A lithium ion battery abnormality identification and diagnosis method based on historical data

Abstract

The application relates to a lithium ion battery abnormality identification and diagnosis method based on historical data, which can be executed by one or more processors and comprises the following steps: S1, the one or more processors acquire historical data of a battery system, and the historical data is cleaned by deleting invalid data and complementing missing data; S2, the one or more processors acquire a charging process in the historical data to further obtain a voltage threshold of a normal battery; S3, the one or more processors determine an abnormal deviation index of the battery based on the voltage threshold, and complete screening of abnormal batteries based on the abnormal deviation index; and S4, the one or more processors can determine a fault type of the abnormal battery based on a change trend of the abnormal deviation index, and judge a fault degree of the abnormal battery based on a change speed of the abnormal deviation index.

Description

Technical Field

[0001] This invention relates to the field of new energy storage, and in particular to a method for identifying and diagnosing anomalies in lithium-ion batteries based on historical data. Background Technology

[0002] With strong government support, the electric vehicle and energy storage industries related to lithium-ion batteries have flourished. In recent years, lithium-ion batteries have also begun to be applied in the rail transit sector. In the past two years, the energy density of power batteries has continuously increased, and battery assembly methods and processes have further developed. The driving range and power capacity of battery systems basically meet user needs, and range anxiety is no longer a major concern in the industry. However, due to the increase in battery energy density and the continuous improvement of system energy through integrated design during assembly, potential safety hazards exist in the battery itself. At the same time, the system's safety protection functions are not perfect, leading to frequent battery safety issues and attracting public attention. As the service life of batteries increases, the new energy used car market is gradually expanding, and the reasonable valuation of battery systems and the determination of insurance costs have become urgent issues that the industry needs to address.

[0003] The degradation of lithium-ion batteries stems from two main causes: firstly, manufacturing process issues during cell production, leading to gradually emerging safety hazards over time; and secondly, over-aging and internal short circuits caused by overuse and stress. These are all types of battery failures that gradually increase in safety risk due to aging during use. Additionally, some cases involve sudden safety issues caused by mechanical stress, external short circuits, or BMS (Battery Management System) malfunctions. Sudden safety issues are typically difficult to predict in advance and generally require improvements in battery pack manufacturing processes and high-strength system design to reduce the risk rate, or system-level passive safety protections such as adding fuses, passive fire suppression systems, and designs to prevent heat spread, thus mitigating the safety hazards caused by battery failures. However, safety issues caused by gradual battery degradation during use can be detected early through health monitoring and human intervention, reducing the probability of safety risks occurring.

[0004] However, BMS online management typically only addresses transient overvoltage, overtemperature, and overcurrent issues in the battery system, making it difficult to locate and protect slowly deteriorating batteries throughout the entire operation. In recent years, the rapid development of big data technology has spurred methods for assessing battery health using real-world vehicle operating data. Current fault analysis based on historical data often employs data-driven approaches, neglecting the inherent characteristics of the battery itself; it relies too heavily on laboratory-built models, placing high demands on the testing requirements of new batteries and the applicability of models under real-world conditions; while signal decomposition-based methods rely less on data or equivalent models, they do not provide diagnostics for fault types.

[0005] Therefore, this application aims to diagnose and identify abnormal batteries in vehicle power batteries, provide rational suggestions for vehicle maintenance, predict battery safety risks in advance, or determine the remaining value of batteries through fault screening, thereby ensuring the safe operation of vehicles and guaranteeing the healthy and orderly development of the industry. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for identifying and diagnosing lithium-ion battery anomalies based on historical data.

[0007] The objective of this invention is achieved through the following technical solution: a method for identifying and diagnosing lithium-ion battery anomalies based on historical data, characterized in that the method can be executed by one or more processors, including:

[0008] S1, the one or more processors acquire historical data of the battery system and clean the historical data by deleting invalid data and filling in missing data;

[0009] S2, the one or more processors acquire the charging process from the historical data to further obtain the voltage threshold of the normal battery;

[0010] S3, the one or more processors determine the abnormal deviation index of the battery based on the voltage threshold, and complete the screening of abnormal batteries based on the abnormal deviation index;

[0011] S4, the one or more processors can determine the fault type of the abnormal battery based on the changing trend of the abnormal deviation index, and judge the fault degree of the abnormal battery based on the changing rate of the abnormal deviation index.

[0012] Preferably, the one or more processors are based on the formula The historical data is cleaned, wherein, V(i,t) represents the voltage value of battery i at time t, k is the first time when the voltage of a single battery cell reaches 0, and m is the first non-zero time after the voltage reaches 0.

[0013] Preferably, the one or more processors obtain the voltage threshold by following these steps:

[0014] Obtain all battery voltages v at time t from the historical data after cleaning. t ;

[0015] Based on formula Obtain the mean value μ of the battery voltage at time t. t,n ;

[0016] Based on formula Obtain the standard value σ of the battery voltage at time t. t,n ;

[0017] The v t,i The voltage threshold range is represented by μ. t,n -3σ t,n ≤v t,i ≤μ t,n +3σ t,n i = 1, 2, 3...n, v t,i Let be the voltage value of battery i at time t, and n be the number of batteries.

[0018] Preferably, the one or more processors are capable of determining the abnormal deviation index d based on the voltage threshold and the standard value. i,t The abnormal deviation index can be expressed as

[0019] Preferably, the one or more processors acquire the normal battery charging process according to the following steps:

[0020] According to the formula The charging end time is filtered, where SOC(t) is the remaining charge value at time t, and I(k) is the current at time k;

[0021] Starting from the end of charging, the SOC(i) at time i is compared with the SOC(i) at the previous time in reverse order of time. The sampling point where the first SOC(i) is less than the SOC(i) at the previous time is taken as the beginning of the charging segment.

[0022] The charging terminals with a SOC(i) difference of less than 40% between the beginning and end of the charging segment are removed.

[0023] Preferably, the one or more processors complete the screening of abnormal batteries according to the following steps:

[0024] N equally spaced sampling points are selected, and an abnormal deviation index d of the discrete battery is established based on a single charging process. i =[d i,t(SOC1) ,d i,t(SOC2) ,……,d i,t(SOCN) ];d i,t(SOCN) Let N be the point of SOC, and let i be the abnormal deviation index of the i-th battery.

[0025] When the frequency of the abnormal deviation index being greater than 1 exceeds a set threshold of 25%, the discrete battery is determined to be an abnormal battery.

[0026] Preferably, the one or more processors determine the fault type of the abnormal battery according to the following steps:

[0027] A linear fit is performed on the abnormal deviation index to obtain its linear fit curve, which can be represented as follows: m is the slope of the linearly fitted line, b0 is the intercept of the linearly fitted line, and SOC is the slope of the linearly fitted line. s and SOC e These are the initial SOC and the final SOC, respectively.

[0028] The fault types are categorized into two types: insufficient remaining power and insufficient battery capacity. The criteria for determining the fault type are as follows: Where m i Let be the slope of the linear fit of the i-th battery.

[0029] Preferably, the one or more processors determine the changing trend of the anomaly deviation index by the following steps: establishing a three-dimensional matrix of the anomaly deviation index for the entire historical process.

[0030] In the above formula, c represents the c-th charge in the entire historical process, i represents the battery number, t represents the selected SOC corresponding to the time point, and d i,t(SOCj) This represents the abnormal deviation index of battery i at the j-th state of charge during the c-th charge.

[0031] Obtain the overall deviation index COV for each battery t(SOC),c =σ t(SOC),c / μ t(SOC),c Obtain the weighting coefficients of the deviation index at different SOC points. The deviation from the exponential weighted matrix for each charging process is represented by ω. c =[w SOC1,c ,w SOC2,c ,…,w SOCN,c ], where w SOCN,c The weighting coefficients at time SOCN during the c-th charge; the weighted matrix of deviation index for the entire historical process is represented as follows.

[0032] Obtain the overall deviation index of each battery during a single charge. Constructing a comprehensive deviation index matrix for the entire historical process

[0033] Preferably, the one or more processors determine the battery degradation rate by the following steps: constructing a mutation rate matrix. The criteria for determining a battery that changes too quickly are set as follows: Where, k i,c =f i,c+1 -f i,c .

[0034] The present invention has the following advantages: It solves the current problem that lithium-ion power batteries in service require extensive laboratory testing or complex model construction in the early stage. It can effectively screen out abnormal batteries and identify abnormalities as SOC abnormalities or capacity abnormalities, providing quick operation suggestions for on-site maintenance personnel and reducing the professional level requirements for practitioners; at the same time, it screens out batteries with significant safety risks and provides timely warning information to prevent safety accidents. Attached Figure Description

[0035] The present invention includes the following figures:

[0036] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein:

[0037] Figure 1 A flowchart illustrating the lithium battery anomaly identification and diagnosis method described in this invention.

[0038] Figure 2 The process of cleaning voltage and current data.

[0039] Figure 3 Dynamic threshold filtering process.

[0040] Figure 4 Abnormal battery screening process based on deviation frequency.

[0041] Figure 5 Adaptation to changes in the deviation index of batteries with different abnormality types. Detailed Implementation

[0042] The present invention will be further described in detail below with reference to the accompanying drawings. This detailed description is an illustration in conjunction with exemplary embodiments of the invention, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0043] like Figure 1As shown, this application provides a lithium battery anomaly identification and diagnosis method based on historical data. First, the historical data is cleaned, invalid data is deleted, and missing data is filled in. Second, the charging curve of a normal battery is reconstructed using statistical methods. Third, the anomaly deviation index is calculated, and abnormal batteries are identified by threshold screening. Then, the anomaly type of the abnormal battery is determined according to the changing trend of the deviation index. Finally, the degree of battery failure is diagnosed based on the long-term rate of change of the anomaly index, and a warning message is given. Specifically, first, to address issues such as acquisition noise, packet loss (records are empty or zero), and time disorder in the historical data of the battery system, the individual cell voltage data is cleaned according to formulas (1) and (2).

[0044]

[0045] Where V(i,t) represents the voltage value of battery i at time t. Current data are similarly cleaned, with an absolute value exceeding 200A as the threshold. Figure 2 This document describes the process of cleaning voltage and current data for a specific vehicle. Addressing the issue that many anomalies in historical data are due to data transmission problems resulting in zero or null values, and considering that the voltage range of individual cells for the same battery type is generally well-defined, and that battery voltage rarely abruptly drops to 0V, this application designs a linear interpolation method for zero and null values. First, zero and null values ​​in individual cell voltages are identified and their quantities are counted. If the number of zero and null values ​​occurs across all cells, and the corresponding recording time exceeds 30 minutes, it indicates that a large amount of operational data was not effectively transmitted, and this data is invalid and directly deleted. If the zero and null values ​​only occur at a specific time in a specific cell, and the corresponding time is less than 30 minutes, then interpolation is performed using the linear interpolation method.

[0046] Secondly, all charging processes in the historical data are extracted. During charging, the SOC data continuously increases; at the end of charging, the SOC data is usually high, mostly at 100%. Charging exceeding 90% SOC is considered a valid charging segment, and the selection process for charging segments is as follows:

[0047] 1) First, select the charging end time according to formula (3);

[0048]

[0049] 2) Starting from the end of charging, compare the SOC value at this moment with the previous moment in reverse chronological order, and take the sampling point where the first SOC value is less than the previous moment as the beginning of the charging segment.

[0050] 3) Remove charging segments where the SOC difference between the beginning and end of the charging segment is less than 40%.

[0051] Next, a standard battery reference voltage curve is constructed. Constructing the standard battery reference voltage curve involves the following steps: First, using the traditional 3σ criterion method, the charging data at each time point in a single process is filtered, and batteries with voltages outside the 3σ range are marked; the marked batteries are removed, and the 3σ criterion method is used again for battery filtering; the first two steps are repeated until no batteries are marked; the average voltage of the remaining batteries is calculated as the standard battery reference voltage curve. Specifically, a segment of battery pack charging data v is extracted. t The Gaussian distribution parameters of the data matrix are calculated as shown in equations (4) and (5), and the voltage threshold of the healthy battery is determined as shown in equation (6).

[0052]

[0053] μ t,n -3σ t,n ≤v t,i ≤μ t,n +3σ t,n ,i=1,2,3...n(6)

[0054] Where v t This refers to the total battery voltage at time t in the cleaned data, v t,i The voltage of the i-th battery at time t, μ t,n σ represents the average battery voltage at time t. t,n The standard deviation of the battery voltage at time t is used. An iterative process is added, and the screening threshold is dynamically adjusted based on the actual situation of the samples. The already screened outlier batteries are removed, and equations (4) to (6) are repeated. At this time, n in the formula is updated to n, which is the number of batteries after removing outliers. The threshold is continuously reduced to determine newly appearing outliers, such as... Figure 3 As shown. After removing abnormal voltage values ​​from the series-connected battery pack, the remaining batteries can be considered as batteries that have aged normally during vehicle use. Therefore, the unselected batteries are defined as healthy batteries. The average voltage of the healthy batteries can be used as a reference voltage value for diagnosing subsequent abnormal batteries.

[0055] The voltage deviation is defined as the distance between the voltage of a single cell and the average voltage of healthy cells. The standard deviation of the voltage of healthy cells in the battery pack is used to characterize the voltage dispersion of an ideal battery pack. The ratio of the voltage deviation to three times the standard deviation of the voltage of healthy cells is the voltage deviation index d at each moment. i,t , as in equation (7).

[0056]

[0057] d i =[d i,t(SOC1) ,d i,t(SOC2) ,……,di,t(SOCN) (8)

[0058] Selecting equally spaced SOC points, the deviation exponential vector d of the discrete battery is established based on a single charging process. i As shown in equation (8). Statistical d i Batteries with a frequency greater than 1, or a frequency exceeding a set threshold of 25%, are considered abnormal batteries. For example... Figure 4 As shown.

[0059] Then, the abnormal batteries selected in the previous step are classified as faults. A linear fit is performed on the degree of deviation, as shown in equation (9):

[0060]

[0061] The fault is judged by the trend of the fitted curve. The judgment basis for batteries with low SOC and small capacity is Equation (10):

[0062]

[0063] The battery deviation index is mostly negative due to low voltage, indicating a low SOC fault. Figure 5 As shown in (a), the battery deviation shows a zero-crossing point and gradually increases at the end of charging. The slope of the fitted curve is positive, indicating a capacity deficiency fault. Figure 5 As shown in (b).

[0064] Finally, the anomaly trend is determined by the changes in the deviation index throughout the entire historical process. Typically, historical data from electric vehicles spanning 1 to 12 months are selected for analysis. A deviation index matrix D for a single charging process is then established. c As shown in equation (11).

[0065]

[0066] In equation (11), c represents the c-th charge in the entire historical process, i represents the battery number, t represents the selected SOC corresponding to the time point, and d i,t(SOCj) This represents the abnormal deviation index of battery i at the j-th state of charge during the c-th charge.

[0067] The coefficient of variation method was used to calculate the comprehensive deviation index of each battery. The coefficient of variation was calculated as shown in equation (12):

[0068] COV t(SOC),c =σ t(SOC),c / μ t(SOC) c (12)

[0069] In equation (12) μ t(SOC),c and σ t(SOC),cLet be the mean voltage and standard deviation at time t(SOC) during the c-th charging process, respectively.

[0070] Where, σ t,c μ represents the standard deviation of all batteries from the exponent at time t during the c-th charge. t,c The average value of all battery deviation indices is represented by equations (13) and (14).

[0071]

[0072] The weighting coefficients of the deviation index at different SOC points are shown in Equation (15), the weighted matrix of the deviation index for each charging process is shown in Equation (16), and the weighted matrix W of the deviation index for the entire historical process is shown in Equation (17). In Equation (15), COV... t(SOC),c From equation (12), w is the coefficient of variation at time t, and in equation (16) w SOCN，c For the c-th charge, the weight coefficient at time SOCN is given by equation (17), where W is the weight coefficient matrix constructed from multiple charge processes.

[0073]

[0074] ω c =[w SOC1,c ,w SOC2,c ,…,w SOCN,c (16)

[0075]

[0076] Therefore, the overall deviation index f for each battery during a single charging process i,c As shown in equation (18), the comprehensive deviation index matrix F of the entire historical process is constructed as shown in equation (19). In equation (19), D represents the deviation index matrix of all charging processes, which is a 3-dimensional matrix; W is the weight coefficient matrix constructed by equation (17) for multiple charging processes, f i,c The overall deviation index for the nth charge of battery i.

[0077]

[0078] To determine the rate of battery degradation, a rate of change matrix K is set, as shown in equation (20), where k represents the voltage deviation of each battery. The location and time of abrupt changes in the battery are located and determined using the rate of change matrix. Since battery failure directly manifests as voltage deviation, the faster the deviation changes (i.e., the larger k is), the faster the degradation rate and the higher the risk of failure. Therefore, this application focuses on k... i,c The threshold is set to 2. At the same time, in order to prevent the battery pack from undergoing collective changes in the series battery pack after being left idle for a long time, the judgment condition for the battery with excessively rapid changes is Equation (22).

[0079]

[0080] k i,c =f i,c+1 -f i,c (twenty one)

[0081]

[0082] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0083] Obviously, the above examples of the present invention are merely illustrative of the present invention and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. Any obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. A method for lithium-ion battery anomaly identification and diagnosis based on historical data, characterized in that, The lithium ion battery anomaly identification and diagnosis method can be executed by one or more processors, comprising: S1, obtaining historical data of a battery system, and cleaning the historical data by deleting invalid data and completing missing data; S2, obtaining the charging process in the historical data to further obtain the voltage threshold of a normal battery; S3, determining the abnormal deviation index of the battery based on the voltage threshold, and completing the screening of abnormal batteries based on the abnormal deviation index; S4, determining the fault type of the abnormal battery based on the change trend of the abnormal deviation index, and judging the fault degree of the abnormal battery based on the change speed of the abnormal deviation index; The voltage threshold in S2 is obtained according to the following steps: Obtaining all battery voltages v at time t in the cleaned historical data t ; Based on the formula Obtaining the mean value μ of the battery voltage at time t t,n ; Based on the formula Obtaining a standard value σ of the battery voltage at time t t,n ; determining the abnormal deviation index d based on the voltage threshold value and the standard value i,t , the abnormal deviation index d is expressed as The v t,i The voltage threshold range in which the v t,n -3σ t,n ≤v t,i ≤μ t,n +3σ t,n i = 1, 2, 3...n; v t,i Vt is the voltage value of the i-th battery at time t, n is the number of batteries; The screening of abnormal batteries is completed according to the following steps: N equidistant sampling points are selected and the abnormal deviation index d of the discrete battery is established based on a charging process i = [d i,t(SOC1) ,d i,t(SOC2) ,……,d i,t(SOCN) ]; d i,t(SOCN) is the abnormal deviation index of the i-th battery at the Nth SOC point When the frequency of the abnormal deviation index being greater than 1 is greater than 25% of the set threshold, the discrete battery is determined as an abnormal battery; The fault type of the abnormal battery is determined according to the following steps: The abnormal deviation index is linearly fitted in a single charging process to obtain a linear fitting curve, which can be expressed as m is a linear fitting slope, b0 is a linear fitting intercept, SOC s and SOC e are the starting SOC and the ending SOC, respectively; The fault type is divided into a too-small remaining power and a small battery capacity, wherein a judgment basis of the fault type is: wherein m i is a linear fitting slope of the i-th battery.

2. The lithium-ion battery abnormality recognition and diagnosis method of claim 1, wherein, Based on the formula The historical data is cleaned, wherein, V(i, t) represents the voltage value of the i-th battery at time t, k is the first time when the battery monomer voltage appears 0, and m is the first non-0 time after 0 appears. 3.The lithium ion battery abnormality identification and diagnosis method of claim 1, wherein, The charging process of the normal battery is obtained according to the following steps: According to the formula The end of charging time is filtered, wherein SOC(t) is the remaining power value at time t, and I(k) is the current at time k. From the end of charging, the SOC value at the end of charging and the SOC value at the previous time are compared in reverse order according to time, and the first SOC value less than the sampling point at the previous time is taken as the beginning of the charging section; The charging section with the SOC difference between the beginning and the end less than 40% is removed.

4. The lithium-ion battery abnormality recognition and diagnosis method of claim 1, wherein, The change trend of the abnormal deviation index is determined according to the following steps: Establishing a three-dimensional matrix of abnormal deviation indices for the full history process In the above formula, c represents the cth charge in the entire history process, i represents the battery number, t represents the time point corresponding to the selected SOC, d i,t(SOCj) represents the abnormal deviation index of the ith battery at the jth SOC at the time point of the cth charge; obtaining a comprehensive deviation index COV for each battery t(SOC),c = σ t(SOC),c / μ t(SOC),c ; In the above formula, μ t(SOC),c and σ t(SOC),c are the mean and standard deviation of the voltage at time t (SOC) during the cth charge. The weight coefficient of the deviation index at different SOC points is obtained The deviation index weighting matrix of each charging process is represented as ω c =[w SOC1,c ,w SOC2,c ,…,w SOCN,c ] , wherein w SOCN,c is the weight coefficient of the SOC at time N when charging for the cth time; The full history process deviation indicator weighted matrix is represented as obtaining a comprehensive deviation index for each battery during a single charging process constructing a comprehensive deviation index matrix for the entire historical process 5. The lithium-ion battery abnormality recognition and diagnosis method of claim 4, wherein, The degradation speed of the battery is judged according to the following steps: Constructing an aberration velocity matrix The determination condition of the battery whose change in setting is too fast is set to where k i,c = f i,c+1 -f i,c .

Images