation To determine, the window range is ;

[0110] S22, the data of feature points is divided into intervals using the equal-frequency binning principle, and the number of bins is... k According to the formula Determine this for discretization processing;

[0111] S23, Calculate each feature point X thermal runaway state Y joint probability distribution and their respective marginal probability distributions and And substitute into the mutual information formula Calculations are performed, and the mutual information value is ultimately selected. MI The top 8 feature points are designated as key feature points, which are consistent with those described in Example 1.

[0112] Subsequently, in optional step S3, the early warning effect evaluation can be based on a preset evaluation system, using the entropy weight method to quantitatively evaluate the early warning effect of the aforementioned key feature points. This system includes response speed. A 1. The calculation formula is as follows: Response accuracy A 2. The specific calculation formula is as follows: System reliability A 3. The specific calculation formula is as follows: A3 =0.5 × Failure Rate + 0.25 × Lifespan + 0.25 × Environmental Tolerance and Rationality of Classification A 4. The specific calculation formula is as follows: .

[0113] The calculation sub-steps of the entropy weight method include:

[0114] First, the values ​​of each indicator are normalized. The normalization formula is as follows: ;

[0115] Then calculate the entropy value and information entropy redundancy of each indicator;

[0116] Finally, the weights of each indicator were determined and the weighted sum was calculated to obtain the overall score. The scoring criteria were divided into "Good" (0.8 ≤ Score ≤1), "Good (0.7≤)" Score <0.8), "General (0.6≤ Score <0.7), "Poor ( Score The four levels are "<0.6". The output of this step is used to guide the allocation of subsequent warning levels and ensure the rationality of the grading strategy.

[0117] The next step is step S3, generating the early warning strategy. This step is based on data mining techniques. Apriori The algorithm mines the correlation between key feature points and warning levels in historical data, sets a minimum support of 0.2 and a minimum confidence of 0.8, thereby generating a four-level warning strategy with thermal-electrical-mechanical-gas coupling as detailed in Example 1, and forming explicit Boolean logic rules.

[0118] Then, step S4 is executed to construct the early warning model. Based on the four-level early warning strategy generated in step S4, the following is adopted: Matlab / Simulink The tool constructs a hierarchical early warning model. This model integrates signal processing and real-time feature point extraction, such as using wavelet transform for extraction. VppExtracting various rates of change using differentiation and filtering modules; inflection point of expansion force. Ftp The basis for judgment is ;based on Stateflow The system incorporates logical judgment of early warning levels and a table-based method for predicting the remaining time of thermal runaway, ultimately forming a software model capable of receiving real-time data and outputting early warning levels and remaining time.

[0119] Finally, step S5 is executed: based on the real-time warning level output by the warning model, the system automatically executes or prompts the operator to perform corresponding emergency response measures; optionally:

[0120] When the output is a Level 1 warning, immediately stop charging the battery.

[0121] When the output is a Level 2 warning, charging is immediately stopped and nitrogen is injected into the battery compartment to reduce the ambient oxygen concentration.

[0122] When the output is a Level 3 warning, immediately stop charging and start the forced ventilation exhaust system to remove flammable and explosive gases such as hydrogen and carbon monoxide.

[0123] When the output is a Level 4 warning, charging is immediately stopped and liquid carbon dioxide is injected into the battery compartment. The vaporization and heat absorption effect of the carbon dioxide is used to quickly reduce the temperature of the battery and the environment, thus preventing thermal runaway.

[0124] In summary, the method described in this embodiment provides an implementable and high-precision early warning scheme for lithium-ion battery thermal runaway through a complete process from data acquisition, intelligent feature extraction, effect evaluation, strategy mining to model construction. The core of this method lies in utilizing mutual information and... Apriori The algorithm performs in-depth analysis of multi-physics parameters to establish accurate hierarchical early warning logic, and ultimately achieves proactive protection against thermal runaway through remaining time prediction and hierarchical handling.

[0125] Example 3:

[0126] This embodiment provides a computer-readable storage medium, such as a USB flash drive, hard disk, or server memory, on which a computer program or instructions are stored. When the program is executed by a processor, such as the CPU of a host computer, it can automatically implement all or part of the process of the method described in Embodiment 2 above, thereby solidifying and promoting the early warning method of the present invention in the form of a software product.

[0127] Experimental example:

[0128] This experimental example provides a verification test based on the energy storage battery compartment experimental platform in Example 1. It records the trigger times of various warnings and the battery status under the condition of 1C overcharge rate triggering battery thermal runaway. Please refer to [link to relevant documentation]. Figure 9 .

[0129] The Level 1 warning was triggered at 2858 seconds, with the battery back temperature at 31.5℃ and the SOC reaching 117.3%.

[0130] The Level 2 warning was triggered at 3725 seconds, with the battery back temperature at 55.2℃ and the SOC reaching 128.5%.

[0131] The Level 3 warning was triggered at 4631 seconds, with the battery back temperature at 75.2℃ and the SOC reaching 140.9%.

[0132] The Level 4 warning was triggered at 5153 seconds, with the battery back temperature reaching 93.1℃. SOC It reached 148.3%.

[0133] Thermal runaway occurred at 5207 s, with a critical temperature of 116.8 ℃. SOC It is 150.1%.

[0134] Experimental data show that the trigger temperatures for all levels of warnings are significantly lower than the critical temperature for thermal runaway. The safety margins for Level 1 to Level 4 are 85.3℃, 61.6℃, 41.6℃ and 23.7℃, respectively.

[0135] Remaining time prediction accuracy analysis shows that:

[0136] Level 1 prediction error was 11.6 seconds, with a relative error of 0.41%.

[0137] Level 2 prediction error is 16.5 seconds, with a relative error of 0.44%.

[0138] The Level 3 prediction error was 17.8 seconds, with a relative error of 0.31%.

[0139] The Level 4 prediction error is 12.2 seconds, with a relative error of 0.24%.

[0140] The prediction error for all warning levels is less than 18 seconds, and the relative error does not exceed 0.5%.

[0141] Verification of the effectiveness of emergency response measures:

[0142] Level 1 was triggered and charging was stopped immediately, successfully preventing the early thermal runaway from developing.

[0143] After Level 2 is triggered, nitrogen is injected, causing the oxygen concentration in the battery compartment to drop from 20.9% to below 12% within 30 seconds.

[0144] After Level 3 is triggered, the ventilation system is activated, causing the hydrogen concentration to drop from a peak of 1580 ppm to below 50 ppm within 100 seconds.

[0145] Upon Level 4 triggering, liquid CO2 was injected, causing the temperature inside the battery compartment to plummet from 31.5°C to -80.7°C within 15 seconds, a cooling rate of 7.48°C / s. All levels of response measures effectively halted the thermal runaway chain reaction.

[0146] Combination Figure 9 As can be seen, before thermal runaway occurs, there is a nearly linear relationship between the trigger times of different warning levels and the temperature on the back of the battery, indicating that the intervals between warning levels are stable and the warning model is reasonably graded. The figure also simultaneously shows the opening of the battery safety valve, the occurrence of thermal runaway, and the highest temperature. Tmax Infrared images at three different times.

[0147] In summary, the above experiments verify that the system of this invention achieves early and accurate multi-level early warning. When each level of warning is triggered, the battery state parameters are far below the thermal runaway critical value, reserving sufficient safety margin. The remaining time prediction error is small and the accuracy is high, providing a reliable time window for emergency response. The multi-parameter coupled early warning strategy effectively solves the problem of low reliability of single-parameter early warning, and the overall system performance is significantly better than existing technical solutions.

Claims

1. A multi-stage pre-warning system for thermal runaway of a lithium-ion battery, characterized in that, The system comprises: a data acquisition device for acquiring multi-dimensional characteristic parameters of thermal, electrical, force, and gas fields during thermal runaway of a lithium ion battery; a feature extraction module in communication with the data acquisition device, configured to receive the multi-dimensional characteristic parameters and screen a plurality of key feature points with the highest correlation with the thermal runaway state from the characteristic parameters of the plurality of physical fields by using a mutual information method, the mutual information method comprising the steps of determining an effective time window of each feature point, dividing the feature point data into intervals by using an equal frequency binning principle, and calculating a joint probability distribution and mutual information value of each feature point and the thermal runaway state; a pre-warning strategy module in communication with the feature extraction module, configured to mine the correlation between the key feature points and different pre-warning levels based on an Apriori algorithm, and generate a four-level pre-warning strategy including Level 1 to Level 4; a pre-warning model module in communication with the pre-warning strategy module, configured to construct a pre-warning model based on the multi-level pre-warning strategy, the pre-warning model adopting a hierarchical architecture, including a data processing layer, a feature point extraction layer, a pre-warning level evaluation layer, and a thermal runaway residual time prediction layer, and capable of simultaneously outputting a pre-warning level and a predicted thermal runaway residual time.

2. The lithium-ion battery thermal runaway multi-stage early warning system of claim 1, wherein: The data acquisition device is specifically configured to acquire voltage, temperature, expansion force, and hydrogen concentration parameters.

3. The lithium-ion battery thermal runaway multi-stage early warning system of claim 2, wherein: The key feature points include at least one selected from the group consisting of a voltage plateau period, an expansion force inflection point, an expansion force change rate, a safety valve temperature, a hydrogen concentration, a battery surface temperature, a voltage change rate, and a battery back temperature rise rate.

4. The lithium-ion battery thermal runaway multi-stage warning system of claim 1, wherein: The system further comprises a pre-warning evaluation module, which is configured to evaluate the pre-warning effect of the key feature points and calculate a comprehensive score based on an evaluation system including response speed, response accuracy, system reliability, and grading rationality by using an entropy weight method.

5. The multi-level pre-warning system for thermal runaway of a lithium ion battery according to claim 1, characterized in that: the association rule mining algorithm is an Apriori algorithm; the multi-level pre-warning strategy is a four-level pre-warning strategy including Level 1 to Level 4; wherein the Level 1 pre-warning is associated with the appearance of the voltage plateau period and the expansion force inflection point, and the Level 4 pre-warning is associated with abnormalities of the voltage change rate, the battery surface temperature, and the battery back temperature rise rate.

6. A multi-stage thermal runaway early warning method for lithium-ion batteries using the system of any one of claims 1-5, characterized in that, The method comprises the following steps: S1. acquiring multi-dimensional characteristic parameters of thermal, electrical, force, and gas fields during thermal runaway of a lithium ion battery by using a multi-sensor; S2. screening a plurality of key feature points with the highest correlation with the thermal runaway state from the characteristic parameters of the plurality of physical fields by using a mutual information method, the mutual information method comprising the steps of determining an effective time window of each feature point, dividing the feature point data into intervals by using an equal frequency binning principle, and calculating a joint probability distribution and mutual information value of each feature point and the thermal runaway state; S3. mining the correlation between the key feature points and different pre-warning levels based on an Apriori algorithm, to generate a four-level pre-warning strategy including Level 1 to Level 4; S4. Constructing a pre-warning model with a hierarchical architecture based on the multi-level pre-warning strategy, the pre-warning model comprising a data processing layer, a feature point extraction layer, a pre-warning level evaluation layer, and a thermal runaway residual time prediction layer, to simultaneously output a pre-warning level and a predicted thermal runaway residual time according to real-time data.

7. The lithium-ion battery thermal runaway multi-stage early warning method according to claim 6, characterized in that: In step S2, the feature point screening by the mutual information method comprises the following sub-steps: S21. Determining the effective time window of each feature point; S22. Dividing the feature point data into intervals by the equal-frequency binning principle; S23. Calculating the joint probability distribution and mutual information value of each feature point and the thermal runaway state, and selecting several feature points with the highest mutual information value as key feature points.

8. The lithium-ion battery thermal runaway multi-stage early warning method of claim 6, wherein, In step S3, the multi-level pre-warning strategy is a four-level pre-warning strategy. The method further comprises the following steps after step S3: Based on a preset evaluation system, the pre-warning effect of the key feature points is evaluated by the entropy weight method, and the comprehensive score of each feature point is calculated.

9. The lithium-ion battery thermal runaway multi-stage early warning method of claim 6, wherein, The method further comprises the following steps after step S4: S5. Executing corresponding emergency disposal measures according to the pre-warning level output by the pre-warning model; The emergency disposal measures comprise one or more of stopping charging, injecting nitrogen, ventilation and exhaust, and injecting liquid carbon dioxide.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program, when executed by a processor, implements the steps of the method of any one of claims 6 to 9.

Images