Lithium ion battery double-layer thermal runaway early warning method, system, device and medium

Abstract

This invention belongs to the field of lithium-ion battery safety early warning technology, specifically involving a two-layer thermal runaway early warning method for lithium-ion batteries. The method includes: collecting voltage data of the battery pack during each charging cycle; obtaining a voltage matrix after linear interpolation and resampling; constructing a voltage envelope using quantile regression; determining a deviation matrix based on the voltage matrix and the lower envelope; extracting deviation statistical features from the deviation matrix to construct a multi-dimensional feature matrix; performing static anomaly detection using a density-based clustering algorithm after dimensionality reduction of the multi-dimensional feature matrix to identify statically abnormal batteries; performing dynamic anomaly detection based on the maximum voltage drop amplitude in the deviation statistical features to identify dynamically abnormal batteries; and fusing the two-layer detection results to output a thermal runaway early warning signal and an abnormal battery number. This invention effectively captures weak early signals of sudden-death thermal runaway by constructing deviation features through quantile regression to extract the voltage envelope and combining a two-layer mechanism of static outlier identification and dynamic cumulative amplification.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion battery safety early warning technology, specifically involving a method, system, device, and medium for early warning of double-layer thermal runaway in lithium-ion batteries. Background Technology

[0002] Lithium-ion battery thermal runaway is one of the most serious accidents threatening the safety of electric vehicles. Based on the differences in the characteristics of the failure evolution, thermal runaway can be divided into two categories: "gradual" and "sudden" types.

[0003] "Gradual" thermal runaway refers to a gradual degradation process in battery performance, where the voltage consistency index deteriorates slowly with the increase of charge-discharge cycles, and the voltage of abnormal batteries gradually deviates from that of normal battery groups. This slow evolution process can be effectively captured by traditional early warning methods based on consistency statistical indicators.

[0004] "Sudden death" thermal runaway is characterized by its abrupt nature. Before ignition, the battery does not show a continuous and obvious deterioration trend, and the voltage consistency remains at a high level across multiple charging cycles, only abruptly changing when the fault is imminent. The early, weak signals of this type of fault are often submerged in normal fluctuations, making it difficult for traditional single-layer early warning mechanisms to identify, leading to delayed or missed warnings and posing a serious safety hazard.

[0005] Existing thermal runaway early warning methods are mostly based on statistical indicators or consistency analysis. They judge the risk of thermal runaway by monitoring the consistent changes in the voltage of each individual cell in the battery pack. They have a good early warning effect for "gradual" thermal runaway, but they are difficult to effectively deal with the sudden anomalies of "sudden death" thermal runaway.

[0006] Therefore, there is an urgent need for an early warning method that can take into account both "gradual" and "sudden" thermal runaway. This method should be able to quickly respond to gradual failures by identifying static outliers, and amplify the early, weak signals of sudden failures through cumulative effects, thus giving drivers and passengers more time to take safe action. Summary of the Invention

[0007] The purpose of this invention is to provide a method, system, device and medium for early warning of double-layer thermal runaway in lithium-ion batteries, so as to solve the problem that existing thermal runaway early warning methods are unable to capture the early weak signals of "sudden death" thermal runaway that lack obvious gradual characteristics before ignition, resulting in delayed or missed warnings.

[0008] The present invention achieves the above objectives through the following technical solutions: Firstly, the present invention proposes a method for early warning of thermal runaway in a double-layer lithium-ion battery, the method comprising: The voltage data of the battery pack in each charging cycle is collected, and after linear interpolation and resampling, the voltage matrix of each individual cell is obtained. Construct the voltage envelope for each target charging cycle, and determine the deviation matrix of each individual cell based on the voltage matrix and the voltage envelope; The deviation statistical features of each individual cell are extracted from the deviation matrix to construct a multidimensional feature matrix. The deviation statistical features are used to characterize the deviation of the voltage drop amplitude of the individual cell. After dimensionality reduction of the multidimensional feature matrix, a density-based clustering algorithm is used to detect static anomalies and identify statically abnormal batteries. Dynamic anomaly detection is performed based on the maximum voltage drop amplitude in the aforementioned deviation statistical characteristics to identify dynamically abnormal batteries. Based on the detection results of static and dynamic anomaly detection, a thermal runaway early warning signal and the corresponding abnormal battery number are output.

[0009] Furthermore, the voltage data of the battery pack in each charging cycle is collected, and after linear interpolation and resampling, the voltage matrix of each individual battery cell is obtained, including: Extract the voltage sequence of all individual cells in each charging cycle from the battery management system cloud data platform; The voltage sequence is resampled to a fixed length T using linear interpolation to obtain the voltage matrix. As shown in the following formula: ; Where N is the number of individual cells and T is the preset fixed length of the sequence; For the first Individual cells at the sampling time voltage value, , .

[0010] Furthermore, the construction of the voltage envelope for each target charging cycle, and the determination of the deviation matrix for each individual cell based on the voltage matrix and the voltage envelope, includes: For sampling time Extract the j-th column vector from the voltage matrix. , which is the set of voltages of all individual cells at the current moment, where Sampling time The voltage value of the Nth individual cell; The voltage set is solved by minimizing the test loss function. quantiles The test loss function is: ,in The preset quantile parameters, It is an indicator function; By iterating through all sampling times, the voltage envelope vector can be obtained. , ; Based on the voltage matrix and the lower envelope vector Calculate the deviation matrix The elements of the deviation matrix are , This indicates the degree to which the i-th individual cell deviates from the group baseline at the j-th sampling time. Represents the j-th sampling time. quantile values.

[0011] Furthermore, the deviation statistical characteristics include a first deviation characterizing the voltage deviation at the beginning of charging, a second deviation characterizing the average deviation throughout the charging process, a third deviation characterizing the voltage deviation at the end of charging, a deterioration rate characterizing the amplification trend of the deviation, a fourth deviation characterizing the degree of fluctuation in the low-voltage stage, and a fifth deviation characterizing the maximum voltage drop amplitude; for the third... One charging cycle, constructing a multidimensional feature matrix Each row corresponds to a six-dimensional feature vector of a battery, and the six-dimensional feature vector of the i-th individual battery is... The formulas for calculating each characteristic component are as follows: First deviation: ,in The first number of sampling points is preset; the second deviation is: Third deviation: ,in The preset number of second sampling points; deterioration rate: , of which To prevent positive numbers with a denominator of zero; Fourth deviation: ,in The number of third sampling points is preset. The mean of the low-range deviation; fifth deviation: ;in .

[0012] Furthermore, after dimensionality reduction of the multidimensional feature matrix, a density-based clustering algorithm is used for static anomaly detection to identify statically abnormal batteries, including: The multidimensional feature matrix is ​​subjected to Z-Score normalization to obtain a normalized feature matrix; Principal component analysis is used to reduce the dimensionality of the standardized feature matrix. The first k principal components are selected such that the cumulative variance contribution rate exceeds a preset threshold to obtain the dimensionality-reduced feature matrix. A density-based clustering algorithm is used to cluster the dimensionality-reduced feature matrix, and points that cannot be assigned to any cluster are marked as noise points. If the i-th cell is marked as a noise point in the current charging cycle, a static anomaly signal is generated. .

[0013] Furthermore, the step of performing dynamic anomaly detection based on the maximum negative deviation in the deviation statistical features to identify dynamically abnormal batteries includes: Obtain the fifth deviation characteristic value of each individual cell in each charging cycle. ; Calculate the cumulative deviation index of the i-th cell in the k-th charging cycle. , ,in The preset attenuation factor, Let be the cumulative deviation index for the (k-1)th charging cycle. This is the fifth deviation characteristic value of the k-th charging cycle; In the k-th charging cycle, the cumulative deviation index of all individual cells is calculated using the following formula. mean and standard deviation : ; ; Calculate the dynamic Z-score of the i-th individual cell. , ; like Then a dynamic cumulative abnormal signal is generated. ;in This is the preset warning threshold.

[0014] Furthermore, the output of the thermal runaway early warning signal and the corresponding abnormal battery number based on the detection results of static anomaly detection and dynamic anomaly detection includes: For the i-th individual cell in the k-th charging cycle, a logical OR operation is used for decision fusion: ; If any single cell satisfies If so, it is determined that the current battery pack is at risk of thermal runaway; Output a warning signal and the abnormal battery number that meets the conditions.

[0015] Secondly, this invention proposes a lithium-ion battery double-layer thermal runaway early warning system to implement the lithium-ion battery double-layer thermal runaway early warning method described above. The system includes: The acquisition module is used to acquire voltage data of the battery pack in each charging cycle, wherein the charging cycle is a continuous charging cycle obtained based on the battery management system cloud data platform; The preprocessing module is used to perform linear interpolation resampling on the voltage data of each charging cycle to obtain the voltage matrix of each individual battery cell. The feature construction module is used to construct the voltage envelope of each target charging cycle, determine the deviation matrix of each individual cell based on the voltage matrix and the voltage envelope, and extract the deviation statistical features of each individual cell from the deviation matrix to construct a multi-dimensional feature matrix. The deviation statistical features are used to characterize the deviation of the voltage drop amplitude of the individual cell. The static detection module is used to perform dimensionality reduction processing on the multidimensional feature matrix and then use a density-based clustering algorithm to perform static anomaly detection to identify statically abnormal batteries. The dynamic detection module is used to perform dynamic anomaly detection based on the maximum negative deviation in the deviation statistical features, and to identify dynamically abnormal batteries. The decision fusion module is used to output thermal runaway early warning signals and corresponding abnormal battery numbers based on the detection results of static anomaly detection and dynamic anomaly detection.

[0016] Thirdly, the present invention proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described lithium-ion battery double-layer thermal runaway early warning method.

[0017] Fourthly, the present invention proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described lithium-ion battery double-layer thermal runaway early warning method.

[0018] The beneficial effects of this invention are as follows: (1) This invention constructs a voltage envelope through quantile regression, which can extract multiple statistical features characterizing the battery consistency state from the deviation matrix, avoiding the problem of extreme value interference in the traditional mean regression method. The first deviation, second deviation, and third deviation reflect the degree of voltage deviation at the beginning, the whole cycle, and the end of the charging cycle, respectively. The deterioration rate quantifies the amplification trend of the deviation with the charging process. The fourth deviation measures the fluctuation degree in the low voltage stage. The fifth deviation captures the maximum voltage drop amplitude during the charging process. The six features characterize the battery health state from different dimensions. Static anomaly detection uses principal component analysis for dimensionality reduction combined with density-based clustering algorithm to identify outlier batteries in the feature space. Dynamic anomaly detection calculates the cumulative deviation index based on the maximum voltage drop amplitude and combines it with the Z-score dynamic threshold to identify abnormal batteries, amplifying the early weak signals through the cumulative effect.

[0019] (2) In the verification of vehicles with gradual thermal runaway, this invention can trigger an early warning when the abnormal battery deviates slightly; in the verification of vehicles with sudden thermal runaway, this invention can trigger an early warning 18 charging cycles before the failure, effectively solving the problem of high false negative rate of traditional methods for sudden failures. This invention is based entirely on voltage data collected by existing battery management systems, requires no additional hardware investment, has controllable computational complexity, and is easy to deploy in the cloud or on the vehicle. Attached Figure Description

[0020] Figure 1 A flowchart of a lithium-ion battery double-layer thermal runaway early warning method provided by the present invention; Figure 2 This is another flowchart of the lithium-ion battery double-layer thermal runaway early warning method provided by the present invention; Figure 3 A comparison diagram of charging and discharging current loads provided in an embodiment of the present invention; Figure 4 The voltage curve of vehicle 1, a type of gradual thermal runaway vehicle, provided in an embodiment of the present invention; Figure 5 The voltage curve of vehicle2, a vehicle experiencing sudden thermal runaway, provided in an embodiment of the present invention; Figure 6 This is a three-dimensional scatter plot of the feature space after PCA dimensionality reduction for the first warning of vehicle1 provided in an embodiment of the present invention; Figure 7 A trend graph showing the change of the cumulative deviation index of vehicle1 with charging cycles in an embodiment of the present invention; Figure 8 This is a three-dimensional scatter plot of the feature space after PCA dimensionality reduction during the first warning of Vehicle2 provided in this embodiment of the invention; Figure 9 This is a trend graph showing the cumulative deviation index of vehicle2 changing with charging cycles, provided in an embodiment of the present invention. Detailed Implementation

[0021] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content. Example 1

[0022] Please see Figure 1 and Figure 2This disclosure proposes a two-layer thermal runaway early warning method for lithium-ion batteries. This method extracts the voltage envelope of the battery pack through quantile regression, constructs a multi-dimensional feature matrix characterizing battery consistency, and combines a two-layer mechanism of static clustering detection and dynamic cumulative detection to achieve early warning of both gradual and sudden thermal runaway. The method includes: collecting voltage data of the battery pack in each charging cycle; obtaining the voltage matrix of each individual battery after linear interpolation and resampling; constructing the voltage envelope for each target charging cycle; determining the deviation matrix of each individual battery based on the voltage matrix and the voltage envelope; extracting the deviation statistical features of each individual battery from the deviation matrix to construct a multi-dimensional feature matrix, where the deviation statistical features characterize the deviation of the voltage drop amplitude of the individual battery; performing dimensionality reduction on the multi-dimensional feature matrix; using a density-based clustering algorithm for static anomaly detection to identify statically abnormal batteries; performing dynamic anomaly detection based on the maximum voltage drop amplitude in the deviation statistical features to identify dynamically abnormal batteries; and outputting a thermal runaway early warning signal and the corresponding abnormal battery number based on the detection results of static and dynamic anomaly detection.

[0023] Specifically, the implementation of this method may include the following steps: S1: Data Acquisition and Preprocessing The system collects voltage data from the battery pack during each charging cycle. Specifically, it obtains the target vehicle's battery pack operating data from the battery management system's cloud data platform. For example, the platform samples the vehicle's operating data every 10 seconds and transmits the data to the cloud platform in real time for continuous archiving via a communication protocol. The battery pack consists of N individual cells connected in series, and the length of the original voltage sequence collected in each charging cycle varies depending on the charging duration.

[0024] The voltage sequence is resampled to a fixed length T using linear interpolation to obtain the voltage matrix. As shown in the following formula: ; Where N is the number of individual cells and T is the preset fixed length of the sequence; For the first Individual cells at the sampling time voltage value, , .

[0025] It should be noted that in this embodiment, T represents the fixed sequence length after linear interpolation resampling, i.e., the total number of sampling points for each charging cycle. Through resampling, the original charging curve is uniformly mapped to T equally spaced sampling times, eliminating the impact of differences in charging cycle durations on subsequent feature extraction. In this embodiment, T can be 1000; in practical applications, it can be flexibly set according to the sampling frequency and charging duration range.

[0026] S2: Consistency Feature Extraction Based on Quantile Regression Ideally, the voltage curves of all individual cells should highly overlap. However, when internal short circuits or material degradation occur, the voltage of abnormal cells will deviate from the overall distribution. To address the slight deviation in the early low-voltage region of sudden thermal runaway, this invention introduces quantile regression to construct the voltage envelope of the battery pack.

[0027] Specifically, regarding the sampling time Extract the j-th column vector from the voltage matrix , which is the set of voltages of all individual cells at the current moment, where Sampling time The voltage value of the Nth individual cell; Solving the voltage set by minimizing the test loss function. quantiles The test loss function is: ,in The preset quantile parameters, It is an indicator function; By iterating through all sampling times j=1,2,3,...,T, the voltage envelope vector is obtained. , ; Based on voltage matrix and lower envelope vector Calculate the deviation matrix The elements of the deviation matrix are , This indicates the degree to which the i-th individual cell deviates from the group baseline at the j-th sampling time. Represents the j-th sampling time. quantile values.

[0028] S3: Deviation Statistical Feature Extraction and Multidimensional Feature Matrix Construction To comprehensively characterize the evolution of battery health status, especially to capture early signals of sudden death faults, this embodiment extracts six deviation statistical features from the deviation matrix, including: a first deviation characterizing the voltage deviation at the beginning of charging, a second deviation characterizing the average deviation throughout the charging process, a third deviation characterizing the voltage deviation at the end of charging, a deterioration rate characterizing the amplification trend of the deviation, a fourth deviation characterizing the degree of fluctuation in the low voltage stage, and a fifth deviation characterizing the maximum voltage drop.

[0029] The above six deviation statistical characteristics comprehensively characterize the battery's consistency state from four dimensions: the first, second, and third deviations constitute the time-domain distribution dimension, representing the degree of voltage deviation at the beginning, end, and end of the charging cycle, respectively, enabling the location of the charging stage where anomalies occur; the deterioration rate constitutes the evolution trend dimension, quantifying the amplification speed of the deviation with the charging process through linear regression slope, reflecting the evolution of the fault; the fourth deviation constitutes the fluctuation characteristic dimension, measuring the stability of consistency during the low-voltage stage and identifying intermittent abnormal fluctuations; the fifth deviation constitutes the extreme value dimension, capturing the maximum voltage drop at a single point during charging, directly related to sudden faults such as micro-short circuits. Specifically, for the first One charging cycle, constructing a multidimensional feature matrix Each row corresponds to a six-dimensional feature vector of a battery, and the six-dimensional feature vector of the i-th individual battery is... The formulas for calculating each characteristic component are as follows: First deviation: ,in The preset number of first sampling points (in this embodiment, the number is taken as...) =200); Second deviation: Third deviation: ,in The preset number of second sampling points (in this embodiment, it is taken as...) =400), the third deviation corresponds to the internal short-circuit risk under high SOC conditions; deterioration rate: , of which To prevent positive numbers with a denominator of zero, this feature quantifies the rate of change of the deviation as the charging process progresses through the slope of a linear regression; Fourth deviation: ,in The preset number of third sampling points (in this embodiment) =500), The low-range deviation mean reflects the degree of consistency fluctuation during the first 50% of the charging phase; fifth deviation: This feature is used to capture the most severe voltage drop at a single point and is a key indicator for identifying micro-short circuits. .

[0030] In this embodiment, , , These are all preset sampling point number parameters, corresponding to the calculation window lengths of the first, third, and fourth deviations, respectively, and all are less than the total number of sampling points T. Used for retrieval The deviation characteristics at the beginning of charging are calculated using individual sampling points; Used after retrieval The deviation characteristics at the charging end are calculated using individual sampling points; Used for retrieval Each sampling point is used to calculate the fluctuation characteristics of the low-voltage stage; in this embodiment, they correspond to the charging start stage (first 20%), the end stage (last 40%), and the low-voltage stage (first 50%), respectively. In practical applications, these can be adjusted according to the battery type, charging strategy, and sampling frequency.

[0031] Understandably, step S3 in this embodiment constructs a voltage envelope through quantile regression, which can capture low-voltage anomaly signals in weak cells within the battery pack. Compared to the traditional mean regression method, it avoids extreme value interference and has higher sensitivity to subtle deviations in the early stages of sudden-death faults. Specifically, six statistical features extracted from the deviation matrix comprehensively characterize the battery's consistency state from four dimensions: time-domain distribution (initial segment, full cycle, and final segment), evolution trend (deterioration rate), fluctuation characteristics (low-range volatility), and extreme values ​​(maximum voltage drop), forming a three-dimensional representation of the battery's health status. Among them, the deterioration rate uses linear regression slope to quantify the deviation amplification trend, reflecting the speed of fault evolution; the maximum voltage drop is directly related to sudden faults such as micro-short circuits, providing a key input for dynamic cumulative detection. The above feature construction method is entirely based on voltage data collected by existing BMS, requiring no additional sensors, with controllable computational complexity, and is easy to deploy in cloud or vehicle-mounted embedded systems.

[0032] S4: Two-tiered early warning mechanism To overcome the limitations of a single early warning mechanism in dealing with complex failure modes, especially to address the problem of the lack of gradual characteristics in sudden thermal runaway, this embodiment adopts a dual-layer early warning mechanism, including two parallel channels: static anomaly detection and dynamic anomaly detection.

[0033] (a) Static anomaly detection Static anomaly detection aims to quickly identify significant outliers using the feature matrix of the current charging cycle, primarily targeting gradual faults.

[0034] First, for the multidimensional feature matrix Z-score normalization is performed to obtain the normalized feature matrix. ; Then, Principal Component Analysis (PCA) is used to reduce the dimensionality of the standardized feature matrix to eliminate collinearity among features and reduce computational complexity. The eigenvalues ​​of the covariance matrix are solved, and the first k principal components are selected such that the cumulative variance contribution rate exceeds a preset threshold. In this embodiment, the first three principal components are solved to obtain the dimensionality-reduced feature matrix. ; Finally, a density-based clustering algorithm is used to cluster the dimensionality-reduced feature matrix. The DBSCAN algorithm does not require a preset number of clusters and can effectively identify clusters of arbitrary shapes and noise points. of Neighborhood is .like ,but The core point is defined as follows: Points that cannot be assigned to any cluster based on density reachability are marked as noise points (labeled -1). The first-level warning criterion is defined as follows: if the... Each battery is in cycle If a point is marked as noise, a static anomaly signal is generated: ; If the i-th cell is marked as a noise point in the current charging cycle, a static anomaly signal is generated. In cases of gradual thermal runaway, anomalous cells exhibit a clear outlier distribution in the 3D feature space after PCA dimensionality reduction, which can be effectively identified by the DBSCAN algorithm.

[0035] (ii) Dynamic anomaly detection Dynamic anomaly detection aims to capture sudden-death faults that lack a clear gradual process, amplifying early weak signals through cumulative deviation exponential amplification.

[0036] First, obtain the fifth deviation characteristic value of each individual cell in each charging cycle. This refers to the maximum voltage drop. Then, the cumulative deviation index of the i-th cell in the k-th charging cycle is calculated according to the recursive formula. , ,in The preset attenuation factor, This is the fifth deviation characteristic value of the k-th charging cycle; Let be the cumulative deviation index for the (k-1)th charging cycle, with an initial value of the cumulative deviation index for the 0th charging cycle. .

[0037] In the k-th charging cycle, the cumulative deviation index of all individual cells is calculated using the following formula. mean and standard deviation : ; ; Calculate the dynamic Z-score of the i-th individual cell. , ; like Then a dynamic cumulative abnormal signal is generated. ;in The preset warning threshold is taken in this embodiment. =3; This dynamic threshold strategy can adapt to the aging baseline of different vehicles and batteries. Even if the battery pack as a whole ages, as long as the individual differences do not show statistically significant outliers, it will not give false alarms and has strong robustness.

[0038] (III) Decision Integration and Early Warning Output To ensure high system sensitivity and coverage of all fault types, a logical OR operation is used to fuse the dual-layer detection results. For the i-th cell in the k-th charging cycle: ; ; If any single cell satisfies If the current battery pack is at risk of thermal runaway, a warning signal and the abnormal battery number that meets the conditions will be output.

[0039] Experimental verification To verify the effectiveness of this invention, a verification experiment was conducted using real-vehicle data collected from a cloud platform of a BMS R&D company. This platform samples vehicle operating data every 10 seconds and transmits the data to the cloud platform in real time for continuous archiving via a communication protocol. The experimental data included two vehicles (vehicle1 and vehicle2) that had experienced thermal runaway, both equipped with ternary lithium-ion battery packs. Vehicle1 experienced a gradual thermal runaway failure, while vehicle2 experienced a sudden thermal runaway failure. The location of the abnormal battery was confirmed based on maintenance records.

[0040] (a) Data preprocessing The charging process of new energy vehicles is strictly regulated by equipment, resulting in a relatively stable current. Voltage fluctuations primarily reflect the battery's health. However, during discharge, short periods of high load occur, and drastic voltage fluctuations do not necessarily indicate battery degradation. Figure 3 As shown, the comparison of current load during charging and discharging shows that the charging phase has smaller fluctuations and is less affected by external interference; the voltage differences between individual cells mainly reflect their health status. Therefore, this chapter extracts the voltage sequence from each complete charging cycle for analysis to avoid discharge interference. The preprocessing steps include: extracting the voltage sequence of all individual cells in each charging cycle from the BMS cloud platform. ,in For the number of battery cells, The number of sampling points for each charging cycle; linear interpolation is used to resample the sequence to a fixed length (e.g., 1000 points) to eliminate duration differences and ensure consistency in subsequent analyses.

[0041] (II) Analysis of Real Vehicle Data Features Taking vehicle1 (with an abnormal battery of size 4) as an example, the charging curve of its size 4 battery gradually changing from a normal state to an abnormal state within one month during the period of abnormality is as follows: Figure 4 As shown in the diagram, the red curve represents the voltage trajectory of the abnormal cell (cell #4), while the green curve represents the voltage of the other normal cells in the battery pack. Observations revealed that the battery pack maintained good voltage consistency in the early stages, but as time progressed, the voltage of the abnormal cell gradually deviated from the normal battery group, exhibiting a gradual deterioration in performance. This progressive change ultimately triggered a thermal runaway event.

[0042] A thorough analysis of voltage data from vehicles experiencing sudden thermal runaway, using Vehicle 2 as an example, revealed distinct characteristics. According to confirmation from the vehicle repair company, battery number 17 in the battery pack was the source of the thermal runaway failure. Figure 5 As shown, the voltage evolution trajectory of battery No. 17 is extremely similar to that of normal batteries. Throughout the multiple charging cycles prior to thermal runaway, the voltage consistency of the entire battery pack remained at a high level, lacking significant gradual deviations. This makes traditional early warning methods based on thresholds or single-cycle consistency judgments ineffective in distinguishing abnormal batteries, further highlighting the necessity of developing a cumulative detection mechanism for sudden-death faults.

[0043] The above data analysis shows that the differences in voltage characteristics between gradual and sudden thermal runaway require early warning methods to have multi-level detection capabilities to cover the early signal capture of different fault modes.

[0044] (III) Early warning results of vehicles with gradual thermal runaway This chapter focuses on verifying the early warning performance of the dual-layer frame in vehicles experiencing gradual thermal runaway. Gradual faults are characterized by a slow deterioration of battery consistency indicators (such as voltage deviation). The first layer of the dual-layer frame, static detection (DBSCAN), can quickly capture immediate deviations, while the second layer, dynamic cumulative detection (cumulative deviation index and Z-score), amplifies early, weak signals through long-term accumulation, achieving earlier warnings. The warning process is explained in detail below using Vehicle 1 as an example.

[0045] Taking vehicle1 (with battery number 4 as an example), this vehicle experienced progressive voltage degradation within a month prior to the fire. The dual-layer frame triggered the first warning on the 29th charging cycle, at which point the voltage of the abnormal battery had already begun to show a slight deviation. The first-layer DBSCAN algorithm effectively captured outliers of the abnormal battery (such as...) in the 3D feature space after PCA dimensionality reduction. Figure 6As shown, anomalies are marked in red and clearly separated from normal battery clusters. This is thanks to the multi-dimensional representation of deviation statistics, including the first deviation reflecting the deviation at the beginning of charging and the deterioration rate reflecting the evolution trend, making immediate anomalies easy to identify. Meanwhile, the second-layer cumulative deviation index trend chart ( Figure 7 The data shows that the cumulative deviation index of the abnormal battery continues to climb, far exceeding the stable level of the normal battery, confirming the cumulative effect of gradual deterioration. This dual-layer early warning mechanism ensures the reliability of the warning, triggering alarms even in the early stages when the deviation is not significant.

[0046] (iv) Early warning results for vehicles experiencing sudden thermal runaway This chapter verifies the performance of a double-layer frame for vehicles experiencing sudden thermal runaway. A typical characteristic of sudden-runaway faults is the lack of a clear gradual transition process; voltage consistency remains high before ignition, making it difficult for traditional static methods to capture early, weak signals. The double-layer frame effectively amplifies these hidden cumulative effects through a second layer of dynamic cumulative detection. Even with insignificant instantaneous deviations, it can trigger an early warning several cycles before the fault occurs, thus solving the problem of missed detection in traditional methods. The following analysis, using Vehicle 2 (with an abnormal battery of size 17) as an example, details the early warning process.

[0047] For Vehicle2, the thermal runaway occurred suddenly before the fire, and the voltage curve was highly similar to that of a normal battery (as shown in the aforementioned data analysis). The dual-layer frame triggered the first warning at the 100th charging cycle (approximately 18 cycles before the failure). At this point, the first-layer DBSCAN static detection did not capture any significant outliers because the consistency deviation was still in a weak stage (e.g., ...). Figure 8 As shown, all points in the PCA 3D space are basically clustered, with no obvious anomalous separation. However, the second-layer cumulative deviation index mechanism plays a key role: the cumulative deviation index of anomalous cells is amplified exponentially, rising sharply in the short term, and the Z score exceeds the dynamic threshold (e.g., Figure 9 As shown, the abnormal battery curve exhibits an accelerated rise, while the normal battery remains stable. This result verifies the unique advantage of the cumulative effect of the cumulative deviation index in sudden death scenarios, which can transform scattered weak signals into detectable risk trends, providing users with a valuable early intervention window (such as parking inspection or battery replacement).

[0048] In summary, the dual-layer early warning mechanism of this invention has good early warning performance for both gradual thermal runaway and sudden thermal runaway, verifying the effectiveness and versatility of the method. Example 2

[0049] A lithium-ion battery dual-layer thermal runaway early warning system is used to implement the lithium-ion battery dual-layer thermal runaway early warning method as described in Example 1. This system can be deployed on a cloud server or vehicle-side battery management system and can be implemented through software modules or a combination of software and hardware.

[0050] The system includes the following modules: The data acquisition module is used to collect voltage data of the battery pack during each charging cycle, which is a continuous charging cycle obtained based on the battery management system cloud data platform. In specific implementation, the data acquisition module connects to the cloud platform or the vehicle-mounted BMS through a communication interface to acquire raw voltage data of each charging cycle in real time or periodically retrieve it.

[0051] The preprocessing module performs linear interpolation and resampling on the voltage data of each charging cycle to obtain the voltage matrix of each individual cell. In practice, the preprocessing module maps the original voltage sequences of varying lengths in each charging cycle to equally spaced sampling times of a fixed length T, eliminating the impact of differences in charging duration on subsequent analysis.

[0052] The feature construction module is used to construct the voltage envelope for each target charging cycle, determine the deviation matrix of each individual cell based on the voltage matrix and the voltage envelope, and extract the deviation statistical features of each individual cell from the deviation matrix to construct a multi-dimensional feature matrix. These deviation statistical features characterize the deviation of the voltage drop amplitude of each individual cell. Specifically, the feature construction module first uses quantile regression to calculate the τ quantile at each sampling time to form the voltage envelope; then it calculates the difference between each individual cell and the voltage envelope to obtain the deviation matrix; finally, it extracts the first to fifth deviations and the degradation rate from the deviation matrix to form a six-dimensional feature matrix.

[0053] The static detection module performs dimensionality reduction on the multidimensional feature matrix and then uses a density-based clustering algorithm to detect static anomalies and identify statically anomalous batteries. Specifically, the module first performs Z-score normalization on the feature matrix, then uses principal component analysis to reduce the dimensionality until the cumulative variance contribution rate exceeds 95%, and finally uses the DBSCAN algorithm to identify outliers in the feature space, marking batteries corresponding to noise points as statically anomalous batteries.

[0054] The dynamic detection module is used to perform dynamic anomaly detection based on the maximum negative deviation in the deviation statistical characteristics, and to identify dynamically abnormal batteries. In specific implementation, the dynamic detection module performs exponential weighted accumulation of the maximum negative deviation for each charging cycle, calculates the cumulative deviation index, and identifies abnormal batteries based on the Z-score dynamic threshold.

[0055] The decision fusion module is used to output a thermal runaway early warning signal and the corresponding abnormal battery number based on the detection results of static anomaly detection and dynamic anomaly detection. In specific implementation, the decision fusion module uses logical OR operation to fuse the two-layer detection results. When any detection channel determines that a battery is abnormal, it outputs an early warning signal and the abnormal battery number.

[0056] It should be noted that each module in the aforementioned early warning device corresponds to a step in implementing the aforementioned early warning method. Multiple modules may have the same instances and application scenarios as their corresponding steps, but are not limited to the content disclosed in Embodiment 1. In actual deployment, these modules can be integrated into a single physical device, such as an in-vehicle BMS controller or a cloud server, or they can be distributed across the vehicle and the cloud, working collaboratively through a communication network. For example, the data acquisition module and preprocessing module can be deployed on the vehicle to process raw data in real time; the feature construction module, static detection module, dynamic detection module, and decision fusion module can be deployed on the cloud to utilize the cloud platform's storage and computing resources for in-depth analysis and push the early warning results to the user terminal or the vehicle.

[0057] In another embodiment of the present invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the lithium-ion battery double-layer thermal runaway early warning method as described in Embodiment 1.

[0058] In another embodiment of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the lithium-ion battery double-layer thermal runaway early warning method as described in Embodiment 1.

[0059] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated.

[0060] The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0061] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0062] In addition, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0063] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for early warning of thermal runaway in a double-layer lithium-ion battery, characterized in that, The method includes: The voltage data of the battery pack in each charging cycle is collected, and after linear interpolation and resampling, the voltage matrix of each individual cell is obtained. Construct the voltage envelope for each target charging cycle, and determine the deviation matrix of each individual cell based on the voltage matrix and the voltage envelope; The deviation statistical features of each individual cell are extracted from the deviation matrix to construct a multidimensional feature matrix. The deviation statistical features are used to characterize the deviation of the voltage drop amplitude of the individual cell. After dimensionality reduction of the multidimensional feature matrix, a density-based clustering algorithm is used to detect static anomalies and identify statically abnormal batteries. Dynamic anomaly detection is performed based on the maximum voltage drop amplitude in the aforementioned deviation statistical characteristics to identify dynamically abnormal batteries. Based on the detection results of static and dynamic anomaly detection, a thermal runaway early warning signal and the corresponding abnormal battery number are output.

2. The lithium-ion battery double-layer thermal runaway early warning method according to claim 1, characterized in that, The voltage data of the battery pack during each charging cycle is collected, and after linear interpolation and resampling, the voltage matrix of each individual battery cell is obtained, including: Extract the voltage sequence of all individual cells in each charging cycle from the battery management system cloud data platform; The voltage sequence is resampled to a fixed length T using linear interpolation to obtain the voltage matrix. As shown in the following formula: ； Where N is the number of individual cells and T is the preset fixed length of the sequence; For the first Individual cells at the sampling time voltage value, , .

3. The lithium-ion battery double-layer thermal runaway early warning method according to claim 1, characterized in that, The process of constructing the voltage envelope for each target charging cycle and determining the deviation matrix for each individual cell based on the voltage matrix and the voltage envelope includes: For sampling time Extract the j-th column vector from the voltage matrix. , which is the set of voltages of all individual cells at the current moment, where Sampling time The voltage value of the Nth individual cell; The voltage set is solved by minimizing the test loss function. quantiles The test loss function is: ,in The preset quantile parameters, It is an indicator function; By iterating through all sampling times, the voltage envelope vector can be obtained. , ; Based on the voltage matrix and the lower envelope vector Calculate the deviation matrix The elements of the deviation matrix are , This indicates the degree to which the i-th individual cell deviates from the group baseline at the j-th sampling time. Represents the j-th sampling time. quantile values.

4. The lithium-ion battery double-layer thermal runaway early warning method according to claim 3, characterized in that, The deviation statistical features include a first deviation characterizing the voltage deviation at the beginning of charging, a second deviation characterizing the average deviation throughout the entire charging process, a third deviation characterizing the voltage deviation at the end of charging, a deterioration rate characterizing the amplification trend of the deviation, a fourth deviation characterizing the degree of fluctuation in the low voltage stage, and a fifth deviation characterizing the maximum voltage drop. For the One charging cycle, constructing a multidimensional feature matrix Each row corresponds to a six-dimensional feature vector of a battery, and the six-dimensional feature vector of the i-th individual battery is... The formulas for calculating each characteristic component are as follows: First deviation: ,in The preset number of first sampling points; Second deviation: Third deviation: ,in The preset number of second sampling points; deterioration rate: , of which To prevent positive numbers with a denominator of zero; Fourth deviation: ,in The number of third sampling points is preset. The mean of the low-range deviation; fifth deviation: ;in .

5. The lithium-ion battery double-layer thermal runaway early warning method according to claim 1, characterized in that, After dimensionality reduction of the multidimensional feature matrix, a density-based clustering algorithm is used for static anomaly detection to identify statically abnormal batteries, including: The multidimensional feature matrix is ​​subjected to Z-Score normalization to obtain a normalized feature matrix; Principal component analysis is used to reduce the dimensionality of the standardized feature matrix. The first k principal components are selected such that the cumulative variance contribution rate exceeds a preset threshold to obtain the dimensionality-reduced feature matrix. A density-based clustering algorithm is used to cluster the dimensionality-reduced feature matrix, and points that cannot be assigned to any cluster are marked as noise points. If the i-th cell is marked as a noise point in the current charging cycle, a static anomaly signal is generated. .

6. The lithium-ion battery double-layer thermal runaway early warning method according to claim 4, characterized in that, The step of performing dynamic anomaly detection based on the maximum negative deviation in the deviation statistical features to identify dynamically abnormal batteries includes: Obtain the fifth deviation characteristic value of each individual cell in each charging cycle. ; Calculate the cumulative deviation index of the i-th cell in the k-th charging cycle. , ,in The preset attenuation factor, Let be the cumulative deviation index for the (k-1)th charging cycle. This is the fifth deviation characteristic value of the k-th charging cycle; In the k-th charging cycle, the cumulative deviation index of all individual cells is calculated using the following formula. mean and standard deviation : ； ； Calculate the dynamic Z-score of the i-th individual cell. , ; like Then a dynamic cumulative abnormal signal is generated. ;in This is the preset warning threshold.

7. The lithium-ion battery double-layer thermal runaway early warning method according to claims 5 and 6, characterized in that, The thermal runaway early warning signal and the corresponding abnormal battery number are output based on the detection results of static anomaly detection and dynamic anomaly detection, including: For the i-th individual cell in the k-th charging cycle, a logical OR operation is used for decision fusion: ； If any single cell satisfies If so, it is determined that the current battery pack is at risk of thermal runaway; Output a warning signal and the abnormal battery number that meets the conditions.

8. A lithium-ion battery double-layer thermal runaway early warning system, used to implement the lithium-ion battery double-layer thermal runaway early warning method as described in any one of claims 1-7, characterized in that, The system includes: The acquisition module is used to acquire voltage data of the battery pack in each charging cycle, wherein the charging cycle is a continuous charging cycle obtained based on the battery management system cloud data platform; The preprocessing module is used to perform linear interpolation resampling on the voltage data of each charging cycle to obtain the voltage matrix of each individual battery cell. The feature construction module is used to construct the voltage envelope of each target charging cycle, determine the deviation matrix of each individual cell based on the voltage matrix and the voltage envelope, and extract the deviation statistical features of each individual cell from the deviation matrix to construct a multi-dimensional feature matrix. The deviation statistical features are used to characterize the deviation of the voltage drop amplitude of the individual cell. The static detection module is used to perform dimensionality reduction processing on the multidimensional feature matrix and then use a density-based clustering algorithm to perform static anomaly detection to identify statically abnormal batteries. The dynamic detection module is used to perform dynamic anomaly detection based on the maximum negative deviation in the deviation statistical features, and to identify dynamically abnormal batteries. The decision fusion module is used to output thermal runaway early warning signals and corresponding abnormal battery numbers based on the detection results of static anomaly detection and dynamic anomaly detection.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the lithium-ion battery double-layer thermal runaway early warning method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the lithium-ion battery double-layer thermal runaway early warning method as described in any one of claims 1-7.

Images