Battery thermal runaway early warning method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of battery safety, and discloses a battery thermal runaway early warning method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining a vehicle group feature library which comprises a charging and discharging feature set of at least one vehicle and a state label of the vehicle, and the state label is used for representing whether the vehicle is a thermal runaway vehicle; screening a target charging and discharging feature associated with the state tag from the charging and discharging feature set of each vehicle; and calculating an abnormal score of the vehicle based on the outlier degree between the target charging and discharging characteristics of the vehicle and the vehicle group characteristics in the vehicle group characteristic library, and performing risk early warning operation on the vehicle based on the abnormal score. According to the method, the problems that in existing sudden death type thermal runaway early warning, due to the fact that vehicle end signals such as single voltage are not obviously abnormal, a traditional model depends on the historical characteristic evolution trend of a single vehicle and single threshold setting, and the characteristic association between the single vehicle and a vehicle group is ignored, risk characteristics cannot be effectively captured are solved.

Description

Technical Field

[0001] This invention relates to the field of battery safety technology, specifically to battery thermal runaway early warning methods, devices, computer equipment, and storage media. Background Technology

[0002] With the continuous growth of electric vehicle ownership, safety accidents caused by battery thermal runaway are frequent. Among them, sudden-death thermal runaway has become a key focus of safety prevention and control due to its suddenness and high destructiveness. Sudden-death thermal runaway usually occurs abruptly when conventional monitoring signals such as battery voltage and SOC do not show significant abnormalities, leading to violent combustion or even explosion of the battery pack within a short period of time, causing significant property damage and personal injury. Traditional early warning models mainly rely on historical trend analysis of single parameters such as cell voltage and temperature, for example, setting a cell SOC growth threshold as a risk threshold. However, these models cannot capture the latent characteristic changes before sudden-death thermal runaway and are greatly affected by vehicle operating conditions, resulting in a high risk of false alarms.

[0003] The main shortcomings of existing technical solutions are twofold: First, traditional early warning models assess risk solely based on the evolutionary trends of historical data for individual vehicles. For example, they set thresholds by monitoring historical fluctuations in individual cell voltage differences or the rate of SOC degradation. However, the risk characteristics of sudden thermal runaway are not reflected in the historical signal evolution of a single vehicle, leading to model failure. Second, existing methods neglect the correlation between individual vehicles and vehicle groups. For instance, they judge risk solely based on the SOC growth index of a single vehicle, without considering the characteristic distribution patterns of similar models or batches of batteries within the same vehicle group. This results in a lack of dynamic adaptability in threshold setting, making it prone to misjudgment due to environmental changes or differences in battery batches. For example, a batch of batteries may share common risks due to manufacturing process issues, but traditional models cannot identify such potential hazards through comparison of vehicle group data. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a battery thermal runaway early warning method, device, computer equipment, and storage medium to solve the problems in existing sudden death type thermal runaway early warning systems, such as the lack of obvious abnormalities in vehicle-end signals such as individual cell voltage, the reliance of traditional models on the historical evolution trend of individual vehicle characteristics and a single threshold setting, and the neglect of the correlation between individual vehicle and vehicle group characteristics, thus failing to effectively capture risk characteristics.

[0005] In a first aspect, embodiments of the present invention provide a battery thermal runaway early warning method, the method comprising:

[0006] Obtain a vehicle group feature library, wherein the vehicle group feature library includes a charging and discharging feature set of at least one vehicle and a status label of the vehicle, the status label being used to indicate whether the vehicle is a thermal runaway vehicle.

[0007] Select target charge / discharge features associated with the status tag from the charge / discharge feature set of each vehicle;

[0008] Based on the outlier degree between the target charging and discharging characteristics of the vehicle and the vehicle group characteristics in the vehicle group feature library, the anomaly score of the vehicle is calculated, and a risk warning operation is performed on the vehicle based on the anomaly score. The outlier degree is used to characterize the degree of deviation between the target charging and discharging characteristics of a single vehicle and the vehicle group characteristics in the vehicle group feature library in terms of distribution pattern.

[0009] Furthermore, the acquisition of the vehicle group feature database includes:

[0010] Thermal runaway events and multiple charge / discharge features are extracted from the historical charge / discharge data of each vehicle. These features include charging relaxation features, discharge fluctuation features, temperature accumulation features, and SOC usage features.

[0011] The vehicle is labeled based on the thermal runaway event to obtain a status label;

[0012] The vehicle group feature library is constructed based on the charging and discharging characteristics and status labels of each vehicle.

[0013] Furthermore, methods for extracting charging relaxation features include:

[0014] Extract the first voltage data from the historical charge and discharge data that is in the relaxation phase after the charging cycle;

[0015] Calculate the time constant sequence corresponding to each relaxation stage based on the first voltage data, and perform linear fitting on the time constant sequence corresponding to each relaxation stage to obtain the first fitting vector.

[0016] Obtain the first mean vector of the vehicle group feature library, and use the first mean vector to truncate the first fitted vector to obtain the relaxation vector;

[0017] The first quantile parameter is extracted from the relaxation vector according to a preset ratio range, and the first moment parameter of the relaxation vector is calculated.

[0018] The charging relaxation characteristics of the vehicle are generated using the first quantile parameter and the first moment parameter.

[0019] Furthermore, methods for extracting discharge wave characteristics include:

[0020] Extract the second voltage data that is in the discharge cycle from the historical charge and discharge data;

[0021] Calculate the single-cell entropy value sequence corresponding to each discharge cycle based on the second voltage data, and convert the single-cell entropy value sequence into a coefficient of variation sequence;

[0022] The second quantile parameter is extracted from the coefficient of variation sequence according to a preset ratio range, and the second moment parameter of the coefficient of variation sequence is calculated.

[0023] The discharge fluctuation characteristics of the vehicle are generated using the second quantile parameter and the second moment parameter.

[0024] Furthermore, the step of calculating the sequence of cell entropy values ​​corresponding to each discharge cycle based on the second voltage data includes:

[0025] Calculate the individual cell voltage deviation value corresponding to each discharge cycle based on the second voltage data;

[0026] The number of hit frames and the total number of frames in the discharge cycle are obtained for the single-cell voltage deviation value, and the probability value corresponding to the single-cell voltage deviation value is calculated based on the number of hit frames and the total number of frames.

[0027] The entropy value of a single cell is calculated based on the single cell voltage deviation value and the corresponding probability value, thus obtaining a sequence of single cell entropy values ​​for each discharge cycle.

[0028] Further methods for extracting temperature accumulation features include:

[0029] Extract the temperature data during the charging cycle from the historical charge and discharge data;

[0030] The first frame count accumulation sequence corresponding to each charging cycle is calculated based on the temperature data. The first frame count accumulation sequence is obtained by acquiring the first frame count in the temperature data where the battery temperature is lower than the temperature threshold for each charging cycle, and accumulating the first frame count up to the current charging cycle.

[0031] A linear fit is performed on the first frame number accumulation sequence corresponding to each charging cycle to obtain a second fitting vector, and the second fitting vector is used as the temperature accumulation feature of the vehicle.

[0032] Furthermore, extract the SOC data that is in the charging cycle from the historical charge and discharge data;

[0033] The second frame count accumulation sequence is calculated based on the SOC data for each SOC interval. The second frame count accumulation sequence is obtained by dividing the SOC data into multiple SOC intervals, and for each SOC interval, obtaining the number of second frames in the SOC data that are lower than the SOC value or lower than the SOC threshold, and accumulating the number of second frames up to the current charging cycle.

[0034] Linear fitting is performed on the second frame count accumulation sequence corresponding to each SOC interval to obtain a third fitting vector, and the second frame count accumulation sequence and the third fitting vector are used as the SOC usage features of the vehicle.

[0035] Furthermore, the step of filtering target charge / discharge features associated with the status tag from the charge / discharge feature set of each vehicle includes:

[0036] Obtain the correlation between each charging / discharging feature in the charging / discharging feature set of the vehicle and the status label;

[0037] Select target charging and discharging features from the vehicle's charging and discharging feature set that have a correlation with the status tag greater than or equal to a preset threshold.

[0038] Furthermore, the calculation of the vehicle's anomaly score based on the outlier degree between the vehicle's target charging / discharging characteristics and the vehicle group characteristics in the vehicle group feature database includes:

[0039] A corresponding histogram is constructed based on the target charging and discharging characteristics of each vehicle.

[0040] The number of samples whose target charging and discharging characteristics hit a preset equal-width box is counted in the histogram;

[0041] The outlier degree between the target charging / discharging feature and the vehicle group features in the vehicle group feature library is determined based on the number of samples, and the outlier degree corresponding to the target charging / discharging feature is fused to obtain the anomaly score corresponding to each vehicle.

[0042] In a second aspect, embodiments of the present invention provide a battery thermal runaway early warning device, the device comprising:

[0043] The acquisition module is used to acquire a vehicle group feature library, wherein the vehicle group feature library includes a charging and discharging feature set of at least one vehicle and a status tag of the vehicle, and the status tag is used to indicate whether the vehicle is a thermal runaway vehicle.

[0044] A filtering module is used to filter target charging and discharging features associated with the status tag from the charging and discharging feature set of each vehicle;

[0045] The calculation module is used to calculate the outlier score of the vehicle based on the outlier degree between the target charging and discharging characteristics of the vehicle and the vehicle group characteristics in the vehicle group feature library, and to perform risk warning operation on the vehicle based on the outlier score. The outlier degree is used to characterize the degree of deviation between the target charging and discharging characteristics of a single vehicle and the vehicle group characteristics in the vehicle group feature library in terms of distribution pattern.

[0046] Thirdly, embodiments of the present invention provide a computer device, including: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method described in the first aspect or any corresponding embodiment thereof.

[0047] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer instructions that cause a computer to perform the method described in the first aspect or any of its corresponding embodiments.

[0048] The method provided in this application has the following beneficial effects:

[0049] The method provided in this application acquires a vehicle group feature library containing charging and discharging characteristics and status tags of multiple vehicles. This enables the construction of a multi-dimensional feature system based on vehicle group-level data, overcoming the limitations of single-vehicle historical data and providing a more universal feature benchmark for risk assessment. By filtering target charging and discharging features strongly correlated with thermal runaway status tags from the charging and discharging feature set, redundant information can be effectively eliminated, focusing on key risk features and improving the mapping accuracy between features and thermal runaway status. Anomaly scores are calculated based on the outlier degree of single-vehicle target features and vehicle group features. This allows for the discovery of hidden anomaly features of individual vehicles within the group through horizontal comparison of vehicle group data, solving the problem that traditional models rely on single-vehicle historical trend analysis and cannot capture the non-progressive risk of sudden thermal runaway. Finally, risk warning operations are implemented based on the anomaly scores, achieving quantitative assessment and early identification of sudden thermal runaway risks without obvious individual cell voltage or SOC anomalies, providing a data-driven technical path for accurate early warning. Attached Figure Description

[0050] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0051] Figure 1 This is a schematic flowchart of a battery thermal runaway early warning method according to an embodiment of the present invention;

[0052] Figure 2 This is a flowchart illustrating another battery thermal runaway early warning method according to an embodiment of the present invention;

[0053] Figure 3 This is a structural block diagram of a battery thermal runaway early warning device according to an embodiment of the present invention;

[0054] Figure 4 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] According to embodiments of the present invention, a battery thermal runaway early warning method, apparatus, computer device, and storage medium are provided. It should be noted that the steps shown in the flowcharts in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0057] This embodiment provides a battery thermal runaway early warning method. Figure 1 This is a flowchart of a battery thermal runaway early warning method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0058] Step S101: Obtain the vehicle group feature library, wherein the vehicle group feature library includes a charging and discharging feature set of at least one vehicle and a status label of the vehicle, the status label being used to indicate whether the vehicle is a thermal runaway vehicle.

[0059] In this embodiment of the application, obtaining a vehicle group feature library includes: extracting thermal runaway events and multiple charging and discharging features from the historical charging and discharging data of each vehicle, wherein the charging and discharging features include charging relaxation features, discharging fluctuation features, temperature accumulation features, and SOC usage features; labeling vehicles according to thermal runaway events to obtain status labels; and constructing a vehicle group feature library based on the charging and discharging features and status labels corresponding to each vehicle.

[0060] It should be noted that thermal runaway events are extracted from the historical charge and discharge data of each vehicle. Simultaneously, charging relaxation features (calculating the time constant from the voltage data of the relaxation phase after the charging cycle, linearly fitting and extracting the first quantile parameter and the first moment parameter), discharge fluctuation features (calculating the cell entropy sequence from the voltage data of the discharge cycle, converting it into a coefficient of variation sequence and extracting the second quantile parameter and the second moment parameter), temperature accumulation features (calculating the first frame number accumulation sequence that meets the preset low temperature conditions from the temperature data of the charging cycle, linearly fitting the first frame number accumulation sequence to obtain the second fitted feature vector), and SOC usage features (dividing the SOC data of the charging cycle into intervals, calculating the second frame number accumulation sequence, linearly fitting to obtain the third fitted vector); and vehicle status labels are assigned based on the thermal runaway events (indicating whether it is a thermal runaway vehicle). Finally, a vehicle group feature library is constructed based on the charging and discharging feature set and status labels of each vehicle.

[0061] By extracting multi-dimensional charge and discharge features such as charging relaxation, discharge fluctuations, temperature accumulation, and SOC usage from historical charge and discharge data, and combining these with thermal runaway event status labels, a vehicle group feature library containing rich battery behavior features and fault state correlations can be constructed. This process ensures the comprehensiveness of the feature set and the accuracy of the labels, providing statistically significant benchmark data for subsequent thermal runaway risk analysis and solving the problem of insufficient early warning reliability caused by single features or ambiguous labels in existing technologies.

[0062] Step S102: Select target charging and discharging features associated with status tags from the charging and discharging feature set of each vehicle.

[0063] In this embodiment, the charge / discharge feature set refers to a collection of multi-dimensional features extracted from the vehicle during the charge / discharge process, such as charging relaxation features and discharge fluctuation features. The state label is used to indicate whether the vehicle is in thermal runaway mode. The screening process includes: firstly, calculating the correlation between each feature in the charge / discharge feature set and the state label using methods such as the maximum information coefficient (MIC). This correlation is used to quantify the degree of association between the feature and the thermal runaway state; then, setting a preset threshold, and filtering out features with a correlation greater than or equal to the threshold to form a target charge / discharge feature set, thereby achieving efficient screening of key features.

[0064] Step S103: Based on the outlier degree between the target charging and discharging characteristics of the vehicle and the vehicle group characteristics in the vehicle group feature library, calculate the vehicle's anomaly score, and perform risk warning operation on the vehicle based on the anomaly score. The outlier degree is used to characterize the degree of deviation between the target charging and discharging characteristics of a single vehicle and the vehicle group characteristics in the vehicle group feature library in terms of distribution pattern.

[0065] It should be noted that outlier degree refers to the degree of deviation between the target charging and discharging characteristics of an individual vehicle and the distribution patterns of vehicle features in the vehicle group feature library. It is used to quantify the difference between the battery charging and discharging behavior of the vehicle and the normal characteristics of the entire vehicle group. This can be achieved by constructing a unified histogram of the vehicle's target charging and discharging characteristics and similar features in the vehicle group feature library, and statistically analyzing the sample distribution of both in a preset equal-width bin. If the number of samples of a vehicle's target charging and discharging characteristics in a specific interval is significantly lower than the normal distribution of vehicle group features in that interval (i.e., it is at the tail end of the distribution or in a rare interval), it indicates a high degree of outlier, reflecting a significant deviation between the vehicle's battery behavior and the characteristic patterns of most normal vehicles. This degree of deviation is the core basis for subsequent calculation of anomaly scores and assessment of thermal runaway risk.

[0066] In this embodiment of the application, the method for calculating the anomaly score is as follows: First, the target charging and discharging features are constructed into a histogram. The number of samples hit by the target charging and discharging features is counted according to the preset equal-width bins. The outlier degree of each target charging and discharging feature is determined by the sample distribution. Then, the outlier degree of different target charging and discharging features is fused to obtain the anomaly score. This score quantifies the degree to which the vehicle features deviate from the normal state of the vehicle group.

[0067] Furthermore, the risk warning operation based on anomaly scores is implemented as follows: First, multiple warning thresholds are set (e.g., anomaly score ≥ 80 indicates high risk, 60 ≤ score < 80 indicates medium risk, and score < 60 indicates low risk). When the calculated anomaly score exceeds the corresponding threshold, different levels of warning mechanisms are triggered. Specifically, this includes: sending alarm signals to the vehicle management system, pushing risk alerts (including anomaly characteristic types and suggested measures) to users through the vehicle terminal, automatically activating real-time monitoring mode (encrypted data collection frequency) for high-risk vehicles, and synchronizing the anomaly data to the cloud server for secondary verification, forming a closed-loop warning process from feature outlier analysis, risk classification, multi-level warnings to real-time monitoring.

[0068] As an example, Figure 2 This is a flowchart illustrating another battery thermal runaway early warning method, such as... Figure 2As shown, the process includes: first, cleaning the raw data to obtain historical charging and discharging data of the vehicles; second, extracting user degradation features (i.e., charging relaxation, discharge fluctuation, temperature accumulation, and SOC usage) from the historical charging and discharging data, and labeling them with status tags based on thermal runaway events to construct a vehicle group feature library containing at least one set of charging and discharging features and status tags corresponding to each vehicle; then, filtering out target charging and discharging features with a correlation higher than a preset threshold through MIC correlation analysis; then, constructing a histogram based on the target features and calculating the probability density of each single feature (i.e., the target charging and discharging feature); determining the outlier degree of the vehicle based on the probability density, and calculating the vehicle's anomaly score based on the outlier degree; finally, triggering corresponding risk warning operations based on the anomaly score.

[0069] In this embodiment of the application, the method for extracting charging relaxation features includes the following steps A1-A5:

[0070] Step A1: Extract the first voltage data from the historical charge / discharge data that is in the relaxation phase after the charging cycle.

[0071] Specifically, historical charge / discharge data refers to data related to the charging and discharging process recorded during vehicle use; a charging cycle refers to a complete charging process; and the relaxation phase refers to the stage after charging is completed, when the battery terminal voltage is in a decreasing state, reflecting the dynamic changes inside the battery. The first voltage data is the sequence of battery terminal voltage measurements within this relaxation phase. The extraction method is as follows: based on the data acquisition time field, voltage data entering the relaxation phase after each charging cycle is selected from the historical charge / discharge data, and the voltage sample value of this phase is retained as the first voltage data.

[0072] Step A2: Calculate the time constant sequence corresponding to each relaxation stage based on the first voltage data, and perform linear fitting on the time constant sequence corresponding to each relaxation stage to obtain the first fitting vector.

[0073] Specifically, for the first voltage data sequence {U} in each relaxation stage t U t-1 The time constants at each moment of this stage are calculated using the following formula:

[0074]

[0075] Where, τ t U is the time constant at time t; t-1 U is the time constant at time (t-1); t Let t be the time constant at time t.

[0076] Finally, by combining the time constants at each moment in this stage, a time constant sequence {τ1,τ2,...,τ} is formed.n Subsequently, the time constant sequence for each relaxation stage is fitted using a linear function:

[0077] Y = aX + b

[0078] Where Y is a vector composed of time constants and X is the number of charging times. The slope a and intercept b are solved by fitting methods such as the least squares method, forming the first fitting vector (a,b), which represents the changing trend of the time constant during the relaxation stage.

[0079] Step A3: Obtain the first mean vector of the vehicle group feature library, and use the first mean vector to truncate the first fitted vector to obtain the relaxation vector.

[0080] Specifically, the first mean vector (ag, bg) is obtained from the vehicle group feature database, and the calculation formula is as follows:

[0081]

[0082] Where, ag is the average slope 'a' of the fitted curves across all vehicles in the charging cycle, i.e., the average slope in the first mean vector; bg is the average intercept 'b' of the fitted curves across all vehicles in the charging cycle, i.e., the average intercept in the first mean vector; N is the number of vehicles in the vehicle group; n is the number of charging cycles for vehicle i; a ij b is the slope of the first fitted vector in the j-th iteration of vehicle i; ij Let A be the intercept of the first fitted vector in the j-th cycle of vehicle i; A is the number of charging cycles of the vehicle group.

[0083] Then, the first fitted vector of vehicle i is truncated using the first mean vector (ag, bg) to obtain the relaxation vector (ar, br), according to the following rules:

[0084]

[0085] Among them, ar ij br represents the slope corresponding to the j-th cycle of vehicle i in the relaxation vector. ij Let a be the intercept of vehicle i in the j-th cycle of the relaxation vector; ag is the mean slope in the first mean vector; bg is the mean intercept of the first mean vector; a ij b is the slope of the first fitted vector in the j-th iteration of vehicle i; ij Let be the intercept of the first fitted vector in the j-th iteration of vehicle i.

[0086] Step A4: Extract the first quantile parameter from the relaxation vector according to the preset ratio range, and calculate the first moment parameter of the relaxation vector.

[0087] Specifically, firstly, the quantile values ​​of the relaxation vector (ar,br) are calculated according to a preset proportion range (e.g., 10%, 20%, ... 90%), and the first quantile parameter (e.g., the 10th quantile value of ar) is extracted. q10 90th percentile of br q90 wait).

[0088] Then, the first-order moment parameters (first skewness, second skewness, first kurtosis, second kurtosis) are calculated for the relaxed vector, where the formula for skewness calculation is:

[0089]

[0090] Among them, sa i The first skewness corresponds to the slope in the first-order moment parameters; sb i The second skewness corresponds to the intercept in the first-order moment parameters; n is the number of charging cycles for vehicle i; ar ij μ1 represents the slope of vehicle i in the j-th cycle of the relaxation vector. i σ1 is the mean slope of the relaxation vector; i Let be the standard deviation of the slope of the relaxation vector.

[0091] Similarly, the formula for calculating kurtosis is:

[0092]

[0093] Among them, ka i The first kurtosis corresponding to the slope in the first-order moment parameters; kb i The second kurtosis corresponds to the intercept in the first-order moment parameter; n is the number of charging cycles for vehicle i; br ij μ2 is the slope intercept corresponding to the j-th cycle of vehicle i in the relaxation vector. i σ² is the mean intercept of the relaxation vector; i Let be the standard deviation of the intercept of the relaxation vector.

[0094] Step A5: Generate the charging relaxation characteristics of the vehicle using the first quantile parameter and the first moment parameter.

[0095] In this embodiment of the application, the first quantile parameter (e.g., ar) of the preset proportion interval (e.g., 10%, 20%, ... 90%) extracted from the relaxation vector is used. q10 ar q20 , ...,ar q90 and br q10 , br q20 , ...,br q90 ), and the first moment parameter (first skewness sa) of the calculated relaxation vector. i sb i And the first peak ka i,kb i The features are integrated to form a multidimensional feature vector containing distribution and shape features, which is the charging relaxation feature of the vehicle.

[0096] The method for extracting charging relaxation features involves linearly fitting a time constant sequence, truncating the fitted vector using the vehicle group mean vector, and combining quantile and moment parameters to generate features. This effectively captures the voltage decay pattern and abnormal fluctuations of the battery during the charging relaxation phase. This process filters out noise interference from the vehicle group data, retains key parameters reflecting the battery's internal polarization characteristics, and makes the extracted features more representative of the battery's health status, providing a microscopic quantitative basis for identifying early signs of thermal runaway.

[0097] In this embodiment of the application, the method for extracting discharge fluctuation characteristics includes the following steps B1-B4:

[0098] Step B1: Extract the second voltage data that is in the discharge cycle from the historical charge and discharge data.

[0099] Specifically, a discharge cycle refers to a complete discharge process; the second voltage data is a sequence of voltage measurements for each individual cell within the discharge cycle. The extraction method is as follows: based on the data acquisition time field, the individual cell voltage data for each discharge cycle is selected from historical charge-discharge data and arranged in chronological order to form the second voltage data sequence. This sequence is used for subsequent calculations of discharge fluctuation characteristics.

[0100] Step B2: Calculate the cell entropy value sequence corresponding to each discharge cycle based on the second voltage data, and convert the cell entropy value sequence into a coefficient of variation sequence.

[0101] In this embodiment of the application, the sequence of cell entropy values ​​corresponding to each discharge cycle is calculated based on the second voltage data, including the following steps B21-B23:

[0102] Step B21: Calculate the individual cell voltage deviation value corresponding to each discharge cycle based on the second voltage data.

[0103] Specifically, for the voltage value U(n,t) of the nth cell in vehicle i at time t, calculate its value along with the median voltage U of all cells at the same time. med The formula for calculating the difference (t), i.e., the individual unit voltage deviation, is as follows:

[0104] dU(n,t)=U(n,t)-U med (t)

[0105] Where dU(n,t) is the voltage deviation of the nth cell in vehicle i at time t; U(n,t) is the voltage value of the nth cell in vehicle i at time t; U med(t) represents the median voltage of the nth cell in vehicle i at time t.

[0106] This deviation value is used to reflect the fluctuation of the individual cell voltage relative to the population during the discharge process, providing basic data for subsequent calculation of the individual cell entropy value.

[0107] Step B22: Obtain the number of hit frames of the individual cell voltage deviation value within the discharge cycle and the total number of frames in the discharge cycle, and calculate the probability value corresponding to the individual cell voltage deviation value based on the number of hit frames and the total number of frames.

[0108] Specifically, for each discharge cycle, the number of times the individual cell voltage deviation value dU(n,t) occurs is counted, i.e., the number of hit frames count(dU). i Simultaneously, the total number of time steps T (i.e., the total number of frames) for that discharge cycle is obtained. The probability value of each deviation value is calculated using the formula:

[0109]

[0110] Wherein, p(dU) i ) represents the individual unit voltage deviation value dU i The probability value that occurs during the discharge cycle; this probability value is used to calculate the subsequent single-cell entropy value; dU i For vehicle i, the voltage deviation value; count(dU) i ) represents the individual unit voltage deviation value dU i The number of hit frames within a discharge cycle; T is the total number of frames in the discharge cycle.

[0111] Step B23: Calculate the cell entropy value based on the cell voltage deviation value and the corresponding probability value to obtain the cell entropy value sequence for each discharge cycle.

[0112] Specifically, for each discharge cycle, all different individual cell voltage deviation values ​​dU are... i and its corresponding probability value p(dU) i Substituting into the entropy calculation formula, the single-unit entropy value for this period is calculated:

[0113]

[0114] Where E is the entropy value of a single cell in this discharge cycle; M is the number of all possible values ​​of the single cell voltage deviation value in this discharge cycle; p(dU) i ) represents the individual cell voltage deviation value for this discharge cycle; dU i This represents the voltage deviation value of a single unit.

[0115] The above calculations are performed sequentially on each individual cell, ultimately forming a sequence of individual cell entropy values ​​for each discharge cycle, E = [E1, E2, ..., E...]. n (n is the number of individual units).

[0116] The calculation of the single-cell entropy sequence transforms the voltage distribution characteristics within the discharge cycle into a thermodynamic entropy index by quantifying voltage deviation, hit frame count, and probability values. This index characterizes the degree of disorder in the battery's internal reactions. From an information theory perspective, it quantifies the uncertainty of voltage fluctuations, making implicit performance degradation during discharge explicit in the form of entropy changes. This provides theoretical support for the early detection of irreversible reactions within the battery and solves the problem that traditional voltage monitoring cannot capture microscopic reaction changes.

[0117] In this embodiment of the application, converting the monomer entropy value sequence into a coefficient of variation sequence includes:

[0118] First, for each discharge cycle, the entropy sequence of the individual cells is E = [E1, E2, ..., E...]. n Calculate its standard deviation σe and mean μe. Next, calculate the coefficient of variation using the following formula:

[0119]

[0120] Among them, C i σe is the coefficient of variation for this discharge cycle, representing the relative dispersion of the individual cell voltage fluctuations; μe is the standard deviation of the individual cell entropy value sequence for this discharge cycle; and μe is the mean of the individual cell entropy value sequence for this discharge cycle.

[0121] Iterate through all discharge cycles and set C for each cycle. i Arranged in order, forming a sequence of coefficients of variation [C1, C2, ..., C m (m is the number of discharge cycles).

[0122] Step B3: Extract the second quantile parameter from the coefficient of variation sequence according to the preset ratio range, and calculate the second moment parameter of the coefficient of variation sequence.

[0123] Specifically, the coefficient of variation sequence is sorted numerically, and quantiles are calculated according to a preset proportion range (e.g., 10%, 20%, ... 90%) to obtain the second quantile parameter (e.g., C). iq10 C iq20 C iq90 Secondly, the second-order moment parameters are calculated using the following formula:

[0124]

[0125] Among them, sc i kc is the third skewness of the coefficient of variation sequence in the second-order moment parameters, used to measure the symmetry of the sequence distribution; i is the third kurtosis of the coefficient of variation sequence in the second-order moment parameters, used to measure the peakedness of the sequence distribution; m is the length of the coefficient of variation sequence, C ijLet μ3 be the j-th coefficient of variation. i σ3 is the mean of the coefficient of variation sequence. i is the standard deviation of the coefficient of variation series.

[0126] Step B4: Generate the discharge fluctuation characteristics of the vehicle using the second quantile parameter and the second moment parameter.

[0127] Specifically, the quantile parameters (e.g., C) of a predetermined proportion interval (e.g., 10%, 20%...90%) extracted from the coefficient of variation sequence are used to determine the quantile parameters. iq10 C iq20 C iq90 ), and the second moment parameter (sc) of the calculated coefficient of variation sequence. i 、kc i The data are integrated to obtain the vehicle's discharge fluctuation characteristics, which characterize the distribution and dispersion of voltage fluctuations during the discharge process.

[0128] The extraction of discharge fluctuation characteristics involves calculating the entropy sequence of individual cells and converting it into a coefficient of variation sequence. This is then combined with quantile parameters and second-order moment parameters to generate features, quantifying the degree of voltage fluctuation dispersion during discharge. This method transforms the probability distribution of voltage deviation into a calculable statistic, highlighting the stability differences of batteries during discharge. It is particularly suitable for identifying abnormal voltage fluctuations caused by internal micro-short circuits or active material decay, improving the sensitivity to battery performance degradation before thermal runaway.

[0129] In this embodiment of the application, the method for extracting temperature accumulation features includes the following steps C1-C4:

[0130] Step C1: Extract temperature data from historical charge / discharge data during the charging cycle.

[0131] Specifically, the temperature data is a sequence of battery temperature measurements within a charging cycle. Based on the data acquisition time field, battery temperature measurements for each charging cycle are selected from historical charge and discharge data and arranged in chronological order to form a temperature data sequence. This sequence is used for subsequent calculations of temperature cumulative characteristics.

[0132] Step C2: Calculate the first frame count accumulation sequence corresponding to each charging cycle based on the temperature data. The first frame count accumulation sequence is obtained by acquiring the first frame count in the temperature data where the battery temperature is lower than the temperature threshold for each charging cycle, and accumulating the first frame count up to the current charging cycle.

[0133] Specifically, for each charging cycle, the number of frames where the battery temperature is below a temperature threshold (e.g., -10℃ or 0℃) is counted from the temperature data, i.e., the first frame number n. i Then, the number of the first frame is accumulated up to the current charging cycle to obtain the accumulated sequence of the first frame number. The accumulation formula is:

[0134]

[0135] Where CM is the accumulated value of the first frame; n1 i The number of frames in the i-th charging cycle where the battery temperature is below the temperature threshold; N is the number of charging cycles up to the current time.

[0136] For example, the accumulated value in the first charging cycle is n11, the accumulated value in the second charging cycle is n11+n12, and so on, forming a sequence [CM1,CM2,...,CM] that reflects the cumulative effect of temperature. N ].

[0137] Step C3: Perform linear fitting on the first frame number accumulation sequence corresponding to each charging cycle to obtain the second fitting vector, and use the second fitting vector as the temperature accumulation feature of the vehicle.

[0138] Specifically, the sequence of the first frame count corresponding to each charging cycle [CM1,CM2,...,CM] is accumulated. N The dependent variable is the number of charging cycles X = [1, 2, ..., N], and the independent variable is the number of charging cycles. A linear function is used for fitting.

[0139] Y = cX + d

[0140] Where Y is the first frame count accumulation sequence; X is the number of charging cycles; c is the slope of the second fitted curve; and d is the intercept of the second fitted curve.

[0141] The slope c and intercept d are solved by fitting methods such as least squares, forming a result from (c i ,d i The second fitting vector (c) is used to characterize the trend of the frame number accumulation sequence with the number of charging cycles, and the second fitting vector (c) is used to characterize the trend of the frame number accumulation sequence with the number of charging cycles. i ,d i This is a characteristic of vehicle temperature accumulation.

[0142] The method for extracting temperature accumulation features transforms the historical cumulative effect of battery temperature into a linearly fitted vector by accumulating the number of low-temperature frames during the charging cycle and performing linear fitting. This vector can characterize the temperature management efficiency of the battery during long-term operation. This method focuses on the cumulative impact of low-temperature conditions, avoiding interference from instantaneous high-temperature fluctuations. It is suitable for identifying temperature accumulation anomalies caused by aging of the heat dissipation system or long-term overcharging, providing time-dimensional temperature evolution features for thermal runaway risk assessment.

[0143] In this embodiment of the application, the feature extraction method used by SOC includes the following steps D1-D4:

[0144] Step D1: Extract SOC data from historical charge / discharge data that is in a charging cycle.

[0145] Specifically, SOC data is a sequence of measured values ​​of the state of charge (SOC) within a charging cycle. The extraction method is as follows: based on the data acquisition time field, SOC measurements for each charging cycle are selected from historical charge / discharge data and arranged chronologically to form an SOC data sequence. This sequence is used for subsequent calculations of SOC usage characteristics.

[0146] Step D2: Calculate the second frame count accumulation sequence corresponding to each SOC interval based on the SOC data. The second frame count accumulation sequence is obtained by dividing the SOC data into multiple SOC intervals, and for each SOC interval, obtaining the number of second frames in the SOC data that are below the SOC value or below the SOC threshold, and accumulating the number of second frames up to the current charging cycle.

[0147] Specifically, the SOC data is divided into intervals of 10% each from 0% to 100% (e.g., 0-10%, 10%-20%...90%-100%). For each SOC interval, the number of frames in each charging cycle where the SOC value is lower than the upper threshold of that interval (i.e., the second frame count) is counted and accumulated to the current charging cycle according to the formula, forming the second frame count accumulation sequence:

[0148]

[0149] Among them, S i This is the accumulated value for the second frame; n2 i is the number of frames in the i-th charging cycle where the SOC is lower than the corresponding interval threshold; N is the number of charging cycles up to the current time.

[0150] For example, the accumulated value of the first charging cycle is the number of frames below a certain threshold within that cycle; the accumulated value of the second charging cycle is the sum of the frame counts from the previous two cycles; and so on, to obtain the second frame count accumulation sequence [S]. i10 ,S i20 ,...,S i90 This sequence is used to characterize the cumulative usage habits of users in different SOC intervals.

[0151] Step D3: Perform linear fitting on the second frame number accumulation sequence corresponding to each SOC interval to obtain the third fitting vector, and use the second frame number accumulation sequence and the third fitting vector as the SOC usage features of the vehicle.

[0152] Specifically, the second frame count sequence corresponding to each SOC interval (e.g., 0-10%, 10%-20%, etc.) is accumulated [S]. i10 ,S i20 ,...,S i90The dependent variable is the number of charging cycles X = [1, 2, ..., N], and the independent variable is the number of charging cycles. A linear function is used for fitting.

[0153] Y = eX + f

[0154] Where Y is the cumulative sequence of the second frame count; X is the number of charging cycles; e is the slope in the third fitting vector; and f is the intercept in the third fitting vector.

[0155] Solving for the slope e and intercept f using the least squares method yields a result consisting of (e... i ,f i The third fitting vector, composed of (e) and (e) elements, is used to characterize the trend of the cumulative frame count sequence within the corresponding SOC interval as a function of the number of charging cycles. i ,f i ) and the second frame number accumulation sequence [S i10 ,S i20 ,...,S i90 These features are integrated into a multi-dimensional feature vector to form the vehicle's SOC usage characteristics, which characterize the cumulative changing trend of user habits within different SOC ranges.

[0156] The method for extracting State of Charge (SOC) usage features quantifies the impact of user habits on the battery under different states of charge by dividing SOC intervals and accumulating the number of frames below a threshold, combined with linear fitting to generate features. This method transforms SOC usage patterns into a statistically significant sequence of frames, reflecting the cumulative exposure of the battery in the low SOC interval. This helps identify battery degradation caused by frequent deep discharges or improper charging cut-off strategies, providing a user behavior-based analytical perspective for thermal runaway early warning.

[0157] In this embodiment of the application, the process of filtering target charging and discharging features associated with status tags from the charging and discharging feature set of each vehicle includes: obtaining the correlation between each charging and discharging feature in the charging and discharging feature set of the vehicle and the status tag; and filtering target charging and discharging features from the charging and discharging feature set of the vehicle whose correlation with the status tag is greater than or equal to a preset threshold.

[0158] Specifically, firstly, for each feature in the vehicle's charging and discharging feature set (including charging relaxation features, discharge fluctuation features, temperature accumulation features, and SOC usage features), the nonlinear correlation R between it and the state label (1 for thermal runaway vehicles and 0 for normal vehicles) is calculated using the maximum information coefficient (MIC). i The correlation value ranges from 0 to 1. Then, a preset threshold (such as R) is set. iIf the correlation of each feature is greater than or equal to a preset threshold (≥0.6), the correlation of each feature is compared with the preset threshold, and features with a correlation greater than or equal to the preset threshold are selected to form a target charge-discharge feature set. This process achieves the selection of key features by quantifying the correlation between features and thermal runaway state.

[0159] By calculating the correlation between charge / discharge characteristics and state tags and filtering highly correlated features, redundant features unrelated to thermal runaway can be eliminated, reducing feature dimensionality. This process focuses on key features in a data-driven manner, reducing the computational complexity of subsequent anomaly analysis and avoiding the risk of misjudgment introduced by irrelevant features. This makes the target charge / discharge feature set more targeted for thermal runaway early warning, improving the generalization ability and real-time performance of the early warning model.

[0160] In this embodiment of the application, the anomaly score of the vehicle is calculated based on the degree of outlier between the target charging and discharging characteristics of the vehicle and the vehicle group characteristics in the vehicle group feature database, including the following steps:

[0161] Step E1: Construct a corresponding histogram based on the target charging and discharging characteristics of each vehicle.

[0162] It should be noted that the target charge / discharge characteristics refer to features selected from the vehicle charge / discharge characteristic set whose correlation with the status label is greater than or equal to a preset threshold (e.g., 0.6). These features include charging relaxation characteristics, discharge fluctuation characteristics, temperature accumulation characteristics, and SOC usage characteristics, which have undergone correlation analysis. The histogram is used to display the distribution of the target charge / discharge characteristics within a numerical range. It is created by dividing the characteristic value range into several equal-width bins and counting the number of samples within each interval.

[0163] Specifically, for each target charge / discharge characteristic O i First, Z-score standardization is performed to obtain the standardized target charge-discharge characteristics Z. i The formula is:

[0164] Z i =(O i -u4 i ) / σ4 i

[0165] Among them, Z i The standardized target charge / discharge characteristics; O i Target charge / discharge characteristics; u4 i It is the mean of the target charge and discharge characteristics; σ4 i It is the standard deviation of the target charge and discharge characteristics.

[0166] Standardized target charge / discharge characteristics Z i Within the range of values, it is divided into k equal-width boxes (k is a preset parameter), and each box has the same width.

[0167] Step E2: Count the number of samples in the histogram that match the target charging and discharging features with the preset equal-width box.

[0168] Specifically, for each target charge / discharge feature, the bin range in which its value lies is determined based on the histogram. All target charge / discharge features are traversed, and the number of feature values ​​contained in each bin is counted, i.e., the number of samples that hit the preset equal-width bin.

[0169] Step E3: Determine the outlier degree between the target charging / discharging feature and the vehicle group feature in the vehicle group feature library based on the number of samples, and fuse the outlier degree corresponding to the target charging / discharging feature to obtain the anomaly score corresponding to each vehicle.

[0170] Specifically, for each target charging / discharging feature, the probability density pi (i.e., the ratio of the number of samples to the total number of samples) of that target charging / discharging feature is calculated based on the number of samples hitting the preset equal-width bin. The degree of outlier behavior is reflected through the logarithmic transformation of the probability density: outlier values ​​have lower probability densities, and their logarithm log(pi) is smaller (or more negative); taking a negative number amplifies the impact of outlier behavior. The outlier behavior (i.e., -log(pi)) of all target charging / discharging features is summed to obtain a comprehensive anomaly score. The higher this score, the greater the degree to which the vehicle characteristics deviate from the vehicle group, and the higher the risk of thermal runaway.

[0171] The formula for calculating the anomaly score for a vehicle is as follows:

[0172]

[0173] Where HBOS(pi) is the anomaly score of the current vehicle; m is the number of target charging and discharging features; and pi is the probability density of the target charging and discharging feature in the histogram.

[0174] By combining histogram statistics and outlier fusion to calculate anomaly scores, the outlier degree of a single feature can be transformed into a comprehensive quantitative anomaly score, enabling multi-dimensional risk assessment of battery status. The degree to which a feature deviates from the normal range of vehicles is quantified through the distribution of samples in an equal-width bin. Combined with a multi-feature fusion strategy, the limitations of misjudgment based on a single feature are avoided, making the anomaly score more comprehensively reflect the battery's thermal runaway risk level. This provides accurate quantitative basis for graded early warning and maintenance decisions.

[0175] This embodiment also provides a battery thermal runaway early warning device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0176] This embodiment provides a battery thermal runaway early warning device, such as... Figure 3 As shown, it includes:

[0177] The acquisition module 31 is used to acquire a vehicle group feature library, wherein the vehicle group feature library includes a charging and discharging feature set of at least one vehicle and a status label of the vehicle, and the status label is used to indicate whether the vehicle is a thermal runaway vehicle.

[0178] The filtering module 32 is used to filter target charging and discharging features associated with status tags from the charging and discharging feature set of each vehicle;

[0179] The calculation module 33 is used to calculate the outlier score of a vehicle based on the outlier degree between the target charging and discharging characteristics of the vehicle and the vehicle group characteristics in the vehicle group feature library, and to perform risk warning operation on the vehicle based on the outlier score. The outlier degree is used to characterize the degree of deviation between the target charging and discharging characteristics of a single vehicle and the vehicle group characteristics in the vehicle group feature library in terms of distribution pattern.

[0180] Furthermore, the acquisition module 31 is used to extract thermal runaway events and multiple charging and discharging features from the historical charging and discharging data of each vehicle. The charging and discharging features include charging relaxation features, discharging fluctuation features, temperature accumulation features, and SOC usage features. The vehicle is labeled according to the thermal runaway events to obtain a status label. A vehicle group feature library is constructed based on the charging and discharging features and status labels corresponding to each vehicle.

[0181] Furthermore, the acquisition module 31 also includes a first extraction submodule, a second extraction submodule, and a third extraction submodule;

[0182] The first extraction submodule is used to extract the first voltage data in the relaxation phase after the charging cycle from historical charging and discharging data; calculate the time constant sequence corresponding to each relaxation phase based on the first voltage data, and perform linear fitting on the time constant sequence corresponding to each relaxation phase to obtain a first fitting vector; obtain the first mean vector of the vehicle group feature library, and use the first mean vector to truncate the first fitting vector to obtain a relaxation vector; extract the first quantile parameter from the relaxation vector according to a preset ratio interval, and calculate the first moment parameter of the relaxation vector; and generate the charging relaxation features of the vehicle using the first quantile parameter and the first moment parameter.

[0183] The second extraction submodule is used to extract the second voltage data in the discharge cycle from the historical charge and discharge data; calculate the single-cell entropy value sequence corresponding to each discharge cycle based on the second voltage data, and convert the single-cell entropy value sequence into a coefficient of variation sequence; extract the second quantile parameter from the coefficient of variation sequence according to a preset ratio range, and calculate the second moment parameter of the coefficient of variation sequence; and generate the vehicle's discharge fluctuation characteristics using the second quantile parameter and the second moment parameter.

[0184] The third extraction submodule is used to extract temperature data in the charging cycle from historical charge and discharge data; calculate the first frame number accumulation sequence corresponding to each charging cycle based on the temperature data, wherein the first frame number accumulation sequence is obtained by acquiring the first frame number in the temperature data where the battery temperature is lower than the temperature threshold for each charging cycle, and accumulating the first frame number up to the current charging cycle; perform linear fitting on the first frame number accumulation sequence corresponding to each charging cycle to obtain the second fitting vector, and use the second fitting vector as the temperature accumulation feature of the vehicle.

[0185] The fourth extraction submodule is used to extract SOC data in the charging cycle from historical charging and discharging data; calculate the second frame number accumulation sequence corresponding to each SOC interval based on the SOC data. The second frame number accumulation sequence is obtained by dividing the SOC data into multiple SOC intervals, and for each SOC interval, obtaining the number of second frames below the SOC value and below the SOC threshold in the SOC data, and accumulating the number of second frames up to the current charging cycle. Linear fitting is performed on the second frame number accumulation sequence corresponding to each SOC interval to obtain a third fitting vector, and the second frame number accumulation sequence and the third fitting vector are used as the SOC usage features of the vehicle.

[0186] Furthermore, the second extraction submodule includes: a calculation unit, used to calculate the individual cell voltage deviation value corresponding to each discharge cycle based on the second voltage data; obtain the number of hit frames of the individual cell voltage deviation value in the discharge cycle and the total number of frames in the discharge cycle, and calculate the probability value corresponding to the individual cell voltage deviation value based on the number of hit frames and the total number of frames; calculate the individual cell entropy value based on the individual cell voltage deviation value and the corresponding probability value, and obtain the individual cell entropy value sequence for each discharge cycle.

[0187] Furthermore, the filtering module 32 is used to obtain the correlation between each charging and discharging feature in the vehicle's charging and discharging feature set and the status tag; and to filter out target charging and discharging features from the vehicle's charging and discharging feature set whose correlation with the status tag is greater than or equal to a preset threshold.

[0188] Furthermore, the calculation module 33 is used to construct a corresponding histogram based on the target charging and discharging features of each vehicle; to count the number of samples in the histogram that the target charging and discharging features hit the preset equal-width bins; to determine the outlier degree between the target charging and discharging features and the vehicle group features in the vehicle group feature library based on the number of samples; and to fuse the outlier degree corresponding to the target charging and discharging features to obtain the anomaly score corresponding to each vehicle.

[0189] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 4As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).

[0190] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0191] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0192] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0193] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0194] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0195] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0196] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for early warning of battery thermal runaway, characterized in that, The method includes: Obtain a vehicle group feature library, wherein the vehicle group feature library includes a charging and discharging feature set of at least one vehicle and a status label of the vehicle, the status label being used to indicate whether the vehicle is a thermal runaway vehicle. Select target charge / discharge features associated with the status tag from the charge / discharge feature set of each vehicle; Based on the outlier degree between the target charging and discharging characteristics of the vehicle and the vehicle group characteristics in the vehicle group feature library, the anomaly score of the vehicle is calculated, and a risk warning operation is performed on the vehicle based on the anomaly score. The outlier degree is used to characterize the degree of deviation between the target charging and discharging characteristics of a single vehicle and the vehicle group characteristics in the vehicle group feature library in terms of distribution pattern.

2. The method according to claim 1, characterized in that, The acquisition of the vehicle group feature database includes: Thermal runaway events and multiple charge / discharge features are extracted from the historical charge / discharge data of each vehicle. These features include charging relaxation features, discharge fluctuation features, temperature accumulation features, and SOC usage features. The vehicle is labeled based on the thermal runaway event to obtain a status label; The vehicle group feature library is constructed based on the charging and discharging characteristics and status labels of each vehicle.

3. The method according to claim 2, characterized in that, Methods for extracting charging relaxation features include: Extract the first voltage data from the historical charge and discharge data that is in the relaxation phase after the charging cycle; Calculate the time constant sequence corresponding to each relaxation stage based on the first voltage data, and perform linear fitting on the time constant sequence corresponding to each relaxation stage to obtain the first fitting vector. Obtain the first mean vector of the vehicle group feature library, and use the first mean vector to truncate the first fitted vector to obtain the relaxation vector; The first quantile parameter is extracted from the relaxation vector according to a preset ratio range, and the first moment parameter of the relaxation vector is calculated. The charging relaxation characteristics of the vehicle are generated using the first quantile parameter and the first moment parameter.

4. The method according to claim 2, characterized in that, Methods for extracting discharge wave characteristics include: Extract the second voltage data that is in the discharge cycle from the historical charge and discharge data; Calculate the single-cell entropy value sequence corresponding to each discharge cycle based on the second voltage data, and convert the single-cell entropy value sequence into a coefficient of variation sequence; The second quantile parameter is extracted from the coefficient of variation sequence according to a preset ratio range, and the second moment parameter of the coefficient of variation sequence is calculated. The discharge fluctuation characteristics of the vehicle are generated using the second quantile parameter and the second moment parameter.

5. The method according to claim 4, characterized in that, The step of calculating the sequence of cell entropy values ​​corresponding to each discharge cycle based on the second voltage data includes: Calculate the individual cell voltage deviation value corresponding to each discharge cycle based on the second voltage data; The number of hit frames and the total number of frames in the discharge cycle are obtained for the single-cell voltage deviation value, and the probability value corresponding to the single-cell voltage deviation value is calculated based on the number of hit frames and the total number of frames. The entropy value of a single cell is calculated based on the single cell voltage deviation value and the corresponding probability value, thus obtaining a sequence of single cell entropy values ​​for each discharge cycle.

6. The method according to claim 2, characterized in that, Methods for extracting temperature accumulation features include: Extract the temperature data during the charging cycle from the historical charge and discharge data; The first frame count accumulation sequence corresponding to each charging cycle is calculated based on the temperature data. The first frame count accumulation sequence is obtained by acquiring the first frame count in the temperature data where the battery temperature is lower than the temperature threshold for each charging cycle, and accumulating the first frame count up to the current charging cycle. A linear fit is performed on the first frame number accumulation sequence corresponding to each charging cycle to obtain a second fitting vector, and the second fitting vector is used as the temperature accumulation feature of the vehicle.

7. The method according to claim 2, characterized in that, The feature extraction methods used in SOC include: Extract the SOC data that is in the charging cycle from the historical charge and discharge data; The second frame count accumulation sequence is calculated based on the SOC data for each SOC interval. The second frame count accumulation sequence is obtained by dividing the SOC data into multiple SOC intervals, and for each SOC interval, obtaining the number of second frames in the SOC data that are lower than the SOC value or lower than the SOC threshold, and accumulating the number of second frames up to the current charging cycle. Linear fitting is performed on the second frame count accumulation sequence corresponding to each SOC interval to obtain a third fitting vector, and the second frame count accumulation sequence and the third fitting vector are used as the SOC usage features of the vehicle.

8. The method according to claim 1, characterized in that, The step of filtering target charge / discharge features associated with the status tag from the charge / discharge feature set of each vehicle includes: Obtain the correlation between each charging / discharging feature in the charging / discharging feature set of the vehicle and the status label; Select target charging and discharging features from the vehicle's charging and discharging feature set that have a correlation with the status tag greater than or equal to a preset threshold.

9. The method according to claim 1, characterized in that, The calculation of the vehicle's anomaly score based on the outlier degree between the vehicle's target charging / discharging characteristics and the vehicle group characteristics in the vehicle group feature database includes: A corresponding histogram is constructed based on the target charging and discharging characteristics of each vehicle. The number of samples whose target charging and discharging characteristics hit a preset equal-width box is counted in the histogram; The outlier degree between the target charging / discharging feature and the vehicle group features in the vehicle group feature library is determined based on the number of samples, and the outlier degree corresponding to the target charging / discharging feature is fused to obtain the anomaly score corresponding to each vehicle.

10. A battery thermal runaway early warning device, characterized in that, The device includes: The acquisition module is used to acquire a vehicle group feature library, wherein the vehicle group feature library includes a charging and discharging feature set of at least one vehicle and a status tag of the vehicle, and the status tag is used to indicate whether the vehicle is a thermal runaway vehicle. A filtering module is used to filter target charging and discharging features associated with the status tag from the charging and discharging feature set of each vehicle; The calculation module is used to calculate the outlier score of the vehicle based on the outlier degree between the target charging and discharging characteristics of the vehicle and the vehicle group characteristics in the vehicle group feature library, and to perform risk warning operation on the vehicle based on the outlier score. The outlier degree is used to characterize the degree of deviation between the target charging and discharging characteristics of a single vehicle and the vehicle group characteristics in the vehicle group feature library in terms of distribution pattern.

Images