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

Abstract

The invention discloses a battery thermal runaway early warning model generation method and device, equipment and a storage medium. The generation method of the battery thermal runaway early warning model comprises the following steps: acquiring real vehicle data of a plurality of batteries; and extracting battery monomer voltage data meeting a preset condition from the real vehicle data. And sorting the voltage data of the single batteries to obtain sorting time sequence change data of the voltage of the single batteries. And calculating a plurality of time sequence characteristics based on the sorting time sequence change data of the battery monomers. And splicing the plurality of time sequence characteristics obtained by calculation according to the sequence of the battery monomers to form a characteristic data matrix. And inputting the characteristic data matrix into a model for training, and generating a battery thermal runaway early warning model. The battery thermal runaway early warning model can improve the determination accuracy of battery thermal runaway, and improve the stability and accuracy of thermal runaway determination, thereby improving the safety of the battery.

Description

Technical Field

[0001] The present invention relates to the technical field of battery equipment, and in particular to a method, device, equipment and storage medium for generating a battery thermal runaway warning model. Background Art

[0002] The number of new energy vehicles (NEVs) on the market is rapidly increasing. Batteries, as core components of NEVs, have a direct impact on vehicle performance and lifespan. Due to factors such as temperature, charge and discharge rates, and cell consistency, battery anomalies can occur during use, impacting the performance of the entire battery pack. In severe cases, thermal runaway can occur, resulting in property damage and personal injury. Therefore, developing early warning methods for NEV batteries experiencing thermal runaway, and promptly identifying and addressing batteries at potential thermal runaway risk, is crucial to ensuring the safe operation of NEVs.

[0003] In related technologies, battery thermal runaway faults are diagnosed based on threshold-based judgments. By calculating statistical indicators of relevant battery parameters, corresponding thermal runaway thresholds are set, and batteries that exceed the thermal runaway thresholds are determined to be at risk. However, this threshold-based approach to battery thermal runaway judgment is unstable, and thresholds are difficult to determine. These thresholds are often set manually, which lacks scientific basis. This results in low accuracy in battery thermal runaway judgments and reduces battery safety. Currently, a single algorithm model is established based on thermal runaway labels, and battery parameter characteristics are input for training to produce an early warning model. However, this early warning model is a single model, and the extraction of battery cell information is relatively one-sided, reducing the accuracy of battery thermal runaway judgments and, consequently, the safety of the battery system. Summary of the Invention

[0004] The main purpose of the present invention is to provide a method, device, equipment and storage medium for generating a battery thermal runaway warning model, aiming to solve the technical problem that the threshold-based method for determining battery thermal runaway is very unstable, resulting in low accuracy in battery thermal runaway determination.

[0005] In order to achieve the above-mentioned object of the invention, the present invention proposes a method for generating a battery thermal runaway warning model, comprising the following steps:

[0006] Obtain real vehicle data for multiple batteries;

[0007] Extracting battery cell voltage data that meets preset conditions from the actual vehicle data;

[0008] Sorting the battery cell voltage data to obtain sorting time series change data of the battery cell voltage;

[0009] Calculating a plurality of time series features based on the time series variation data of the battery cells;

[0010] The multiple calculated time series features are spliced ​​according to the order of the battery cells to form a feature data matrix;

[0011] The characteristic data matrix is ​​input into the model training to generate a battery thermal runaway warning model.

[0012] In one embodiment, the calculation of the plurality of timing features includes at least two of the following:

[0013] Absolute sum of first-order differences, approximate entropy, Benford correlation, timing complexity, bucket entropy, permutation information entropy, average of absolute change values, and proportion less than a preset voltage threshold.

[0014] In one embodiment, the step of calculating the approximate entropy based on the time series variation data of the battery cells comprises:

[0015] A vector sequence of dimension m is formed by sequence number. in, X(i) is the i-th value of the sequence;

[0016] Calculating vectors and The Chebyshev distance of Among them, x(i+k) is a vector The kth value in x(j+k) is a vector The kth value in

[0017] calculate and Distance between The number of j (1≤j≤N-m+1) less than or equal to the threshold r, the number is c i , and calculate the approximate ratio

[0018] Define φ m,r for,

[0019] Increase the dimension to m+1 and repeat the above steps to get φ m+1,r ;

[0020] Calculate the approximate entropy A p En(m,r)=φ m,r -φ m+1,r .

[0021] In one embodiment, the step of calculating bucket entropy based on the sorting time series change data of the battery cells includes:

[0022] Divide [min(X),max(X)] into multiple intervals;

[0023] distributing the values ​​of the sorted time series change data in the plurality of intervals;

[0024] Calculating bucket entropy Among them, p k It represents the probability that the value of sorted data X falls into the kth bucket, maxbin represents the number of buckets, and len(X) represents the sequence length.

[0025] In one embodiment, the step of calculating the permutation information entropy based on the time series variation data of the battery cell ranking includes:

[0026] Setting an embedding dimension m and a step size t, reconstructing the order of the battery cells to obtain I subsequences, where I = n-(m-1)t;

[0027] Convert each of the subsequences into a permutation of size relationships;

[0028] Calculate the probability p of each size relationship arrangement, where the probability p is equal to the number of times the arrangement is divided by 1;

[0029] Calculate the permutation information entropy P of the probability e En=-∑ m！ plog(p).

[0030] In one embodiment, after the step of splicing the calculated multiple time series features according to the order of the battery cells to form a feature data matrix, the following steps are included:

[0031] Delete feature columns in the feature data matrix that are greater than a missing rate threshold and less than a variance threshold.

[0032] In one embodiment, after the step of deleting the feature columns in the feature data matrix that are greater than a missing rate threshold and less than a variance threshold, the step includes:

[0033] Initialize the feature columns that are less than or equal to the missing rate threshold and greater than or equal to the variance threshold, let S be the condition set, and let S = φ, where the feature columns that are less than or equal to the missing rate threshold and greater than or equal to the variance threshold are the feature columns obtained by processing the non-thermal runaway battery cell samples in the actual vehicle data;

[0034] Based on each instance (x i ,y i ), calculate x i The distance between each instance in S;

[0035] Find the nearest neighbor (x j ,y j );

[0036] If y i ≠yj , and (x i ,y i ) is added to S;

[0037] When all datasets T can be correctly classified using instances in S, the algorithm terminates and the dataset used for training is obtained.

[0038] In one embodiment, the step of inputting the characteristic data matrix into model training to generate a battery thermal runaway warning model includes:

[0039] Split the feature data matrix into training and test sets;

[0040] Using multiple models to train the training set and obtain the optimal prediction result probability for each model;

[0041] Setting a two-layer neural network having an input layer and an output layer, wherein the input layer receives the prediction result probabilities of the multiple models, and the output layer generates a final prediction label;

[0042] Training the neural network using the actual labels of the feature data matrix to obtain optimal weights;

[0043] Applying the optimal weights to an ensemble learning model;

[0044] After the learning model is trained, iterative optimization and verification are performed using the test set until the learning model is stable, thereby generating a battery thermal runaway warning model.

[0045] In one embodiment, the preset condition includes one or any combination of the following:

[0046] The voltage of the battery cell is within a preset voltage range;

[0047] The current of the battery cell is within a preset current range;

[0048] The temperature of the battery cell is within a preset temperature range;

[0049] The state of charge of the battery cell is within a preset range.

[0050] A second aspect of the present invention provides a device for generating a battery thermal runaway warning model, comprising:

[0051] Acquisition module, used to obtain real vehicle data of multiple batteries;

[0052] An extraction module, configured to extract battery cell voltage data that meets preset conditions from the actual vehicle data;

[0053] a sorting module, configured to sort the battery cell voltage data to obtain sorting time sequence change data of the battery cell voltage;

[0054] A calculation module, configured to calculate a plurality of time series features based on the time series variation data of the battery cells;

[0055] A splicing module is used to splice the multiple calculated time series features according to the order of the battery cells to form a feature data matrix;

[0056] The training module is used to input the characteristic data matrix into the model training to generate a battery thermal runaway warning model.

[0057] A third aspect of the present invention provides a battery thermal runaway early warning method, comprising the following steps:

[0058] Obtain real-time data of the battery cells of the vehicle to be tested;

[0059] Importing the real-time data into the above-mentioned battery thermal runaway warning model;

[0060] Based on the battery thermal runaway warning model, it is determined whether a battery cell of the vehicle to be inspected has thermal runaway.

[0061] In a fourth aspect, the present invention provides a computer device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for generating a battery thermal runaway warning model as described above is implemented.

[0062] In a fifth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements the steps of the method for generating a battery thermal runaway warning model as described above.

[0063] Beneficial effects:

[0064] The method for generating a battery thermal runaway warning model of the present invention includes the following steps: obtaining real-world vehicle data of multiple batteries; extracting battery cell voltage data that meets preset conditions from the real-world vehicle data; sorting the battery cell voltage data to obtain sorted time-series variation data of the battery cell voltages; calculating multiple time-series features based on the sorted time-series variation data of the battery cells; concatenating the calculated multiple time-series features according to the sorting of the battery cells to form a feature data matrix; and inputting the feature data matrix into model training to generate a battery thermal runaway warning model. The battery thermal runaway warning model designs multiple time-series features to effectively extract the intrinsic characteristics of the battery cell voltage sorting data, including time-series complexity, nonlinearity, periodicity, and discreteness, and comprehensively extracts battery cell voltage information, thereby improving the accuracy of battery thermal runaway determination, the stability and accuracy of thermal runaway determination, and thus improving battery safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 The figure is a flow chart of a method for generating a battery thermal runaway warning model according to an embodiment of the present invention.

[0066] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0067] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0068] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0069] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections, direct connections, or indirect connections through an intermediate medium; they may refer to internal communication between two components or the interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0070] In the present invention, unless otherwise expressly specified or limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features being in contact not directly but through another feature between them. Furthermore, a first feature being "above," "above," and "above" a second feature may include the first feature being directly above or obliquely above the second feature, or may simply mean that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature may include the first feature being directly below or obliquely below the second feature, or may simply mean that the first feature is lower in level than the second feature.

[0071] like Figure 1 As shown, in some embodiments, a method for generating a battery thermal runaway warning model includes the following steps:

[0072] S100: Acquire actual vehicle data of multiple batteries.

[0073] S200: extracting battery cell voltage data that meets preset conditions from actual vehicle data.

[0074] S300 , sorting battery cell voltage data to obtain sorting time sequence change data of battery cell voltages.

[0075] S400 , calculating multiple time series features based on the time series variation data of the battery cells.

[0076] S500 , combining the multiple calculated time series features according to the order of the battery cells to form a feature data matrix.

[0077] S600: Input the feature data matrix into the model training to generate a battery thermal runaway warning model. This battery thermal runaway warning model incorporates multiple time series features to effectively extract the intrinsic characteristics of battery cell voltage sequencing data, including time series complexity, nonlinearity, periodicity, and discreteness. This comprehensive extraction of battery cell voltage information improves the accuracy of battery thermal runaway determination, as well as the stability and accuracy of thermal runaway determination, thereby enhancing battery safety.

[0078] In some embodiments, such as S100 described above: obtaining real vehicle data of multiple batteries. Obtaining real vehicle driving data, that is, historical operating data of new energy vehicles on the market, needs to include the following relevant parameter fields: frame number, time, mileage, voltage (total voltage), current, temperature, SOC, battery cell voltage, and battery cell temperature. At the same time, based on the results of the disassembly and judgment of the thermal runaway vehicle, training label annotation is performed, that is, which battery cell of each thermal runaway vehicle has thermal runaway, and the battery cell sample is labeled as 1, representing a positive example, and the remaining battery cell samples are labeled as 0, representing a negative example. Specifically, for failed vehicles, historical data before the time node of thermal runaway occurrence should be selected to capture data change characteristics.

[0079] Specifically, a battery cell is a basic unit that makes up a battery. A battery contains multiple battery cells that are connected in series to form a battery.

[0080] Specifically, electric vehicles are typically equipped with a battery management system (BMS), which monitors and manages various battery parameters. During vehicle operation, the voltage, current, temperature, and SOC (State of Charge) of multiple batteries continuously change, reflecting the real-time status and performance of the batteries. The BMS collects this data through sensors and stores it in the vehicle's memory or transmits it via the CAN bus. Acquiring real-time data from multiple batteries involves leveraging the vehicle's own data acquisition and transmission systems to extract this data, reflecting the actual operating conditions of the batteries, for subsequent analysis and processing. A battery pack consists of multiple battery cells, and the status of each cell impacts the performance and safety of the entire pack. The BMS collects real-time data from each cell, such as voltage and temperature. This data is transmitted to the vehicle's central controller via the CAN bus. External devices then retrieve this real-time data from the central controller using a specific UDS protocol, thus enabling the collection of real-time data from multiple batteries.

[0081] This step provides a comprehensive and authentic data source for subsequent model training. The accuracy of the battery thermal runaway warning model relies on a large amount of real-world data. By acquiring real-world data from multiple batteries, we can cover the various battery states under different operating conditions, thereby improving the model's accuracy.

[0082] In some embodiments, as in S200 above, battery cell voltage data meeting preset conditions is extracted from the actual vehicle data. The preset conditions include one or any combination of the following: the battery cell voltage is within a preset voltage range; the battery cell current is within a preset current range; the battery cell temperature is within a preset temperature range; and the battery cell state of charge is within a preset range.

[0083] Real-world vehicle data typically contains a wealth of information, not all of which is directly relevant to battery thermal runaway warnings. By filtering battery cell voltage data using preset conditions, we can focus on data potentially related to thermal runaway risks, eliminate interfering data, and improve the efficiency and specificity of data processing. During vehicle operation, battery parameters can fluctuate due to a variety of factors, but not all fluctuations indicate a thermal runaway risk. By setting preset conditions, we can filter out the data that truly requires attention, reduce unnecessary data processing, and make subsequent analysis more efficient.

[0084] Specifically, this step processes abnormal values ​​and erroneous values ​​in the acquired real vehicle data. The criterion for determining abnormal and erroneous values ​​is whether the parameter conforms to the specified value range. In this embodiment, the value range of the battery cell voltage is 2V to 5V, the value range of the battery cell temperature is -40°C to 100°C, the value range of the battery cell current is -500A to 500A, the value range of the battery cell SOC is 0% to 100%, and the value range of the battery cell total voltage is determined by multiplying the value range of the battery cell voltage by the corresponding number of battery cell strings. Data outside this data range is discarded.

[0085] In some embodiments, as described above at S300 : sorting the battery cell voltage data to obtain sorting time sequence change data of the battery cell voltages.

[0086] After the data is cleaned, the battery cell voltages of each vehicle are sorted. The battery cell voltage values ​​for each frame are sorted from smallest to largest, with the smallest value assigned a ranking value of 1 and the largest value assigned a ranking value corresponding to the number of battery cell strings. This setup results in as many ranking values ​​as there are frames of data per day. The mode of the values ​​for each battery cell string is then grouped by day, and this is used as the daily ranking result for that battery cell. This ultimately yields daily time-series ranking data for the battery cells.

[0087] It should be noted that ordered time-series change data is more conducive to the calculation and analysis of various time-series features. In subsequent steps, multiple time-series features, such as voltage change rate, voltage fluctuation amplitude, etc., need to be calculated based on these data. The sorted data makes these calculations more logical and accurate. When calculating the voltage change rate, since the data is arranged in chronological order, it is convenient to subtract the voltage at the previous moment from the voltage at the next moment, and then divide it by the time interval to obtain the accurate voltage change rate. This calculation based on ordered data can more accurately reflect the dynamic change characteristics of battery cell voltage and provide a data basis for establishing an accurate battery thermal runaway early warning model.

[0088] In some embodiments, as in S400 above, multiple time series features are calculated based on the time series variation data of the battery cells.

[0089] Specifically, the calculation of multiple time series features includes at least two of the following:

[0090] Absolute sum of first-order differences, approximate entropy, Benford correlation, timing complexity, bucket entropy, permutation information entropy, average of absolute change values, and proportion less than a preset voltage threshold.

[0091] In some embodiments, the absolute sum of first-order differences is Where x is the sort value and n represents the number of days.

[0092] In some embodiments, S210, the step of calculating the approximate entropy based on the time series variation data of the battery cell ranking, includes:

[0093] S211. Form a vector sequence of dimension m according to the sequence number. in, X(i) is the i-th value of the sequence.

[0094] S212, calculation vector and That is, the maximum absolute value of the difference between the numerical values ​​of each element, Among them, x(i+k) is a vector The kth value in x(j+k) is a vector The kth value in .

[0095] S213, calculation and Distance between The number of j (1≤j≤N-m+1) less than or equal to the threshold r is c i , and calculate the approximate ratio

[0096] S214. Define φ m,r for,

[0097] S215, increase the dimension to m+1, repeat the above steps to get φ m+1,r .

[0098] S216. Calculate approximate entropy A p En(m,r)=φ m,r -φ m+1,r The larger the difference in approximate entropy, the more irregular and complex the sequence. The smaller the difference in approximate entropy, the more regular the sequence.

[0099] In some embodiments, S220, the step of calculating bucket entropy based on the sorting time series change data of the battery cells, includes:

[0100] S221. Bucket the values ​​of the sorted data X, and divide [min(X), max(X)] into multiple intervals, specifically 10 intervals.

[0101] S222. Disperse the values ​​of the sorted time-series change data into multiple intervals. Based on the equally spaced buckets, the entropy of the probability distribution can be calculated.

[0102] S223. Calculate bucket entropy Among them, p k It represents the probability that the value of sorted data X falls into the kth bucket, maxbin represents the number of buckets, and len(X) represents the sequence length.

[0103] In some embodiments, S230, the step of calculating the permutation information entropy based on the time series variation data of the battery cell ranking, includes:

[0104] S231 . Set an embedding dimension m and a step size t, reconstruct the order of the battery cells, and obtain I subsequences, where I=n-(m-1)t.

[0105] S232. Convert each subsequence into a permutation of size relationship.

[0106] S233. Calculate the probability p of each size relationship arrangement, where the probability p is equal to the number of arrangements divided by 1.

[0107] S234. Calculate the permutation information entropy P of the probability e En=-∑ m！ plog(p).

[0108] In some embodiments, the time series feature further includes Benford correlation. The distribution formula of Benford's law for the first digit d (d ranges from 1 to 9) is: P(d) = log 10 (d+1)-log 10 (d), record it as the expected frequency, then calculate the actual frequency of the first digit in the data set, and finally calculate the Pearson correlation coefficient between the expected and actual frequencies.

[0109] In some embodiments, the timing characteristics also include timing complexity. The calculation formula for timing complexity can be

[0110] In some embodiments, the timing feature further includes a proportion of cells less than a certain battery cell voltage threshold. For example, the battery cell voltage threshold is set to t, and the proportion of the sorted data X less than t is calculated.

[0111] In some embodiments, the time series feature also includes an average value of the absolute change value. The calculation formula for the average value of the absolute change value is:

[0112] In some embodiments, the time series features further include conventional statistical indicators, such as minimum value, maximum value, average value, median, variance, and coefficient of variation.

[0113] In some embodiments, as in the above S500 : the multiple calculated time series features are spliced ​​according to the order of the battery cells to form a feature data matrix.

[0114] It should be noted that for each battery cell, its different timing features are arranged in sequence to form a row of data. These data from all battery cells are combined to form a two-dimensional feature data matrix. For example, assuming there are a battery cells and b timing features are calculated for each battery cell, the size of the resulting feature data matrix is ​​a × b. The previously dispersed multiple timing features for each battery cell are integrated into a unified matrix structure. This structured data format facilitates subsequent data processing, storage, and analysis, improving data organization and readability.

[0115] In some embodiments, after the step of S500, combining the calculated multiple time series features according to the order of the battery cells to form a feature data matrix, the following steps are included:

[0116] S600. Delete feature columns in the feature data matrix that have a missing rate greater than the threshold and a variance less than the threshold. That is, delete feature columns with a missing rate greater than the threshold and less than the variance threshold and exclude them from model training. By deleting feature columns with a missing rate greater than the threshold, the problem of incomplete information caused by missing data can be avoided, making the data used for model training more accurate and reliable. Removing feature columns with a variance less than the threshold can eliminate features that cannot distinguish between positive and negative samples, select the most relevant and useful features, and thus improve model performance, reduce computational complexity, and improve model training efficiency and generalization ability.

[0117] In some embodiments, after S600, the step of deleting feature columns with a value greater than a missing rate threshold and less than a variance threshold in the feature data matrix, the step includes:

[0118] S610: Initialize the feature columns that are less than or equal to the missing rate threshold and greater than or equal to the variance threshold. Let S be the condition set, and let S = φ, indicating that the initial value is set to an empty set. The feature columns that are less than or equal to the missing rate threshold and greater than or equal to the variance threshold are the feature columns obtained by processing the non-thermal runaway battery cell samples in the actual vehicle data.

[0119] S620, based on each instance (xi, y i ), calculate x i The distance between each instance in S.

[0120] S630, find the nearest neighbor (x j ,y j ).

[0121] S640, if y i ≠y j , and (x i ,y i ) is added to S.

[0122] S650. When all data sets T can be correctly classified using instances in S, the algorithm terminates and a data set for training is obtained.

[0123] It should be noted that, considering the serious imbalance of data samples, that is, the proportion of thermal runaway battery cells with label 1 is very small, the Condensed Nearest Neighbor algorithm is used to downsample the non-thermal runaway battery cell samples and select the most representative samples to reduce the number of data sets. The final balanced data set that can be used for training is obtained, which effectively eliminates data redundancy and noise, and retains representative samples for training to the greatest extent, thereby improving the accuracy of the training model.

[0124] In some embodiments, S500, the step of inputting the feature data matrix into model training to generate a battery thermal runaway warning model, includes:

[0125] S510: Split the feature data matrix into a training set and a test set. Specifically, the dataset obtained by downsampling the non-thermal runaway battery cell samples using the condensed nearest neighbor rule algorithm is split into a training set and a test set. The split ratio of the training set to the test set can be 7:3 or 6:4.

[0126] S520. Use multiple models to adjust the parameters of the training set and obtain the optimal prediction result probability of each. Specifically, the multiple models can be Xgboost models, RF models and LR models. Furthermore, Xgboost is an extreme gradient boosting model. Xgboost is an ensemble learning algorithm based on a gradient boosting framework. It gradually improves the performance of the model by continuously adding new weak learners (usually decision trees). In each round of iteration, it adjusts the weight of the sample according to the prediction error of the previous round of model, so that subsequent weak learners can pay more attention to those samples that are incorrectly predicted. At the same time, Xgboost uses a second-order Taylor expansion to approximate the loss function during training, so that the gradient can be calculated more accurately and the convergence of the model can be accelerated.

[0127] RF stands for Random Forest model. Random Forest is an ensemble learning algorithm based on decision trees. It constructs multiple decision trees by randomly sampling with replacement from the original training data. The predictions from these trees are then combined (typically using voting for classification tasks and averaging for regression tasks) to arrive at the final prediction. When constructing each decision tree, Random Forest randomly selects a subset of features for splitting. This increases diversity among the decision trees, reduces model variance, and improves model generalization.

[0128] LR stands for logistic regression. Logistic regression is a linear model used for classification problems. It represents the probability that a sample belongs to a certain category by mapping the output of a linear function to the interval [0, 1] through a logistic function (usually a sigmoid function). The goal of logistic regression is to learn the model parameters by minimizing a loss function (usually a cross-entropy loss function) so that the model can predict the sample category as accurately as possible.

[0129] S530: Setting a two-layer neural network with an input layer and an output layer, wherein the input layer receives the prediction result probabilities of multiple models, and the output layer generates a final prediction label.

[0130] S540. Train the neural network using the actual labels of the feature data matrix to obtain optimal weights; and apply the optimal weights to the ensemble learning model.

[0131] S550: After the learning model is trained, it is iteratively optimized and verified using the test set until the learning model is stable, generating a battery thermal runaway warning model.

[0132] It should be noted that this embodiment adopts a weighted ensemble learning model and automatically learns weights through a neural network, avoiding the shortcomings and limitations of manually setting weights, further improving the accuracy of the model, effectively identifying potential battery cell thermal runaway, and generating an accurate early warning model.

[0133] To ensure model effectiveness and prevent overfitting or underfitting, after model training, it is iteratively optimized and validated using a test set until the model stabilizes, generating a final early warning model. This early warning model can then be used to assess new data and identify battery cells experiencing thermal runaway.

[0134] In another embodiment, a device for generating a battery thermal runaway warning model is provided, comprising an acquisition module, an extraction module, a sorting module, a calculation module, a splicing module, and a training module. The acquisition module is used to acquire actual vehicle data of multiple batteries. The extraction module is used to extract battery cell voltage data that meets preset conditions from the actual vehicle data. The sorting module is used to sort the battery cell voltage data to obtain the sorted time series change data of the battery cell voltage. The calculation module is used to calculate multiple time series features based on the sorted time series change data of the battery cell. The splicing module is used to splice the multiple calculated time series features according to the sorting of the battery cells to form a feature data matrix. The training module is used to input the feature data matrix into model training to generate a battery thermal runaway warning model.

[0135] In another embodiment, a battery thermal runaway early warning method is provided, comprising the following steps:

[0136] Obtain real-time data of the battery cells of the vehicle to be tested;

[0137] Importing real-time data into the above-mentioned battery thermal runaway warning model;

[0138] Based on the battery thermal runaway warning model, determine whether the battery cells of the vehicle to be tested have thermal runaway.

[0139] In another embodiment, a computer device is provided, which may be a server. The computer device includes a processor, a memory, a network interface, and a database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as a predicted battery thermal runaway model. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for generating a battery thermal runaway warning model is implemented.

[0140] In another embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a method for generating a battery thermal runaway warning model is implemented.

[0141] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0142] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for generating a battery thermal runaway warning model, characterized in that: The following steps are involved: Obtain real vehicle data for multiple batteries; Extracting battery cell voltage data that meets preset conditions from the actual vehicle data; Sorting the battery cell voltage data to obtain sorting time series change data of the battery cell voltage; Calculating a plurality of time series features based on the time series variation data of the battery cells; The multiple calculated time series features are spliced ​​according to the order of the battery cells to form a feature data matrix; The characteristic data matrix is ​​input into the model training to generate a battery thermal runaway warning model.

2. The generation method according to claim 1, characterized in that The calculation of the plurality of time series features includes at least two of the following: Absolute sum of first-order differences, approximate entropy, Benford correlation, timing complexity, bucket entropy, permutation information entropy, average of absolute change values, and proportion less than a preset voltage threshold.

3. The generation method according to claim 2, characterized in that The step of calculating the approximate entropy based on the time series variation data of the battery cell ranking includes: A vector sequence of dimension m is formed by sequence number. in, X(i) is the i-th value of the sequence, X is the sequence set, and N is the total number of sorted time series change data; Calculating vectors and The Chebyshev distance of Among them, j (1≤j≤N-m+1), x (i+k) is a vector The kth value in x(j+k) is a vector Here, the kth value, j represents the jth sequence; calculate and Distance between The number of j (1≤j≤N-m+1) less than or equal to the threshold r, the number is c i , and calculate the approximate ratio r is the self-set similarity threshold; Define φ m,r for, Increase the dimension to m+1 and repeat the above steps to get φ m+1,r ; Calculate the approximate entropy A p En(m,r)=φ m,r -φ m+1,r .

4. The generation method according to claim 2, characterized in that The step of calculating bucket entropy based on the sorting time series change data of the battery cells includes: Divide [min(X), max(X)] into multiple intervals, where X is the sorted time series change data; distributing the values ​​of the sorted time series change data in the plurality of intervals; Calculating bucket entropy Among them, p k It represents the probability that the value of sorted data X falls into the kth bucket, maxbin represents the number of buckets, and len(X) represents the sequence length.

5. The generation method according to claim 2, characterized in that The step of calculating the permutation information entropy based on the time series variation data of the battery cells comprises: Setting an embedding dimension m and a step size t, reconstructing the order of the battery cells to obtain I subsequences, where I = n-(m-1)t; Convert each of the subsequences into a permutation of size relationships; Calculate the probability p of each size relationship arrangement, where the probability p is equal to the number of times the arrangement is divided by 1; Calculate the permutation information entropy P of the probability e En=-∑ m！ plog(p).

6. The generation method according to claim 1, characterized in that After the step of splicing the calculated multiple time series features according to the order of the battery cells to form a feature data matrix, the method includes: Delete feature columns in the feature data matrix that are greater than a missing rate threshold and less than a variance threshold.

7. The generation method according to claim 6, characterized in that After the step of deleting the feature columns in the feature data matrix that are greater than the missing rate threshold and less than the variance threshold, the method further includes: Initialize the feature columns that are less than or equal to the missing rate threshold and greater than or equal to the variance threshold, let S be the condition set, and let S = φ, where the feature columns that are less than or equal to the missing rate threshold and greater than or equal to the variance threshold are the feature columns obtained by processing the non-thermal runaway battery cell samples in the actual vehicle data; Based on each instance (x i ,y i ), calculate x i The distance between each instance in S; Find the nearest neighbor (x j ,y j ); If y i ≠y j , and (x i ,y i ) is added to S; When all datasets T can be correctly classified using instances in S, the algorithm terminates and the dataset used for training is obtained.

8. The generation method according to claim 1, characterized in that The step of inputting the characteristic data matrix into model training to generate a battery thermal runaway warning model includes: Split the feature data matrix into training and test sets; Using multiple models to train the training set and obtain the optimal prediction result probability for each model; Setting a two-layer neural network having an input layer and an output layer, wherein the input layer receives the prediction result probabilities of the multiple models, and the output layer generates a final prediction label; Training the neural network using the actual labels of the feature data matrix to obtain optimal weights; Applying the optimal weights to an ensemble learning model; After the learning model is trained, iterative optimization and verification are performed using the test set until the learning model is stable, thereby generating a battery thermal runaway warning model.

9. The generation method according to claim 1, characterized in that The preset conditions include one or any combination of the following: The voltage of the battery cell is within a preset voltage range; The current of the battery cell is within a preset current range; The temperature of the battery cell is within a preset temperature range; The state of charge of the battery cell is within a preset range.

10. A device for generating a battery thermal runaway warning model, characterized in that: include: Acquisition module, used to obtain real vehicle data of multiple batteries; An extraction module, configured to extract battery cell voltage data that meets preset conditions from the actual vehicle data; a sorting module, configured to sort the battery cell voltage data to obtain sorting time sequence change data of the battery cell voltage; A calculation module, configured to calculate a plurality of time series features based on the time series variation data of the battery cells; A splicing module is used to splice the multiple calculated time series features according to the order of the battery cells to form a feature data matrix; The training module is used to input the characteristic data matrix into the model training to generate a battery thermal runaway warning model.

11. A battery thermal runaway early warning method, characterized in that: The following steps are involved: Obtain real-time data of the battery cells of the vehicle to be tested; Importing the real-time data into the battery thermal runaway warning model according to any one of claims 1 to 9; Based on the battery thermal runaway warning model, it is determined whether a battery cell of the vehicle to be inspected has thermal runaway.

12. A computer device, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for generating a battery thermal runaway warning model according to any one of claims 1 to 9 is implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for generating a battery thermal runaway warning model as claimed in any one of claims 1 to 9 are implemented.

Images