A lithium ion battery short circuit fault diagnosis method based on fuzzy Kalman filter

Abstract

The application discloses a lithium ion battery short circuit fault diagnosis method based on fuzzy Kalman filtering. First, the offline data of the battery pack is acquired, and then parameter estimation and a benchmark health model are constructed according to the offline data of the battery pack; next, the online data of the battery pack is acquired, a health difference feedback model is constructed and solved by using a fuzzy Kalman filtering method, and the SOC difference values of the batteries are obtained; then, the short circuit current of each battery is calculated by using a threshold RLS, and then the resistance is calculated and obtained, and the short circuit current is input into the health difference feedback model as a feedback quantity to form a closed loop; finally, the battery is diagnosed for short circuit according to the short circuit current and the resistance of the battery. The application solves the problems of estimation error under extreme fault, inaccurate estimation in the low SOC area and the influence of short circuit current on the battery model, and the estimation accuracy is at least 50% higher than that of the traditional algorithm, and the smaller the short circuit fault is, the more obvious the advantage is.

Description

Technical Field

[0001] This invention relates to a battery short-circuit fault diagnosis method in the field of battery fault diagnosis, specifically a soft short-circuit fault diagnosis method for series lithium-ion battery packs based on fuzzy Kalman filtering and a health difference feedback model. Background Technology

[0002] Lithium-ion batteries (LIBs) offer advantages such as high energy density, long cycle life, and environmental friendliness. To meet the extended range requirements of electric vehicles (EVs), battery capacity and size have been continuously increased, but this practice can threaten battery safety. After prolonged use, unintentional misuse, or harsh operating environments, lithium-ion batteries may experience internal short circuits. This can significantly reduce battery life and, in severe cases, lead to thermal runaway, resulting in more serious safety incidents. Therefore, early detection and diagnosis of battery faults, especially short circuit faults, are particularly important.

[0003] Short-circuit faults can be categorized into internal short circuits (ISC) and external short circuits (ESC) from a fault type perspective. Although the fault mechanisms of ISC and ESC differ, their external characteristics are very similar, so they can be described using the same equivalent circuit model. From a fault severity perspective, short-circuit faults can be classified into soft short circuits (SSC) and hard short circuits (HSC). Hard short circuits are considered late-stage short-circuit faults, where fault characteristics are obvious and the condition is rapidly deteriorating and irreversible. In contrast, soft short-circuit faults are considered early-stage faults, with less obvious fault characteristics and difficult to diagnose using conventional methods. In recent years, mainstream short-circuit fault diagnosis algorithms have mainly fallen into two categories: Residual Charge Capacity (RCC) and Inconsistency Detection. The RCC algorithm diagnoses faults by comparing the charge difference between healthy and faulty batteries after one charge-discharge cycle, thus requiring a long detection time. The Inconsistency Detection method retrieves battery fault characteristics by comparing various battery characteristics. While this reduces detection time, it demands higher model accuracy. However, using a high-precision model can lead to problems such as the inability to perform real-time online measurements or excessive online computation. Summary of the Invention

[0004] To address the problems in the background technology while applying the algorithm to online series battery packs, this invention proposes a short-circuit fault diagnosis method for lithium-ion batteries based on fuzzy Kalman filtering. Based on the proposed Health Difference Feedback Model, combined with the Fuzzy Kalman Filter algorithm, a closed-loop Health Difference Feedback Model is designed. By collecting partial information from the series battery pack offline and monitoring the battery health status in real time online, the invention can quickly and accurately diagnose soft or hard short-circuit faults in some batteries (more than one) within the battery pack. This invention solves the estimation error problem under extreme faults, the inaccurate estimation in the low SOC region, and the impact of short-circuit current on the battery model. The estimation accuracy is at least 50% higher than traditional algorithms, and the advantage is more pronounced for smaller short-circuit faults.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] Step 1: Offline acquisition of current, voltage, SOC-OCV curves, and battery capacity of all batteries in the battery pack;

[0007] Step 2: Based on the SOC-OCV curves of all batteries collected offline, and using the TS Fuzzy system principle, a genetic algorithm is used to linearize the SOC-OCV curves of all batteries, obtaining the slope and intercept corresponding to each SOC value of each battery. The 3D lookup table curve of all batteries is composed of each SOC value and the corresponding slope and intercept. Based on the current and voltage of all batteries in the battery pack collected offline, the average internal resistance of the battery pack and the difference in initial internal resistance of each battery are calculated using the particle swarm optimization algorithm.

[0008] Step 3: Select one battery in the battery pack as the healthy battery. Calculate the internal resistance of the healthy battery based on the difference between the average internal resistance of the battery pack and the initial internal resistance of the healthy battery. Then, construct a battery model of the healthy battery and use it as a benchmark healthy model.

[0009] Step 4: Collect the voltage and current of all batteries in the battery pack online. Then, update the current baseline health model and construct the battery models of the remaining batteries based on the collected voltage and current of all batteries. Then, based on the current baseline health model, the battery models of the remaining batteries, the corresponding initial internal resistance differences, 3D lookup table curves, the average internal resistance of the battery pack, and the current short-circuit current and current SOC difference value of all batteries, construct a health difference feedback model. Solve the health difference feedback model using the fuzzy Kalman filter algorithm to obtain and update the SOC difference value of each battery.

[0010] Step 5: Multiply the SOC difference value of each battery by the corresponding battery capacity to obtain the battery power difference of each battery. Use the least squares recursive algorithm to calculate the battery power difference of each battery, obtain the short circuit current of each battery and update it, and then update the health difference feedback model of each battery. Diagnose whether the corresponding battery has a short circuit fault based on the short circuit current of each battery.

[0011] Step 6: Repeat steps 4 and 5 to continuously update the short-circuit current of each battery, thereby continuously diagnosing whether the corresponding battery has a short-circuit fault.

[0012] The battery pack in step 1 consists of several batteries with identical specifications.

[0013] In step 3, the battery voltages collected offline are sorted, and the battery with the median offline collected battery voltage in the battery pack is selected as the healthy battery.

[0014] The formula for the health difference feedback model of each battery is as follows:

[0015] ΔU i =f(ΔSOC) i )-IΔR i -I sc (R health +ΔR i )

[0016] f(ΔSOC i )=k1(ΔSOC i +SOC health )+b1-k2SOC health -b2

[0017]

[0018] Where, ΔU i ΔR i and ΔSOC i R represents the voltage difference, internal resistance difference, and SOC difference between the i-th battery and the baseline health model, respectively. health I represents the internal resistance of a healthy battery. sc Let I and I represent the short-circuit current of the i-th battery and the series current of the battery pack, respectively, and f(ΔSOC) i () represents the open-circuit voltage difference of the i-th battery, k1 and b1 represent the slope and intercept of the current SOC value of the i-th battery in the corresponding 3D lookup table curve, respectively, and k2 and b2 represent the slope and intercept of the current SOC value of the healthy battery in the corresponding 3D lookup table curve, respectively. health K represents the current SOC value of a healthy battery. d This represents the SOC difference parameter.

[0019] In step 5, the least squares recursive algorithm has a response threshold. When the calculated short-circuit current is less than the response threshold, the calculated short-circuit current is set to zero and then output.

[0020] Step 5 or step 6 further includes calculating the short-circuit resistance of the corresponding battery based on the short-circuit current of each battery, and judging the severity of the short-circuit fault of the corresponding battery based on the short-circuit resistance of each battery.

[0021] The beneficial effects of this invention are as follows:

[0022] (1) This invention proposes a health difference feedback model that combines the advantages of the mean and median values. The model parameters remain highly stable even under extreme value interference. Simultaneously, it can promptly detect the fault status of the reference battery used in the model itself.

[0023] (2) This invention utilizes a fuzzy Kalman filter algorithm based on the introduction of the TS-fuzzy system. By fuzzifying the SOC-OCV curve into a 3D lookup table curve, not only is the online computation of the algorithm reduced to a certain extent, but the diagnostic system can also obtain more accurate SOC difference estimates across the entire SOC range.

[0024] (3) This invention introduces a least-squares recursive algorithm with a threshold constraint. The short-circuit current obtained by the algorithm is used as a feedback input in the diagnostic process at the next time step. By setting the threshold, the feedback system will not oscillate or fail to converge. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention.

[0026] Figure 2 This is a diagram illustrating the blurring effect of the present invention.

[0027] Figure 3 This is a comparison chart of the experimental results of the present invention. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0029] like Figure 1 As shown, the present invention includes the following steps:

[0030] Step 1: Offline data acquisition of current, voltage, SOC-OCV curves (SOC, State of Charge; OCV, Open Circuit Voltage), and battery capacity for all cells in the battery pack. The SOC-OCV curves need to be obtained through offline (static) testing using a slow charge / discharge current of 1 / 30C. Online (dynamic) testing requires completing a set of DST (Discharge-Side Test) experiments using the entire series-connected battery pack, recording the voltage of each cell and the current flowing through the battery pack.

[0031] The battery pack in step 1 consists of several batteries with identical specifications.

[0032] Step 2: 1. Offline Static Testing: Based on the SOC-OCV curves of all batteries collected offline, and using the T-SFuzzy system principle, a genetic algorithm is used to linearize the SOC-OCV curves of all batteries, obtaining the slope and intercept corresponding to each SOC value of each battery. A unique 3D lookup table curve for each battery is then formed by combining each SOC value with its corresponding slope and intercept. 2. Offline Dynamic Testing: Based on the current and voltage of all batteries in the battery pack collected offline, a particle swarm optimization algorithm is used to calculate the average internal resistance of the battery pack and the difference in initial internal resistance of each battery.

[0033] The offline static testing method is as follows:

[0034] The original SOC-OCV curve is blurred using the TS Fuzzy system. n line segments are selected as subsystems of the SOC-OCV curve. The SOC value on the horizontal axis is used as an antecedent variable to determine the steady-state point of the system corresponding to each basic subsystem line segment. A weight function is assigned to each selected basic subsystem line segment. When their respective membership degrees reach their maximum value, it indicates that the subsystem is at a steady-state point. To ensure a smooth transition between the linear sub-segments, all membership functions are Gaussian. The mean of the function is the steady-state point, and the function is entirely determined by its mean and variance. All membership functions are normalized to obtain their weighted average, resulting in the TS-fuzzed SOC-OCV curve expression:

[0035] E linear (SOC)=k f SOC+b f

[0036] Among them, E linear (SOC) represents the battery open-circuit voltage at the current SOC, k f and b fThis represents the slope and intercept parameters after fuzzification. The original nonlinear SOC-OCV curve is fuzzified into a linear straight line, and the slope k of this line is... f and intercept b f Each line segment is calculated by adding its slope and intercept according to its respective weight, based on the slope and intercept of the subsystem line segments selected at the beginning. These weights change in real time as the SOC value changes. The expression for each line segment is:

[0037] E linear (n) (SOC)=u·SOC+v

[0038] Where the superscript (n) denotes the nth line segment forming the line, and u and v represent the slope and intercept of the selected line segment. Each weight function must satisfy the following relationship:

[0039] Φ(SOC)≥0, ∑Φ(SOC)>0

[0040] Where Φ(SOC) represents the weight under the current SOC. A Gaussian membership function can be expressed as:

[0041] Φ(SOC)=Φ(μ,σ 2 )

[0042] Where μ represents the steady-state point, i.e., the selected SOC value, and σ 2 The variance is represented by a genetic algorithm (GA). GA is selected as the primary optimization algorithm to optimize the parameters of the variance in the Gaussian membership function. The objective function used is shown below:

[0043]

[0044]

[0045]

[0046] θ=σ 2

[0047] Where E(SOC) represents the value of the original SOC-OCV curve at SOC. θ is the first optimization objective, which is the variance of each Gaussian membership function. st This indicates the number of SOC-OCV curve data points used in the optimization process. After optimization, the corresponding k and b for each SOC are obtained and converted into a 3D query curve for import into an offline database, facilitating easy access by the online system. The final effect obtained by fuzzification is shown below. Figure 2 As shown.

[0048] After obtaining all the weight functions, k f and b f They can be represented as:

[0049] k f =∑[ω(SOC)·u]

[0050] b f =∑[ω(SOC)·v]

[0051]

[0052] Where ω(SOC) is the normalized weight, which is the result obtained after normalizing the original weight.

[0053] Offline dynamic testing steps and methods:

[0054] The particle filter algorithm is used to optimize two objective parameters.

[0055] In the first objective function, the optimization objective is the difference ΔR between the internal resistance of a single cell and the average internal resistance of the battery pack. (i) The function expression is as follows:

[0056]

[0057] ΔR (i) =[ΔR (i) d ΔR (i) c ]

[0058] in, ΔR represents an estimate of the difference between the internal resistance of a single cell and the average internal resistance of the battery pack. (i) d ΔR is the difference between the internal resistance of the battery when it is in a discharged state and the average internal resistance of the battery pack. (i) c This represents the difference between the internal resistance of the battery when it is charging and the average internal resistance of the battery pack. Both are initially set to 0. n represents the total number of data points for a single battery cell in the dynamic test, I represents the current flowing through the battery pack, and ΔU... (i) This represents the difference between the battery terminal voltage and the average voltage of the battery pack.

[0059] In the second objective function, the optimization objective is the average internal resistance R of the battery pack. mean The function expression is as follows:

[0060]

[0061] Among them, R mean The initial value is set to 0. E(SOC) mean ) indicates that the SOC-OCV curve is derived from the SOC curve at SOC. mean The function value obtained at that point. I represents the current flowing through the battery pack. U meanThis represents the average voltage of the battery pack. After the two objective functions are optimized, the data will be stored in the offline system so that the online system can access it at any time.

[0062] Step 3: Select one battery in the battery pack as the healthy battery. Calculate the internal resistance of the healthy battery based on the difference between the average internal resistance of the battery pack and the initial internal resistance of the healthy battery. Then, construct a battery model of the healthy battery based on its various parameters (online collected voltage and current; offline collected 3D lookup table curve and internal resistance) and use it as a benchmark healthy model.

[0063] In step 3, the offline collected battery voltages are sorted, and the battery with the median offline collected battery voltage in the battery pack is selected as the healthy battery. The formula for calculating the internal resistance of the healthy battery is as follows:

[0064] R health =R mean +ΔR L

[0065] Among them, R health R represents the internal resistance of a healthy battery. mean ΔR represents the average internal resistance of the battery pack. L This indicates the initial internal resistance difference of a healthy battery.

[0066] In step 3, the health model is not a fixed battery model; it changes at each moment based on the current data, and its data does not only consider the average values ​​of battery pack parameters. The average SOC is recursively obtained using the following formula:

[0067]

[0068] Where η represents the battery's charge / discharge efficiency, T represents the time of one sampling period, C represents the battery's rated capacity, and I... k-1 This represents the current magnitude at the previous sampling time.

[0069] Step 4: Collect the voltage and current of all batteries in the battery pack online. Then, update the current baseline health model and construct the battery models of the remaining batteries (excluding healthy batteries in the battery pack) based on the collected voltage and current of all batteries. Then, based on the current baseline health model, the battery models of the remaining batteries, the corresponding initial internal resistance differences, 3D lookup table curves, the average internal resistance of the battery pack, and the current short-circuit current (the initial value of the short-circuit current is 0) and the current SOC difference value (the initial SOC difference value is set to 0) of all batteries, construct a health difference feedback model. Use the fuzzy Kalman filter algorithm to solve the health difference feedback model to obtain the SOC difference value of each battery.

[0070] The formula for the health difference feedback model for each battery is as follows:

[0071] ΔU i =f(ΔSOC) i )-IΔR i -I sc (R health +ΔR i )

[0072] f(ΔSOC i )=k1(ΔSOC i +SOc health )+b1-k2SOC health -b2

[0073]

[0074] k1=k f (SOC health +ΔSOC i )=∑[ω(SOC health +ΔSOC i )·u]

[0075] b1 = b f (SOC health +ΔSOC i )=∑[ω(SOC health +ΔSOC i )·v]

[0076] k2=k f (SOC health )=∑[ω(SOC health )·u]

[0077] b2 = b f (SO Chealth )=∑[ω(SOC health )·v]

[0078] Where, ΔU i ΔR i and ΔSOC i R represents the voltage difference, internal resistance difference, and SOC difference between the i-th battery and the baseline health model, respectively. health I represents the internal resistance of a healthy battery. sc Let I and I represent the short-circuit current of the i-th battery and the series current of the battery pack, respectively, and f(ΔSOC) i () represents the open-circuit voltage difference of the i-th battery, k1 and b1 represent the slope and intercept of the current SOC value of the i-th battery in the corresponding 3D lookup table curve, respectively, and k2 and b2 represent the slope and intercept of the current SOC value of the healthy battery in the corresponding 3D lookup table curve, respectively. healthK represents the current SOC value of a healthy battery. d The SOC difference parameter represents the difference between the open-circuit voltage and the SOC, as a function of f(ΔSOC). i The linear relationship between the dependent and independent variables in ().

[0079] The fuzzy Kalman filter equation is as follows:

[0080]

[0081]

[0082] z k =ΔU i，k +I k ΔR i +I sc，k-1 (R health +ΔR i )

[0083] Where, x k Let x represent the state variable in k solutions. k-1 This represents the state input obtained from k-1 solutions, i.e., the SOC difference value ΔSOC of the i-th battery in k-1 solutions. i，k-1 z k The measurement input for the kth solution iteration is ΔU, derived from a modified formula of the health difference feedback model. i，k I represents the voltage difference between the i-th battery and the healthy battery in k calculations. k Let ΔR represent the current of the i-th battery in k solutions. i I represents the difference in internal resistance between the i-th battery and the healthy battery. sc，k-1 R represents the short-circuit current of the i-th battery in the (k-1)th solution iteration. health This indicates the internal resistance of a healthy battery. K represents d In ΔSOC i =ΔSOC i，k- The value obtained by calculating w using the formula at time 1. k and v k This represents a value with a mean of zero and a covariance of ∑. w and ∑ v Gaussian noise was used to calculate the Kalman gain.

[0084] Step 5: After obtaining the SOC difference values ​​of each battery in k calculations, convert the calculated ΔSOC into a quantity that can quantitatively analyze the degree of short-circuit fault. When a battery experiences a short-circuit fault, due to the short-circuit current, this battery will release more electricity than other healthy batteries in the same amount of time. Therefore, the difference in battery power consumption can serve as a quantitative reference standard for the degree of battery short circuit. Assuming that the batteries in the battery pack have the same maximum capacity, multiply this by the corresponding battery capacity to obtain the battery power difference for each battery in k calculations. The formula for calculating the battery power difference is as follows:

[0085] ΔC k =CΔSOC k

[0086] Where C represents the battery capacity measured offline, ΔC k This represents the difference in battery capacity between the battery and the baseline health model at the k-th solution.

[0087] The least squares recursive RLS algorithm is used to calculate the battery capacity difference of each battery in k solutions, obtain the short-circuit current of each battery and update it, and then update the health difference feedback model of each battery. Based on the short-circuit current of each battery, the short-circuit fault of the corresponding battery is diagnosed. In specific implementation, if the short-circuit current of the battery is not 0, the corresponding battery has a short-circuit fault.

[0088] The RLS algorithm filters the data and uses the linear change in battery charge over time to determine battery health and short-circuit current. The specific recursive expression for the algorithm is shown below:

[0089]

[0090]

[0091] e k =Y k -X k T ω k-1

[0092] ω k =ω k-1 +θ k e k

[0093] X k =t k -t0

[0094] Y k =ΔC k -ΔC0

[0095] ωk =-I sc，k

[0096] The input quantities of the algorithm include X k and Y k X k The change in time is represented by T, which represents the transpose operation, and t is the value of t. k Y represents the current time relative to the initial time, where t0 represents the preset initial time; k ΔC represents the change in the difference in electrical charge. k Let ω represent the current battery level difference, and ΔC0 represent the battery level difference at time t0. The algorithm output is ω. k , representing the short-circuit current I sc，k The opposite of e. k This represents the error between the measured and estimated values ​​of the current change in battery level. λ is the forgetting factor, which is taken as 0.98 in this paper. In the algorithm, P... k This is an intermediate parameter that is updated in real time; its initial value is 0. θ k The error coefficient is P. k It changes with updates.

[0097] In step 5, to ensure that the diagnosed short-circuit current is not a product of measurement noise, and since the measured value of the short-circuit current in actual testing is almost never completely equal to 0, a response threshold is set in the least squares recursive algorithm to prevent the system from making false diagnoses or reacting too sensitively, and to eliminate the influence of noise. When the calculated short-circuit current is less than the response threshold, the calculated short-circuit current is set to zero and output, and will not be counted as a short-circuit current, thus improving the feedback stability of the short-circuit current.

[0098] Step 6: Repeat steps 4 and 5 to continuously update the short-circuit current of each battery, thereby continuously diagnosing whether the corresponding battery has a short-circuit fault.

[0099] Step 5 or step 6 further includes calculating the short-circuit resistance of each battery based on its short-circuit current, and determining the severity of the short-circuit fault based on the short-circuit resistance of each battery. In practice, the smaller the short-circuit resistance of a battery, the more severe the short-circuit fault in that battery.

[0100] The formula for calculating short-circuit resistance is as follows:

[0101]

[0102] Among them, R sc I represents the short-circuit resistance of the battery. SC U represents the short-circuit current of the battery, and U represents the voltage across the battery terminals measured at the current moment.

[0103] The formula for the short-circuit resistance, which includes the response threshold, is shown below:

[0104]

[0105] Among them, R sc，k U represents the short-circuit resistance obtained in the k-th solution, and its magnitude is used in this paper as a quantitative indicator to define the degree of battery short circuit. k This represents the battery's terminal voltage, used to calculate the short-circuit resistance according to Ohm's law. In this paper, the threshold for short-circuit current is defined as 0.05% of the battery pack current I. k This prevents misdiagnosis while ensuring high detection accuracy. When the detected short-circuit current is less than this threshold, the corresponding battery will be considered healthy, and the short-circuit resistance will be represented as infinite.

[0106] The method proposed in this invention is compared with the traditional average difference model method. Fault conditions of a battery string consisting of 10 cells under the same operating conditions are tested to compare the diagnostic performance of the two algorithms. Figure 3 As shown, resistors of 53.89 ohms, 2.643 ohms, and 102 ohms are connected to the two ends of batteries 2, 8, and 10 respectively to simulate a battery pack fault. The figure compares this invention with the traditional average difference model method and the traditional average difference model method after removing extreme errors. It can be seen that even after removing the influence of extreme errors from traditional diagnostic methods, the method used in this invention significantly outperforms the traditional method in terms of SOC difference estimation accuracy and short-circuit resistance (fault level) testing accuracy. This demonstrates the superior performance of the method in this invention.

Claims

1. A method for diagnosing short-circuit faults in lithium-ion batteries based on fuzzy Kalman filtering, characterized in that, Includes the following steps: Step 1: Offline acquisition of current, voltage, SOC-OCV curves, and battery capacity of all batteries in the battery pack; Step 2: Based on the SOC-OCV curves of all batteries collected offline, and using the TS Fuzzy system principle, a genetic algorithm is used to linearize the SOC-OCV curves of all batteries, obtaining the slope and intercept corresponding to each SOC value of each battery. The 3D lookup table curve of all batteries is composed of each SOC value and the corresponding slope and intercept. Based on the current and voltage of all batteries in the battery pack collected offline, the average internal resistance of the battery pack and the difference in initial internal resistance of each battery are calculated using the particle swarm optimization algorithm. Step 3: Select one battery in the battery pack as the healthy battery. Calculate the internal resistance of the healthy battery based on the difference between the average internal resistance of the battery pack and the initial internal resistance of the healthy battery. Then, construct a battery model of the healthy battery and use it as a benchmark healthy model. Step 4: Collect the voltage and current of all batteries in the battery pack online. Then, update the current baseline health model and construct the battery models of the remaining batteries based on the collected voltage and current of all batteries. Then, based on the current baseline health model, the battery models of the remaining batteries, the corresponding initial internal resistance differences, 3D lookup table curves, the average internal resistance of the battery pack, and the current short-circuit current and current SOC difference value of all batteries, construct a health difference feedback model. Solve the health difference feedback model using the fuzzy Kalman filter algorithm to obtain and update the SOC difference value of each battery. The formula for the health difference feedback model of each battery is as follows: in, and Let represent the voltage difference, internal resistance difference, and SOC difference values ​​between the i-th battery and the baseline health model, respectively. This indicates the internal resistance of a healthy battery. and Let represent the short-circuit current of the i-th battery and the series current of the battery pack, respectively. This represents the difference in open-circuit voltage of the i-th battery. Let S and t represent the slope and intercept of the SOC value of the i-th battery in the corresponding 3D lookup table curve, respectively. These represent the slope and intercept of the current SOC value of a healthy battery in the corresponding 3D lookup table curve, respectively. This indicates the current SOC value of a healthy battery. Indicates the SOC difference parameter; Step 5: Multiply the SOC difference value of each battery by the corresponding battery capacity to obtain the battery power difference of each battery. Use the least squares recursive algorithm to calculate the battery power difference of each battery, obtain the short circuit current of each battery and update it, and then update the health difference feedback model of each battery. Diagnose whether the corresponding battery has a short circuit fault based on the short circuit current of each battery. Step 6: Repeat steps 4 and 5 to continuously update the short-circuit current of each battery, thereby continuously diagnosing whether the corresponding battery has a short-circuit fault.

2. The lithium-ion battery short-circuit fault diagnosis method based on fuzzy Kalman filtering according to claim 1, characterized in that, The battery pack in step 1 consists of several batteries with identical specifications.

3. The lithium-ion battery short-circuit fault diagnosis method based on fuzzy Kalman filtering according to claim 1, characterized in that, In step 3, the battery voltages collected offline are sorted, and the battery with the median offline collected battery voltage in the battery pack is selected as the healthy battery.

4. The method for diagnosing short-circuit faults in lithium-ion batteries based on fuzzy Kalman filtering according to claim 1, characterized in that, In step 5, the least squares recursive algorithm has a response threshold. When the calculated short-circuit current is less than the response threshold, the calculated short-circuit current is set to zero and then output.

5. The method for diagnosing short-circuit faults in lithium-ion batteries based on fuzzy Kalman filtering according to claim 1, characterized in that, Step 5 or step 6 further includes calculating the short-circuit resistance of the corresponding battery based on the short-circuit current of each battery, and judging the severity of the short-circuit fault of the corresponding battery based on the short-circuit resistance of each battery.

Images