Battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy

Abstract

The application discloses a battery fault diagnosis method based on DBSCAN, an empirical distribution function and information entropy, and comprises the following steps: pre-processing acquired battery voltage data to obtain battery voltage data with normal time stamps and no obvious abnormal values, and dividing the data; adopting information entropy and empirical distribution data to statistically determine battery threshold states, and determining adaptive abnormal early warning thresholds of different battery packs; using a DBSCAN algorithm to calculate voltage outlier degrees, and calculating outlier weights; and jointly early warning according to the adaptive abnormal early warning thresholds and the outlier weights. The application adopts the above method, determines adaptive abnormal early warning thresholds of different battery packs through an empirical distribution function to adapt to individual differences of different batteries, then judges whether the batteries are in an abnormal state through information entropy and an abnormal coefficient, and performs fault early warning according to the combination of DBSCAN and information entropy, so that the accuracy of fault early warning is improved, and the safe operation of a battery system is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery diagnosis, in particular to a battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy combination. BACKGROUND

[0002] Battery systems may fail during operation due to reasons such as electrochemical aging, thermal runaway, mechanical damage, overcharging / overdischarging, etc., which may cause performance degradation, or even safety accidents such as fire or explosion. Therefore, it is necessary to diagnose the faults of the battery system. The model-based battery anomaly diagnosis method has very limited migration ability in the battery anomaly diagnosis process. The battery anomaly diagnosis method based on machine learning requires a large amount of running computing power and high data quality, which restricts its further development. Therefore, the battery anomaly diagnosis method based on statistics is still a hot research topic today. Among many statistical algorithms, the information entropy algorithm stands out due to its strong interpretability and high robustness.

[0003] The method for battery fault diagnosis and early warning based on information entropy often combines standard normalization and 3σ method to calculate the abnormal coefficient and determine the abnormal state of the battery. Existing researches believe that when the abnormal coefficient exceeds 3, it is a small probability event, which means that the battery is in an abnormal state. However, due to differences in battery manufacturing and different operating conditions, the abnormal coefficient of the battery during normal operation may still exceed 3. A constant threshold cannot meet the early warning needs of the battery under various conditions, and it is necessary to determine an adaptive abnormal threshold.

[0004] In the calculation process of information entropy, due to the limitation of its calculation principle, when the abnormality of the battery becomes more and more serious, multiple battery monomers may become outliers, and the calculation result of information entropy may decrease. Relying on a single abnormal coefficient to determine the abnormal state of the battery may result in a smaller abnormal coefficient when the abnormality of the battery becomes more and more serious, which does not match the actual expected effect. It is necessary to improve the abnormal diagnosis method of information entropy in the whole life cycle of the battery.

[0005] The early warning effect based on the calculation of battery inconsistency information based on information entropy is based on the consistency of battery voltage. However, although the consistency of battery voltage is strong, the battery voltage value itself changes strongly during the alternation of charging and discharging conditions, and the inconsistency between batteries also changes. Therefore, without dividing the battery voltage state, directly calculating the information entropy of the battery voltage may cause false positives of battery abnormal information. SUMMARY

[0006] The application aims to provide a battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy combination, to determine the adaptive abnormal early warning threshold of different battery packs through the empirical distribution function, to judge whether the battery is in an abnormal state through the information entropy and abnormal coefficient, and to perform fault early warning according to the combination of DBSCAN and information entropy, thereby improving the accuracy of fault early warning and ensuring the safe operation of the battery system.

[0007] To achieve the above-mentioned purpose, the application provides a battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy combination, comprising the following steps:

[0008] S1, obtaining battery voltage data, preprocessing the battery voltage data to obtain battery voltage data with normal time stamp and no obvious abnormal value, and dividing the battery voltage data into threshold training data set and algorithm application data set;

[0009] S2, using information entropy and empirical distribution data to statistically determine the battery threshold state in the threshold training data set, and determining the adaptive abnormal early warning threshold of different battery packs;

[0010] S3, using DBSCAN algorithm to calculate the voltage outlying degree under the same calculation window in the algorithm application data set, and calculating the outlying weight;

[0011] S4, performing joint early warning according to the adaptive abnormal early warning threshold and the discrete weight.

[0012] Preferably, step S2 specifically comprises:

[0013] S21, calculating the information entropy and abnormal coefficient of the threshold training data set;

[0014] S22, observing the abnormal coefficient by using the empirical distribution function;

[0015] S23, selecting the abnormal coefficient corresponding to the observation value 0.998 as the adaptive abnormal early warning threshold of different battery packs.

[0016] Preferably, the step S3 of using DBSCAN algorithm to calculate the voltage outlying degree under the same calculation window in the algorithm application data set comprises:

[0017] Using DBSCAN algorithm to perform clustering analysis on the battery voltage data, identifying the battery charging and discharging interval, judging whether the battery voltage under the calculation window appears obvious outlying, if there is outlying voltage information, counting the voltage column name, and calculating the difference between the average value of abnormal battery voltage data and the average value of normal battery voltage data, and according to the DBSCAN clustering result, dividing different information entropy calculation windows.

[0018] Preferably, the step S3 calculates the discrete weight α based on the difference, and the formula is:

[0019] a = e 1+x1 ;

[0020] In the formula, x1 represents the absolute value of the difference.

[0021] Preferably, step S4 specifically comprises:

[0022] S41, calculate the information entropy of different information entropy calculation windows in the algorithm application data set;

[0023] S42, calculate the abnormal coefficient of the battery voltage based on the calculated information entropy;

[0024] S43, multiply the abnormal coefficient of the outlier battery voltage by the outlier weight, and when the calculation result exceeds the adaptive abnormal early warning threshold, determine that the battery under the current window is in an abnormal state.

[0025] Preferably, step S41 specifically comprises:

[0026] All battery monomer voltages under the same calculation window are regarded as a whole to construct a probability matrix, and the formula is:

[0027]

[0028] In the formula, l is the information entropy resolution, c i,j represents the number of all the jth battery monomer in the ith equal interval under the information entropy time window, and n represents the number of battery monomers;

[0029] The value range of the ith equal interval in the information entropy matrix is calculated, and the formula is:

[0030]

[0031] In the formula, x max and x min are the maximum and minimum values in the calculation window CW respectively;

[0032] Based on the probability matrix, a probability matrix is constructed, and the formula is:

[0033]

[0034] In the formula, p i,j represents the probability of c i,j relative to the battery monomer i under the calculation window, and the information entropy h j of the jth battery monomer in the information entropy sequence;

[0035]

[0036] Based on the probability matrix, the information entropy sequence H under the information entropy calculation window is calculatedm Calculate the information entropy h based on the j-th battery cell in the information entropy sequence. j The formula is:

[0037] H m =[h1,h2,......,h j ......h n ]

[0038]

[0039] The calculation window iterates through all battery voltage data within the batch processing window according to the sliding step size to obtain the information entropy matrix H, as shown in the formula:

[0040] H = [H1,H2,......,H] m ......H batch_end ] T ;

[0041] In the formula, H batch_end This represents the last value obtained from the information entropy sequence after traversing the batch processing window.

[0042] Preferably, step S42 specifically includes:

[0043] Based on the obtained information entropy h j Calculate the coefficient of variation z for each battery cell. j The coefficient of variation Z is composed of the coefficients of variation of all individual battery cells under the information entropy calculation window. m The formula is:

[0044]

[0045] Z m =[z1,z2,......,z j ......z n ];

[0046] In the formula, h ave with h stdj The battery information entropy sequence H is calculated under the same information entropy calculation window. m The average value and the standard deviation of the information entropy of the j-th battery cell are given by the following formula:

[0047]

[0048] Calculate the coefficient of variation sequence for each information entropy sequence in the information entropy matrix to obtain the coefficient of variation matrix Z, using the following formula:

[0049] Z = [Z1, Z2, ..., Z] m ......Z batch_end ]T ;

[0050] wherein Z batch_end represents the last value of the coefficient of variation sequence obtained after traversing the batch window, calculated by H batch_end .

[0051] Therefore, the battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy is adopted, and the following beneficial effects are obtained:

[0052] (1) The abnormal early warning threshold is adaptively determined by the empirical distribution function, which can adapt to the individual differences of different batteries, reduce the false positive rate, and improve the accuracy of diagnosis;

[0053] (2) The DBSCAN clustering algorithm based on voltage and the outlier weight alpha are introduced, and the information entropy calculation result is further amplified, which effectively solves the problem that the information entropy cannot normally identify in the later stage of battery failure, further optimizes the fault diagnosis result, and at the same time, the battery voltage fluctuation is clustered, avoiding the problem that the entropy value calculation result is inaccurate due to the battery voltage fluctuation itself;

[0054] (3) By combining information entropy and clustering analysis, the early abnormal state and the later state of the battery can be quickly and accurately identified, the diagnosis efficiency is improved, and the safe operation of the battery system is ensured.

[0055] The technical solutions of the embodiments of the present application will be further described in detail below with reference to the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 The method flowchart of the embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. In the description of the present application, it should be noted that the terms "upper", "lower", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, or the orientation or positional relationship commonly used when the product is used, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0058] EMBODIMENT

[0059] As Figure 1 shown, the application provides a battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy, comprising the following steps:

[0060] S1, obtaining battery voltage data, preprocessing the battery voltage data, setting voltage data, eliminating outliers, so that there is no abnormal sampling point, obtaining battery voltage data with normal timestamp and no obvious outliers. The battery voltage data with normal timestamp is sorted in ascending order, the battery voltage data one week before the battery is shipped is used as threshold training data set, and the remaining battery voltage data is divided into algorithm application data set.

[0061] S2, using information entropy and empirical distribution data to determine the adaptive abnormal early warning threshold of different battery packs. Specifically, it includes:

[0062] S21, calculating the information entropy and abnormal coefficient of the threshold training data set;

[0063] S22, using empirical distribution function to observe the empirical distribution of abnormal coefficient;

[0064] S23, selecting the abnormal coefficient corresponding to the observation value 0.998 as the adaptive abnormal early warning threshold of different battery packs.

[0065] S3, using DBSCAN algorithm to calculate the voltage outlier degree under the same calculation window in the algorithm application data set, and calculating the outlier weight. Through the outlier weight, the abnormal coefficient calculation result of the outlier battery monomer is amplified, so that when multiple battery monomers are out of range, the abnormal coefficient calculated based on information entropy can still well reflect the abnormal state of the battery.

[0066] Specifically:

[0067] Using DBSCAN algorithm to cluster analyze the battery voltage data, identifying the battery charging and discharging interval, judging whether the battery voltage under the calculation window is obviously out of range, if there is out of range voltage information, counting the voltage column name, and calculating the difference between the average value of abnormal battery voltage data and the average value of normal battery voltage data, and according to the DBSCAN clustering result, different information entropy calculation windows are divided.

[0068] Based on the difference, the discrete weight a is calculated, and the formula is:

[0069] a = e 1+x1 ;

[0070] In the formula, x1 represents the absolute value of the difference.

[0071] S4, joint early warning according to the adaptive abnormal early warning threshold and the discrete weight, specifically:

[0072] S41, calculate the information entropy of different information entropy calculation windows in the algorithm application data set. For a battery pack with n battery monomers, the information entropy calculation detailed steps are as follows:

[0073] Construct the probability matrix: based on the number of probabilities matrix, construct the probability matrix, the formula is:

[0074]

[0075] In the formula, l is the information entropy resolution, c i,j represents the number of all the jth battery monomer in the ith equal interval under the information entropy time window, and n represents the number of battery monomers;

[0076] Calculate the value range of the ith equal interval in the information entropy matrix, the formula is:

[0077]

[0078] In the formula, x max and x min are the maximum and minimum values in the calculation window CW;

[0079] Construct the probability matrix: based on the number of probabilities matrix, construct the probability matrix, the formula is:

[0080]

[0081] In the formula, p i,j represents the probability of c i,j relative to the battery monomer i under the calculation window, and the formula is:

[0082]

[0083] Information entropy calculation: based on the probability matrix, calculate the information entropy sequence H m under the information entropy calculation window, and the information entropy of the jth battery monomer in the information entropy sequence h j The calculation method is shown in the formula:

[0084] H m = [h1, h2,..., h j ..., h n ]

[0085]

[0086] The calculation window is traversed according to the sliding step length to obtain the information entropy matrix H under all the battery voltage data in the batch processing window, and the formula is:

[0087] H = [H1,H2,......,H] m ......H batch_end ] T ;

[0088] In the formula, H batch_end This represents the last value obtained from the information entropy sequence after traversing the batch processing window.

[0089] S42. Calculate the abnormal coefficient of the battery voltage based on the calculated information entropy, specifically: based on the obtained information entropy h... j Calculate the coefficient of variation z for each battery cell. j The coefficient of variation Z is composed of the coefficients of variation of all individual battery cells under the information entropy calculation window. m The formula is:

[0090]

[0091] Z m =[z1,z2,......,z j ......z n ];

[0092] In the formula, h ave with h stdj The battery information entropy sequence H is calculated under the same information entropy calculation window. m The average value and the standard deviation of the information entropy of the j-th battery cell are given by the following formula:

[0093]

[0094] Calculate the coefficient of variation sequence for each information entropy sequence in the information entropy matrix to obtain the coefficient of variation matrix Z, using the following formula:

[0095] Z = [Z1, Z2, ..., Z] m ......Z batch_end ] T ;

[0096] In the formula, Z batch_end This represents the last value of the coefficient of variation sequence obtained after traversing the batch processing window, denoted by H. batch_end Calculated.

[0097] S43. In the process of calculating the abnormal coefficient in step S42, the abnormal coefficient of the outlier battery voltage is multiplied by the outlier weight. When the calculation result exceeds the adaptive abnormal warning threshold, the battery under the current window is considered to be in an abnormal state.

[0098] Therefore, the application adopts the above-mentioned battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy combination, determines the adaptive abnormal early warning threshold of different battery packs through the empirical distribution function, judges whether the battery is in an abnormal state through the information entropy and the abnormal coefficient, and performs fault early warning according to the combination of DBSCAN and information entropy, thereby improving the accuracy of fault early warning and guaranteeing the safe operation of the battery system.

[0099] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit them, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: it can still modify or equivalently replace the technical solutions of the present application, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. A battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy combination, characterized by the steps of The method comprises the following steps: S1, acquiring battery voltage data, preprocessing the battery voltage data to obtain battery voltage data with normal time stamp and no obvious abnormal value, and dividing the battery voltage data into threshold training data set and algorithm application data set; S2, training the battery threshold state in the threshold training data set by using information entropy and empirical distribution data statistics to determine the adaptive abnormal early warning threshold of different battery packs; S3, calculating the voltage outlier degree in the same calculation window in the algorithm application data set by using DBSCAN algorithm, and calculating the outlier weight, specifically including: using DBSCAN algorithm to perform cluster analysis on the battery voltage data, identifying the battery charging and discharging interval, judging whether the battery voltage in the calculation window appears obvious outlier, if there is outlier voltage information, counting the voltage column name, and calculating the difference between the average value of abnormal battery voltage data and the average value of normal battery voltage data, and dividing different information entropy calculation windows according to the DBSCAN clustering result; Calculating outlier weights based on difference ， The formula is: ； In the formula, x1 represents the absolute value of the difference; S4, joint early warning according to the adaptive abnormal early warning threshold and the outlier weight, specifically including: S41, calculating the information entropy of different information entropy calculation windows in the algorithm application data set; S42, calculating the abnormal coefficient of battery voltage based on the calculated information entropy; S43, multiplying the abnormal coefficient of outlier battery voltage by the outlier weight, when the calculation result exceeds the adaptive abnormal early warning threshold, it is determined that the battery in the current window is in abnormal state. 2.The battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy combination according to claim 1, characterized in that, Step S2 specifically includes: S21, calculating the information entropy and abnormal coefficient of the threshold training data set; S22, observing the empirical distribution of the abnormal coefficient by using the empirical distribution function; S23, selecting the abnormal coefficient corresponding to the observation value 0.998 as the adaptive abnormal early warning threshold of different battery packs. 3.The battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy combination according to claim 1, characterized in that, Step S41 specifically includes: regarding all battery monomer voltages in the same calculation window as a whole, constructing a probability matrix, and the formula is: ； In the formula, is the information entropy resolution, represents the number of all the battery monomers in the th equal interval of the th battery monomer under the information entropy time window, n represents the number of battery monomers; Calculate the first information entropy matrix. i The range of values ​​between the equal partitions is given by the formula: ； wherein and are the maximum and minimum values within the calculation window CW, respectively. based on the probability matrix, constructing a probability matrix, and the formula is: ； wherein represents with respect to the probability of the lower battery cell of the calculation window i is given by the formula: ； Calculate information entropy sequence in the window based on the probability matrix , the information entropy of the first j battery monomer in the information entropy sequence The calculation method formula is: ； The calculation window traverses all battery voltage data under the batch processing window according to a sliding step, to obtain an information entropy matrix , and the formula is: ； wherein represents the last value of the sequence of information entropy obtained after traversing the batch window.

4. The battery fault diagnosis method based on DBSCAN, empirical distribution function and information entropy combination of claim 3, characterized in that, Step S42 specifically includes: According to the information entropy obtained The coefficient of variation of each battery monomer is calculated The coefficient of variation of all battery monomers under the information entropy calculation window is calculated The formula is: ； ； In the formula, and are the average value of the battery information entropy sequence and the standard deviation of the information entropy of the first j battery cell under the same information entropy calculation window, and the formula is: ； ； A variation coefficient sequence of each information entropy sequence in the information entropy matrix is calculated to obtain a variation coefficient matrix , and the formula is: ； wherein, represents the last value of the coefficient of variation sequence obtained after traversing the batch window, calculated by .

Images