Power battery thermal runaway early warning judgment method and device and vehicle

Abstract

The present application relates to power battery fault monitoring, in particular to a kind of power battery thermal runaway early warning judging method, device and vehicle, comprising: obtaining the running state data of battery in first preset time length, the first preset time length includes several sub preset time lengths;According to thermal runaway early warning model, the running state data collected in each sub preset time length is analyzed, and the corresponding several thermal runaway probabilities are obtained;Several thermal runaway probabilities are sorted according to time, and the sliding window parameter is set, the thermal runaway probability in first preset time length is analyzed by sliding window, whether each sliding window is warned is evaluated, and the warning number of battery is obtained by statistics;If warning number is greater than or equal to preset number, it is judged that battery is in sudden thermal runaway risk state, if warning number is less than preset number, it is judged that battery is in non-sudden thermal runaway risk state.It can realize the accurate early warning of sudden thermal runaway, improve the safety of power battery.

Description

TECHNICAL FIELD

[0001] The present application relates to power battery fault monitoring, in particular to a power battery thermal runaway early warning judgment method and device and vehicle. BACKGROUND

[0002] Developing a power battery thermal runaway early warning algorithm is a core measure to ensure the safety and reliability of the battery system. In recent years, overheat events of new energy vehicles have occurred frequently, causing a large amount of property loss every year. One type of overheat problem that plagues the entire industry is particularly difficult. From the existing inspection methods and collected signals, there is no obvious rule to follow before thermal runaway, which brings great challenges to monitoring and early warning. This special thermal runaway condition is defined as "sudden thermal runaway" in the industry.

[0003] In related technologies, the temperature, voltage and insulation resistance related data of the power battery are obtained through a big data platform, and after screening, they are used for neural network training to build a vehicle power battery safety prediction model. After the model is built, the current power battery safety related data of the vehicle connected to the big data platform is obtained, and the safety early warning result is output by substituting the model. However, this scheme does not consider the influence of the whole vehicle signal on the battery, and the sudden thermal runaway is difficult to accurately identify through the conventional voltage, temperature, insulation and other signals, which is easy to cause the problem of false negatives. SUMMARY

[0004] The purpose of the present application is to provide a power battery thermal runaway early warning judgment method, device and vehicle, which can realize accurate early warning of sudden thermal runaway and improve the safety of the power battery.

[0005] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0006] In a first aspect, the present application discloses a power battery thermal runaway early warning judgment method, which comprises:

[0007] obtaining the running state data of the battery within a first preset time period;

[0008] The first preset time period includes a plurality of sub-pre-set time periods, and the running state data collected in each sub-pre-set time period is analyzed by the thermal runaway early warning model to obtain a plurality of thermal runaway probabilities corresponding to the plurality of sub-pre-set time periods;

[0009] The plurality of thermal runaway probabilities are sorted by time, a sliding window parameter is set, the thermal runaway probabilities within the first preset time period are analyzed by the sliding window, it is evaluated whether each sliding window is early warned, and the early warning times of the battery are counted;

[0010] If the number of warnings is greater than or equal to the preset number, it is determined that the battery is in a sudden thermal runaway risk state; if the number of warnings is less than the preset number, it is determined that the battery is in a non-sudden thermal runaway risk state.

[0011] Furthermore, the determination of the preset number of times includes:

[0012] Obtaining operating status data of a plurality of abnormal samples within a first preset time period before the occurrence of sudden thermal runaway, wherein the abnormal samples are battery module samples determined to have experienced sudden thermal runaway;

[0013] Use the thermal runaway warning model to perform early warning analysis on the collected operating status data to obtain the number of warnings for each abnormal sample;

[0014] The minimum value of the warning times of all abnormal samples is used as the preset number.

[0015] Furthermore, the evaluation of whether each sliding window should issue an early warning specifically includes:

[0016] At least one of the calculated center position, distribution range, and fluctuation of the thermal runaway probability within each sliding window is used as an evaluation indicator. If the evaluation indicator is greater than or equal to a first preset evaluation threshold, a warning is issued; if the evaluation indicator is less than the first preset evaluation threshold, no warning is issued.

[0017] Or, based on y = ax 2 + b A quadratic curve is fitted to the probability of thermal runaway within the sliding window, where: y is the probability of thermal runaway, x 1~ c The data consists of a natural number multiplied by the scaling factor, c is the sliding window length, a and b is the fitting coefficient; if the fitting coefficient obtained is a If the fitting coefficient is greater than or equal to the second preset evaluation threshold, an early warning is issued; a If it is less than the second preset evaluation threshold, no warning will be issued.

[0018] Furthermore, the thermal runaway warning model is obtained by training a training sample set;

[0019] The training sample set includes a historical operating status data set of several normal samples accumulated within a second preset time period and a historical operating status data set of several abnormal samples accumulated within a third preset time period before sudden thermal runaway occurs.

[0020] Furthermore, the thermal runaway warning model is trained by the training sample set and specifically includes:

[0021] Obtain a training sample set and divide the collected historical operating status data set by time nodes to obtain several groups of data segments; the starting time node of each group of data segments is the time when charging starts, and the ending time node is the time when the next charging starts;

[0022] Perform data aggregation on each set of data segments to obtain a feature vector matrix, and set labels for the feature vector matrix according to the sample attributes corresponding to the feature vector matrix;

[0023] Combine the feature vector matrix with the label to generate a feature dataset;

[0024] The feature data sets corresponding to several normal samples and several abnormal samples are divided proportionally to obtain training sets and test sets, and the training sets are standardized;

[0025] The thermal runaway warning model is obtained after being trained with the training set and tested with the test set.

[0026] Furthermore, the data aggregation includes horizontal aggregation and vertical aggregation;

[0027] The horizontal aggregation includes: obtaining the maximum value, minimum value, mean value, median value and range of the operating parameters of each battery cell in the battery module within the data segment; the operating parameters include the voltage, temperature, current and insulation resistance of the battery cell;

[0028] The longitudinal aggregation includes: calculating the maximum value of the vehicle speed in the data segment and the operating parameters of all battery cells in the battery module x max , minimum value x min , mean x mean , median x med , effective value x rms , standard deviation x std , extreme x ffz , Difference Ratio x dr , skewness x skew and crest factor x crestf ;

[0029] The calculation formula of the effective value is: ;

[0030] The peak value is calculated as follows: ;

[0031] The calculation formula of the difference ratio is: ;

[0032] The calculation formula for the skewness is ;

[0033] The calculation formula of the peak factor is: ;

[0034] Where, X is the data vector before aggregation, i is the data vector X A single element in N is the number of data vectors before aggregation.

[0035] Further, in response to the battery cell operating parameters being voltage and temperature, the initial values ​​of the operating parameters in the data segment are also obtained. x beg , end value x end , working condition change x cha and the maximum value of working condition change x chamax ;

[0036] The calculation formula for the working condition change is: ;

[0037] The calculation formula for the maximum value of the working condition change is: , where diff ( X ) represents the data vector X Make a difference.

[0038] Furthermore, in response to the battery cell operating parameter being the current under the energy recovery condition, the proportion of the current under the energy recovery condition in different preset current intervals is also calculated.

[0039] Furthermore, each set of data segments includes a charging condition data set and a non-charging condition data set, wherein the non-charging condition data set includes a discharging condition data set and an energy recovery condition data set;

[0040] Data aggregation is performed on the charging condition dataset, the discharging condition dataset, and the energy recovery condition dataset respectively to obtain the eigenvector I corresponding to the charging condition dataset, the eigenvector II corresponding to the discharging condition dataset, and the eigenvector III corresponding to the energy recovery condition dataset;

[0041] The eigenvector I is used as the first row vector of the eigenvector matrix, and the eigenvector II and the eigenvector III are concatenated as the second row vector of the eigenvector matrix.

[0042] Furthermore, the thermal runaway warning model is a neural network algorithm model, a logistic regression classification algorithm model, a random forest classification algorithm model, a decision tree classification algorithm model, a support vector machine classification algorithm model, an XGBoost classification algorithm model or a Light GBM classification algorithm model;

[0043] The Bayesian search method is used to find the optimal values ​​of the hyperparameters of the thermal runaway warning model.

[0044] In a second aspect, the present invention discloses a power battery thermal runaway early warning and judgment device, which includes:

[0045] an acquiring unit, configured to acquire operating status data of the battery within a first preset time period, where the first preset time period includes a plurality of sub-preset time periods;

[0046] The prediction unit performs a warning analysis on the operating status data collected within each sub-preset time period according to the thermal runaway warning model to obtain a number of thermal runaway probabilities corresponding to the sub-preset time periods;

[0047] an analysis unit, configured to sort the thermal runaway probabilities by time, set sliding window parameters, perform sliding window analysis on the thermal runaway probabilities within a first preset time period, evaluate whether to issue an early warning for each sliding window, and obtain a statistical number of early warnings for the battery;

[0048] A comparison unit is used to compare the obtained number of warnings with a preset number. If the number of warnings is greater than or equal to the preset number, it is determined that the battery is in a sudden thermal runaway risk state; if the number of warnings is less than the preset number, it is determined that the battery is in a non-sudden thermal runaway risk state.

[0049] In a third aspect, the present invention discloses a vehicle comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the above-mentioned method for determining thermal runaway warning of a power battery.

[0050] The present invention has the following beneficial effects:

[0051] 1. The early warning judgment method described in the present invention divides the first preset time length into several sub-preset time lengths, and calculates the thermal runaway probability of each time period separately through the thermal runaway early warning model, thereby avoiding the limitations of single time dimension data. Then, the sliding window parameter is used to perform a rolling evaluation of the time-series thermal runaway probability, which can effectively filter out the interference caused by short-term fluctuations and focus on the continuity of risk trends. Through multi-period probability calculation and sliding window trend analysis, this method can capture the initial signal of sudden risk through multiple warning accumulations even if it is weak, thus avoiding safety accidents caused by the omission of a single indicator. In addition, by setting the preset number of times, unnecessary intervention triggered by a single abnormal data is avoided, and safety and economy are balanced.

[0052] 2. The preset number of times described in the present invention is determined by using the minimum value of the number of warnings of multiple abnormal samples within the first preset time period as the warning threshold, which avoids the subjectivity and limitations of manually setting the threshold in the traditional solution. Compared with the problem that the fixed manual threshold cannot be updated adaptively and it is difficult to achieve differentiated control, the preset number of times determined by the method described in the present invention, that is, the warning threshold, can more accurately reflect the warning law before the occurrence of sudden thermal runaway, so that the preset number setting is more in line with the actual scenario of the actual thermal runaway. In addition, since the preset number is derived from the minimum number of warnings before the thermal runaway of the abnormal sample, it means that in actual applications, when the number of warnings reaches the preset number, it can cover the warning signal before the sudden thermal runaway occurs to the greatest extent, effectively reducing the problem of missed reports caused by unreasonable settings of the preset number of times. Compared with the missed reports caused by the difficulty of accurately identifying sudden thermal runaway with conventional signals in the prior art, this method can significantly improve the warning sensitivity for sudden thermal runaway. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the implementation methods or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention.

[0054] Figure 1 The figure shows a flow chart of the method for early warning and judging thermal runaway of a power battery according to an embodiment of the present invention.

[0055] Figure 2 A schematic diagram of the process of determining the preset number of times according to an embodiment of the present invention is shown.

[0056] Figure 3 A flow chart illustrating an implementation of evaluating whether each sliding window should issue an early warning according to an embodiment of the present invention is shown.

[0057] Figure 4 A flow chart illustrating another implementation of evaluating whether each sliding window should issue an early warning according to an embodiment of the present invention is shown.

[0058] Figure 5 A schematic diagram of the training process of the thermal runaway warning model according to an embodiment of the present invention is shown.

[0059] Figure 6 The figure shows a schematic structural diagram of a power battery thermal runaway warning and judgment device according to an embodiment of the present invention.

[0060] Figure 7 A schematic diagram of the hardware entity of a vehicle provided in an embodiment of the present application. DETAILED DESCRIPTION

[0061] The following describes the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art will readily appreciate the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the various details in this specification may be modified or altered based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are intended only to illustrate the present invention and are not intended to limit the scope of protection of the present invention.

[0062] In one embodiment, the present invention provides a method for early warning and judging thermal runaway of a power battery, see Figure 1 As shown, the method includes:

[0063] Obtaining operating status data of the battery within a first preset time period;

[0064] The first preset time period includes a plurality of sub-preset time periods, and a warning analysis is performed on the operating status data collected within each sub-preset time period according to the thermal runaway warning model to obtain a plurality of thermal runaway probabilities corresponding to the plurality of sub-preset time periods;

[0065] Sort several thermal runaway probabilities by time, set sliding window parameters, perform sliding window analysis on the thermal runaway probabilities within a first preset time period, evaluate whether to issue an early warning for each sliding window, and calculate the number of early warnings for the battery;

[0066] If the number of warnings is greater than or equal to the preset number, it is determined that the battery is in a sudden thermal runaway risk state; if the number of warnings is less than the preset number, it is determined that the battery is in a non-sudden thermal runaway risk state.

[0067] The early warning judgment method described in the present invention divides the first preset time length into several sub-preset time lengths, and calculates the thermal runaway probability of each time period separately through the thermal runaway early warning model, thus avoiding the limitations of single time dimension data. Then, the sliding window parameters are used to perform a rolling evaluation of the time-series thermal runaway probability, which can effectively filter out the interference caused by short-term fluctuations and focus on the continuity of risk trends. Through multi-period probability calculation and sliding window trend analysis, this method can capture the initial signal of sudden risk through multiple early warning accumulations even if it is weak, thus avoiding safety accidents caused by the omission of a single indicator. And by setting the preset number of times, unnecessary intervention triggered by a single abnormal data is avoided, and a balance is achieved between safety and economy.

[0068] As a preferred embodiment of the present invention, see Figure 2 As shown, the determination of the preset number of times includes:

[0069] Obtain operating status data of a plurality of abnormal samples within a first preset time period before the occurrence of sudden thermal runaway, wherein the abnormal samples are battery module samples determined to have experienced sudden thermal runaway.

[0070] The thermal runaway warning model is used to perform early warning analysis on the collected operating status data to obtain the number of warnings for each abnormal sample.

[0071] The minimum value among the warning times is used as the preset number, that is, the minimum value among the warning times of all abnormal samples is used as the preset number.

[0072] In this preferred embodiment, the minimum number of warnings generated by multiple abnormal samples within a first preset time period is used as the warning threshold, avoiding the subjectivity and limitations of manually setting thresholds in traditional solutions. Compared to the problem of fixed manual thresholds that cannot be adaptively updated and difficult to achieve differentiated control, the preset number of times determined by the method described in this invention, i.e., the warning threshold, can more accurately reflect the warning patterns before sudden thermal runaway occurs, making the preset number setting more closely aligned with actual thermal runaway scenarios.

[0073] Furthermore, because the preset number of warnings is derived from the minimum number of warnings before thermal runaway in abnormal samples, this means that in practical applications, when the number of warnings reaches this preset number, it can provide the greatest coverage of warning signals before sudden thermal runaway, effectively reducing the problem of missed warnings caused by improperly set preset times. Compared to existing technologies where conventional signals cannot accurately identify sudden thermal runaway and thus result in missed warnings, this method significantly improves the warning sensitivity for sudden thermal runaway.

[0074] At the same time, this preferred implementation method can determine the threshold value by only analyzing the operating status data of abnormal samples and combining the warning times of the thermal runaway warning model. It does not require complex equipment investment or huge computing resources, meets the demand of solving problems at the lowest cost, and has strong practicality and economy while ensuring the warning effect.

[0075] As a preferred embodiment of the present invention, see Figure 3 As shown, the evaluation of whether each sliding window should issue an early warning specifically includes:

[0076] At least one of the calculated center position, distribution range, and fluctuation of the thermal runaway probability within each sliding window is used as an evaluation indicator. If the evaluation indicator is greater than or equal to a first preset evaluation threshold, an early warning is issued; if the evaluation indicator is less than the first preset evaluation threshold, no early warning is issued.

[0077] This preferred implementation uses central location, distribution range, and fluctuation as evaluation indicators, breaking through the limitations of single-dimensional data analysis. It explores abnormal characteristics from multiple perspectives, including probability concentration trend, dispersion, and change trend, enabling more comprehensive detection of thermal runaway precursors. For example, an abnormal distribution range may indicate the potential risk of localized battery overheating, while abnormal fluctuations may reflect unstable battery status. The coordinated evaluation of multiple indicators can improve the rationality of early warning triggering.

[0078] Exemplarily, the center position includes the median or mean of multiple thermal runaway probabilities within each sliding window, and the first evaluation threshold is set to 0.65. In response to the calculated median or mean exceeding the first evaluation threshold, an early warning is issued, otherwise no early warning is issued.

[0079] The distribution range includes the extreme values ​​and ranges of multiple thermal runaway probabilities within each sliding window. The extreme values ​​include the maximum and minimum values ​​of the thermal runaway probabilities within each sliding window, and the range is the difference between the maximum and minimum values ​​of the thermal runaway probabilities within each sliding window. The distribution range directly displays the boundaries of the data range and reflects the extreme differences in the data.

[0080] The fluctuation includes the standard deviation of multiple thermal runaway probabilities within each sliding window, which is used to measure the average dispersion of the data relative to the mean: the larger the standard deviation, the more dispersed the data, that is, the greater the fluctuation; conversely, the more concentrated the data, the smaller the fluctuation.

[0081] As a preferred embodiment of the present invention, see Figure 4 As shown, the evaluation of whether each sliding window should issue an early warning specifically includes: based on y = ax 2 + b A quadratic curve is fitted to the probability of thermal runaway within the sliding window, where: y is the probability of thermal runaway, x 1~ c The data consists of a natural number multiplied by the scaling factor, c is the sliding window length, a and b is the fitting coefficient; if the fitting coefficient obtained is a If the fitting coefficient is greater than or equal to the second preset evaluation threshold, an early warning is issued; a If it is less than the second preset evaluation threshold, no warning will be issued.

[0082] In this preferred embodiment, quadratic curve fitting is used y = ax 2 + b Model the probability of thermal runaway within the sliding window and convert discrete probability data into a continuous curve trend. Fitting coefficienta The curvature characteristics that can reflect probability changes. Compared with simple probability value judgment, curve fitting can more sensitively capture the subtle change trend of thermal runaway probability, identify the gradual process of battery status evolving towards thermal runaway in advance, and make up for the shortcomings of relying solely on single probability or simple statistical feature judgment.

[0083] Furthermore, the probability of thermal runaway y It is usually a proportional value calculated based on the model, such as a 0-1 interval or other probability value range. x It is composed of 1~ c A natural number multiplied by a scaling factor, if the sliding window length c Larger, x The value may be much greater than y If scaling is not performed, the dimension difference will cause the independent variable to x The dominant position of the curve shape is too strong, and the dependent variable y The changes in are masked, and the fitting results are likely to deviate from the real data trend. When fitting, the horizontal coordinate x and the vertical coordinate y are set to the same order of magnitude by setting the scaling coefficient. By scaling the two to unify the order of magnitude, the independent variable and the dependent variable can be balanced to fit the model. y = ax 2 + b The influence of a 、 b Can more accurately reflect the probability of thermal runaway y Follow x The real law of change can improve the accuracy of quadratic curve fitting. In addition, in the numerical calculation of curve fitting, if x and y Large differences in order of magnitude will lead to a deterioration in the condition number of the calculated matrix (too high correlation between the matrix row and column vectors), increase the difficulty of iterative convergence, and even lead to problems such as non-convergence and result oscillation. After scaling, the order of magnitude is unified, which can improve the numerical characteristics of the calculated matrix and make the fitting algorithm more stable and efficient in solving reasonable a and b , to ensure the reliability of the fitting process. Exemplarily, the scaling factor is 0.1.

[0084] At the same time, when x and y After the order of magnitude is unified, the fitting coefficient a The numerical value of can more intuitively reflect the rate of change and trend strength of the probability of thermal runaway. aCompared with the second preset evaluation threshold, the threshold setting does not need to consider the dimension conversion, and can be formulated based on the probability-position change law of the same order of magnitude, making the judgment logic of whether to issue an early warning more concise and reasonable, and reducing the risk of misjudgment of the threshold due to dimensional confusion. For example, the second preset evaluation threshold is set to 0.015, in response to the fitting coefficient a If the second preset assessment threshold is exceeded, an early warning is issued; otherwise, no early warning is issued.

[0085] In summary, the settings are as follows Figure 3 The sliding window indicator shown is compared with Figure 4 The curve fitting coefficient comparison shown here forms a redundant verification mechanism. Different paths, based on differentiated data processing logic, assess thermal runaway risk from different dimensions, reducing the probability of misjudgment or omission by a single algorithm.

[0086] Capturing data patterns that precede sudden thermal runaway is difficult. This preferred implementation, through sliding window analysis and dynamic time series segmentation, and curve fitting to identify trend changes, effectively adapts to complex data characteristics such as brief abnormal fluctuations and hidden trend changes that may occur before sudden thermal runaway. Compared to conventional static threshold judgments, this method can more accurately identify subtle anomalies before sudden runaway, addressing the industry's pain point of difficult early warning of sudden thermal runaway.

[0087] As a preferred embodiment of the present invention, the preset number itself can be dynamically updated based on newly accumulated thermal runaway cases. For example, if it is found that the number of warnings before thermal runaway of a batch of power batteries is too high, the preset number can be recalibrated to enable the solution to have continuous optimization capabilities. Specifically, by selecting different types of thermal runaway warning models, or adjusting the thermal runaway warning model parameters, or adjusting the first preset evaluation threshold and the preset second preset evaluation threshold, the preset number can be made as small as possible without reducing the F1 score, so as to reduce the disposal cost. The F1 score is the harmonic mean of precision and recall, and is used to evaluate the performance of the thermal runaway warning model.

[0088] The final output of the conclusion on whether there is a risk can be fed back to the model optimization link, forming a full-process closed loop of data collection, feature extraction, model warning, threshold judgment, and model iteration, and promoting the continuous upgrading of the warning judgment plan.

[0089] As a preferred embodiment of the present invention, the thermal runaway warning model is obtained by training a training sample set; the training sample set includes a historical operating status data set of several normal samples accumulated within a second preset time period and a historical operating status data set of several abnormal samples accumulated within a third preset time period before the occurrence of sudden thermal runaway.

[0090] This preferred embodiment incorporates the historical operating status data of normal samples and the data before thermal runaway of abnormal samples, allowing the model training to simultaneously learn the characteristics of the battery health status and the thermal runaway evolution status. Compared with training with only single state data, this dual-scenario data allows the model to accurately distinguish between normal fluctuations of the battery and thermal runaway precursors, build a more complete battery state feature space, and lay a solid data foundation for subsequent early warning analysis. Among them, the abnormal sample data deliberately selects the historical operating status data accumulated within the third preset time period before the sudden thermal runaway, and accurately anchors the key data interval of the sudden thermal runaway. During model training, it can deeply explore the subtle anomalies of the battery operating status before the sudden runaway, allowing the thermal runaway warning model to adapt to the pain points of sudden thermal runaway data that are difficult to capture regular patterns. Compared with the general thermal runaway model, it has a stronger early warning capability for difficult industry problems.

[0091] The trained thermal runaway warning model serves as the core tool for subsequently calculating the number of warnings for abnormal samples. Its training data covers both normal and abnormal scenarios, ensuring that the model's output of thermal runaway probabilities and warning determinations more closely reflects actual risks. When calculating the number of warnings for abnormal samples based on the thermal runaway warning model and further determining the warning threshold, the threshold effectively distinguishes between normal, risk-free conditions and abnormal, risk-indicating conditions. This ensures a self-consistent logic and a more rigorous data closed-loop for the entire warning threshold determination and application system.

[0092] As a preferred embodiment of the present invention, see Figure 5 As shown, the thermal runaway warning model is obtained by training a training sample set and specifically includes the following steps.

[0093] A training sample set is obtained, and the collected historical operating status data set is divided according to time nodes to obtain several groups of data fragments; the starting time node of each group of data fragments is the time when charging starts, and the ending time node is the time when charging starts next time.

[0094] Data aggregation is performed on each group of data segments to obtain a feature vector matrix, and labels are set for the feature vector matrix according to the sample attributes corresponding to the feature vector matrix.

[0095] Combine the feature vector matrix with the labels to generate a feature dataset.

[0096] The feature data sets corresponding to several normal samples and several abnormal samples are divided proportionally to obtain training sets and test sets, and the training sets are standardized.

[0097] The thermal runaway warning model is obtained after being trained with the training set and tested with the test set.

[0098] In this preferred embodiment, the data segments are divided into "the moment when charging starts - the moment when charging starts next" to accurately match the actual usage rhythm of the power battery. In the charge and discharge cycle of the power battery, the state changes of the operating parameters have periodic characteristics. Cutting the data according to the charging cycle can fully retain the state evolution within a charge and discharge cycle, allowing the thermal runaway warning model to learn the battery operation rules that are more in line with the actual working conditions. Compared with random cutting or fixed-time cutting, it can capture the state mutations of key nodes such as charging start and discharge end, and improve the data's restoration of the real behavior of the battery. Each data segment obtained by cutting corresponds to an independent charge and discharge cycle, which can avoid mixed interference of data between different cycles. For example, after a certain charge, the high temperature of the battery has not fully recovered. If the data is merged across cycles, the model will misjudge that high temperature is the normal state of the next cycle. After dividing by charging cycle, the state characteristics of each cycle are independent and clear. During model training, the evolution of thermal runaway risks within a single charge and discharge cycle can be learned more accurately, especially the feature capture of sudden thermal runaway is more accurate.

[0099] Each data segment is then aggregated to generate a feature vector matrix. This transforms dispersed time-series operational data, such as continuous temperature and voltage sequences, into structured features that represent the battery state during the charge and discharge cycle. This data aggregation preserves key status indicators within the charge and discharge cycle while compressing redundant data, making the model input more concise and representative. Compared to directly inputting raw time series data, the feature vector matrix improves model training efficiency while strengthening the weighting of features that are valuable for thermal runaway risk.

[0100] At the same time, labels are set according to sample attributes (normal / abnormal) to form positive samples corresponding to abnormal samples and negative samples corresponding to normal samples, providing a clear training target for supervised learning. During model training, the difference boundary between normal cycle features and abnormal cycle features can be accurately learned by pairing the eigenvector matrix + label. For example, the label of the abnormal sample can be associated with the cycle in which thermal runaway occurs, allowing the model to focus on learning which eigenvectors in the cycle indicate thermal runaway. Compared with unsupervised learning or weakly supervised learning, the model's ability to identify thermal runaway risks is greatly improved, providing a more reliable algorithm foundation for subsequent early warnings. For example, the eigenvector matrix corresponding to the abnormal sample, i.e., the sample that sends thermal runaway, is set to label 1, and the eigenvector matrix corresponding to the normal sample is set to label 0.

[0101] Proportionally dividing the training set and test set, and then standardizing the training set, is a classic method for optimizing the machine learning model training process. Dividing the training set and test set allows for verification of model generalization through a closed-loop training-testing process, preventing overfitting. Standardization eliminates dimensional differences between feature dimensions, resulting in a more balanced impact of each feature on the loss function during model training, improving training efficiency and model stability. For example, the training set and test set are divided in a 7:3 ratio. The training set is standardized using the Z-score. Given the relatively small number of abnormal samples, resulting in a significant difference in the ratio of positive and negative samples, the training set is augmented using an adaptive synthetic sampling method using oversampling technology, resulting in a 1:1 ratio of positive and negative samples.

[0102] Furthermore, through an iterative process of training with a training set and validating with a test set, model parameters can be continuously adjusted, continuously optimizing the accuracy of the thermal runaway warning model in identifying thermal runaway risks. Independent validation of the test set promptly identifies flaws in the model's performance on unknown data, driving iterative model upgrades. The resulting thermal runaway warning model is more reliable in actual warning applications and better adapted to the complex operating conditions of power batteries.

[0103] As a preferred embodiment of the present invention, the data aggregation includes horizontal aggregation and vertical aggregation;

[0104] The horizontal aggregation includes: obtaining the maximum value, minimum value, mean value, median value and range of the operating parameters of each battery cell in the battery module within the data segment; the operating parameters include the voltage, temperature, current and insulation resistance of the battery cell;

[0105] The longitudinal aggregation includes: calculating the maximum value of the vehicle speed in the data segment and the operating parameters of all battery cells in the battery module x max , minimum value x min , mean x mean , median x med , effective value x rms , standard deviation x std , extreme x ffz , Difference Ratio x dr , skewness x skew and crest factor x crestf ;

[0106] The calculation formula of the effective value is: ;

[0107] The peak value is calculated as follows: ;

[0108] The calculation formula of the difference ratio is: ;

[0109] The calculation formula for the skewness is ;

[0110] The calculation formula of the peak factor is: ;

[0111] Where, X is the data vector before aggregation, i is the data vector X A single element in N is the number of data vectors before aggregation.

[0112] In this preferred embodiment, the horizontal aggregation can accurately capture the differences between cells. By extracting the maximum, minimum, mean, median and range of the voltage, temperature, current and insulation resistance of each battery cell in the battery module, the state fluctuations of the single cell during the charge and discharge cycle can be accurately characterized. For example, a large voltage range may reflect abnormal internal resistance of the single cell, such as local overheating, a precursor to thermal runaway, which will cause a sudden voltage change; a high temperature mean can be an early warning of the potential thermal runaway risk of the single cell. Compared with the solution that ignores cell differences and only focuses on the module as a whole, horizontal aggregation allows the model to identify the risk of overall thermal runaway caused by a cell abnormality in the module, solving the pain point of local abnormalities being difficult to capture in sudden thermal runaway.

[0113] The vertical aggregation can globally control the module characteristics, calculate the extreme value, mean, median, effective value, standard deviation, range, difference ratio, skewness, peak factor and other indicators of vehicle speed and single cell parameters for the entire battery module, and build a module-level multi-dimensional feature space. For example: effective value The energy equivalent value of a parameter can identify hidden fluctuations in current and voltage. Skewness measures the asymmetry of the data distribution. Abnormal skewness in the temperature distribution may indicate localized heating prior to thermal runaway. The peak factor highlights the impact of extreme values ​​on the overall situation and is adapted to the transient peak anomalies of sudden thermal runaway. By vertically aggregating battery cell characteristics to the battery module level, the thermal runaway warning model can assess thermal runaway risks from a global perspective, avoiding missed detections caused by focusing solely on cells and ignoring synergistic effects.

[0114] As a preferred embodiment of the present invention, in response to the battery cell operating parameters being voltage and temperature, the initial value of the operating parameter in the data segment is also obtained. x beg , end value x end, working condition change x cha and the maximum value of working condition change x chamax ;

[0115] The calculation formula for the working condition change is: ;

[0116] The calculation formula for the maximum value of the working condition change is: , where diff ( X ) represents the data vector X Make a difference.

[0117] This preferred embodiment adds initial and final values ​​for voltage and temperature to characterize the difference between the beginning and end of a charge and discharge cycle. For example, if the initial temperature value in a cycle is 30°C and the final value is 40°C, the heating rate can be calculated based on the duration, identifying the risk of thermal runaway caused by a slow temperature rise.

[0118] The difference between the initial and final voltage values ​​can indicate abnormal changes in the battery's SOC. For example, if the voltage fails to return to a reasonable range after discharge, it could indicate an internal short circuit. Compared to focusing solely on intra-cycle statistics (such as mean and range), the initial and final values ​​provide additional state anchors at the start and end of the cycle, allowing the model to perceive the "start-to-end" evolution of the battery's state.

[0119] The operating condition variation reflects the overall fluctuation of the parameter, while the maximum operating condition variation focuses on local extreme mutations. For example, when the vehicle accelerates suddenly, the current differential increases instantly, corresponding to a sudden temperature change. The combination of these two measures not only covers slow changes but also captures sudden changes, adapting to the state fluctuations of the power battery under complex operating conditions (acceleration, climbing, and sudden braking).

[0120] In horizontal aggregation, the initial and final values ​​of battery cell voltage and temperature can refine the state differences of battery cells within a cycle. In vertical aggregation, the module-level operating condition variation and maximum operating condition variation can enhance the battery module's overall response to operating condition fluctuations. With this addition, the feature system has been upgraded from statistical coverage to a combination of full-cycle status and dynamic trends, providing more complete input for thermal runaway warning models.

[0121] As a preferred embodiment of the present invention, in response to the battery cell operating parameter being the current under the energy recovery condition, the proportion of the current under the energy recovery condition in different preset current intervals is further calculated.

[0122] During energy recuperation conditions (such as reverse charging during vehicle braking or coasting), the battery current direction is opposite to that of conventional charging and discharging, and the current characteristics differ significantly from those under conventional conditions. The instantaneous increase in current during energy recuperation can easily lead to short-term thermal stress concentration. If the current is improperly controlled, this can cause battery overcharging and sudden temperature rise, posing a risk of thermal runaway. This preferred embodiment analyzes the current distribution during energy recuperation conditions separately, precisely focusing on the risks associated with this specific condition and avoiding being overwhelmed by data from conventional charging and discharging conditions. By calculating the proportion of the energy recuperation current in different preset ranges, the current distribution pattern under these conditions can be quantified. For example, if the high energy recuperation current range (e.g., >20A) accounts for 30% of the cycle, this indicates that the battery is frequently experiencing high current recuperation, posing a high risk of thermal runaway. If the proportion is concentrated in the low range (e.g., <5A), the impact of energy recuperation on the battery is minimal, and the risk is low. Compared to focusing solely on the mean and peak current values, the range proportions provide a more nuanced picture of the ongoing distribution of the energy recuperation current, identifying hidden risks such as long-term, high-frequency, high-current recuperation.

[0123] As a preferred embodiment of the present invention, each group of data segments includes a charging condition data set and a non-charging condition data set, and the non-charging condition data set includes a discharging condition data set and an energy recovery condition data set; data aggregation is performed on the charging condition data set, the discharging condition data set, and the energy recovery condition data set respectively to obtain an eigenvector I corresponding to the charging condition data set, an eigenvector II corresponding to the discharging condition data set, and an eigenvector III corresponding to the energy recovery condition data set; the eigenvector I is used as the first row vector of the eigenvector matrix, and the eigenvector II and the eigenvector III are concatenated as the second row vector of the eigenvector matrix.

[0124] In this preferred embodiment, the data segments are split into three types of working condition data sets: charging, discharging, and energy recovery, to accurately match the core operating scenarios of the power battery. Among them, under the charging condition, the power battery passively accepts electrical energy, and the current and voltage changes have specific rules, such as constant current charging and constant voltage charging stages). Under the discharging condition, the power battery actively releases electrical energy and needs to match the vehicle's power requirements, such as large current discharge during acceleration. Under the energy recovery condition, that is, reverse charging during braking / coasting, the current direction and characteristics are significantly different from conventional charging. Compared with mixed working condition processing, the battery status characteristics under each working condition can be captured separately after stratification, avoiding the problem of mixing characteristics of different working conditions and confusion in model learning. These three types of working conditions cover the main operating modes of power batteries, namely charging energy storage, discharging output, and energy recovery. As a characteristic working condition of new energy vehicles, energy recovery can enhance the risk identification of this scenario after being split separately.

[0125] This preferred embodiment uses eigenvector I as the first row of the eigenvector matrix, and concatenates eigenvectors II and III as the second row. The first row independently reflects the static, fundamental characteristics of the charging condition, while the second row integrates the dynamic output and recovery characteristics of discharge and energy recovery, strengthening the hierarchical associations between operating conditions. Discharge and energy recovery are both non-charging scenarios, and their concatenation allows the model to learn their synergistic risks. Furthermore, the matrix structure enhances the layered nature of the feature system, adapting to the model's need to learn about the synergistic risks of multiple operating conditions.

[0126] As a preferred embodiment of the present invention, the thermal runaway warning model is a neural network algorithm model, a logistic regression classification algorithm model, a random forest classification algorithm model, a decision tree classification algorithm model, a support vector machine classification algorithm model, an XGBoost (i.e., extreme gradient boosting) classification algorithm model or a LightGBM (i.e., lightweight gradient boosting machine) classification algorithm model; and a Bayesian search method is used to find the optimal values ​​of the hyperparameters of the thermal runaway warning model.

[0127] In one embodiment, the present invention provides a power battery thermal runaway early warning judgment device, see Figure 6 As shown, the early warning judgment device 10 includes an acquisition unit 11, a prediction unit 12, an analysis unit 13 and a comparison unit 14.

[0128] The acquisition unit 11 is used to acquire the operating status data of the battery within a first preset time period, where the first preset time period includes a plurality of sub-preset time periods.

[0129] The prediction unit 12 performs a warning analysis on the operating status data collected within each sub-preset time period according to the thermal runaway warning model to obtain a plurality of thermal runaway probabilities corresponding to the plurality of sub-preset time periods;

[0130] The analysis unit 13 is configured to sort the thermal runaway probabilities by time, set sliding window parameters, perform sliding window analysis on the thermal runaway probabilities within a first preset time period, evaluate whether to issue an early warning for each sliding window, and obtain a statistical number of early warnings for the battery;

[0131] The comparison unit 14 is used to compare the obtained number of warnings with a preset number. If the number of warnings is greater than or equal to the preset number, it is determined that the battery is in a sudden thermal runaway risk state; if the number of warnings is less than the preset number, it is determined that the battery is in a non-sudden thermal runaway risk state.

[0132] In one embodiment, the present invention provides a vehicle comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the above-mentioned method for determining thermal runaway warning of a power battery.

[0133] like Figure 7 As shown, the hardware entity of the vehicle 20 includes: a processor 21, a memory 22 and a communication interface 23, wherein:

[0134] The processor 21 generally controls the overall operation of the electronic device 20 .

[0135] The memory 22 is configured to store instructions and applications executable by the processor 21 , and may also cache data to be processed or already processed by the processor 21 and various modules in the electronic device 20 .

[0136] The communication interface 23 enables the electronic device to communicate with other terminals or servers through a network.

[0137] Data can be transmitted between the processor 21 , the memory 22 and the communication interface 23 via the bus 24 .

[0138] It should be noted that the vehicle may be, but is not limited to, a pure electric vehicle (Pure Electric Vehicle / Battery Electric Vehicle, PEV / BEV), a hybrid electric vehicle (Hybrid Electric Vehicle, HEV), a range extended electric vehicle (Range Extended Electric Vehicle, REEV), a plug-in hybrid electric vehicle (Plug-in Hybrid Electric Vehicle, PHEV), etc.

[0139] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0140] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0141] In addition, all functional units in the embodiments of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above-mentioned integrated units can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0142] Those skilled in the art will understand that all or part of the steps of the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, and other media that can store program codes.

[0143] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling an electronic device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: various media that can store program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0144] The above embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Any equivalent substitution or modification made by those skilled in the art based on the present invention is within the protection scope of the present invention.

Claims

1. A method for early warning and judging thermal runaway of a power battery, characterized in that: include: Obtaining operating status data of the battery within a first preset time period; The first preset time period includes a plurality of sub-preset time periods, and a warning analysis is performed on the operating status data collected within each sub-preset time period according to the thermal runaway warning model to obtain a plurality of thermal runaway probabilities corresponding to the plurality of sub-preset time periods; Sort several thermal runaway probabilities by time, set sliding window parameters, perform sliding window analysis on the thermal runaway probabilities within a first preset time period, evaluate whether to issue an early warning for each sliding window, and calculate the number of early warnings for the battery; If the number of warnings is greater than or equal to the preset number, the battery is judged to be in a state of sudden thermal runaway risk; if the number of warnings is less than the preset number, the battery is judged to be in a state of non-sudden thermal runaway risk; Determining the preset number of times includes: Obtaining operating status data of a plurality of abnormal samples within a first preset time period before the occurrence of sudden thermal runaway, wherein the abnormal samples are battery module samples determined to have experienced sudden thermal runaway; Use the thermal runaway warning model to perform early warning analysis on the collected operating status data to obtain the number of warnings for each abnormal sample; The minimum value of the warning times of all abnormal samples is used as the preset number.

2. The power battery thermal runaway early warning and judgment method according to claim 1, characterized in that: The evaluation of whether each sliding window issues an early warning specifically includes: At least one of the calculated center position, distribution range, and fluctuation of the thermal runaway probability within each sliding window is used as an evaluation indicator. If the evaluation indicator is greater than or equal to a first preset evaluation threshold, a warning is issued; if the evaluation indicator is less than the first preset evaluation threshold, no warning is issued. Or, based on y = ax 2 + b A quadratic curve is fitted to the probability of thermal runaway within the sliding window, where: y is the probability of thermal runaway, x 1~ c The data consists of a natural number multiplied by the scaling factor, c is the sliding window length, a and b is the fitting coefficient; if the fitting coefficient obtained is a If the fitting coefficient is greater than or equal to the second preset evaluation threshold, an early warning is issued; a If it is less than the second preset evaluation threshold, no warning will be issued.

3. The power battery thermal runaway early warning and judgment method according to claim 1, characterized in that: The thermal runaway warning model is obtained by training a training sample set; The training sample set includes a historical operating status data set of several normal samples accumulated within a second preset time period and a historical operating status data set of several abnormal samples accumulated within a third preset time period before sudden thermal runaway occurs.

4. The power battery thermal runaway early warning and judgment method according to claim 3, characterized in that: The thermal runaway warning model is trained by the training sample set and specifically includes: Obtain a training sample set and divide the collected historical operating status data set by time nodes to obtain several groups of data segments; the starting time node of each group of data segments is the time when charging starts, and the ending time node is the time when the next charging starts; Perform data aggregation on each set of data segments to obtain a feature vector matrix, and set labels for the feature vector matrix according to the sample attributes corresponding to the feature vector matrix; Combine the feature vector matrix with the label to generate a feature dataset; The feature data sets corresponding to several normal samples and several abnormal samples are divided proportionally to obtain training sets and test sets, and the training sets are standardized; The thermal runaway warning model is obtained after being trained with the training set and tested with the test set.

5. The power battery thermal runaway early warning and judgment method according to claim 4, characterized in that: The data aggregation includes horizontal aggregation and vertical aggregation; The horizontal aggregation includes: obtaining the maximum value, minimum value, mean value, median value and range of the operating parameters of each battery cell in the battery module within the data segment; the operating parameters include the voltage, temperature, current and insulation resistance of the battery cell; The longitudinal aggregation includes: calculating the maximum value of the vehicle speed in the data segment and the operating parameters of all battery cells in the battery module x max , minimum value x min , mean x mean , median x med , effective value x rms , standard deviation x std , extreme x ffz , Difference Ratio x dr , skewness x skew and crest factor x crestf ; The calculation formula of the effective value is: ; The peak value is calculated as follows: ; The calculation formula of the difference ratio is: ; The calculation formula for the skewness is ; The calculation formula of the peak factor is: ; Where, X is the data vector before aggregation, i is the data vector X A single element in N is the number of data vectors before aggregation.

6. The power battery thermal runaway early warning and judgment method according to claim 5, characterized in that: In response to the battery cell operating parameters being voltage and temperature; also obtaining the initial values ​​of the operating parameters in the data segment x beg , end value x end , working condition change x cha and the maximum value of the working condition change x chamax ; The calculation formula for the working condition change is: ; The calculation formula for the maximum value of the working condition change is: , where diff ( X ) represents the data vector X Make a difference.

7. The power battery thermal runaway early warning and judgment method according to claim 5, characterized in that: In response to the battery cell operating parameter being the current under the energy recovery operating condition, proportions of the current under the energy recovery operating condition in different preset current intervals are also calculated.

8. The power battery thermal runaway early warning and judgment method according to claim 4, characterized in that: Each set of data segments includes a charging condition data set and a non-charging condition data set, wherein the non-charging condition data set includes a discharging condition data set and an energy recovery condition data set; Data aggregation is performed on the charging condition dataset, the discharging condition dataset, and the energy recovery condition dataset respectively to obtain the eigenvector I corresponding to the charging condition dataset, the eigenvector II corresponding to the discharging condition dataset, and the eigenvector III corresponding to the energy recovery condition dataset; The eigenvector I is used as the first row vector of the eigenvector matrix, and the eigenvector II and the eigenvector III are concatenated as the second row vector of the eigenvector matrix.

9. The power battery thermal runaway early warning and judgment method according to claim 1, characterized in that: The thermal runaway warning model is a neural network algorithm model, a logistic regression classification algorithm model, a random forest classification algorithm model, a decision tree classification algorithm model, a support vector machine classification algorithm model, an XGBoost classification algorithm model or a Light GBM classification algorithm model; The Bayesian search method is used to find the optimal values ​​of the hyperparameters of the thermal runaway warning model.

10. A power battery thermal runaway early warning and judgment device, characterized in that: include: an acquiring unit, configured to acquire operating status data of the battery within a first preset time period, where the first preset time period includes a plurality of sub-preset time periods; The prediction unit performs a warning analysis on the operating status data collected within each sub-preset time period according to the thermal runaway warning model to obtain a number of thermal runaway probabilities corresponding to the sub-preset time periods; an analysis unit, configured to sort the thermal runaway probabilities by time, set sliding window parameters, perform sliding window analysis on the thermal runaway probabilities within a first preset time period, evaluate whether to issue an early warning for each sliding window, and obtain a statistical number of early warnings for the battery; A comparison unit is used to compare the obtained number of warnings with a preset number. If the number of warnings is greater than or equal to the preset number, it is determined that the battery is in a sudden thermal runaway risk state; if the number of warnings is less than the preset number, it is determined that the battery is in a non-sudden thermal runaway risk state.

11. A vehicle, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the power battery thermal runaway warning judgment method according to any one of claims 1 to 9.

Images