Battery fault prediction method and device, cloud, medium, program product and vehicle

Abstract

The invention relates to the technical field of vehicles, and discloses a battery fault prediction method and device, a cloud, a medium, a program product and a vehicle, and the method comprises the steps: obtaining full sample feature data, determining a plurality of first quantiles according to the distribution of the feature value of each feature in the full sample feature data, determining a plurality of second quantiles according to the distribution of the characteristic value of each characteristic in the fault battery sample data; the multiple first quantiles and the multiple second quantiles are combined and then sorted, and the feature weight of each feature is determined according to the position relation between the multiple second quantiles and the multiple first quantiles; according to the distribution position of the fault battery sample data in the full sample feature data, generating a weight coefficient corresponding to a feature value of each feature in the full sample feature data; and determining a fault risk score of the target battery according to the weight coefficient and the corresponding feature weight. Compared with the prior art, the accuracy of battery fault prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a battery failure prediction method, device, equipment, medium, program product and vehicle. Background Art

[0002] As the popularity of electric vehicles continues to rise, electric vehicle accidents caused by thermal runaway of power batteries are increasing, so the safety of power batteries has become one of the key issues that need to be urgently addressed in the new energy vehicle industry. In related technologies, the early warning function of sudden failures such as self-heating runaway of power batteries can be realized based on big data recognition technology; however, there are often many factors that affect the failure of power batteries. Due to the technical limitations of related technologies, it is difficult to accurately predict failures and timely warn before power batteries fail. Therefore, related technologies have the problem of inaccurate prediction of power battery failures. Summary of the invention

[0003] In view of this, the present invention provides a battery fault prediction method, device, cloud, medium, program product and vehicle to improve the accuracy of vehicle battery fault prediction.

[0004] In a first aspect, the present invention provides a battery failure prediction method, the method comprising:

[0005] Acquire full sample feature data, where the full sample feature data includes faulty battery sample data and target battery sample data, where the target battery sample data is sample data of the target battery for which fault probability prediction is to be performed;

[0006] Determine a plurality of first quantiles according to the distribution of the characteristic value of each characteristic in the full sample characteristic data, and determine a plurality of second quantiles according to the distribution of the characteristic value of each characteristic in the faulty battery sample data;

[0007] For each feature, multiple first quantiles and multiple second quantiles are combined and sorted, and a feature weight of each feature is determined according to a positional relationship between the multiple second quantiles and the multiple first quantiles;

[0008] Generate a weight coefficient corresponding to the characteristic value of each characteristic in the full amount of sample characteristic data according to the distribution position of the faulty battery sample data in the full amount of sample characteristic data;

[0009] The failure risk score of the target battery is determined based on the weight coefficient corresponding to the characteristic value of each characteristic in the target battery sample data and the corresponding characteristic weight; the failure risk score is used to characterize the failure probability of the target battery.

[0010] Based on the processing of the full - scale sample feature data and the faulty battery sample data therein, the present invention determines the first quantile representing the distribution of the full - scale sample data and the second quantile representing the distribution of the faulty battery sample data. Furthermore, based on this, the feature weights and weight coefficients are determined, so that the failure probability of each target battery can be obtained. It can be seen that the present invention can comprehensively consider the relationship between the faulty battery sample data and the full - scale sample feature data, and predict the failure of the non - faulty batteries based on the faulty battery samples, thereby improving the accuracy of battery failure prediction. Furthermore, it better realizes the early warning function for sudden - type failures such as self - heating out - of - control of power batteries, and improves the safety of power batteries of electric vehicles.

[0011] In an alternative embodiment, the failure risk score includes an initial risk score and a final risk score. The failure risk score of the target battery is determined according to the weight coefficient and the corresponding feature weight corresponding to the feature value of each feature in the target battery sample data, including:

[0012] For each target battery, multiply the feature weight of each feature by the corresponding weight coefficient and then accumulate them to obtain the initial risk score of each target battery;

[0013] Select the target batteries whose initial risk scores are within a preset range and use them as potential faulty batteries;

[0014] Perform penalty adjustment on the initial risk scores of the potential faulty batteries to obtain the final risk scores of the potential faulty batteries.

[0015] Based on the screening of the target batteries according to the initial risk scores, the present invention can screen out the potential faulty batteries with a greater probability of failure, so as to more accurately predict battery failures.

[0016] In an alternative embodiment, performing penalty adjustment on the initial risk scores of the potential faulty batteries includes:

[0017] Select a target feature from all the features in the faulty battery sample data;

[0018] Determine the feature distance between the target feature of the potential faulty battery and the target feature of the faulty battery;

[0019] Generate a penalty value corresponding to the feature distance, and the penalty value is positively correlated with the feature distance;

[0020] Determine the product of the initial risk score and the penalty value as the final risk score.

[0021] Based on the above penalty adjustment method, the present invention considers the characteristic distance between the target battery and the faulty battery, thereby generating a corresponding penalty value. This method can more accurately determine the fault risk score of the target battery and improve the accuracy of battery fault prediction.

[0022] In an alternative embodiment, according to the distribution of the feature values of each feature in the full sample feature data, a plurality of first quantiles are determined, including: sorting the feature values of each feature in the full sample feature data according to a preset sorting method, and determining a plurality of first quantiles for each feature at a first preset interval;

[0023] According to the distribution of the feature values of each feature in the faulty battery sample data, a plurality of second quantiles are determined, including: sorting the feature values of each feature in the faulty battery sample data according to a preset sorting method, and determining a plurality of second quantiles for each feature at a second preset interval;

[0024] Wherein, the preset sorting method is from small to large or from large to small, and the first preset interval and the second preset interval are the same or different.

[0025] The present invention can sort the feature values of the full sample feature data and the faulty battery sample data in a manner from large to small or from small to large, thereby determining corresponding multiple quantiles at a preset interval. This method analyzes the relationship between the faulty battery sample data and the full sample feature data through quantiles, and thus realizes the function of the distribution characteristics of the faulty battery sample data in the full sample feature data without increasing the amount of calculation, thereby providing a reliable basis for the accurate identification of faulty batteries.

[0026] In an alternative embodiment, according to the positional relationship between the plurality of second quantiles and the plurality of first quantiles, the feature weight of each feature is determined, including:

[0027] According to the positional relationship, a plurality of preset discrimination degrees for characterizing the distribution of the faulty battery sample data in the full sample feature data are determined;

[0028] Determine the average discrimination degree for characterizing the average of the plurality of preset discrimination degrees, and determine the comprehensive discrimination degree for characterizing the weighted average of the average discrimination degree and the plurality of preset discrimination degrees;

[0029] Generate the feature weight corresponding to the comprehensive discrimination degree, and eliminate the features with feature weights less than the set threshold.

[0030] The present invention comprehensively considers multiple preset discrimination degrees, as well as the corresponding average discrimination degrees and comprehensive discrimination degrees, so as to be able to generate feature weights more accurately, and exclude the influence of features with smaller feature weights on the subsequent calculation of risk scores. Therefore, the present invention further improves the accuracy of battery fault prediction.

[0031] In an alternative embodiment, generating the feature weight corresponding to the comprehensive discrimination degree includes:

[0032] Determine the initial weight corresponding to the comprehensive discrimination degree according to the range of the comprehensive discrimination degree;

[0033] Obtain multiple penalty coefficients corresponding one by one to multiple preset discrimination degrees;

[0034] Use the multiple penalty coefficients to attenuate the initial weight to obtain the feature weight.

[0035] For the preset discrimination degrees with poor discrimination, the present invention can reduce the influence of the preset discrimination degrees with poor discrimination on the calculation of the fault risk score by attenuating the initial weight, so as to improve the accuracy of the calculation of the fault risk score, and further improve the accuracy of battery fault prediction.

[0036] In an alternative embodiment, the weight coefficient includes a first coefficient, a second coefficient and a third coefficient; according to the distribution position of the faulty battery sample data in the full sample feature data, generating the weight coefficient corresponding to the feature value of each feature in the full sample feature data includes:

[0037] According to the distribution position of the faulty battery sample data in the full sample feature data, respectively assign the first coefficient to the feature values corresponding to multiple second quantiles, and respectively assign the second coefficient to the feature values corresponding to multiple first quantiles;

[0038] Based on the method of linear interpolation, assign the third coefficient to the remaining feature values;

[0039] Among them, the first coefficient, the second coefficient and the third coefficient are all different.

[0040] The present invention first assigns weight coefficients to the feature values corresponding to the quantiles, and then assigns weight coefficients by means of feature differences on the basis of the already assigned weight coefficients. It can be seen that the present invention can determine the weight coefficients in a gradient manner, so as to be able to calculate the risk score more accurately and improve the accuracy of battery fault prediction.

[0041] In an alternative embodiment, the first coefficient includes a first sub-coefficient and a second sub-coefficient; according to the distribution position of the faulty battery sample data in the full sample feature data, respectively assign the first coefficient to the feature values corresponding to multiple second quantiles, including:

[0042] According to the concentration degree of the faulty battery sample data reaching a first preset degree, assign first sub-coefficients to the eigenvalue corresponding to the second quantile respectively, and according to the dispersion degree of the faulty battery sample data reaching a second preset degree, assign second sub-coefficients to the eigenvalue corresponding to the second quantile respectively;

[0043] The first sub-coefficient is greater than or less than the second sub-coefficient.

[0044] The present invention can also accurately assign first coefficients to the eigenvalues corresponding to the second quantile according to the concentration or dispersion degree of the faulty battery sample data in the full sample feature data, so as to calculate the risk score more accurately and improve the accuracy of battery fault prediction.

[0045] In an optional implementation manner, based on the linear interpolation method, assign third coefficients to the remaining eigenvalues, including:

[0046] Select the third coefficient used to be assigned to the remaining eigenvalues from the data interval composed of the first sub-coefficient, the second sub-coefficient and the second coefficient based on the linear interpolation method.

[0047] The present invention customizes the weight coefficient gradient near the boundary point of the minority class samples based on the linear interpolation method to determine the weight coefficient, significantly improving the separation effect of the minority class samples from the full sample, so as to more accurately identify the target battery samples close to the minority class samples and improve the accuracy of battery fault prediction.

[0048] In a second aspect, the present invention provides a battery fault prediction device, which includes:

[0049] A data acquisition module, configured to acquire full sample feature data, where the full sample feature data includes faulty battery sample data and target battery sample data, and the target battery sample data is the sample data of the target battery to be predicted for the fault probability;

[0050] A data analysis module, configured to determine a plurality of first quantiles according to the distribution of the eigenvalues of each feature in the full sample feature data, and configured to determine a plurality of second quantiles according to the distribution of the eigenvalues of each feature in the faulty battery sample data;

[0051] A weight determination module, configured to, for each feature, sort the plurality of first quantiles and the plurality of second quantiles after merging, and determine the feature weight of each feature according to the positional relationship between the plurality of second quantiles and the plurality of first quantiles;

[0052] A coefficient generation module, configured to generate weight coefficients corresponding to the eigenvalues of each feature in the full sample feature data according to the distribution position of the faulty battery sample data in the full sample feature data;

[0053] A fault prediction module, configured to determine a fault risk score of a target battery according to weight coefficients corresponding to feature values of each feature in target battery sample data and corresponding feature weights; the fault risk score is used to characterize the fault probability of the target battery.

[0054] In a third aspect, the present invention provides a cloud, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the battery fault prediction method according to the first aspect or any corresponding embodiment thereof.

[0055] In a fourth aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored. The computer instructions are used to cause a computer to execute the battery fault prediction method according to the first aspect or any corresponding embodiment thereof.

[0056] In a fifth aspect, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the battery fault prediction method according to the first aspect or any corresponding embodiment thereof.

[0057] In a sixth aspect, the present invention provides a vehicle, which includes a vehicle controller, and the vehicle controller is configured to execute the battery fault prediction method according to the first aspect or any corresponding embodiment thereof. Description of the Drawings

[0058] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related art, the following will briefly introduce the drawings required to be used in the description of the specific embodiments or the related art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0059] Figure 1 is a schematic diagram of the communication architecture between the vehicle end and the cloud according to an embodiment of the present invention.

[0060] Figure 2 is a schematic flowchart of the battery fault prediction method according to an embodiment of the present invention.

[0061] Figure 3 is a schematic flowchart of another battery fault prediction method according to an embodiment of the present invention.

[0062] Figure 4 is a schematic diagram of the initial risk score before feature penalty according to an embodiment of the present invention.

[0063] Figure 5 is a schematic diagram of the final risk score after feature penalty according to an embodiment of the present invention.

[0064] Figure 6 It is a schematic flowchart of yet another battery fault prediction method according to an embodiment of the present invention.

[0065] Figure 7 It is a schematic flowchart of still another battery fault prediction method according to an embodiment of the present invention.

[0066] Figure 8 It is a structural block diagram of a battery fault prediction device according to an embodiment of the present invention.

[0067] Figure 9 It is a schematic diagram of the hardware structure of the cloud in an embodiment of the present invention. Detailed implementation manners

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0069] As Figure 1 shown, the vehicle end 100 can upload battery-related data to the cloud 200 through a wireless communication network. The vehicle end 100 includes vehicle end 1, vehicle end 2, ……, vehicle end n. Among them, the value of n can be 10,000 or more (for example, 100,000), indicating that a large number of vehicle ends 100 upload battery-related data to the cloud 200. The cloud 200 realizes the early warning function for sudden faults such as self-heating out-of-control of power batteries based on big data recognition technology. However, conventional big data recognition methods not only require building an effective feature system but also need to deeply integrate the experience and knowledge of experts in related fields. Traditional machine learning algorithms (such as tree models, logistic regression, etc.) and deep learning models often have difficulty distinguishing minority class samples from the overall samples during the application to imbalanced samples. Even after parameter tuning, there are still generally underfitting problems or overfitting problems, resulting in a large gap between the model performance and the upper limit, that is, the problem of low accuracy of battery fault prediction results.

[0070] According to an embodiment of the present invention, an embodiment of a battery fault prediction method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0071] In this embodiment, a battery fault prediction method is provided, which can be used in the above-mentioned cloud or vehicle end. Figure 2 It is a flowchart of the battery fault prediction method according to an embodiment of the present invention, as Figure 2 shown. The process includes the following steps:

[0072] Step S201, obtain full-scale sample feature data, where the full-scale sample feature data includes faulty battery sample data and target battery sample data, and the target battery sample data is the sample data of the target battery for which the fault probability is to be predicted.

[0073] Among them, the batteries involved in this embodiment include faulty batteries and target batteries, and specifically can be the power batteries configured on electric vehicles or the battery cells in battery packs. The faulty battery sample data is minority-class sample data, which is the sample data of the batteries that have already failed, and the full-scale sample feature data includes minority-class sample data.

[0074] Specifically, a large amount of battery-related data is uploaded from the vehicle end to the cloud. For example, signals such as current, voltage, and SOC (State Of Charge) are uploaded to the cloud, and these signals serve as the original data of the cloud. The cloud constructs the original data into full-scale sample feature data based on the battery cell mechanism, that is, processes the original data into full-scale sample feature data through a data processing method based on feature engineering. The full-scale sample feature data in this embodiment includes, but is not limited to, features in mechanism dimensions such as self-discharge characteristics, capacity characteristics, and internal resistance. For the case where the battery fault prediction method of the present invention is executed at the vehicle end, the cloud also needs to send the full-scale sample feature data to the vehicle end for executing the battery fault prediction function. The vehicle end involved in the present invention is the vehicle end, and the vehicles include, but are not limited to, electric vehicles and hybrid vehicles, etc.

[0075] Step S202, determine a plurality of first quantiles according to the distribution of the feature values of each feature in the full-scale sample feature data, and determine a plurality of second quantiles according to the distribution of the feature values of each feature in the faulty battery sample data.

[0076] Each feature in this embodiment includes, but is not limited to, features in mechanism dimensions such as self-discharge characteristics, capacity characteristics, and internal resistance. The self-discharge characteristics specifically include features such as the magnitude of the pressure difference, the change in the pressure difference, and the slope of the pressure difference. The feature value refers to the value of these sample signals. Taking the magnitude of the pressure difference feature as an example, for example, the magnitude of the pressure difference of a certain battery sample is 50 mV (millivolt). In this embodiment, the plurality of first quantiles represent the quantiles of the full-scale sample feature data, and the plurality of second quantiles represent the quantiles of the faulty battery sample data.

[0077] Specifically, this embodiment can separately count the distributions of the eigenvalues of the full - scale sample feature data and the faulty battery sample data. Then, according to the magnitudes of the eigenvalues, taking the features as dimensions, the full - scale sample feature data and the faulty battery sample data are intervalized, thereby obtaining multiple first quantiles and multiple second quantiles. For example, in this embodiment, the interval can be set at 10% quantiles from the minimum value to the maximum value of the eigenvalues. Then, the multiple first quantiles and the multiple second quantiles can respectively include 0%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100%.

[0078] Step S203: For each feature, after merging and sorting the multiple first quantiles and the multiple second quantiles, determine the feature weight of each feature according to the positional relationship between the multiple second quantiles and the multiple first quantiles.

[0079] For any feature, by sorting the quantiles of the minority - class sample data and the quantiles of the full - scale sample feature data together, the discriminability of the feature can be evaluated. Discriminability, that is, the ability to separate as many minority - class samples as possible from the full - scale samples, is used to evaluate the discrimination effect of the feature.

[0080] Among them, the feature weight is used to characterize the discriminability of the current feature from other features. For example, the higher the feature weight, the higher the discriminability can be indicated. Or, in some embodiments, the feature weight is negatively correlated with the discriminability. By means of sorting, the data to be processed has the property of monotonic numerical magnitude.

[0081] For each feature, this embodiment respectively merges the corresponding multiple first quantiles and multiple second quantiles. Based on the positional relationship between the quantiles of the sorted minority - class sample data and the quantiles of the full - scale sample feature data, by analyzing whether the quantiles of the minority - class sample data are distributed at the top or bottom, it can be determined whether the minority - class samples are distributed at the top or bottom of the full - scale samples. This method can significantly reduce the computational amount and then determine the feature weight.

[0082] Step S204: Generate a weight coefficient corresponding to the eigenvalue of each feature in the full - scale sample feature data according to the distribution position of the faulty battery sample data in the full - scale sample feature data.

[0083] Among them, the weight coefficient is used to represent the score or performance of a certain battery sample compared to the faulty battery sample in the current feature.

[0084] For example, a larger weight coefficient is assigned to the features in the faulty battery sample data. In this embodiment, the more similar the feature of a certain battery sample is to the feature of the faulty battery, the higher the weight coefficient. For example, the more similar the feature of the battery sample is to the feature of the faulty battery, the closer the weight coefficient is to 1.

[0085] Step S205: Determine the failure risk score of the target battery according to the weight coefficient corresponding to the eigenvalue of each feature in the target battery sample data and the corresponding feature weight; the failure risk score is used to represent the failure probability of the target battery.

[0086] In specific implementation, the failure risk score of each battery in the full sample can be determined according to the weight coefficient corresponding to the eigenvalue of each feature in the full sample feature data and the corresponding feature weight.

[0087] Among them, the failures of the batteries involved in the embodiments of the present invention include, for example, but are not limited to, power battery thermal runaway failures.

[0088] In some optional implementation manners, the greater the failure risk score, the higher the failure probability of the target battery. The target battery with a higher failure probability, that is, a potential failure battery, has a failure risk score closer to that of a failed battery.

[0089] The battery failure prediction method provided in this embodiment specifically determines the first quantile representing the distribution of the full sample data and the second quantile representing the distribution of the failed battery sample data based on the processing of the full sample feature data and the failed battery sample data therein, and then determines the feature weight and the weight coefficient on this basis, so as to obtain the failure risk score of each target battery. The closer the failure risk score is to that of the target battery of the failed battery, the greater the possibility of failure. It can be seen that this embodiment realizes the technical purpose of predicting the failure probability of the target battery by determining the similarity degree between the full sample and the failed battery sample, thereby achieving the technical purpose of identifying the failure probability of the target battery. It can be seen that the present invention can comprehensively consider the relationship between the failed battery sample data and the full sample feature data, and perform failure prediction on the batteries that have not failed based on the failed battery samples, thereby improving the accuracy of battery failure prediction, and further better realizing the early warning function for sudden failures such as power battery self-heating runaway, and improving the safety of the power batteries of electric vehicles.

[0090] In this embodiment, a battery failure prediction method is provided, which can be used in the above-mentioned cloud or vehicle terminal. Figure 3 It is a flowchart of the battery failure prediction method according to an embodiment of the present invention, as Figure 3 shown, and this process includes the following steps:

[0091] Step S301: Obtain full sample feature data, where the full sample feature data includes failed battery sample data and target battery sample data, and the target battery sample data is the sample data of the target battery to be predicted for the failure probability. For details, please refer to Figure 2 Step S201 of the embodiment shown herein, which will not be elaborated herein.

[0092] Step S302: Determine multiple first quantiles according to the distribution of the feature values of each feature in the full sample feature data, and determine multiple second quantiles according to the distribution of the feature values of each feature in the faulty battery sample data. For details, please refer to Figure 2 Step S202 of the embodiment shown, which will not be elaborated here.

[0093] Step S303: For each feature, sort the combined multiple first quantiles and multiple second quantiles, and determine the feature weight of each feature according to the positional relationship between the multiple second quantiles and the multiple first quantiles. For details, please refer to Figure 2 Step S203 of the embodiment shown, which will not be elaborated here.

[0094] Step S304: Generate a weight coefficient corresponding to the feature value of each feature in the full sample feature data according to the distribution position of the faulty battery sample data in the full sample feature data. For details, please refer to Figure 2 Step S204 of the embodiment shown, which will not be elaborated here.

[0095] Step S305: Determine the fault risk score of the target battery according to the weight coefficient corresponding to the feature value of each feature in the target battery sample data and the corresponding feature weight; the fault risk score is used to characterize the fault probability of the target battery.

[0096] Specifically, the fault risk score includes an initial risk score and a final risk score, and the above step S305 includes:

[0097] Step S3051: For each target battery, multiply the feature weight of each feature by the corresponding weight coefficient and then accumulate to obtain the initial risk score of each target battery.

[0098] In this embodiment, the initial risk score is denoted as pre_value, the number of features is n, the feature weight of each feature is denoted as weight i and the corresponding weight coefficient is denoted as coef i , then the initial risk score

[0099] Step S3052: Screen out the target batteries whose initial risk scores are within a preset range and use them as potentially faulty batteries.

[0100] In this embodiment, screen out the target batteries whose initial risk scores are greater than a preset threshold as potentially faulty batteries, and the preset threshold can be 60 points, for example.

[0101] Step S3053: Perform penalty adjustment on the initial risk scores of the potentially faulty batteries to obtain the final risk scores of the potentially faulty batteries.

[0102] Specifically, the initial risk score is penalized and adjusted according to the eigenvalue situation of the characteristics of the potentially faulty battery, so that the battery with characteristics closer to the target battery of the faulty battery has a greater final risk score.

[0103] Based on the calculation of the initial risk score and the screening of the target battery according to the initial risk score, this embodiment can screen out potentially faulty batteries with a greater probability of failure, so as to more accurately predict battery failures.

[0104] In some optional embodiments, the above step S3053 includes:

[0105] Step a1, select target characteristics from all characteristics in the faulty battery sample data.

[0106] Among them, the target characteristics are characteristics with a feature weight greater than a specified value, that is, the target characteristics include several characteristics with relatively large feature weights; the several characteristics are represented as n characteristics, where n≥3.

[0107] Step a2, determine the feature distance between the target characteristics of the potentially faulty battery and the target characteristics of the faulty battery.

[0108] Step a3, generate a penalty value corresponding to the feature distance, and the penalty value is positively correlated with the feature distance.

[0109] Among them, the n characteristics correspond to n penalty values, which can be expressed as p_coef1, p_coef2, ……, p_coefn. If the feature distance between the target characteristics of a potentially faulty battery and the target characteristics of the faulty battery is large, a relatively small penalty value can be generated correspondingly, for example, the penalty value is 0.

[0110] Step a4, determine the product of the initial risk score and the penalty value as the final risk score.

[0111] In this embodiment, the initial risk score is represented as pre_value, and the final risk score is represented as final_value, then final_value = pre_value * p_coef1 * p_coef2 * …… * p_coefn.

[0112] Figure 4 Schematically shows the initial risk score before feature penalty, Figure 5 Schematically shows the final risk score after feature penalty, Figure 4 The abscissa in is the initial risk score of dimension A, and the ordinate is the initial risk score of dimension B, Figure 5The abscissa therein is the final risk score of Dimension A, and the ordinate is the final risk score of Dimension B. Among them, Dimension A and Dimension B represent the characteristics of two different business dimensions. For example, the abscissa is the initial risk score or the final risk score of the self-discharge dimension, and the ordinate is the initial risk score or the final risk score of the capacity dimension; or the abscissa is the initial risk score or the final risk score of the capacity dimension, and the ordinate is the initial risk score or the final risk score of the internal resistance dimension. Figure 4 and Figure 5 The L2 curve involved in Figure 5 represents the threshold line for screening potential fault groups according to the initial risk score. The points above the L2 curve represent potential fault groups, and the risk of the battery samples is relatively high. Figure 5 Compared with Figure 4 , after feature penalty by the method with a penalty value of 0, most of the points return to the origin.

[0113] Based on the above penalty adjustment method, this embodiment generates corresponding penalty values based on the feature distance between the target battery and the faulty battery. This method can more accurately determine the fault risk score of the target battery and improve the accuracy of battery fault prediction.

[0114] This embodiment can adjust the initial risk score based on the relative size of the feature values. Specifically, the target battery group with the initial risk score greater than the preset threshold is selected, and after penalty adjustment by the relative size of the feature values, the potential risk group is further classified to obtain the final risk score of the full sample. This method can more accurately identify high-risk samples, provide a reliable basis for subsequent decision-making, and can output the risk list of the battery or vehicle in combination with the final risk score to guide the after-sales recall activity.

[0115] In this embodiment, a battery fault prediction method is provided, which can be used for the above-mentioned cloud or vehicle end. Figure 6 is the flowchart of the battery fault prediction method according to the embodiment of the present invention, as Figure 6 shown, and this process includes the following steps:

[0116] Step S601, obtain the full sample feature data. The full sample feature data includes faulty battery sample data and target battery sample data. The target battery sample data is the sample data of the target battery to be predicted for the fault probability. For details, please refer to step S201 of the embodiment shown in Figure 2 or step S301 of the embodiment shown in Figure 3 , which will not be elaborated here.

[0117] Step S602: Determine multiple first quantiles according to the distribution of the feature values of each feature in the full sample feature data, including: sorting the feature values of each feature in the full sample feature data according to a preset sorting method, and determining multiple first quantiles of each feature according to a first preset interval; determine multiple second quantiles according to the distribution of the feature values of each feature in the faulty battery sample data, including: sorting the feature values of each feature in the faulty battery sample data according to a preset sorting method, and determining multiple second quantiles of each feature according to a second preset interval.

[0118] Specifically, the preset sorting method is from small to large or from large to small, and the first preset interval and the second preset interval may be the same or different.

[0119] In this embodiment, both the first preset interval and the second preset interval may be 10%. 10% can help separate the minority class samples while avoiding increasing the computational complexity. If the interval is too dense, such as 5%, the difference between the feature values corresponding to adjacent quantiles is small and the subsequent computational complexity increases; while if the interval is too sparse, such as 30%, it is likely that there is no difference in the weight coefficient gradient, which is not conducive to separating the minority class samples. In this embodiment, with an interval of 10%, the feature values corresponding to each quantile node are obtained. For example, full sample - pressure difference magnitude - 10% (feature1 - all_sample_10%) can be used to represent that the pressure difference magnitude of 10% of the samples in the full sample is < 10 mV.

[0120] By sorting the feature values of the full sample feature data and the faulty battery sample data respectively in the way of sorting the feature values from large to small or from small to large, multiple corresponding quantiles are determined according to the preset interval in this embodiment. This method analyzes the relationship between the faulty battery sample data and the full sample feature data through quantiles, and thus realizes the function of the distribution characteristics of the faulty battery sample data in the full sample feature data without increasing the computational complexity, providing a reliable basis for the accurate identification of faulty batteries.

[0121] Step S603: For each feature, sort the multiple first quantiles and the multiple second quantiles after merging, and determine the feature weight of each feature according to the positional relationship between the multiple second quantiles and the multiple first quantiles.

[0122] In some optional implementation manners, the above step S603 includes:

[0123] Step b1: Determine multiple preset discrimination degrees for characterizing the distribution of the faulty battery sample data in the full sample feature data according to the positional relationship.

[0124] The multiple preset discrimination degrees in this embodiment may include, but are not limited to, 70% discrimination degree, 80% discrimination degree, and 90% discrimination degree. Among them, the 70% discrimination degree means that 70% of the minority class samples are distributed in the bottom 30% (i.e., 0% - 30%) or the top 30% (i.e., 70% - 100%) of the total samples, the 80% discrimination degree means that 80% of the minority class samples are distributed in the bottom 20% (i.e., 0% - 20%) or the top 20% (i.e., 80% - 100%) of the total samples, and the 90% discrimination degree means that 90% of the minority class samples are distributed in the bottom 10% (i.e., 0% - 10%) or the top 10% (i.e., 90% - 100%) of the total samples.

[0125] Specifically, the 70% discrimination degree can be expressed as discrim_70%, the 80% discrimination degree can be expressed as discrim_80%, and the 90% discrimination degree can be expressed as discrim_90%. The specific value of the discrimination degree can be determined according to the actual positional relationship, and the typical value range of the discrimination degree can be set according to the actual situation.

[0126] Step b2, determine the average discrimination degree for characterizing the average of multiple preset discrimination degrees, and determine the comprehensive discrimination degree for characterizing the weighted average of the average discrimination degree and multiple preset discrimination degrees.

[0127] Among them, the average discrimination degree represents the mean of multiple preset discrimination degrees, and the comprehensive discrimination degree represents the weighted average of multiple average discrimination degrees and multiple preset discrimination degrees.

[0128] In this embodiment, the average discrimination degree can be expressed as discrim_avg, the comprehensive discrimination degree can be expressed as discrim_comp, discrim_avg = (discrim_90% + discrim_80% + discrim_70%) / 3, and discrim_comp = discrim_avg * 0.4 + discrim_90% * 0.3 + discrim_80% * 0.2 + discrim_70% * 0.1.

[0129] Step b3, generate the feature weights corresponding to the comprehensive discrimination degree, and eliminate the features with feature weights less than the set threshold.

[0130] Among them, the generation of the feature weights depends on the comprehensive discrimination degree and is specifically determined according to the value of the comprehensive discrimination degree; for example, the larger the comprehensive discrimination degree, the relatively larger the feature weights, and the smaller the comprehensive discrimination degree, the relatively smaller the feature weights.

[0131] For the features with smaller feature weights, in this embodiment, if the feature weights are less than the set threshold, then these features will not be considered in the subsequent battery fault prediction.

[0132] This embodiment can compare the fault battery sample data with the full sample feature data. For a certain feature in the fault battery sample data, if the feature in the full sample feature data is similar to it, it indicates that the discrimination of this feature is not good, the corresponding feature weight is small, and the impact on the fault risk score is small; if the feature in the full sample feature data is significantly different from it, it indicates that the discrimination of this feature is good, the corresponding feature weight is large, and it will significantly affect the fault risk score.

[0133] Therefore, this embodiment comprehensively considers multiple preset discrimination degrees, as well as the corresponding average discrimination degree and comprehensive discrimination degree, so as to be able to generate feature weights more accurately, and exclude the influence of features with smaller feature weights on the subsequent risk score calculation, thereby further improving the accuracy of battery fault prediction.

[0134] In some optional implementation manners, the above step b3 includes:

[0135] Step c1, determine the initial weight corresponding to the comprehensive discrimination degree according to the range of the comprehensive discrimination degree.

[0136] In this embodiment, the initial weight can be expressed as Initial_fewt. If the range of the comprehensive discrimination degree is discrim_comp < 0.45, then the initial weight Initial_fewt = 0; if the range of the comprehensive discrimination degree is discrim_comp ≥ 0.45, the initial weight is calculated according to the value of the comprehensive discrimination degree, then the initial weight Initial_fewt = 4*(discrim_comp + 0.05)*(discrim_comp + 0.05) - 4*(discrim_comp + 0.05) + 1, so as to convert the value of the initial weight in the range of 0.45 to 1 to the range of 0 to 1.

[0137] Step c2, obtain multiple penalty coefficients corresponding to the multiple preset discrimination degrees one by one.

[0138] Specifically, each preset discrimination degree corresponds to a penalty coefficient. Combining the foregoing embodiments, in this embodiment, the penalty coefficient corresponding to 70% discrimination degree can be expressed as discrim_70%_attecoef, the penalty coefficient corresponding to 80% discrimination degree can be expressed as discrim_80%_attecoef, and the penalty coefficient corresponding to 90% discrimination degree can be expressed as discrim_90%_attecoef; if the 70% discrimination degree discrim_70% ≥ 0.7, then discrim_70%_attecoef = 1, if the 70% discrimination degree discrim_70% < 0.7, then discrim_70%_attecoef = discrim_70% / 0.7; if the 80% discrimination degree discrim_80% ≥ 0.6, then discrim_80%_attecoef = 1, if the 80% discrimination degree discrim_80% < 0.6, then discrim_80%_attecoef = discrim_80% / 0.6; if the 90% discrimination degree discrim_90% ≥ 0.5, then discrim_90%_attecoef = 1, if the 90% discrimination degree discrim_90% < 0.5, then discrim_90%_attecoef = discrim_90% / 0.5.

[0139] Step c3, decay the initial weights using multiple penalty coefficients to obtain the feature weights.

[0140] Specifically, multiply all the multiple penalty coefficients by the initial weights to obtain the feature weights. In this embodiment, the feature weights are expressed as fewt, then the feature weights fewt = Initial_fewt * discrim_70%_attecoef * discrim_80%_attecoef * discrim_90%_attecoef.

[0141] In Table 1, a calculation example of the feature weights is illustrated; among them, the features include feature1, feature2, feature3, feature4, feature5, feature6, feature7, feature8, feature9, CASE is used to represent the faulty battery samples, and cell is used to represent all samples.

[0142]

[0143] Table 1

[0144] In this embodiment, for a preset discrimination degree with poor discrimination, by attenuating the initial weight, the influence of the preset discrimination degree with poor discrimination on the calculation of the fault risk score can be significantly reduced, so as to improve the accuracy of the fault risk score calculation, and further improve the accuracy of battery fault prediction.

[0145] This embodiment realizes the function of automatically generating feature weights based on the distribution of minority-class samples in the full-scale sample. Compared with the feature weights generated by traditional machine learning algorithms (such as tree models, logistic regression, etc.), the feature weights generated in this embodiment are more accurate and more in line with the actual business needs, thus significantly improving the prediction accuracy of the model constructed by the battery fault prediction method based on the present invention. It performs well in practical applications, and key model evaluation indicators such as recall rate and F1 score (harmonic mean) have reached the leading level in the industry, fully verifying its effectiveness and practicality.

[0146] Step S604: Generate weight coefficients corresponding to the feature values of each feature in the full-scale sample feature data according to the distribution position of the faulty battery sample data in the full-scale sample feature data.

[0147] In some optional implementation manners, the weight coefficients include a first coefficient, a second coefficient, and a third coefficient. The above step S604 includes:

[0148] Step d1: According to the distribution position of the faulty battery sample data in the full-scale sample feature data, respectively assign the first coefficient to the feature values corresponding to multiple second quantiles, and respectively assign the second coefficient to the feature values corresponding to multiple first quantiles.

[0149] In this embodiment, the distribution position of the faulty battery sample data in the full-scale sample feature data represents the distribution characteristics of the faulty battery sample data in the full-scale sample feature data.

[0150] Step d2: Based on the method of linear interpolation, assign the third coefficient to the remaining feature values.

[0151] Among them, the first coefficient, the second coefficient, and the third coefficient are all different.

[0152] This embodiment can set weight coefficients with a certain gradient according to expert experience. For example, the range of the weight coefficients is from 0 to 1, or from -1 to 1, etc.

[0153] For example, a set of feature values of differential pressure magnitudes are 10mV, 20mV, 30mV, 40mV, 50mV respectively, and their corresponding weight coefficients are 1, 0.9, 0.8, 0.7, 0.6 respectively. For a sample with a differential pressure magnitude feature value of 25mV, the weight coefficient corresponding to the differential pressure magnitude of this sample is determined to be 0.85 by the method of linear interpolation.

[0154] By first assigning weight coefficients to the eigenvalue corresponding to the quantile, and then assigning weight coefficients in the form of eigenvalue differences based on the assigned weight coefficients, this embodiment can determine the weight coefficients in a gradient manner, so as to calculate the risk score more accurately and improve the accuracy of battery fault prediction.

[0155] In some alternative embodiments, the first coefficient includes a first sub - coefficient and a second sub - coefficient; step d1 includes: according to the concentration degree of the faulty battery sample data reaching a first preset degree, respectively assigning the first sub - coefficient to the eigenvalue corresponding to the second quantile, and according to the dispersion degree of the faulty battery sample data reaching a second preset degree, respectively assigning the second sub - coefficient to the eigenvalue corresponding to the second quantile; the first sub - coefficient is greater than or less than the second sub - coefficient.

[0156] Taking the first sub - coefficient being greater than the second sub - coefficient as an example, the difference between the first sub - coefficient and the second sub - coefficient can be a preset value, and this preset value can be 1 for example.

[0157] It can be adjusted according to the distribution of minority class samples, feature weights, and business experience. The purpose and adjustment direction are to make the risk scores of minority class samples as high as possible, and at the same time, try to reduce the possibility of high risk scores of normal samples. For example, let the weight coefficient of the quantile node where minority class samples are concentrated be 1, and let the weight coefficient of the quantile node where minority class samples are sparse be 0. The weight coefficients of the nodes between 1 and 0 are adjusted according to the distribution of minority class samples, feature weights, and business experience.

[0158] This embodiment accurately assigns the first coefficient to the eigenvalue corresponding to the second quantile according to the concentration or dispersion degree of the faulty battery sample data in the full - volume sample feature data, so as to calculate the risk score more accurately and improve the accuracy of battery fault prediction.

[0159] In some alternative embodiments, step d2 includes: selecting a third coefficient for assigning to the remaining eigenvalues from the data interval composed of the first sub - coefficient, the second sub - coefficient, and the second coefficient based on linear interpolation.

[0160] For two adjacent coefficients (such as two adjacent first coefficients or two adjacent second coefficients or adjacent first and second coefficients), assuming the corresponding eigenvalues are 0.8 and 0.85 respectively, the third coefficient assigned to these two coefficients by linear interpolation can be 0.85.

[0161] In Table II, an example of generating weight coefficients in combination with feature weights is shown; where feature1 can represent a certain feature of each battery sample, the statistical nodes include the aforementioned multiple first quantiles and second quantiles, and CASE is used to represent faulty battery samples.

[0162] features statistical node eigenvalue feature weight CASE distribution position in the full volume weight coefficient feature1 lower boundary -0.18324 0.662 top 0.00 feature1 all_sample_0% -0.15270 0.662 top -1.00 feature1 all_sample_10% -0.00293 0.662 top -1.00 feature1 all_sample_20% -0.00193 0.662 top -0.90 feature1 all_sample_30% -0.00122 0.662 top -0.60 feature1 all_sample_40% -0.00063 0.662 top -0.30 feature1 all_sample_50% -0.00007 0.662 top 0.00 feature1 all_sample_60% 0.00050 0.662 top 0.45 feature1 case_sample_0% 0.00098 0.662 top 0.90 feature1 all_sample_70% 0.00113 0.662 top 0.93 feature1 all_sample_80% 0.00193 0.662 top 0.97 feature1 case_sample_10% 0.00219 0.662 top 1.00 feature1 all_sample_90% 0.00312 0.662 top 1.00 feature1 case_sample_20% 0.00391 0.662 top 1.00 feature1 case_sample_30% 0.00454 0.662 top 1.00 feature1 case_sample_40% 0.00548 0.662 top 1.00 feature1 case_sample_50% 0.00643 0.662 top 1.00 feature1 case_sample_60% 0.00800 0.662 top 1.00 feature1 case_sample_70% 0.00840 0.662 top 1.00 feature1 case_sample_80% 0.00868 0.662 top 1.00 feature1 case_sample_90% 0.00924 0.662 top 1.00 feature1 case_sample_100% 0.00953 0.662 top 1.00 feature1 all_sample_100% 0.02130 0.662 top 1.00 feature1 upper boundary 0.02556 0.662 top 0.00

[0163] Table II

[0164] In this embodiment, according to the distribution of the minority class samples, the weight coefficients corresponding to the feature weights are automatically generated, achieving flexible adjustment of the gradient transition near the boundary point of the minority class samples. Combining with specific business requirements, smoother or steeper gradient changes can be customized, thus significantly improving the recognition effect of the minority class samples.

[0165] Step S605: Determine the fault risk score of the target battery according to the weight coefficient corresponding to the eigenvalue of each feature in the target battery sample data and the corresponding feature weight; the fault risk score is used to characterize the fault probability of the target battery. For details, please refer to Figure 2 Step S205 of the illustrated embodiment or Figure 3 Step S305 of the illustrated embodiment, which will not be elaborated here.

[0166] Combined with Figure 2 、 Figure 3 and Figure 6 's embodiments, in this embodiment, a battery fault prediction method is provided. This battery fault prediction method is specifically a method for calculating the risk score based on features and is applied to the above-mentioned cloud.

[0167] Figure 7 is a flowchart of the battery fault prediction method according to the embodiment of the present invention. As shown in Figure 7 shown, this process includes the following steps:

[0168] Step S701: The vehicle can upload vehicle-end signals (voltage, current, SOC, etc.) to the cloud through wireless communication.

[0169] Step S702: The vehicle-end signals uploaded to the cloud are used as the cloud original data.

[0170] Step S703: Based on methods such as feature engineering, the cloud original data can be processed into full-scale sample feature data.

[0171] Step S704: Process the full-scale sample feature data to obtain the full-scale sample feature quantiles and the minority class sample feature quantiles; in this embodiment, based on the full-scale sample feature data including minority class samples, the eigenvalue distributions of the minority class samples and the full-scale samples are respectively counted at 10% quantile intervals from the minimum value to the maximum value.

[0172] Step S705: Sort the above two types of quantiles respectively; in this embodiment, for each feature, the feature quantiles of the minority class samples and the full-scale samples are combined and sorted.

[0173] Step S706: Generate feature weights. In this embodiment, based on the positional relationship between the quantiles of the minority class and the overall quantiles, the feature weights of this feature are calculated, and the features with feature weights less than the set threshold are removed.

[0174] Step S707: Generate weight coefficients. In this embodiment, according to the distribution characteristics of the minority class, weight coefficients are assigned to some of the quantiles of the minority class and the overall quantiles. When assigning values, the feature weights and business experience are comprehensively considered, and the weight coefficients at the remaining quantiles are filled by linear interpolation.

[0175] Step S708: Calculate the initial risk score. In this embodiment, based on linear interpolation, the weight coefficient corresponding to a certain sample feature value is calculated. The product of the feature weight and this weight coefficient is the risk score of this feature. The risk scores of all features are accumulated to obtain the initial risk score of the sample.

[0176] Step S709: Perform feature penalty processing. In this embodiment, the groups with initial risk scores exceeding the preset threshold are screened and determined as potential failure groups.

[0177] Step S710: Calculate the final risk score. Among the potential failure groups, in this embodiment, the initial risk score is punished and adjusted according to the relative magnitudes of the sample feature values to obtain the final risk score.

[0178] It can be seen that the present invention can provide a method and system for calculating the battery failure risk score based on the constructed features, so as to solve the deficiencies of traditional artificial intelligence algorithms in learning the deep and complex business experience of power battery failure prediction in the professional field. The present invention innovatively combines the distribution characteristics of minority class samples and expert experience, through feature weight assignment and weight coefficient assignment of nodes, performs linear interpolation processing on feature values to generate corresponding weight coefficients; on this basis, the product of the feature weight and the corresponding weight coefficient is accumulated to obtain the initial score, and further optimization can be carried out through a penalty mechanism to finally obtain the power battery failure risk score. The present invention can customize the gradient according to expert experience near the boundary point of the minority class and perform penalty adjustment on the initial risk score, thereby significantly improving the recognition effect of minority class samples, increasing the upper limit of the model constructed based on the present invention, and providing reliable technical support for the accurate early warning of faults such as power battery thermal runaway.

[0179] In this embodiment, a battery failure prediction device is further provided. This device is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0180] This embodiment provides a battery fault prediction device, as Figure 8 shown, including:

[0181] A data acquisition module 801, configured to acquire full - volume sample feature data, where the full - volume sample feature data includes faulty battery sample data and target battery sample data, and the target battery sample data is the sample data of the target battery for which the fault probability is to be predicted.

[0182] A data analysis module 802, configured to determine a plurality of first quantiles according to the distribution of the feature values of each feature in the full - volume sample feature data, and to determine a plurality of second quantiles according to the distribution of the feature values of each feature in the faulty battery sample data.

[0183] A weight determination module 803, configured to, for each feature, sort the combined plurality of first quantiles and plurality of second quantiles, and determine the feature weight of each feature according to the positional relationship between the plurality of second quantiles and the plurality of first quantiles.

[0184] A coefficient generation module 804, configured to generate a weight coefficient corresponding to the feature value of each feature in the full - volume sample feature data according to the distribution position of the faulty battery sample data in the full - volume sample feature data.

[0185] A fault prediction module 805, configured to determine the fault risk score of the target battery according to the weight coefficient and the corresponding feature weight corresponding to the feature value of each feature in the target battery sample data; the fault risk score is used to characterize the fault probability of the target battery.

[0186] In some alternative embodiments, the fault risk score includes an initial risk score and a final risk score, and the fault prediction module 805 includes:

[0187] An accumulation unit, configured to, for each target battery, multiply the feature weight of each feature by the corresponding weight coefficient and then accumulate them to obtain the initial risk score of each target battery.

[0188] A screening unit, configured to screen out the target batteries whose initial risk scores are within a preset range and use them as potential faulty batteries.

[0189] A penalty unit, configured to perform penalty adjustment on the initial risk score of the potential faulty battery to obtain the final risk score of the potential faulty battery.

[0190] In some alternative embodiments, the penalty unit includes:

[0191] A selection subunit, configured to select a target feature from all features in the faulty battery sample data.

[0192] A determining subunit, configured to determine a feature distance between a target feature of a potential faulty battery and a target feature of a faulty battery.

[0193] A generating subunit, configured to generate a penalty value corresponding to the feature distance, where the penalty value is positively correlated with the feature distance.

[0194] A calculating subunit, configured to determine the product of an initial risk score and the penalty value as a final risk score.

[0195] In some alternative embodiments, the data analysis module 802 includes:

[0196] A first determining unit, configured to sort the feature values of each feature in the full - volume sample feature data according to a preset sorting method, and to determine multiple first quantiles of each feature at a first preset interval.

[0197] A second determining unit, configured to sort the feature values of each feature in the faulty battery sample data according to a preset sorting method, and to determine multiple second quantiles of each feature at a second preset interval.

[0198] Wherein, the preset sorting method is from small to large or from large to small, and the first preset interval and the second preset interval are the same or different.

[0199] In some alternative embodiments, the weight determining module 803 includes:

[0200] A third determining unit, configured to determine multiple preset discrimination degrees for characterizing the distribution of the faulty battery sample data in the full - volume sample feature data according to the positional relationship.

[0201] A fourth determining unit, configured to determine an average discrimination degree for characterizing the average of the multiple preset discrimination degrees, and to determine a comprehensive discrimination degree for characterizing the weighted average of the average discrimination degree and the multiple preset discrimination degrees.

[0202] A fifth determining unit, configured to generate a feature weight corresponding to the comprehensive discrimination degree, and to eliminate features with feature weights less than a set threshold.

[0203] In some alternative embodiments, the fifth determining unit includes:

[0204] An initial weight determining subunit, configured to determine an initial weight corresponding to the comprehensive discrimination degree according to the range of the comprehensive discrimination degree.

[0205] A penalty coefficient obtaining subunit, configured to obtain multiple penalty coefficients corresponding one - to - one to the multiple preset discrimination degrees.

[0206] A feature weight determining subunit, configured to attenuate the initial weight by using the multiple penalty coefficients to obtain the feature weight.

[0207] In some alternative embodiments, the weight coefficients include a first coefficient, a second coefficient, and a third coefficient.

[0208] The coefficient generation module 804 includes:

[0209] A first allocation unit, configured to respectively allocate a first coefficient to the feature values corresponding to multiple second quantiles, and respectively allocate a second coefficient to the feature values corresponding to multiple first quantiles according to the distribution positions of the faulty battery sample data in the full sample feature data.

[0210] A second allocation unit, configured to allocate a third coefficient to the remaining feature values based on linear interpolation.

[0211] Among them, the first coefficient, the second coefficient, and the third coefficient are all different.

[0212] In some alternative embodiments, the first coefficient includes a first sub - coefficient and a second sub - coefficient.

[0213] The first allocation unit is specifically configured to respectively allocate the first sub - coefficient to the feature values corresponding to the second quantile according to the concentration degree of the faulty battery sample data reaching a first preset degree, and respectively allocate the second sub - coefficient to the feature values corresponding to the second quantile according to the dispersion degree of the faulty battery sample data reaching a second preset degree; the first sub - coefficient is greater than or less than the second sub - coefficient.

[0214] In some alternative embodiments, the second allocation unit is specifically configured to select, based on linear interpolation, the third coefficient for allocating to the remaining feature values from the data interval formed by the first sub - coefficient, the second sub - coefficient, and the second coefficient.

[0215] The further function descriptions of the above - mentioned various modules and units are the same as those in the corresponding above - mentioned embodiments, and will not be elaborated here.

[0216] The battery fault prediction device in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0217] The embodiment of the present invention further provides a cloud, having the above - mentioned Figure 8 shown battery fault prediction device.

[0218] Please refer to Figure 9 , Figure 9 is a schematic structural diagram of a cloud provided by an alternative embodiment of the present invention. As Figure 9As shown, the cloud includes one or more processors 10, a memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the cloud, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if needed, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple clouds can be connected, and each device provides some necessary operations (such as an array of servers, a set of blade servers, or a multi-processor system). Figure 9 In [0000637], a processor 10 is taken as an example.

[0219] The processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware chip. The above-mentioned hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above-mentioned programmable logic device can be a complex programmable logic device, a field-programmable gate array, a general array logic, or any combination thereof.

[0220] Among them, the memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiments.

[0221] The memory 20 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the cloud. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 can optionally include a memory remotely set relative to the processor 10, and these remote memories can be connected to the cloud through a network. Examples of the above-mentioned network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0222] The memory 20 can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 can also include a combination of the above types of memories.

[0223] The cloud further includes a communication interface 30 for the cloud to communicate with other devices (such as a vehicle terminal) or a communication network.

[0224] An embodiment of the present invention further provides a vehicle, which includes a vehicle controller that can be used to execute the battery fault prediction method provided in any of the above embodiments. The embodiments of the vehicle controller executing the battery fault prediction method are the same as the corresponding above embodiments and will not be elaborated herein.

[0225] Among them, the vehicle controller can be, for example, a VCU (Vehicle Control Unit, vehicle controller). Of course, based on the embodiments of the present invention, the vehicle controller can also be any in-vehicle controller that can execute the above battery fault prediction method.

[0226] An embodiment of the present invention further provides a computer-readable storage medium. The method according to the embodiments of the present invention can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented by downloading through a network and originally stored in a remote storage medium or a non-transitory machine-readable storage medium and will be stored in a local storage medium, so that the method described herein can be stored as such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0227] A part of the present invention can be applied as a computer program product, such as computer program instructions. When executed by a computer, through the operation of the computer, the method and / or technical solution according to the present invention can be called or provided. Those skilled in the art should be able to understand that the forms of existence of computer program instructions in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible by the computer.

[0228] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A battery fault prediction method, characterized in that, The method includes: Obtaining full - volume sample feature data, where the full - volume sample feature data includes faulty battery sample data and target battery sample data, and the target battery sample data is the sample data of the target battery for which the fault probability is to be predicted; Determining a plurality of first quantiles according to the distribution of the feature values of each feature in the full - volume sample feature data, and determining a plurality of second quantiles according to the distribution of the feature values of each feature in the faulty battery sample data; For each feature, after merging and sorting the plurality of first quantiles and the plurality of second quantiles, determining the feature weight of each feature according to the positional relationship between the plurality of second quantiles and the plurality of first quantiles; Generating a weight coefficient corresponding to the feature value of each feature in the full - volume sample feature data according to the distribution position of the faulty battery sample data in the full - volume sample feature data; Determining the fault risk score of the target battery according to the weight coefficient corresponding to the feature value of each feature in the target battery sample data and the corresponding feature weight; the fault risk score is used to characterize the fault probability of the target battery.

2. The method according to claim 1, wherein The fault risk score includes an initial risk score and a final risk score. The determining the fault risk score of the target battery according to the weight coefficient corresponding to the feature value of each feature in the target battery sample data and the corresponding feature weight includes: For each target battery, multiplying the feature weight of each feature by the corresponding weight coefficient and then accumulating to obtain the initial risk score of each target battery; Selecting the target batteries whose initial risk scores are within a preset range and regarding them as potential faulty batteries; Performing penalty adjustment on the initial risk scores of the potential faulty batteries to obtain the final risk scores of the potential faulty batteries.

3. The method according to claim 2, wherein The performing penalty adjustment on the initial risk scores of the potential faulty batteries includes: Selecting a target feature from all features in the faulty battery sample data; Determining the feature distance between the target feature of the potential faulty battery and the target feature of the faulty battery; Generating a penalty value corresponding to the feature distance, where the penalty value is positively correlated with the feature distance; Determining the product of the initial risk score and the penalty value as the final risk score.

4. The method according to any one of claims 1 to 3, wherein The determining a plurality of first quantiles according to the distribution of the feature values of each feature in the full - volume sample feature data includes: sorting the feature values of each feature in the full - volume sample feature data according to a preset sorting method, and determining the plurality of first quantiles of each feature at a first preset interval; The determining a plurality of second quantiles according to the distribution of the feature values of each feature in the faulty battery sample data includes: sorting the feature values of each feature in the faulty battery sample data according to the preset sorting method, and determining the plurality of second quantiles of each feature at a second preset interval; Wherein, the preset sorting method is from small to large or from large to small, and the first preset interval is the same as or different from the second preset interval.

5. The method according to claim 4, wherein The determining of the feature weight of each feature according to the positional relationship between the multiple second quantiles and the multiple first quantiles includes: Determining, according to the positional relationship, a plurality of preset discrimination degrees for characterizing the distribution of the faulty battery sample data in the full sample feature data; Determining an average discrimination degree for characterizing the average of the plurality of preset discrimination degrees, and determining a comprehensive discrimination degree for characterizing the weighted average of the average discrimination degree and the plurality of preset discrimination degrees; Generating a feature weight corresponding to the comprehensive discrimination degree, and eliminating features with feature weights less than a set threshold.

6. The method according to claim 5, wherein The generating of the feature weight corresponding to the comprehensive discrimination degree includes: Determining an initial weight corresponding to the comprehensive discrimination degree according to the range of the comprehensive discrimination degree; Obtaining a plurality of penalty coefficients corresponding one by one to the plurality of preset discrimination degrees; Attenuating the initial weight by using the plurality of penalty coefficients to obtain the feature weight.

7. The method according to claim 4, characterized in that, The weight coefficient includes a first coefficient, a second coefficient and a third coefficient; the generating of the weight coefficient corresponding to the feature value of each feature in the full sample feature data according to the distribution position of the faulty battery sample data in the full sample feature data includes: According to the distribution position of the faulty battery sample data in the full sample feature data, respectively assigning a first coefficient to the feature values corresponding to the multiple second quantiles, and respectively assigning a second coefficient to the feature values corresponding to the multiple first quantiles; Based on linear interpolation, assigning a third coefficient to the remaining feature values; Wherein, the first coefficient, the second coefficient and the third coefficient are all different.

8. The method according to claim 7, wherein The first coefficient includes a first sub-coefficient and a second sub-coefficient; the respectively assigning a first coefficient to the feature values corresponding to the multiple second quantiles according to the distribution position of the faulty battery sample data in the full sample feature data includes: According to the concentration degree of the faulty battery sample data reaching a first preset degree, respectively assigning a first sub-coefficient to the feature values corresponding to the second quantiles, and according to the dispersion degree of the faulty battery sample data reaching a second preset degree, respectively assigning a second sub-coefficient to the feature values corresponding to the second quantiles; The first sub-coefficient is greater than or less than the second sub-coefficient.

9. The method according to claim 8, wherein The assigning of a third coefficient to the remaining feature values based on linear interpolation includes: Selecting, from the data interval formed by the first sub-coefficient, the second sub-coefficient and the second coefficient, the third coefficient for assigning to the remaining feature values based on linear interpolation.

10. A battery fault prediction device, characterized in that, The device includes: A data acquisition module, configured to acquire full sample feature data, where the full sample feature data includes faulty battery sample data and target battery sample data, and the target battery sample data is the sample data of a target battery for which a fault probability is to be predicted; A data analysis module, configured to determine a plurality of first quantiles according to the distribution of the feature values of each feature in the full sample feature data, and configured to determine a plurality of second quantiles according to the distribution of the feature values of each feature in the faulty battery sample data; A weight determination module, configured to, for each feature, sort the plurality of first quantiles and the plurality of second quantiles after merging, and determine the feature weight of each feature according to the positional relationship between the plurality of second quantiles and the plurality of first quantiles; A coefficient generation module, configured to generate a weight coefficient corresponding to the feature value of each feature in the full sample feature data according to the distribution position of the faulty battery sample data in the full sample feature data; A fault prediction module, configured to determine the fault risk score of the target battery according to the weight coefficient corresponding to the feature value of each feature in the target battery sample data and the corresponding feature weight; the fault risk score is used to characterize the fault probability of the target battery.

11. A cloud, characterized in that, Comprising: A memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the battery fault prediction method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, Computer instructions are stored on the computer-readable storage medium, and the computer instructions are used to cause a computer to execute the battery fault prediction method according to any one of claims 1 to 9.

13. A computer program product, characterized in that, Comprising computer instructions, the computer instructions are used to cause a computer to execute the battery fault prediction method according to any one of claims 1 to 9.

14. A vehicle, characterized in that, The vehicle includes a vehicle controller, and the vehicle controller is configured to execute the battery fault prediction method according to any one of claims 1 to 9.

Images