A model- and data-driven approach to fault diagnosis of lithium-ion battery stacks

Abstract

This invention discloses a fault diagnosis method for lithium-ion battery stacks that combines model-based and data-driven approaches, comprising the following steps: S1, based on the operating conditions and environment of the lithium-ion battery stack, a health state dataset of the lithium-ion battery stack is obtained through charge-discharge cycle testing, and a fault state space model with a main circuit current sensor is established; S2, based on the established fault state space model with the main circuit current sensor, an AKF (Alternating Current Function) main circuit current sensor fault estimator is designed; S3, using the health state dataset obtained in step S1, the stable kernel space of the lithium-ion battery stack is identified, and then a data-driven residual generator, residual evaluation function, and detection threshold are designed; S4, based on the characteristics of the model-based AKF fault estimator and the data-driven kernel space identification fault detection method, a fault diagnosis decision logic table is designed. This invention can accurately estimate main circuit current sensor faults and diagnose battery internal resistance faults in parallel battery stacks.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery stack fault diagnosis technology, and in particular to a lithium-ion battery stack fault diagnosis method that combines model and data-driven approaches. Background Technology

[0002] In recent years, lithium-ion batteries have been a popular research area in energy storage applications due to their high energy density. Lithium-ion batteries are monitored and controlled for charge and discharge status using a Battery Management System (BMS). However, lithium-ion batteries operate in complex and harsh environments, critical sensors may malfunction, and the slow crystal growth within the battery can pose serious safety hazards. Furthermore, many large-scale systems require higher voltage and power levels, necessitating the series and parallel connection of individual battery cells to meet practical engineering needs. A failure in the battery stack will disrupt the normal operation of the system.

[0003] Most existing technologies focus on fault diagnosis methods for individual battery cells or series-connected battery stacks. However, many scenarios also require parallel connection of lithium-ion batteries to meet power demands. Current research on fault diagnosis for parallel lithium-ion battery stacks is insufficient, especially regarding the diagnosis of faults in the main circuit current sensor and the internal resistance of the battery stack. From a methodological perspective, most current methods are purely mechanistic model-based or purely data-driven, failing to fully combine the advantages of both—namely, interpretability and the ability to operate without requiring precise models. Furthermore, the existence of external disturbances and inherent uncertainties in the system makes it difficult to achieve the desired false alarm rate (FAR). Therefore, a fault diagnosis method for lithium-ion battery stacks that combines model-based and data-driven approaches is urgently needed. Summary of the Invention

[0004] The purpose of this invention is to provide a fault diagnosis method for lithium-ion battery stacks that combines model and data-driven approaches. This method can accurately estimate faults in the main circuit current sensor and diagnose internal resistance faults in parallel battery stacks, reduce the fault accuracy rate (FAR), provide accurate fault information for the BMS to execute corresponding fault-tolerant control, and improve the safety and reliability of the BMS.

[0005] To achieve the above objectives, this invention provides a fault diagnosis method for lithium-ion battery stacks that combines model-driven and data-driven approaches, comprising the following steps:

[0006] S1. Based on the operating conditions and environment of the lithium-ion battery stack, obtain the health status dataset of the lithium-ion battery stack through charge-discharge cycle testing, and establish a fault state space model with a main circuit current sensor.

[0007] S2. Based on the established fault state space model with main circuit current sensor, design the main circuit current sensor fault estimator for AKF.

[0008] AKF's main circuit current sensor fault estimator is:

[0009] P(k|k-1)=AP(k-1|k-1)A T +∑ w (k)

[0010] ∑(k)=CP(k|k-1)C T +∑ v (k)

[0011] K(k)=P(k|k-1)C T ∑ -1 (k)

[0012] P(k|k=[I n -K(k)C]P(k|k-1)

[0013] Ψ(k)=[I n -K(k)C]AΨ(k-1)-[I n -K(k)C]Θ(k)

[0014] Ξ(k)=CAΨ(k-1)+CΘ(k)

[0015] Φ(k)=[γ∑+Ξ(k)Λ(k-1)Ξ T (k)] -1

[0016] γ(k)=Λ(k-1)Ξ T (k)Φ(k)

[0017]

[0018]

[0019]

[0020]

[0021] Where k represents the discrete-time label, (·) T Let P(k|k) denote the transpose of the matrix, where P(k|k) is the covariance matrix, and ∑(k) is the updated covariance matrix. w (k) and ∑ v (k) represent the covariance matrix of the system process noise w(k) and the measurement noise v(k), respectively. A, B, C, and D are the system matrices of the equivalent circuit model of the lithium-ion battery stack state space. nThe identity matrix is ​​compatible with multiple dimensions, the subscript n represents the number of system state variables, K(k) is the gain matrix of AKF, Ψ(k), Ξ(k), Υ(k), and Λ(k) are all auxiliary recursive iteration matrices, and Θ(k) = -[B+C T (CC T ) -1 D]diag(U(k)), where U(k) represents the measured value of the main circuit current of the lithium-ion battery stack, γ∈(0,1) is the forgetting factor of AKF, and Y(k) is the measured value of the terminal voltage of the lithium-ion battery stack. This represents the voltage estimation error between the measured terminal voltage of the lithium-ion battery stack and the AKF estimate. For estimating the state of lithium-ion battery stacks, Real-time estimation of fault θ of the main circuit current sensor;

[0022] S3. Using the health status dataset obtained in step S1, identify the stable kernel space of the lithium-ion battery stack, and then design a data-driven residual generator, residual evaluation function, and detection threshold.

[0023] S4. Based on the characteristics of the model-based AKF fault estimator and the data-driven kernel space identification fault detection method, design a fault diagnosis decision logic table.

[0024] Preferably, in step S1, the operating mode of the battery stack is constant current charge-discharge mode, constant voltage charge-discharge mode, multi-segment constant current charge-discharge mode, multi-segment constant voltage charge-discharge mode, and constant voltage constant current charge-discharge mode.

[0025] Preferably, in step S1, the operating environment of the lithium-ion battery stack includes altitude, atmospheric pressure, ambient temperature, relative humidity, input and output voltage and current range of the battery stack, and maximum charging and discharging current.

[0026] Preferably, in step S1, the health status dataset includes the main circuit current data of the parallel battery stack, the battery stack terminal voltage data, the ambient temperature data, and the current data of each branch inside the parallel battery stack.

[0027] Preferably, in step S1, establishing a fault state-space model with a main circuit current sensor includes the following steps:

[0028] S11. A lithium-ion battery stack consists of N lithium-ion battery cells connected in parallel. A first-order RC equivalent circuit model is established for each lithium-ion battery cell in the stack according to Kirchhoff's laws.

[0029] S12. Based on the obtained lithium-ion battery stack health status dataset, fit the functional relationship curve between the state of charge and open circuit voltage of each lithium-ion battery cell in the lithium-ion battery stack.

[0030] S13. Based on the acquired lithium-ion battery stack health status dataset, the parameters of the first-order RC equivalent circuit model of each lithium-ion battery cell in the lithium-ion battery stack are obtained through the system identification method, and a single-cell lithium battery state space model with branch current as input and terminal voltage as output is established.

[0031] S14. Using the state extension method, establish an extended state space model of the parallel lithium-ion battery stack with all branch currents as inputs and the terminal voltage of the lithium-ion battery stack as output. According to Kirchhoff's voltage and current law, derive the mapping relationship between the main current and the branch current, and substitute it into the extended state space model of the parallel lithium-ion battery stack to establish a state space model of the parallel lithium-ion battery stack with the main current as input and the terminal voltage as output.

[0032] S15. Based on the obtained state-space model of the parallel lithium-ion battery stack with the main circuit current as input and the terminal voltage as output, a state-space model with multiplicative main circuit current sensor fault is established again.

[0033] Preferably, in step S15, the fault information θ value is multiplied by the main circuit current information to model the fault.

[0034] Preferably, in step S3, the residual generator based on data-driven kernel space identification is:

[0035]

[0036] in, For the core space of a parallel lithium-ion battery stack, U k,s and Y k,s These are data vectors representing the main circuit current and terminal voltage of a lithium-ion battery stack, respectively. and The input vectors U and U are respectively k,s and output vector Y k,s The projection operator, r(k), is the residual signal for detecting internal resistance faults in lithium-ion battery stacks;

[0037] To reduce FAR (Fault Alarm Rate), a fault alarm counter is designed to record the number of fault alarms. The specific form of the fault alarm counter is as follows:

[0038]

[0039] Among them, J th Let |J(k)| be the threshold for fault detection, and O{|J(k)|>J} be the fault detection statistic. th} represents the result recorded by the calculator, which is either 1 or 0; assuming a sampling window length of... The residual evaluation function is then:

[0040]

[0041] Among them, J O This represents a sampling window with a length of 1. The number of inner 1s, i.e., the number of fault alarms within one sampling window period; Ultimately, the new fault detection logic is as follows:

[0042]

[0043] Among them, J O (k) represents the number of fault alarms within a sampling window period, O th This is the upper limit for the number of normal alarms, that is, within a continuous period of time. In the sub-sampling calculation, J(k)>J th The number of times.

[0044] Therefore, the present invention employs a lithium-ion battery stack fault diagnosis method that combines model and data-driven approaches, and its technical effects are as follows:

[0045] (1) This invention fully combines the interpretability of the mechanism model and the advantage of the data-driven method that does not require the establishment of an accurate model, and can accurately estimate the fault of the main circuit current sensor in the parallel lithium-ion battery stack while detecting and locating the internal resistance fault.

[0046] (2) The detection threshold in this invention does not issue a fault alarm every time it exceeds the threshold in the traditional method. Instead, it uses the low frequency of faults to make multiple comparisons and only triggers a fault alarm after exceeding a certain number of alarms, which greatly reduces the FAR of the fault.

[0047] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the equivalent circuit model of the first-order RC model in an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of the equivalent circuit model of the parallel battery pack in an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of the charging and discharging electrical topology of a parallel battery pack in an embodiment of the present invention;

[0051] Figure 4 This is a schematic diagram of a fault diagnosis technology route based on a combination of mechanism model and data-driven approach according to the present invention;

[0052] Figure 5 This is a schematic diagram illustrating the comparison results of the FAR (Fault Arrangement Analysis) technology in an embodiment of the present invention; wherein, Figure 5(a) in the text represents a widely used root mean square residual statistical and threshold design method; Figure 5 (b) in the figure represents the fault detection logic proposed in this invention;

[0053] Figure 6 This is a schematic diagram illustrating the battery SOC (State of Charge) fault-tolerant estimation under a main circuit current sensor failure in an embodiment of the present invention; wherein, Figure 6 (a) in the figure represents the result of real-time estimation of faults in the main circuit current sensor; Figure 6 (b) in the figure represents the estimated result after battery SOC fault tolerance compensation correction;

[0054] Figure 7 This is a flowchart of a fault diagnosis method for lithium-ion battery stacks that combines model-driven and data-driven approaches. Detailed Implementation

[0055] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0056] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0057] Example 1

[0058] Figure 4 This is a schematic diagram illustrating the technical route of a fault diagnosis method based on a combination of mechanistic modeling and data-driven approaches provided by the present invention. In this diagram, T represents the internal temperature of the battery stack, and other symbols represent... Figure 3 Same as above. Figure 7 The flowchart illustrates a fault diagnosis method for lithium-ion battery stacks that combines model-driven and data-based approaches. The specific steps of this method are as follows:

[0059] Step 1: Based on the operating mode and environment of the battery stack, obtain the health status dataset through charge-discharge cycle testing, and establish an equivalent circuit model of the parallel battery stack with faults in the main circuit current sensor.

[0060] Specifically, according to Figure 1 Based on the equivalent circuit model topology and Kirchhoff's laws, a first-order RC equivalent circuit model of a lithium-ion battery cell is established, where U oc U is the open-circuit voltage. RCLet R0 be the polarization voltage, C0 be the polarization resistance and C0 be the polarization capacitance, R be the battery internal resistance, I0 be the battery charging / discharging current, and U be the battery terminal voltage. Based on the acquired battery stack health state dataset, the SOC-OCV (state-of-charge versus open-circuit voltage) curve of each lithium-ion battery cell in the stack is fitted. The first-order RC equivalent circuit model parameters of each cell in the stack are obtained using a system identification method. A state-space model of the parallel battery stack is established using the state extension method, with all branch currents as input and the stack terminal voltage as output. The equivalent circuit model of the parallel battery stack is shown below. Figure 2 As shown, where I cell,1 ...I cell,M U oc,1 ...U oc,M R 0,1 ...R 0,M C 0,1 ...C 0,M R1...R M Each of B1...B in the parallel battery stack M The charging and discharging current, open-circuit voltage, polarization resistance, polarization capacitance, and battery internal resistance are all considered. I1 is the main circuit current, and I... k-1 ...I M U represents the current in each loop of a parallel battery stack. t This represents the output voltage of the battery stack. Based on Kirchhoff's voltage and current laws, the mapping relationship between the main circuit current and branch circuit current within the parallel battery stack is derived, and this relationship is substituted into an extended state-space model to establish a state-space model of the parallel lithium-ion battery stack with the main circuit current as input and the terminal voltage as output:

[0061] X(k)=AX(k-1)+B[I-diag(θ)]U(k)

[0062] Y(k)=CX(k)+D[I-diag(θ)]U(k)

[0063] Where X(k) is the state variable of the battery stack, U(k) = I1 is the main circuit current, and Y(k) = U t θ represents the output voltage of the battery stack, A, B, C, and D are the identified system matrices, and θ represents the multiplicative fault coefficient of the main circuit current sensor.

[0064] Step 2: Based on the established fault state-space model with the main circuit current sensor, design an AKF-based main circuit current sensor fault estimator. Specifically, the AKF-based main circuit current sensor fault estimator is as follows:

[0065] P(k|k-1)=AP(k-1|k-1)A T +∑ w (k)

[0066] ∑(k)=CP(k|k-1)C T +∑ v (k)

[0067] K(k)=P(k|k-1)C T ∑ -1 (k)

[0068] P(k|k=[I n -K(k)C]P(k|k-1)

[0069] Ψ(k)=[I n -K(k)C]AΨ(k-1)-[I n -K(k)C]Θ(k)

[0070] Ξ(k)=CAΨ(k-1)+CΘ(k)

[0071] Φ(k)=[γ∑+Ξ(k)Λ(k-1)Ξ T (k)] -1

[0072] γ(k)=Λ(k-1)Ξ T (k)Φ(k)

[0073]

[0074]

[0075]

[0076]

[0077] Where k represents the discrete-time label, (·) T Let P(k|k) denote the transpose of the matrix, where P(k|k) is the covariance matrix, and ∑(k) is the updated covariance matrix. w (k) and ∑ v (k) represent the covariance matrix of the system process noise w(k) and the measurement noise v(k), respectively. A, B, C, and D are the system matrices of the equivalent circuit model of the lithium-ion battery stack state space. n The identity matrix is ​​compatible with multiple dimensions, the subscript n represents the number of system state variables, K(k) is the gain matrix of AKF, Ψ(k), Ξ(k), γ(k), and Λ(k) are all auxiliary recursive iteration matrices, and Θ(k) = -[B+C T (CC T ) -1D]diag(U(k)), where U(k) represents the measured value of the main circuit current of the lithium-ion battery stack, γ∈(0,1) is the forgetting factor of AKF, and Y(k) is the measured value of the terminal voltage of the lithium-ion battery stack. This represents the voltage estimation error between the measured terminal voltage of the lithium-ion battery stack and the AKF estimate. For estimating the state of lithium-ion battery stacks, This algorithm provides real-time estimation of the fault θ of the main circuit current sensor; in this algorithm, matrix Θ(k) = -[B+C T (CC T ) -1 The design of D]diag(U(k)) is the most obvious difference from other technologies. For example... Figure 6 As shown in (a), the AKF (Adaptive Kalman Filter) of the present invention is mainly applicable to the estimation and diagnosis of faults in multiplicative trunk current sensors in parallel battery stacks, and has strong application specificity.

[0078] Step 3: As Figure 3 As shown, B1, B2, and B3 are individual battery cells connected in parallel in a parallel battery stack. cell，1 I cell，2 I cell，3 The charging and discharging currents of B1, B2, and B3 are respectively, U t Let I represent the output voltage of the battery stack, I1 represent the main circuit current, and A1, A, and V within the circle represent the main circuit ammeter, branch circuit ammeter, and output voltmeter, respectively. Using the health status dataset obtained from experimental tests, the stable kernel space of the lithium-ion battery stack is identified, and then a data-driven residual generator, residual evaluation function, and detection threshold are designed.

[0079] Specifically, the residual generator based on data-driven kernel space identification is as follows:

[0080]

[0081] in, For the core space of the parallel battery stack, U k，s and Y k，s , respectively, are the data vectors of the main circuit current and terminal voltage of the battery stack, and r(k) is the residual signal for detecting internal resistance faults in the battery stack.

[0082] To reduce FAR (Fault Alarm Rate), this invention proposes a fault alarm counter:

[0083]

[0084] Based on this, a sampling window length is The residual evaluation function is:

[0085]

[0086] Ultimately, the new fault detection logic is as follows:

[0087]

[0088] Among them, J O (k) represents the number of fault alarms within a sampling window period, O th This is the upper limit for the number of normal alarms, that is, within a continuous period of time. In the sub-sampling calculation, J(k)>J th The number of times. Figure 5 (a) in the text represents a widely used root mean square residual statistical and threshold design method. Figure 5 (b) in the figure represents the fault detection logic proposed in this invention, which significantly reduces FAR under the same conditions.

[0089] Step 4: Based on the characteristics of the model-based AKF sensor fault estimator and the data-driven kernel space identification fault detection method, design a fault diagnosis decision logic table.

[0090] Specifically, this invention utilizes the characteristics of model-driven and data-driven methods to design diagnostic decision logic for faults in the parallel battery stack main circuit current sensor and battery internal resistance. The proposed AKF exhibits robustness against bounded internal resistance faults, leading to the following fault diagnosis decision logic:

[0091] Table 1 Fault Diagnosis Decision Logic

[0092]

[0093] This invention employs a fault diagnosis method that combines a designed mechanistic model with data-driven approaches. This method can diagnose multiplicative circuit current sensor faults and internal resistance faults in parallel battery stacks in real time, and reports the fault information to the BMS to achieve fault-tolerant estimation of battery SOC (e.g., ...). Figure 6 (as shown in (b)) and fault-tolerant control of battery stack charging and discharging improve the safety and reliability of lithium-ion battery applications.

[0094] Therefore, the present invention adopts a lithium-ion battery stack fault diagnosis method that combines the above-mentioned model and data-driven approach. It fully combines the interpretability of the mechanism model and the advantage of the data-driven approach that does not require the establishment of an accurate model. It can accurately estimate the fault of the main circuit current sensor in the parallel lithium-ion battery stack and simultaneously detect and locate the internal resistance fault. The detection threshold is not issued every time the threshold is exceeded in the traditional method. Instead, it uses the low-frequency characteristics of the fault to make multiple comparisons and only triggers the fault alarm after exceeding a certain number of alarms. This greatly reduces the fault accuracy (FAR).

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A model and data driven combined lithium-ion battery stack fault diagnosis method, characterized in that, Comprise the following steps: S1, according to the working condition mode of lithium ion battery stack and the running environment, through the charge and discharge cycle test obtains lithium ion battery stack health state data set, establishes the state space model with the dry way current sensor fault state; S2, according to the established state space model with the dry way current sensor fault state, design AKF dry way current sensor fault estimator; The dry way current sensor fault estimator of AKF is: ； ； ； ； ； ； ； ； ； ； in, Represents the discrete-time label. Represents the transpose of a matrix. Let covariance matrix be the variance matrix. For the updated covariance matrix, and System process noise covariance matrix and measurement noise The covariance matrix is ​​given by A, B, C, and D, which are the system matrices of the equivalent circuit model of the lithium-ion battery stack state space. For identity matrices with compatible dimensions, subscripts The number of system state variables. Here is the gain matrix of AKF. , , , Both are auxiliary recursive iteration matrices. , This represents the measured value of the main circuit current of a lithium-ion battery stack. Forgetting factor of AKF This is the measured value of the terminal voltage of a lithium-ion battery stack. This represents the voltage estimation error between the measured terminal voltage of the lithium-ion battery stack and the AKF estimate. For estimating the state of lithium-ion battery stacks, The fault is due to a fault in the main circuit current sensor. Real-time estimation; S3, using the health state data set obtained in step S1, identify the stable core space of lithium ion battery stack, and then design data-driven residual generator, residual evaluation function and detection threshold; S4, according to the characteristics of AKF fault estimator based on model and data-driven core space identification fault detection method, design fault diagnosis decision logic table; In step S1, the health state data set includes the dry way current data of parallel battery stack, battery stack terminal voltage data, environmental temperature data, the current data of each branch in parallel battery stack; In step S1, the state space model with dry way current sensor fault state is established, comprising the following steps: S11, the lithium ion battery stack is composed of N one lithium ion battery monomer in parallel, according to Kirchhoff's law, a first-order RC equivalent circuit model is established for each lithium ion battery monomer in the battery stack; S12, according to the obtained lithium ion battery stack health state data set, the function relationship curve of the battery state of charge and open circuit voltage of each lithium ion battery monomer in lithium ion battery stack is fitted; S13, based on the obtained lithium ion battery stack health state data set, the first-order RC equivalent circuit model parameters of each lithium ion battery monomer in lithium ion battery stack are obtained by system identification method, and a monomer lithium battery state space model with branch current as input and terminal voltage as output is established; S14, an extended parallel lithium ion battery stack state space model with all branch currents as input and lithium ion battery stack terminal voltage as output is established by using state expansion method, the mapping relationship between dry way current and branch current is obtained according to Kirchhoff's voltage and current law, and is substituted into the extended parallel lithium ion battery stack state space model to establish a parallel lithium ion battery stack state space model with dry way current as input and terminal voltage as output; S15, according to the obtained parallel lithium ion battery stack state space model with dry way current as input and terminal voltage as output, the state space model with multiplicative dry way current sensor fault is established again; In step S15, the information of the fault The values are multiplied with the line current information to model the fault.

2. The model and data driven combined lithium-ion battery stack fault diagnosis method of claim 1, wherein, In step S1, the working condition mode of battery stack is constant current charge and discharge mode, constant voltage charge and discharge mode, multi-section constant current charge and discharge mode, multi-section constant voltage charge and discharge mode and constant voltage constant current charge and discharge mode.

3. The model and data driven combined lithium-ion battery stack fault diagnosis method of claim 1, wherein, In step S1, the running environment of lithium ion battery stack includes altitude, atmospheric pressure, environmental temperature, relative humidity, battery stack input and output voltage and current range, and maximum charge and discharge current.

4. The model and data driven combined lithium-ion battery stack fault diagnosis method of claim 1, wherein, In step S3, the residual generator based on data-driven core space identification is: ； wherein, is the core space of the parallel lithium-ion battery stack, and are data vectors of the dry current and the terminal voltage of the lithium-ion battery stack, respectively, and are input vectors and output vectors of the projection operator, is the residual signal for detecting the lithium-ion battery stack internal resistance fault; In order to reduce FAR, a fault alarm counter is designed to record the number of fault alarms, and the specific form of the fault alarm counter is: ； wherein, is a threshold for fault detection, represents a fault detection statistic, represents a result recorded by the calculator, the result being 1 or 0; assuming a sampling window length of the residual evaluation function is: ； wherein, represents a sampling window length of the number of the 1s, i.e. the number of fault alarms in a sampling window period; Finally, the new fault detection logic is: ； wherein, is the number of fault alarms in a sampling window period, is the upper limit of the number of normal alarms, i.e. in a continuous sampling calculation, is the number of times.

Images