Early warning method and system for thermal runaway of power battery, storage medium and electronic equipment

Abstract

The invention discloses an early warning method and system for thermal runaway of a power battery, a storage medium and electronic equipment. The early warning method comprises the following steps: preprocessing original real vehicle data to obtain charging stroke data; obtaining strong correlation characteristics between the charging stroke data and thermal runaway, aggregating the strong correlation characteristics, and inputting an aggregation result into a pre-trained grading early warning model; the strong correlation features comprise mechanism features, statistical features and frequency domain features; according to the method, multiple methods such as data preprocessing, feature aggregation, AI model prediction and big data support are combined, efficient and accurate fault early warning can be achieved, the safety of the new energy vehicle is improved, and the service life of the new energy vehicle is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of power batteries for new energy vehicles, and in particular to a method and system for early warning of thermal runaway of power batteries, a storage medium, and electronic equipment. Background Art

[0002] The new energy vehicle industry is developing rapidly. As the core power source of new energy vehicles, the health of power batteries is increasingly valued. When monitoring the safety status of power batteries throughout their life cycle, it is found that due to the continuous aging of batteries, the driving habits of car owners, the external vehicle environment and some potential influencing factors, vehicle thermal runaway events occur frequently and have not been effectively resolved; therefore, early warning of the occurrence of vehicle thermal runaway has become a difficult problem that all automobile companies focus on. Power battery thermal runaway is a problem involving multiple fault modes, which are specifically divided into easily perceived fault modes and not easily perceived fault modes; easily perceived fault modes refer to faults that can be identified in advance by directly analyzing the data collected by the sensor, such as internal short circuits, self-discharge, abnormal voltage differences, abnormal temperature differences, etc.; not easily perceived fault modes refer to faults that cannot be directly identified in advance by analyzing the data collected by the sensor, especially sudden death types.

[0003] In the prior art, characteristic parameters related to thermal runaway data are obtained to determine the training set and the test set, and a linear regression predictor is used to compare the prediction results with the real data to predict whether the battery has thermal runaway. This method also only considers the single cell voltage, module temperature, and air pressure as characteristic parameters. The characteristic parameters cannot cover all types of thermal runaway, and can only identify simple fault modes. In addition, a simple linear regression method is used to achieve prediction. The model accuracy is low and the effect of accurate early warning cannot be achieved.

[0004] In addition to the above problems, a large number of existing technologies also face the following problems:

[0005] (1) Thermal runaway has many modes, which are often a fusion of multiple implicit faults and are difficult to identify using a single or a few parameters, especially for sudden death fault modes. Existing analysis methods only cover a small number of easily identifiable fault types.

[0006] (2) Most existing methods are based on expert experience or mechanism algorithm analysis, and artificially set thresholds to warn of the moment when thermal runaway occurs. This method is highly subjective and limited by expert experience, making it difficult to popularize and use.

[0007] (3) The AI ​​algorithm feature design is simple and has fewer dimensions. Most of the features are invalid, and no effective features are extracted for the battery fast charging stage data, resulting in low model accuracy and poor warning effect.

[0008] How to provide objective and accurate early warning of thermal runaway of power batteries is an issue that needs to be urgently addressed in this field. Summary of the invention

[0009] In order to solve the above-mentioned problem of poor accuracy and objectivity of power battery thermal runaway warning, the present invention provides a power battery thermal runaway warning method and system, storage medium, and electronic device. Based on a large amount of real vehicle data, the present invention extracts strongly correlated features with power battery thermal runaway, and combines them with AI models to achieve accurate graded warning of faulty vehicles through binary classification. The model trained using this method achieves high accuracy and recall rate on the test set.

[0010] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0011] The early warning method for thermal runaway of a power battery comprises the following steps:

[0012] The original real vehicle data is preprocessed to obtain charging trip data;

[0013] Obtain the strong correlation features between charging trip data and thermal runaway, aggregate the strongly correlated features, and input the aggregation results into the pre-trained hierarchical warning model;

[0014] The strong correlation features include: mechanism features, statistical features and frequency domain features;

[0015] The graded warning model determines the warning level of thermal runaway according to the preset classification strategy and executes the warning.

[0016] Preferably, the pretreatment comprises:

[0017] Obtain original real vehicle data of normal vehicles and faulty vehicles;

[0018] Filter the original fields available in the original real vehicle data;

[0019] Clean the original fields according to the preset cleaning strategy;

[0020] The charging data is divided into at least two charging trip data according to the frame number.

[0021] Preferably, the preset cleaning strategy includes:

[0022] Reorganize the original fields by sorting;

[0023] Filter the fields during the vehicle charging trip;

[0024] Filter the data frames in the fast charging phase;

[0025] Fill in the missing frame data.

[0026] Preferably, the mechanism characteristics include: charging overcurrent characteristics, charging rate characteristics, charging internal resistance characteristics, daily average mileage characteristics and monomer consistency characteristics;

[0027] The statistical features include: one or any combination of voltage-temperature extreme difference feature, charging current maximum value feature, large current ratio feature, charging current extreme difference feature and skewness feature;

[0028] The frequency domain features include: one or any combination of pressure difference wavelet analysis features, temperature difference wavelet analysis features, maximum monomer voltage wavelet features and minimum monomer wavelet features.

[0029] Preferably, the preset grading strategy includes at least two warning levels. The warning level is determined and the warning is executed according to the prediction result output by the graded warning model. Each warning accumulates a warning number. When the warning number reaches a specified number, a maintenance prompt is issued.

[0030] Preferably, the prediction result is the proportion of charging trip failures of the vehicle at the current moment.

[0031] The early warning system for thermal runaway of power batteries includes:

[0032] A data acquisition preprocessing subsystem, wherein the data acquisition preprocessing subsystem obtains charging trip data by preprocessing the original real vehicle data;

[0033] A feature aggregation subsystem, wherein the feature aggregation subsystem obtains strongly correlated features between charging trip data and thermal runaway, aggregates the strongly correlated features, and inputs the aggregation results into a pre-trained hierarchical warning model;

[0034] The strong correlation features include: mechanism features, statistical features and frequency domain features;

[0035] The early warning subsystem includes a hierarchical early warning model, which determines the early warning level of thermal runaway according to a preset hierarchical strategy and executes the early warning.

[0036] Preferably, the pretreatment comprises:

[0037] Obtain original real vehicle data of normal vehicles and faulty vehicles;

[0038] Filter the original fields available in the original real vehicle data;

[0039] Clean the original fields according to the preset cleaning strategy;

[0040] Dividing the charging data into at least two charging trip data according to the number of frames;

[0041] Wherein, the preset cleaning strategy includes:

[0042] Reorganize the original fields by sorting;

[0043] Filter the fields during the vehicle charging trip;

[0044] Filter the data frames in the fast charging phase;

[0045] Fill in the missing frame data.

[0046] Preferably, the mechanism characteristics include: charging overcurrent characteristics, charging rate characteristics, charging internal resistance characteristics, daily average mileage characteristics and monomer consistency characteristics;

[0047] The statistical features include: one or any combination of voltage-temperature extreme difference feature, charging current maximum value feature, large current ratio feature, charging current extreme difference feature and skewness feature;

[0048] The frequency domain features include: one or any combination of pressure difference wavelet analysis features, temperature difference wavelet analysis features, maximum monomer voltage wavelet features and minimum monomer wavelet features.

[0049] Preferably, the pre-training of the hierarchical warning model based on real vehicle data includes: training set division, test set division, pre-adjustment of parameters, model training and testing.

[0050] Preferably, the preset classification strategy includes at least two warning levels, and the warning level is determined and the warning is executed according to the prediction result output by the graded warning model. Each warning accumulates a warning number, and a maintenance prompt is issued when the warning number reaches a specified number;

[0051] The prediction result is the proportion of charging trip failures of the vehicle at the current moment.

[0052] Preferably, the early warning of the early warning subsystem includes:

[0053] If the warning result shows that the proportion of charging trip failures of the vehicle at the current moment is greater than or equal to 80%, a Level I warning is implemented, and the number of Level I warnings of the current vehicle is accumulated by 1. If the number of Level I warnings reaches 3 times, a prompt maintenance message is sent;

[0054] If the warning result shows that the proportion of charging trip failures of the vehicle at the current moment is greater than or equal to 50% and less than 80%, a Level II warning is implemented, and the number of Level II warnings of the current vehicle is accumulated by 1. If the number of Level II warnings reaches 6 times, a prompt maintenance message is sent;

[0055] If the warning result shows that the proportion of charging trip failures of the vehicle at the current moment is greater than or equal to 20% and less than 50%, a Level III warning is implemented, and the number of Level III warnings of the current vehicle is accumulated by 1. If the number of Level II warnings reaches 10 times, a maintenance information is sent;

[0056] If the warning result shows that the proportion of charging trip failures of the vehicle at the current moment is less than 20%, a Level IV warning is implemented, and the number of Level IV warnings of the current vehicle is accumulated by 1. If the number of Level IV warnings reaches 15 times, a prompt maintenance message is sent;

[0057] Except for the above situations, no warning will be triggered.

[0058] A computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, a warning method for thermal runaway of a power battery is implemented.

[0059] An electronic device comprises a memory and a processor, wherein the memory is used to store at least one instruction, and the processor is used to execute at least one instruction to implement an early warning method for thermal runaway of a power battery.

[0060] Beneficial effects of the present invention:

[0061] (1) Early warning of various fault modes: The present invention can not only provide early warning of easily perceived fault modes, but also accurately identify fault modes that are not easily perceived, especially those serious faults that may occur suddenly. This capability significantly improves the safety prevention of battery thermal runaway;

[0062] (2) Feature aggregation: This solution uses the aggregated features of the charging trip fast charging data that are strongly correlated with thermal runaway, including the battery’s mechanism, statistics, and frequency domain features. These three features can fully describe the battery’s status and effectively identify the failure mode of thermal runaway.

[0063] (3) Application of AI models: Combining AI models for early warning can more accurately identify and predict fault modes and provide accurate early warning information. The output result is a binary classification, which can achieve graded early warning for thermal runaway of power batteries, providing more accurate early warning information for automakers and users.

[0064] (4) Big data support: Based on a large amount of real vehicle data, we extract features that are strongly correlated with thermal runaway of power batteries and combine them with AI models to achieve accurate fault vehicle classification warnings. With the support of big data, the pre-trained model shows high accuracy and recall rate on the test set;

[0065] (5) Early warning efficiency: Each early warning accumulates the number of early warnings. When the number of early warnings reaches a specified number, a maintenance reminder is issued. This design can capture and handle battery problems in a timely manner to avoid serious consequences caused by battery thermal runaway.

[0066] (6) Practicality of prediction results: The prediction result of the early warning model is the proportion of charging trip failures at the current moment of the vehicle, which is an intuitive metric that helps drivers and maintenance personnel quickly understand and handle problems;

[0067] In general, the present invention combines multiple methods such as data preprocessing, feature aggregation, AI model prediction, and big data support to achieve efficient and accurate fault warning and improve the safety and service life of new energy vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic diagram of the process of the present invention;

[0069] Figure 2 This is a schematic diagram of the distribution of strongly correlated features of the present invention;

[0070] Figure 3 This is a schematic diagram of the deployment of the cloud platform model in the present invention;

[0071] Figure 4 This is a schematic diagram of the prediction result of a normal vehicle in the present invention;

[0072] Figure 5 This is a schematic diagram of the fault vehicle prediction result in the present invention;

[0073] Figure 6 This is a schematic diagram of graded warning in the present invention. DETAILED DESCRIPTION

[0074] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0075] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0076] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0077] In the case of using expressions such as "at least one of A, B, and C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.). In the case of using expressions such as "at least one of A, B, or C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0078] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some boxes or combinations thereof in the block diagrams and / or flow charts can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions can create a device for implementing the functions / operations described in these block diagrams and / or flow charts when executed by the processor. The technology of the present disclosure can be implemented in the form of hardware and / or software (including firmware, microcode, etc.). In addition, the technology of the present disclosure can take the form of a computer program product on a computer-readable storage medium storing instructions, which can be used by an instruction execution system or used in conjunction with an instruction execution system.

[0079] like Figure 1 As shown, the early warning method for thermal runaway of a power battery includes the following steps:

[0080] The original real vehicle data is preprocessed to obtain charging trip data;

[0081] Obtain the strong correlation features between charging trip data and thermal runaway, aggregate the strongly correlated features, and input the aggregation results into the pre-trained hierarchical warning model;

[0082] The strong correlation features include: mechanism features, statistical features and frequency domain features;

[0083] The graded warning model determines the warning level of thermal runaway according to the preset classification strategy and executes the warning.

[0084] Data preprocessing: First, the present invention obtains charging trip data by preprocessing the original real vehicle data. The preprocessing process includes obtaining the original real vehicle data of normal vehicles and faulty vehicles, screening the available original fields in the original real vehicle data, cleaning the original fields according to the preset cleaning strategy, and dividing the charging data into at least two charging trip data according to the number of frames. This process is mainly to prepare data for input into the early warning model.

[0085] Feature aggregation: Secondly, the present invention obtains the strong correlation features between charging trip data and thermal runaway, and aggregates these features. These features include mechanism features, statistical features, and frequency domain features. The three types of features together construct a comprehensive description of the battery state.

[0086] Warning model: Finally, the present invention inputs the aggregated features into the pre-trained hierarchical warning model. Based on the preset classification strategy, this model can determine the warning level of thermal runaway and execute the warning. In this process, the model predicts the possibility of thermal runaway of the power battery through a binary classification method, and at the same time, provides the prediction result as the charging trip failure ratio of the vehicle at the current moment.

[0087] The present invention aims to achieve accurate early warning of thermal runaway of power batteries through a large amount of real vehicle data, precise feature aggregation and advanced AI models. First, the data preprocessing step helps clean and organize the data input to the model, improving the data quality. Secondly, the feature aggregation step ensures that the model receives comprehensive and accurate battery status information by conducting a detailed analysis of the features that are strongly related to thermal runaway. Finally, the pre-trained early warning model achieves accurate prediction of thermal runaway by learning the relationship between these features and thermal runaway.

[0088] The present invention can not only warn of easily perceived fault modes in advance, but also accurately identify fault modes that are not easily perceived, especially serious faults that may occur suddenly. In addition, the prediction result of the early warning model is an intuitive metric, which is convenient for drivers and maintenance personnel to quickly understand and deal with problems. Therefore, the present invention has high practicality and effectiveness.

[0089] Preferably, the pretreatment comprises:

[0090] Obtain original real vehicle data of normal vehicles and faulty vehicles;

[0091] Download a large amount of real vehicle data based on the big data cloud platform, including normal vehicles and faulty vehicles. Normal vehicles have not experienced any faults before downloading the data, and faulty vehicles have experienced faults before downloading the data, covering various fault types, especially sudden death fault types. To ensure the quality of the data, the screening conditions require that the vehicle driving cycle should be more than 3 years as much as possible, and the cumulative mileage should be more than 200,000 kilometers as much as possible;

[0092] Filter the original fields available in the original real vehicle data;

[0093] Use the Python data processing toolkit pandas to loop through the vehicle raw data and filter out the available raw fields, including: vehicle vin, terminal time, state of charge, capacity, cumulative mileage, total voltage, total current, voltage list, temperature list, vehicle operating status, maximum cell voltage, minimum cell voltage, maximum sensor temperature, minimum sensor temperature, etc.

[0094] Clean the original fields according to the preset cleaning strategy;

[0095] Dividing the charging data into at least two charging trip data according to the number of frames;

[0096] Each time the vehicle is charged is recorded as a trip. All the charging data of the vehicle has been screened out. First, based on this, the charging trip to which each frame of data belongs is divided into at least two charging trip data. For example, the data of frames 1-100 belongs to the first charging trip, and the charging trip number of frames 1-100 is 0; similarly, the data of frames 101-200 belongs to the second charging trip, and the charging trip number of frames 101-200 is 1; secondly, the duration of each charging must at least satisfy t≥20min, that is, the charging time is at least greater than or equal to 20 minutes. The value of t can be determined according to the battery model, and the sampling frequency of each frame of charging data is the same, otherwise the charging segment is discarded.

[0097] Preferably, the preset cleaning strategy includes:

[0098] Reorganize the original fields by sorting;

[0099] Since the data collected by the sensor is out of order, it is necessary to reintegrate the data according to the vehicle's VIN number. Specifically, the sort_values ​​function of the pandas toolkit is used to sort the data in ascending order according to the terminal time and cumulative mileage fields.

[0100] Filter the fields during the vehicle charging trip;

[0101] The vehicle discharge trip data has a lot of noise and abnormal fluctuations. The feature availability extracted using the discharge trip data is low and will reduce the model accuracy. Therefore, the present invention only uses the vehicle charging trip data for analysis and filters the charging trip data according to the driving status field.

[0102] Filter the data frames in the fast charging phase;

[0103] According to the battery fast charging strategy used in the present invention, the data frames with charging current I<ρ are regarded as slow charging, and the remaining data frames are regarded as fast charging; because in actual situations, the probability of battery failure is higher during the fast charging stage; therefore, the data frames with charging current I≥ρ are screened out for analysis, and the remaining data frames are eliminated, ρ∈[20,40], unit ampere;

[0104] Fill the missing frame data;

[0105] When the sensor transmits data, due to network signals and other reasons, many field data are missing when uploading; therefore, it is necessary to fill the current missing frame data. There are two main methods for filling missing values. One is if the number of missing frames in a charging trip accounts for the proportion of the total number of data frames. Then the charging segment is directly removed, otherwise the current missing frame data is filled with the sliding average of the charging data of the previous x frames, x∈[5,10].

[0106] like Figure 2 As shown in the figure, the characteristics of the divided charging trip data are strongly correlated with thermal runaway, including mechanism characteristics, statistical characteristics and frequency domain characteristics. These three categories of characteristics reflect the continuous aging process of the battery during use. The combination of the three can effectively identify various modes of thermal runaway and achieve accurate early warning.

[0107] Preferably, the mechanism characteristics include: charging overcurrent characteristics, charging rate characteristics, charging internal resistance characteristics, daily average mileage characteristics and monomer consistency characteristics;

[0108] ①Charging overcurrent characteristics. Continuous overcurrent during charging will cause battery performance aging. The overcurrent judgment method is to compare the calculated theoretical current with the actual current. The calculation formula for the theoretical current of each data frame in the charging trip is: I L =-H×SOH NS ×SOH RC ; where I L is the theoretical current, H is the current data frame charging rate, SOH NS It is the national standard capacity, SOH RC is the rated capacity; subtract the actual current from the theoretical current to obtain a new eigenvector [i1,i2,i3....i n ], if i k >0, the current data frame is over-current, otherwise it means there is no over-current; based on the feature vector [i1,i2,i3....i n ], extract the maximum overcurrent value i of this trip max and the overcurrent ratio value i r And other features.

[0109] ②Charging rate characteristics. The charging rate measures the speed of charging in the current trip. The calculation formula for the charging rate of the trip is: Where e represents the charging rate, Δsoc represents the change in the state of charge during the current charging trip, and t represents the charging time of the current trip.

[0110] ③Charging internal resistance characteristics. The internal resistance reflects whether an internal short circuit occurs in the current trip. The calculation formula for the internal resistance of each data frame of the charging trip is: Where r is the current internal resistance of the monomer, v is the current voltage of the monomer, and v s is the open circuit voltage obtained by looking up the ocv table; a new column of eigenvectors [r1, r2, r3...r n ]; Based on this feature vector [r1,r2,r3...r n ], extract the maximum internal resistance r of this stroke max And other features.

[0111] ④ Daily average mileage characteristics. The longer the vehicle is used, the greater the daily average mileage, the greater the damage to the battery. The daily average mileage is also an important indicator to measure whether the battery has failed. The calculation formula for the current daily average mileage is: Where m is the average daily mileage of the current trip, s is the cumulative mileage of the current trip, and t is the average daily mileage of the current trip. s The current cumulative number of driving days.

[0112] ⑤ Single-cell consistency characteristics. Whether the voltage of a single cell is consistent during a charging trip can also characterize the state of the battery. The coefficient of variation is used to measure the consistency of the single cell. The calculation formula for the current trip coefficient of variation is: where f v is the current trip monomer variation coefficient, σ v is the current trip monomer voltage standard deviation, μ v It is the average voltage of the current stroke monomer.

[0113] The statistical features include: one or any combination of voltage-temperature extreme difference feature, charging current maximum value feature, large current ratio feature, charging current extreme difference feature and skewness feature;

[0114] ① Voltage temperature extreme difference characteristics. Based on the voltage list field [v1,v2,v3...v n ] and the temperature list field [t1,t2,t3...t n ], the calculation formula for the voltage range of each data frame in the charging trip is: δ v =v max -v min ; where δ v is the voltage extreme difference of the current data frame, v max is the maximum value of the single cell voltage in the current data frame, v min is the minimum value of the single cell voltage in the current data frame; the extreme characteristic vector [δv1 ,δ v2 ,δ v3 ,...δ vn ], based on this feature vector [δ v1 ,δ v2 ,δ v3 ,...δ vn ,], extract δ vmax Similarly, the temperature extreme difference of each data frame in the charging trip is calculated as follows: t =t max -t min ; where δ t is the temperature extreme difference of the current data frame, t max is the maximum value of the sensor temperature in the current data frame, t min is the minimum value of the sensor temperature in the current data frame; the temperature extreme difference characteristic vector [δ t1 ,δ t2 ,δ t3 ,...δ tn ], based on this feature vector [δ t1 ,δ t2 ,δ t3 ,...δ tn ], extract δ tmax .

[0115] ②Charging current maximum value feature. Based on the total charging current field, extract the maximum value of the charging current I for each trip max .

[0116] ③ High current ratio characteristics. High current refers to the current with a larger value in the current charging process. If the charging current is above I ampere, it can be regarded as high current charging. Generally, I≥100A. Based on the total charging current field, the current high current ratio calculation formula is: where r bc is the proportion of large current, ε p is the total number of high current frames in the current charging trip, ε t The total number of frames in the current charging trip.

[0117] ④Charging current range characteristics. Based on the total charging current field, extract the range I of the charging current for each trip r =I max -I min ; where I max is the maximum charging current of the current trip, I min It is the minimum charging current of the current trip.

[0118] ⑤ Skewness feature. Based on the voltage list field [v1,v2,v3...v n] and the temperature list field [t1,t2,t3...t n ], the calculation formula for the voltage list skewness of each data frame of the current charging trip is: where ψ v is the voltage list skewness of the current data frame, x i is the single cell voltage value of the current data frame, u v is the voltage list mean value of the current data frame, σ v is the voltage list standard deviation of the current data frame. The voltage skewness vector [ψ v1 ,ψ v2 ,ψ v3 ,...ψ vn ], based on this eigenvector [ψ v1 ,ψ v2 ,ψ v3 ,...ψ vn ], extract ψ vmax ; Similarly, the calculation formula for the temperature list skewness of each data frame of the current charging trip is: where ψ t is the temperature list skewness of the current data frame, x i is the sensor temperature value of the current data frame, u t is the mean of the temperature list of the current data frame, σ t is the standard deviation of the temperature list of the current data frame. The temperature skewness vector [ψ t1 ,ψ t2 ,ψ t3 ,...ψ tn ], based on this eigenvector [ψ t1 ,ψ t2 ,ψ t3 ,...ψ tn ], extract ψ tmax ;

[0119] The frequency domain features include: one or any combination of pressure difference wavelet analysis features, temperature difference wavelet analysis features, maximum monomer voltage wavelet features and minimum monomer wavelet features.

[0120] The voltage difference characteristic vector of each charging trip is [δ v1 ,δ v2 ,δ v3 ,...δ vn ,], temperature difference characteristic vector [δ t1 ,δ t2 ,δ t3 ,...δ tn ,], the maximum single cell voltage characteristic vector is [v max1 ,v max2 ,v max2...v maxn ], the minimum single cell voltage characteristic vector is [v min1 ,v min2 ,v min2 ...v minn ]; for the current stroke pressure difference, temperature difference, maximum and minimum single cell voltage feature vectors, use the Ricker wavelet basis function in CWT (continuous wavelet analysis) to extract features. The specific calculation formula is: Where a is the width parameter, and x is the above four eigenvectors; the corresponding wavelet analysis result τ is calculated by this formula vr , τ tr , τ vmax , τ vmin .

[0121] Data aggregation: The nxm-dimensional time series data of each charging trip is aggregated into 1x h-dimensional aggregated data through feature extraction. Compared with pure time series features, aggregated features can better reflect the current trip charging information and are more conducive to AI model learning. Therefore, according to the above aggregation method, each vehicle is aggregated into vxh-dimensional data; v represents the number of current vehicle charging trips, and h represents the number of aggregated features. The design is based on strongly correlated features, and the aggregated feature vector expression is:

[0122] H=[time,i max ,i r ,e,r max ,m,f v ,δ vmax ,δ tmax ,I max ,r bc ,I r ,Δsoc,ψ vmax ,ψ tmax ,τ vr τ tr ,τ vmax ,τ vmin ,label]

[0123] The present invention can use the python development environment to build code based on any integrated model to train and test the strongly correlated features of the vehicle. The integrated models mainly include Bagging and Boosting. The integrated models of Bagging include: RF; the integrated models of Boosting include GBDT, XGBOOST and LIGHTGBM, etc. Since there are many types of integrated models, in order to avoid redundancy, the present invention specifically takes the XGBOOST integrated model as an example for subsequent modeling and testing. XGBOOST is an end-to-end gradient boosting tree system, which is widely used in classification and regression tasks. Its advantages are fast training speed, automatic processing of missing values, use of the second-order derivative information of the objective function, and introduction of regularization terms to prevent model overfitting.

[0124] AI model building includes training set division, test set division, pre-adjustment of parameters, model training and testing, specifically:

[0125] ① Training and test set division. First, based on the real vehicle data, the aggregated feature vector is completed according to the above method. Secondly, the vehicle data is divided into a training set and an independent test set. The ratio of the two can be set according to the actual situation. About 80% or 90% of the data in the training set is used for training, and 10% or 20% of the data is used for verification. The verification set is used to adjust the hyperparameters of the model to prevent overfitting of the model and improve the generalization ability of the model. The ratio of the training and test sets can be adjusted as needed;

[0126] ② Pre-adjust model parameters. Specifically, use the training and validation data sets that have been divided in step ①, and use the grid search method to determine the optimal values ​​of the model hyperparameters in turn. The order of determining the hyperparameters is learning_rate, max_depth, scale_pos_weight, gamma, colsample_bytree, lambda, alpha, n_estimator, etc.; select binary:logistic as the objective function, and output the failure probability of the current charging trip to achieve binary classification. Parameter pre-adjustment ensures that the model's indicators on the validation set are optimal. The evaluation indicators include accuracy, precision, recall, and F1 value.

[0127] ③ Model training and testing. Specifically, use the pre-trained hyperparameters in step ② to iteratively train a stable version of the XGBOOST model, and perform independent tests on the test set divided in step ①. If the test results show that all faulty vehicles are recalled and there are no false positives for normal vehicles, it means that the model has achieved a relatively good result and can be used for online deployment on the cloud platform; Figure 4 and Figure 5 The model predicts each charging trip of normal vehicles and faulty vehicles in the test set. The results show that the model can achieve accurate early warning and locate the faulty charging cycle for secondary analysis based on the prediction results.

[0128] Preferably, the preset grading strategy includes at least two warning levels. The warning level is determined and the warning is executed according to the prediction result output by the graded warning model. Each warning accumulates a warning number. When the warning number reaches a specified number, a maintenance prompt is issued.

[0129] Preferably, the prediction result is the proportion of charging trip failures of the vehicle at the current moment.

[0130] like Figure 3 As shown in the figure, the model online prediction specifically includes encapsulating the local offline built data processing, feature design code and trained AI model into an API and deploying it to the cloud platform to achieve real-time graded warning of vehicles, and continue to add the faulty vehicle data processed by the model online warning to the offline training set, continuously iterate and optimize the model, and enhance the generalization ability of the model. See the cloud platform deployment model graded warning process for details. Figure 6 , the specific method is:

[0131] ① If the model predicts that the vehicle's current charging trip fault ratio is ≥ 80%, a Level I warning will be implemented, and the number of Level I warnings for the current vehicle will be increased by 1. If the number of Level I warnings reaches 3 times, the cloud platform will send a reminder message to ask the owner to go to the maintenance shop for inspection;

[0132] ② If the model predicts that the vehicle's current charging trip fault ratio is 50% ≤ faultRatio ≤ 80%, a Level II warning will be implemented, and the number of Level II warnings for the current vehicle will be increased by 1. If the number of Level II warnings reaches 6 times, the cloud platform will send a reminder message to ask the owner to go to the maintenance shop for inspection;

[0133] ③ If the model predicts that the vehicle's current charging trip fault ratio is 20% ≤ faultRatio ≤ 50%, a Level III warning will be implemented, and the number of Level III warnings for the current vehicle will be increased by 1. If the number of Level III warnings reaches 10 times, the cloud platform will send a reminder message to ask the owner to go to the maintenance shop for inspection;

[0134] ④ If the model predicts that the vehicle's current charging trip fault ratio is ≤20%, a Level IV warning will be implemented, and the number of Level IV warnings for the current vehicle will be increased by 1. If the number of Level IV warnings reaches 15 times, the cloud platform will send a reminder message to ask the owner to go to the maintenance shop for inspection;

[0135] Except for the above strategies, other strategies do not trigger the graded warning mechanism.

[0136] The early warning system for thermal runaway of power batteries includes:

[0137] A data acquisition preprocessing subsystem, wherein the data acquisition preprocessing subsystem obtains charging trip data by preprocessing the original real vehicle data;

[0138] A feature aggregation subsystem, wherein the feature aggregation subsystem obtains strongly correlated features between charging trip data and thermal runaway, aggregates the strongly correlated features, and inputs the aggregation results into a pre-trained hierarchical warning model;

[0139] The strong correlation features include: mechanism features, statistical features and frequency domain features;

[0140] The early warning subsystem includes a hierarchical early warning model, which determines the early warning level of thermal runaway according to a preset hierarchical strategy and executes the early warning.

[0141] Data preprocessing: First, the present invention obtains charging trip data by preprocessing the original real vehicle data. The preprocessing process includes obtaining the original real vehicle data of normal vehicles and faulty vehicles, screening the available original fields in the original real vehicle data, cleaning the original fields according to the preset cleaning strategy, and dividing the charging data into at least two charging trip data according to the number of frames. This process is mainly to prepare data for input into the early warning model.

[0142] like Figure 2 As shown, feature aggregation: Secondly, the present invention obtains strong correlation features between charging trip data and thermal runaway, and aggregates these features. These features include mechanism features, statistical features, and frequency domain features. The three types of features together construct a comprehensive description of the battery state.

[0143] Warning model: Finally, the present invention inputs the aggregated features into the pre-trained hierarchical warning model. Based on the preset classification strategy, this model can determine the warning level of thermal runaway and execute the warning. In this process, the model predicts the possibility of thermal runaway of the power battery through a binary classification method, and at the same time, provides the prediction result as the charging trip failure ratio of the vehicle at the current moment.

[0144] The present invention aims to achieve accurate early warning of thermal runaway of power batteries through a large amount of real vehicle data, precise feature aggregation and advanced AI models. First, the data preprocessing step helps clean and organize the data input to the model, improving the data quality. Secondly, the feature aggregation step ensures that the model receives comprehensive and accurate battery status information by conducting a detailed analysis of the features that are strongly related to thermal runaway. Finally, the pre-trained early warning model achieves accurate prediction of thermal runaway by learning the relationship between these features and thermal runaway.

[0145] The present invention can not only warn of easily perceived fault modes in advance, but also accurately identify fault modes that are not easily perceived, especially serious faults that may occur suddenly. In addition, the prediction result of the early warning model is an intuitive metric, which is convenient for drivers and maintenance personnel to quickly understand and deal with problems. Therefore, the present invention has high practicality and effectiveness.

[0146] Preferably, the pretreatment comprises:

[0147] Obtain original real vehicle data of normal vehicles and faulty vehicles;

[0148] Download a large amount of real vehicle data based on the big data cloud platform, including normal vehicles and faulty vehicles. Normal vehicles have not experienced any faults before downloading the data, and faulty vehicles have experienced faults before downloading the data, covering various fault types, especially sudden death fault types. To ensure the quality of the data, the screening conditions require that the vehicle driving cycle should be more than 3 years as much as possible, and the cumulative mileage should be more than 200,000 kilometers as much as possible;

[0149] Filter the original fields available in the original real vehicle data;

[0150] Use the Python data processing toolkit pandas to loop through the vehicle raw data and filter out the available raw fields, including: vehicle vin, terminal time, state of charge, capacity, cumulative mileage, total voltage, total current, voltage list, temperature list, vehicle operating status, maximum cell voltage, minimum cell voltage, maximum sensor temperature, minimum sensor temperature, etc.

[0151] Clean the original fields according to the preset cleaning strategy;

[0152] Dividing the charging data into at least two charging trip data according to the number of frames;

[0153] Each time the vehicle is charged is recorded as a trip. All the charging data of the vehicle has been screened out. First, based on this, the charging trip to which each frame of data belongs is divided into at least two charging trip data. For example, the data of frames 1-100 belongs to the first charging trip, and the charging trip number of frames 1-100 is 0; similarly, the data of frames 101-200 belongs to the second charging trip, and the charging trip number of frames 101-200 is 1; secondly, the duration of each charging must at least satisfy t≥20min, that is, the charging time is at least greater than or equal to 20 minutes. The value of t can be determined according to the battery model, and the sampling frequency of each frame of charging data is the same, otherwise the charging segment is discarded.

[0154] Preferably, the preset cleaning strategy includes:

[0155] Reorganize the original fields by sorting;

[0156] Since the data collected by the sensor is out of order, it is necessary to reintegrate the data according to the vehicle's VIN number. Specifically, the sort_values ​​function of the pandas toolkit is used to sort the data in ascending order according to the terminal time and cumulative mileage fields.

[0157] Filter the fields during the vehicle charging trip;

[0158] The vehicle discharge trip data has a lot of noise and abnormal fluctuations. The feature availability extracted using the discharge trip data is low and will reduce the model accuracy. Therefore, the present invention only uses the vehicle charging trip data for analysis and filters the charging trip data according to the driving status field.

[0159] Filter the data frames in the fast charging phase;

[0160] According to the battery fast charging strategy used in the present invention, the data frames with charging current I<ρ are regarded as slow charging, and the remaining data frames are regarded as fast charging; because in actual situations, the probability of battery failure is higher during the fast charging stage; therefore, the data frames with charging current I≥ρ are screened out for analysis, and the remaining data frames are eliminated, ρ∈[20,40], unit ampere;

[0161] Fill the missing frame data;

[0162] When the sensor transmits data, due to network signals and other reasons, many field data are missing when uploading; therefore, it is necessary to fill the current missing frame data. There are two main methods for filling missing values. One is if the number of missing frames in a charging trip accounts for the proportion of the total number of data frames. Then the charging segment is directly removed, otherwise the current missing frame data is filled with the sliding average of the charging data of the previous x frames, x∈[5,10].

[0163] Based on the characteristics of the divided charging trip data that are strongly correlated with thermal runaway, including mechanism characteristics, statistical characteristics and frequency domain characteristics, these three categories of characteristics all reflect the continuous aging process of the battery during use. The combination of the three can effectively identify various modes of thermal runaway and achieve accurate early warning.

[0164] Preferably, the mechanism characteristics include: charging overcurrent characteristics, charging rate characteristics, charging internal resistance characteristics, daily average mileage characteristics and monomer consistency characteristics;

[0165] ①Charging overcurrent characteristics. Continuous overcurrent during charging will cause battery performance aging. The overcurrent judgment method is to compare the calculated theoretical current with the actual current. The calculation formula for the theoretical current of each data frame in the charging trip is: I L =-H×SOH NS ×SOH RC ; where IL is the theoretical current, H is the current data frame charging rate, SOH NS It is the national standard capacity, SOH RC is the rated capacity; subtract the actual current from the theoretical current to obtain a new eigenvector [i1,i2,i3....i n ], if i k >0, the current data frame is over-current, otherwise it means there is no over-current; based on the feature vector [i1,i2,i3....i n ], extract the maximum overcurrent value i of this trip max and the overcurrent ratio value i r And other features.

[0166] ②Charging rate characteristics. The charging rate measures the speed of charging in the current trip. The calculation formula for the charging rate of the trip is: Where e represents the charging rate, Δsoc represents the change in the state of charge during the current charging trip, and t represents the charging time of the current trip.

[0167] ③Charging internal resistance characteristics. The internal resistance reflects whether an internal short circuit occurs in the current trip. The calculation formula for the internal resistance of each data frame of the charging trip is: Where r is the current frame internal resistance of the monomer, v is the current frame voltage of the monomer, v s is the open circuit voltage obtained by looking up the ocv table; a new column of eigenvectors [r1, r2, r3...r n ]; Based on this feature vector [r1,r2,r3...r n ], extract the maximum internal resistance r of this stroke max And other features.

[0168] ④ Daily average mileage characteristics. The longer the vehicle is used, the greater the daily average mileage, the greater the damage to the battery. The daily average mileage is also an important indicator to measure whether the battery has failed. The calculation formula for the current daily average mileage is: Where m is the average daily mileage of the current trip, s is the cumulative mileage of the current trip, and t is the average daily mileage of the current trip. s The current cumulative number of driving days.

[0169] ⑤ Single-cell consistency characteristics. Whether the voltage of a single cell is consistent during a charging trip can also characterize the state of the battery. The coefficient of variation is used to measure the consistency of the single cell. The calculation formula for the current trip coefficient of variation is: where f v is the current trip monomer variation coefficient, σ v is the current trip monomer voltage standard deviation, μ v It is the average voltage of the current stroke monomer.

[0170] The statistical features include: one or any combination of voltage-temperature extreme difference feature, charging current maximum value feature, large current ratio feature, charging current extreme difference feature and skewness feature;

[0171] ① Voltage temperature extreme difference characteristics. Based on the voltage list field [v1,v2,v3...v n ] and the temperature list field [t1,t2,t3...t n ], the calculation formula for the voltage range of each data frame in the charging trip is: δ v =v max -v min ; where δ v is the voltage extreme difference of the current data frame, v max is the maximum value of the single cell voltage in the current data frame, v min is the minimum value of the single cell voltage in the current data frame; the extreme characteristic vector [δ v1 ,δ v2 ,δ v3 ,...δ vn ], based on this feature vector [δ v1 ,δ v2 ,δ v3 ,...δ vn ,], extract δ vmax Similarly, the temperature extreme difference of each data frame in the charging trip is calculated as follows: t =t max -t min ; where δ t is the temperature extreme difference of the current data frame, t max is the maximum value of the sensor temperature in the current data frame, t min is the minimum value of the sensor temperature in the current data frame; the temperature extreme difference characteristic vector [δ t1 ,δ t2 ,δ t3 ,...δ tn ], based on this feature vector [δ t1 ,δ t2 ,δ t3 ,...δ tn ], extract δ tmax .

[0172] ②Charging current maximum value feature. Based on the total charging current field, extract the maximum value of the charging current I for each trip max .

[0173] ③ High current ratio characteristics. High current refers to the current with a larger value in the current charging process. If the charging current is above I ampere, it can be regarded as high current charging. Generally, I≥100A. Based on the total charging current field, the current high current ratio calculation formula is: where r bc is the proportion of large current, ε p is the total number of high current frames in the current charging trip, ε t The total number of frames in the current charging trip.

[0174] ④Charging current range characteristics. Based on the total charging current field, extract the range I of the charging current for each trip r =I max -I min ; where I max is the maximum charging current of the current trip, I min It is the minimum charging current of the current trip.

[0175] ⑤ Skewness feature. Based on the voltage list field [v1,v2,v3...v n ] and the temperature list field [t1,t2,t3...t n ], the calculation formula for the voltage list skewness of each data frame of the current charging trip is: where ψ v is the voltage list skewness of the current data frame, x i is the single cell voltage value of the current data frame, u v is the voltage list mean value of the current data frame, σ v is the voltage list standard deviation of the current data frame. The voltage skewness vector [ψ v1 ,ψ v2 ,ψ v3 ,...ψ vn ], based on this eigenvector [ψ v1 ,ψ v2 ,ψ v3 ,...ψ vn ], extract ψ vmax ; Similarly, the calculation formula for the temperature list skewness of each data frame of the current charging trip is: where ψ t is the temperature list skewness of the current data frame, x i is the sensor temperature value of the current data frame, u t is the mean of the temperature list of the current data frame, σ t is the standard deviation of the temperature list of the current data frame. The temperature skewness vector [ψ t1 ,ψ t2 ,ψ t3 ,...ψ tn], based on this eigenvector [ψ t1 ,ψ t2 ,ψ t3 ,...ψ tn ], extract ψ tmax ;

[0176] The frequency domain features include: one or any combination of pressure difference wavelet analysis features, temperature difference wavelet analysis features, maximum monomer voltage wavelet features and minimum monomer wavelet features.

[0177] The voltage difference characteristic vector of each charging trip is [δ v1 ,δ v2 ,δ v3 ,...δ vn ,], temperature difference characteristic vector [δ t1 ,δ t2 ,δ t3 ,...δ tn ,], the maximum single cell voltage characteristic vector is [v max1 ,v max2 ,v max2 ...v maxn ], the minimum single cell voltage characteristic vector is [v min1 ,v min2 ,v min2 ...v minn ]; for the current stroke pressure difference, temperature difference, maximum and minimum single cell voltage feature vectors, use the Ricker wavelet basis function in CWT (continuous wavelet analysis) to extract features. The specific calculation formula is: Where a is the width parameter, and x is the above four eigenvectors; the corresponding wavelet analysis result τ is calculated by this formula vr , τ tr , τ vmax , τ vmin .

[0178] Data aggregation: The nxm-dimensional time series data of each charging trip is aggregated into 1x h-dimensional aggregated data through feature extraction. Compared with pure time series features, aggregated features can better reflect the current trip charging information and are more conducive to AI model learning. Therefore, according to the above aggregation method, each vehicle is aggregated into vxh-dimensional data; v represents the number of current vehicle charging trips, and h represents the number of aggregated features. The design is based on strongly correlated features, and the aggregated feature vector expression is:

[0179] H=[time,i max ,i r ,e,r max ,m,f v ,δ vmax ,δ tmax ,Imax ,r bc ,I r ,Δsoc,ψ vmax ,ψ tmax ,τ vr τ tr ,τ vmax ,τ vmin ,label]

[0180] The present invention can use the python development environment to build code based on any integrated model to train and test the strongly correlated features of the vehicle. The integrated models mainly include Bagging and Boosting. The integrated models of Bagging include: RF; the integrated models of Boosting include GBDT, XGBOOST and LIGHTGBM, etc. Since there are many types of integrated models, in order to avoid redundancy, the present invention specifically takes the XGBOOST integrated model as an example for subsequent modeling and testing. XGBOOST is an end-to-end gradient boosting tree system, which is widely used in classification and regression tasks. Its advantages are fast training speed, automatic processing of missing values, use of the second-order derivative information of the objective function, and introduction of regularization terms to prevent model overfitting.

[0181] AI model building includes training set division, test set division, pre-adjustment of parameters, model training and testing, specifically:

[0182] ① Training and test set division. First, based on the real vehicle data, the aggregated feature vector is completed according to the above method. Secondly, the vehicle data is divided into a training set and an independent test set. The ratio of the two can be set according to the actual situation. About 80% or 90% of the data in the training set is used for training, and 10% or 20% of the data is used for verification. The verification set is used to adjust the hyperparameters of the model to prevent overfitting of the model and improve the generalization ability of the model. The ratio of the training and test sets can be adjusted as needed;

[0183] ② Pre-adjust model parameters. Specifically, use the training and validation data sets that have been divided in step ①, and use the grid search method to determine the optimal values ​​of the model hyperparameters in turn. The order of determining the hyperparameters is learning_rate, max_depth, scale_pos_weight, gamma, colsample_bytree, lambda, alpha, n_estimator, etc.; select binary:logistic as the objective function, and output the failure probability of the current charging trip to achieve binary classification. Parameter pre-adjustment ensures that the model's indicators on the validation set are optimal. The evaluation indicators include accuracy, precision, recall, and F1 value.

[0184] ③ Model training and testing. Specifically, use the pre-trained hyperparameters in step ② to iteratively train a stable version of the XGBOOST model, and perform independent tests on the test set divided in step ①. If the test results show that all faulty vehicles are recalled and there are no false positives for normal vehicles, it means that the model has achieved a relatively good result and can be used for online deployment on the cloud platform; Figure 4 and Figure 5 The model predicts each charging trip of normal vehicles and faulty vehicles in the test set. The results show that the model can achieve accurate early warning and locate the faulty charging cycle for secondary analysis based on the prediction results.

[0185] Preferably, the preset grading strategy includes at least two warning levels. The warning level is determined and the warning is executed according to the prediction result output by the graded warning model. Each warning accumulates a warning number. When the warning number reaches a specified number, a maintenance prompt is issued.

[0186] Preferably, the prediction result is the proportion of charging trip failures of the vehicle at the current moment.

[0187] The model online prediction specifically includes encapsulating the local offline built data processing, feature design code and trained AI model into API and deploying it to the cloud platform to achieve real-time graded warning of vehicles, and adding the faulty vehicle data processed by the model online warning to the offline training set, continuously iterating and optimizing the model to enhance the generalization ability of the model. See the cloud platform deployment model graded warning process for details. Figure 6 , the specific method is:

[0188] ① If the model predicts that the proportion of charging trip failures of the vehicle at the current moment is greater than or equal to 80%, a Level I warning will be implemented, and the number of Level I warnings for the current vehicle will be increased by 1. If the number of Level I warnings reaches 3 times, the cloud platform will send a reminder message to ask the owner to go to the maintenance shop for inspection;

[0189] ② If the model predicts that the proportion of charging trip failures of the vehicle at the current moment is greater than or equal to 50% and less than 80%, a Level II warning will be implemented, and the number of Level II warnings for the current vehicle will be increased by 1. If the number of Level II warnings reaches 6 times, the cloud platform will send a reminder message to ask the owner to go to the maintenance shop for inspection;

[0190] ③ If the model predicts that the proportion of charging trip failures of the vehicle at the current moment is greater than 20% or equal to and less than 50%, a Level III warning will be implemented, and the number of Level III warnings for the current vehicle will be increased by 1. If the number of Level III warnings reaches 10 times, the cloud platform will send a reminder message to ask the owner to go to the maintenance shop for inspection;

[0191] ④ If the model predicts that the proportion of charging trip failures of the vehicle at the current moment is less than 20%, a Level IV warning will be implemented, and the number of Level IV warnings for the current vehicle will be increased by 1. If the number of Level IV warnings reaches 15 times, the cloud platform will send a reminder message to ask the owner to go to the maintenance shop for inspection;

[0192] Except for the above strategies, other strategies do not trigger the graded warning mechanism.

[0193] A computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, a warning method for thermal runaway of a power battery is implemented.

[0194] An electronic device comprises a memory and a processor, wherein the memory is used to store at least one instruction, and the processor is used to execute at least one instruction to implement an early warning method for thermal runaway of a power battery.

[0195] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0196] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0197] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0198] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0199] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0200] Memory includes non-permanent memory in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0201] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0202] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0203] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. An early warning method for thermal runaway of a power battery, characterized in that: The following steps are involved: The original real vehicle data is preprocessed to obtain charging trip data; Obtain the strong correlation features between charging trip data and thermal runaway, aggregate the strongly correlated features, and input the aggregation results into the pre-trained hierarchical warning model; The strong correlation features include: mechanism features, statistical features and frequency domain features; The graded warning model determines the warning level of thermal runaway according to the preset classification strategy and executes the warning.

2. The early warning method for thermal runaway of a power battery according to claim 1, characterized in that: The pre-processing comprises: Obtain original real vehicle data of normal vehicles and faulty vehicles; Filter the original fields available in the original real vehicle data; Clean the original fields according to the preset cleaning strategy; The charging data is divided into at least two charging trip data according to the frame number.

3. The early warning method for thermal runaway of a power battery according to claim 2, characterized in that: The preset cleaning strategy includes: Reorganize the original fields by sorting; Filter the fields during the vehicle charging trip; Filter the data frames in the fast charging phase; Fill in the missing frame data.

4. The early warning method for thermal runaway of a power battery according to claim 1, characterized in that: The mechanism characteristics include: charging overcurrent characteristics, charging rate characteristics, charging internal resistance characteristics, daily average mileage characteristics and monomer consistency characteristics; The statistical features include: one or any combination of voltage-temperature extreme difference feature, charging current maximum value feature, large current ratio feature, charging current extreme difference feature and skewness feature; The frequency domain features include: one or any combination of pressure difference wavelet analysis features, temperature difference wavelet analysis features, maximum monomer voltage wavelet features and minimum monomer wavelet features.

5. The early warning method for thermal runaway of a power battery according to claim 1, characterized in that: The preset grading strategy includes at least two warning levels. The warning level is determined and the warning is executed according to the prediction result output by the graded warning model. Each warning accumulates a warning number. When the warning number reaches a specified number, a maintenance prompt is issued.

6. The method for early warning of thermal runaway of a power battery according to claim 5, characterized in that: The prediction result is the charging trip failure ratio of the vehicle at the current moment.

7. The early warning system for thermal runaway of power batteries is characterized by: include: A data acquisition preprocessing subsystem, wherein the data acquisition preprocessing subsystem obtains charging trip data by preprocessing the original real vehicle data; A feature aggregation subsystem, wherein the feature aggregation subsystem obtains strongly correlated features between charging trip data and thermal runaway, aggregates the strongly correlated features, and inputs the aggregation results into a pre-trained hierarchical warning model; The strong correlation features include: mechanism features, statistical features and frequency domain features; The early warning subsystem includes a hierarchical early warning model, which determines the early warning level of thermal runaway according to a preset hierarchical strategy and executes the early warning.

8. The power battery thermal runaway warning system according to claim 7, characterized in that: The mechanism characteristics include: charging overcurrent characteristics, charging rate characteristics, charging internal resistance characteristics, daily average mileage characteristics and monomer consistency characteristics; The statistical features include: one or any combination of voltage-temperature extreme difference feature, charging current maximum value feature, large current ratio feature, charging current extreme difference feature and skewness feature; The frequency domain features include: one or any combination of pressure difference wavelet analysis features, temperature difference wavelet analysis features, maximum monomer voltage wavelet features and minimum monomer wavelet features.

9. The power battery thermal runaway warning system according to claim 7, characterized in that: The pre-training of the hierarchical warning model based on real vehicle data includes: training set division, test set division, pre-adjustment of parameters, model training and testing.

10. The power battery thermal runaway warning system according to claim 7, characterized in that: The preset classification strategy includes at least two warning levels. The warning level is determined and the warning is executed according to the prediction results output by the graded warning model. Each warning accumulates a warning number. When the number of warnings reaches a specified number, a maintenance prompt is issued; The prediction result is the proportion of charging trip failures of the vehicle at the current moment.

11. The power battery thermal runaway warning system according to claim 10, characterized in that: The early warning of the early warning subsystem includes: If the warning result shows that the proportion of charging trip failures of the vehicle at the current moment is greater than or equal to 80%, a Level I warning is implemented, and the number of Level I warnings of the current vehicle is accumulated by 1. If the number of Level I warnings reaches 3 times, a prompt maintenance message is sent; If the warning result shows that the proportion of charging trip failures of the vehicle at the current moment is greater than or equal to 50% and less than 80%, a Level II warning is implemented, and the number of Level II warnings of the current vehicle is accumulated by 1. If the number of Level II warnings reaches 6 times, a prompt maintenance message is sent; If the warning result shows that the proportion of charging trip failures of the vehicle at the current moment is greater than or equal to 20% and less than 50%, a Level III warning is implemented, and the number of Level III warnings of the current vehicle is accumulated by 1. If the number of Level III warnings reaches 10 times, a maintenance information is sent; If the warning result shows that the proportion of charging trip failures of the vehicle at the current moment is less than 20%, a Level IV warning is implemented, and the number of Level IV warnings of the current vehicle is accumulated by 1. If the number of Level IV warnings reaches 15 times, a prompt maintenance message is sent; Except for the above situations, no warning will be triggered.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by the processor, the early warning method for thermal runaway of a power battery according to any one of claims 1 to 6 is implemented.

13. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory is used to store at least one instruction, and the processor is used to execute at least one instruction to implement the early warning method for thermal runaway of a power battery according to any one of claims 1 to 6.

Images