Lithium battery fault diagnosis method based on measurement constraint directional multi-cell spatial filtering

Abstract

The invention discloses a lithium battery charge and discharge state fault detection and analysis method based on multi-cell spatial filtering, and belongs to the technical field of lithium battery fault diagnosis. The method comprises the following steps: monitoring and judging the occurrence of a system fault by constructing and observing a residual multi-cell body of a system, so that the accuracy of fault detection is improved, and a fault detection program has more reasonable sensitivity; aiming at the isolation of fault parameters, a measurement constraint directional fault isolation method using a constraint measurement strip is adopted, so that the occurrence direction of a system fault is obtained more clearly, and the positioning of the parameters of the system fault is quicker and more accurate; for the isolation problem of fault parameters, an extended multi-cell spatial filter is designed for parameter identification, and a filter is specially designed for wrapping the fault parameters in combination with the isolation result of the fault parameters, so that the optimal result of fault parameter identification can be obtained, and the estimation of the fault parameters is more accurate and practical.

Description

Technical Field

[0001] The invention relates to a lithium battery fault diagnosis method based on measurement-constrained directional polyhedral space filtering, and belongs to the technical field of lithium battery fault diagnosis. Background Art

[0002] With the increasing development of new energy vehicles, new energy power generation and energy storage, the application of lithium batteries in industries such as industry and transportation is expanding. In order to meet the further demand for safe, reliable and continuous operation of lithium batteries, it is required to analyze and monitor the charging and discharging status of lithium batteries more accurately and timely, and realize real-time fault detection, isolation and diagnosis.

[0003] Traditional lithium battery fault diagnosis technology often uses sensor information fusion algorithms based on Kalman filtering and its derivative algorithms. Kalman filtering is an effective data fusion and state estimation technology that uses the linear system state equation and the system input and output observation data to optimally estimate the system state. This type of method estimates the state of key parameters based on the probability law of noise distribution and a given system parameter model, and detects faults by analyzing the residual between the estimated state and the actual observed value to increase the accuracy of fault detection. However, in a series of working conditions represented by lithium battery charging and discharging, it is often difficult to describe the noise distribution law with a single probability density function, and in practice, the covariance matrix of process noise and observation noise is often difficult to obtain accurately. In this case, the accuracy and sensitivity of fault diagnosis will be limited, and it has certain limitations.

[0004] In order to overcome the limitations of the above traditional algorithms, fault diagnosis methods based on set membership filtering have been widely studied and applied in recent years. Set membership filtering is a non-statistical filtering technology that recursively constructs sets covering real states. These sets are usually represented as geometric shapes such as ellipsoids and boxes (polyhedrons). According to certain iterative optimization criteria, the measured values ​​of system parameters and system parameter models are combined to obtain a feasible set of system parameters to be detected in the state space. Since the set membership filtering algorithm only requires that the system noise is bounded, and the noise energy in the actual system is limited, this requirement can usually be well met.

[0005] However, the choice of geometric shape and iteration rule has a great influence on the accuracy, conservatism and computational complexity of set membership methods. Different geometric shapes will produce different boundary effects and affect the interaction mode between individuals, thus affecting the accuracy and authenticity of the results; and the selection of iteration rules will affect the convergence speed, computational accuracy and stability. At present, among the set membership methods for fault diagnosis, the ellipsoid iteration method has attracted widespread attention due to its simple iteration form and low complexity, but it has the disadvantages of poor accuracy and high conservatism. Therefore, the set membership method based on centrosymmetric polytopes has received increasing attention. The fault tracking and identification algorithm based on the partition global expansion of polytopes is a method that uses polytopes to describe system uncertainty and perform fault detection; its core idea is to express the uncertainty of the system as a polytope and use the properties of the polytope to design a fault detection method. However, this method is computationally complex and model-dependent, and is not universal in practical applications, and is not sufficient for fault detection with high sensitivity requirements. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a lithium battery fault diagnosis method based on measurement-constrained directional multi-cell spatial filtering, comprising:

[0007] Step 1: Establish an uncertain discrete state space model of the key parameters to be monitored;

[0008] Step 2: Implement polyhedral space state estimation based on polyhedral F-norm optimization for the key parameters to be monitored;

[0009] Step 3: Perform fault detection based on residual polytopes, i.e., determine whether a fault occurs;

[0010] Step 4: perform fault isolation based on measurement constraint orientation, i.e. locate the fault source;

[0011] Step five: perform fault identification based on extended polyhedral space filtering, that is, identify the specific type of fault.

[0012] Optionally, the uncertain discrete state space model of the key parameter to be monitored is in the form of:

[0013] θ( k +1)=θ(k)+Δθ f (k)+Ew(k)

[0014]

[0015] Among them, the vector is the parameter vector composed of the parameters to be monitored, To measure the output vector obtained, k represents the time, T represents the transposition, Represents the coefficient matrix between the system's sensing devices and outputs. is the parameter change value indicating the change of the monitored parameter due to the fault. Assume that at the initial stage of the system, that is, when k = 0, the system has no fault. E and F are known coefficient matrices. w(k) and v(k) are process noise and measurement noise respectively, and both satisfy the unknown but bounded condition, that is:

[0016] |w(k)| ∞ ≤w,|v(k)| ∞ ≤v

[0017] Among them, |*| ∞ represents the infinite norm. This condition can also be expressed in polyhedral form as follows

[0018] w(k)∈ < 0,G w >,v(k)∈<0,G v >

[0019] Optionally, the polyhedral space state estimation method based on polyhedral F-norm optimization is:

[0020] Given the optimal estimate at the previous moment in, is the optimal estimated center point vector of the polyhedron at time k, is the generating matrix of the optimally estimated polyhedron at time k, representing the boundary range of the polyhedron. Then the optimal estimate of the monitored parameter at time k+1 is

[0021]

[0022] Among them, L(k) is the optimal gain matrix, which is the filter iteration variable, and its iteration method is:

[0023]

[0024] Where I represents the identity matrix, G w is the generator matrix of the polytope encapsulating the process noise, G v is the generator matrix of the polyhedron encapsulating the measurement noise, and T denotes the transpose.

[0025] Optionally, the residual polyhedron-based fault detection method is:

[0026] Given the state estimate at time k Then the corresponding residual polytope is

[0027]

[0028] Then when the origin of the state space is wrapped by the residual polyhedron, the system has no fault, otherwise it has a fault.

[0029] That is, when:

[0030]

[0031] When , it can be judged that the system has a fault. Among them, p ε , i (k) represents the i-th element of the residual polytope center point vector, G ε,i,j (k) represents the element in the i-th row and j-th column of the residual polytope generator matrix, i∈{1,…,n y}, j∈{1,…,s} represent the number of rows and columns of the generated matrix respectively.

[0032] Optionally, the fault isolation method based on measurement constraint orientation is:

[0033] If the normal working range of the parameter vector at time k is Then when a fault occurs, the diagonal matrix Γ(k) can be used to characterize the direction of fault expansion, which satisfies

[0034] θ f,i (k)=(Γ(k)θ(k)) i ,i=1,…,n θ ,θ f,i (k)≠0

[0035] Γ(k)=diag(Γ i,i (k)),Γ i,i (k)∈{0,1}

[0036] Among them, θ f,i (k) represents the i-th element of the fault parameter vector, n θ represents the total length of the vector, Γ i,i (k) represents the element in the i-th row and i-th column of the diagonal matrix representing the fault expansion direction.

[0037] Optionally, the fault identification method based on extended polyhedral space filtering is:

[0038] If the fault variable satisfies System parameters meet Then when the system fails, the estimated values ​​of the system parameters can be wrapped by the following polyhedron:

[0039]

[0040] Among them, G f (k) represents the generation matrix of the fault polytope, Representing Minkowski and, represents the set of parameters affected by noise disturbance, Represents the set of parameters affected by the fault. The above variables satisfy:

[0041]

[0042]

[0043] in, represents the center point vector of the polytope set formed by the noise disturbance, represents the generator matrix of the above polyhedron, Represents the generator matrix of the above parameter polytope set affected by the fault.

[0044] The beneficial effects of the present invention are:

[0045] Aiming at the fault diagnosis problem of lithium battery charging and discharging, a lithium battery fault diagnosis method based on measurement constraint oriented polyhedral space filtering is proposed. The present invention monitors and judges the state of key technical parameters of the system by constructing and observing the residual polyhedral of the system, and judges the state of key parameters of the system by fault detection based on residual polyhedral, that is, whether the fault occurs; isolates the fault by polyhedral method based on measurement constraint oriented, that is, locates the source of the fault; identifies the fault by extended polyhedral space filtering method, that is, clarifies the specific type of the fault; and combines the isolation result of the fault parameter, specially designs the filter to wrap the fault parameter, so as to obtain the optimal result of fault parameter identification, making the estimation of fault parameters more accurate and practical. The above method is used to analyze the charging and discharging state of the lithium battery, so that the range judgment of normal parameters is more reasonable, the accuracy of fault detection is increased, and the fault detection program has more reasonable sensitivity. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 It is an experimental comparison diagram of estimated curves obtained under a fault-free state at different times respectively by using the method of the present application and the existing method in one embodiment of the present invention;

[0048] Figure 2 It is an experimental comparison diagram of estimated curves under fault conditions at different times obtained by using the method of the present application and the existing method respectively, disclosed in one embodiment of the present invention;

[0049] Figure 3 It is a comparison diagram of the fault detection effect of the state variable θ1(k) obtained by using the method of the present application and the existing method respectively in one embodiment of the present invention;

[0050] Figure 4 It is a comparison diagram of the fault detection effects on the state variable θ2(k) obtained by using the method of the present application and the existing method respectively, disclosed in one embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0052] Embodiment 1:

[0053] This embodiment provides a method for fault detection in a lithium battery charging and discharging process, which is implemented based on measurement-constrained directional multi-cell spatial filtering. The method includes:

[0054] Step 1: Carry out charge and discharge experiments on the lithium battery to be monitored, and establish a mathematical model of lithium battery charge and discharge based on the Thevenin equivalent circuit, as shown in the following figure:

[0055] θ(k+1)=θ(k)+Δθ f (k)+Ew(k)

[0056]

[0057] According to the modeling method adopted in the embodiment, in this model

[0058]

[0059]

[0060] Among them, U OC (k) is the open-loop voltage of the lithium battery, U(k) is the terminal voltage of the lithium battery, is the coefficient matrix between the system sensor devices and outputs, I c (k) is the current passing through the lithium battery, and the time constant τ p =R p C p , where R p is the electrochemical polarization resistance of the lithium battery, C p is the polarization capacitor, R0 is the internal resistance of the lithium battery. T is the sampling period of the computer test, and T is the transposition. The specific values ​​of the above parameters can be measured experimentally or obtained from the parameter description of the battery itself.

[0061] It should be noted that no matter what method of lithium battery modeling and charge and discharge experimental measurement and modeling is adopted, in theory, as long as the modeling method and parameter measurement are correct and reasonable, a lithium battery charge and discharge mathematical model of the above formula should be obtained. It is not limited to the Thevenin equivalent circuit method described in this embodiment, and the correctness of the final result should be independent of the specific modeling method selected.

[0062] Step 2: Implement multi-cell spatial state estimation for the key parameters to be monitored;

[0063] Given the optimal estimate at the previous moment in, is the optimal estimated center coordinate vector of the polyhedron at time k, is the optimal estimate of the polyhedron at time k. The optimal estimate at time k+1 is

[0064]

[0065] Among them, L(k) is the optimal gain matrix, which is the filter iteration variable, and its iteration method is:

[0066]

[0067] Where I represents the identity matrix, G w is the generator matrix of the polytope encapsulating the process noise, G v is the generator matrix of the polyhedron encapsulating the measurement noise, and T denotes the transpose.

[0068] Step 3: Fault detection based on residual polytope fault detection method;

[0069] Given the state estimate at time k Then the corresponding residual polytope is

[0070]

[0071]

[0072] where p ε (k) represents the center point vector of the residual polyhedron, G ε (k) represents the generator matrix of the residual polytope.

[0073] when:

[0074]

[0075] , it can be determined that there is a fault in the system.

[0076] Among them, p ε,i (k) represents the i-th element of the residual polytope center point vector, Gε,i,j (k) represents the element in the i-th row and j-th column of the residual polytope generator matrix, i∈{1,…,n y}, j∈{1,…,s} represent the number of rows and columns of the generated matrix respectively.

[0077] Step 4: Fault isolation based on measurement constraint orientation;

[0078] If the normal working range of the parameter vector at time k is Then when a fault occurs, the diagonal matrix Γ(k) can be used to characterize the direction of fault expansion, which satisfies

[0079] θ f,i (k)=(Γ(k)θ(k)) i ,i=1,…,n θ ,θ f,i (k)≠0

[0080] Γ(k)=diag(Γ i,i (k)),Γ i,i (k)∈{0,1}

[0081] Among them, θ f,i (k) represents the i-th element of the fault parameter vector, n θ represents the total length of the vector, Γ i,i (k) represents the element in the i-th row and i-th column of the diagonal matrix representing the fault expansion direction.

[0082] Step 5: Fault identification based on extended polyhedral space filtering;

[0083] If the fault variable satisfies System parameters meet Then when the system fails, the estimated values ​​of the system parameters can be wrapped by the following polyhedron:

[0084]

[0085] Among them, G f (k) represents the generation matrix of the fault polytope, Representing Minkowski and, represents the set of parameters affected by noise disturbance, Represents the set of parameters affected by the fault. The above variables satisfy:

[0086]

[0087] in, represents the center point vector of the polytope set formed by the noise disturbance, represents the generator matrix of the above polyhedron, Represents the generator matrix of the above parameter polytope set affected by the fault.

[0088] In order to verify the effectiveness of the method of this application, the method of this application is compared with the existing method; the existing method EF is an ellipsoid algorithm, which can be specifically referred to in the introduction of "Robust particle filter for state estimation in presence of bounded but uncertain parameters based on ellipsoidal set membership approach", and EZKF is a general set membership method, which can be specifically referred to in the introduction of "Dynamic event-triggered fault detection via zonotopic residual evaluation and its application to vehicle lateral dynamics". The experimental comparison results under no fault condition are as follows Figure 1 As shown in the figure, the comparison results under fault conditions are as follows Figure 2 As shown, Figure 1 and Figure 2 In the figure, the black dots represent the true parameter values, the red line represents the estimated curve obtained by the method proposed in this invention, and the blue line represents the estimated curve obtained by the traditional ellipsoid method. Figure 3 and Figure 4 As shown, the MCO-EZKF curve represented by the blue line is the curve obtained by the present invention, the EF curve represented by the red line is the curve obtained by the traditional ellipsoid algorithm, and the EZKF curve represented by the purple line is the curve obtained by the traditional extended polyhedral space algorithm; the faulty area is Figure 3 , Figure 4 The peak appears at about k=300; it can be seen that in terms of fault detection, the method disclosed in the present invention can make the polytope expand in the direction of the fault in a timely manner as the fault occurs (i.e., the peak of the image fluctuation appears), and obtain a more accurate fault value estimation than other methods under the fault state. In summary, compared with other traditional algorithms, the method disclosed in the present invention has better accuracy and lower conservatism when performing state estimation (such as Figure 1 , 2 In terms of fault detection, the method disclosed in the present invention can make the polytope expand in the direction of the fault in a timely manner as the fault occurs, and obtain a more accurate fault value estimation than other methods under the fault state (such as Figure 3 , 4 shown).

[0089] Some steps in the embodiments of the present invention may be implemented using software, and the corresponding software program may be stored in a readable storage medium, such as a CD or a hard disk.

[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A lithium battery fault diagnosis method based on measurement-constrained directional multi-cell spatial filtering, characterized in that: The method comprises: Step 1: Establish an uncertain discrete state space model of the key parameters to be monitored; Step 2: Implement polyhedral space state estimation based on polyhedral F-norm optimization for the key parameters to be monitored to obtain the optimal parameter estimation in, represents the center point vector of the optimal estimated polyhedron at time k+1, Represents the generation matrix of the optimal estimated polyhedron at time k+1; Step 3: Perform fault detection based on residual polytopes, that is, determine whether a fault occurs; Step 4: Perform fault isolation based on measurement constraint orientation, i.e. locate the fault source; Step 5: Perform fault identification based on extended polyhedral space filtering, that is, clarify the specific type of fault.

2. The method according to claim 1, characterized in that The uncertain discrete state space model of the key parameters to be monitored established in step 1 is in the form of: θ(k+1 ) =θ(k)+Δθ f (k)+Ew(k) Among them, the vector is the parameter vector composed of the parameters to be monitored, To measure the output vector obtained, represents the parameter change value of the monitored parameter due to a fault, k represents the time, T represents the transposition, E and F are known coefficient matrices, w(k) and v(k) are process noise and measurement noise respectively, and both satisfy the following unknown but bounded conditions: w(k)∈<0,G w >,v(k)∈<0,G v > Represents the coefficient matrix between the system sensor device and the output. Depending on the actual situation, this matrix may be a time-varying parameter or a time-invariant parameter.

3. The method according to claim 2, characterized in that The second step comprises: Given the optimal estimate at the previous moment in, is the optimal estimated center point vector of the polyhedron at time k, is the generator matrix of the optimally estimated polytope at time k, then the optimal estimate of the monitored parameter at time k+1 is: Among them, L(k) is the optimal gain matrix, is the filter iteration variable, and satisfies Where I represents the identity matrix, G w is the generator matrix of the polytope encapsulating the process noise, G v is the generator matrix of the polyhedron encapsulating the measurement noise, and T denotes the transpose.

4. The method according to claim 3, characterized in that The step three comprises: Given the state estimate at time k Then the corresponding residual polytope is Then when the origin of the state space is wrapped by the residual polyhedron, the system has no fault, otherwise it has a fault; specifically, when: When the system fails, it can be judged that there is a fault; Among them, p ε,i (k) represents the i-th element of the residual polytope center point vector, G ε,i,j (k) represents the element in the i-th row and j-th column of the residual polytope generator matrix, i∈{1,…,n y }, j∈{1,…,s} represent the number of rows and columns of the generated matrix respectively.

5. The method according to claim 4, characterized in that The fourth step comprises: If the normal working range of the parameter vector at time k is Then when a fault occurs, the diagonal matrix Γ(k) can be used to characterize the direction of fault expansion, which satisfies i f,i (k)=(Γ(k)θ(k)) i ,i=1,…,n θ ,i f,i (k)≠0 Γ(k)=diag(Γ i,i (k)),C i,i (k)∈{0,1} Among them, θ f,i (k) represents the i-th element of the fault parameter vector, n θ represents the total length of the vector, Γ i,i (k) represents the element in the i-th row and i-th column of the diagonal matrix representing the fault expansion direction.

6. The method according to claim 5, characterized in that The step five comprises: The fault variable satisfies System parameters meet Then when the system fails, the estimated values ​​of the system parameters can be wrapped by the following polyhedron: Among them, G f (k) represents the generation matrix of the fault polytope, Representing Minkowski and, represents the set of parameters affected by noise disturbance, Represents the set of parameters affected by the fault; the above variables satisfy: in, represents the center point vector of the polytope set formed by the noise disturbance, represents the generator matrix of the above polyhedron, Represents the generator matrix of the above parameter polytope set affected by the fault.

7. The method according to claim 6, characterized in that In addition to being applied to lithium battery charge and discharge fault analysis, the method can also be applied to other fields such as charge and discharge fault analysis of storage batteries manufactured using other principles.

Images