Lithium ion power battery fault feature extraction method based on EEMD-AR

Abstract

The invention discloses a lithium ion power battery fault feature extraction method based on EEMD-AR, and belongs to the technical field of lithium ion power battery fault diagnosis. The invention aims to provide a method for extracting fault features of a lithium ion power battery. The method comprises the following steps: step 1, respectively collecting voltage signals of the lithium ion power battery in different working states, adding normally distributed white noise, and then performing EMD decomposition on the voltage signals to obtain IMF components; step 2, carrying out AR modeling on the normalized IMF component; 3, solving a feature vector of the ith IMF component; step 4, calculating the Mahalanobis distance between the ith IMF component of the voltage signal and the feature vector; 5, selecting a proper weighting parameter, and calculating a comprehensive discrimination distance; and 6, comparing the sizes of the distances, taking the state corresponding to the minimum discriminating distance as the type of the voltage signal of the lithium ion power battery, and calculating the recognition accuracy. The method can effectively improve the accuracy of fault feature extraction.

Description

Technical field:

[0001] The present invention relates to the technical field of lithium-ion power battery fault diagnosis, and in particular to a lithium-ion power battery fault feature extraction method based on EEMD-AR. Background technology:

[0002] In recent years, against the backdrop of increasingly serious global environmental pollution and the shortage of fossil energy, countries around the world have gradually shifted their development focus to the new energy industry, and the new energy vehicle industry and the power battery industry have developed rapidly. New energy vehicles are widely used in the field of transportation, playing an important role in global energy conservation and emission reduction, environmental protection, and promoting the advancement and innovation of new energy technologies. However, the safety of lithium-ion power batteries restricts the development of the new energy vehicle industry. How to quickly and accurately diagnose the factors that cause lithium-ion power battery failures and monitor the fault status of electric vehicles and lithium-ion power batteries has become the key to ensuring the safe operation of electric vehicles. Therefore, it is very necessary to study a method for extracting fault characteristics of lithium-ion power batteries.

[0003] Lithium-ion power batteries will have various faults during use. Before a battery fails, the battery parameters will change abnormally, but this change is usually difficult to detect, or hidden in the parameter changes caused by battery aging, and the sensors inside the battery usually measure the parameter values ​​directly, and cannot diagnose the specific source of the fault. Therefore, it is necessary to decompose the original data signal to decompose the fault information component caused by battery aging and the information component that reflects the real fault of the battery, and it is necessary to extract its fault characteristics through a specific algorithm. The dynamic process of lithium-ion power battery failure is complex, and it is impossible to directly analyze the fault state characteristics from the collected signal. The differences in parameters, characteristic functions, and time domain model structures when lithium-ion power batteries collect signals usually reflect changes in the working state of the battery pack, which also hinders the fault diagnosis of lithium-ion power batteries. Summary of the invention:

[0004] The purpose of the present invention is to provide a method for extracting fault features of lithium-ion power batteries in response to the problems currently existing in the fault diagnosis of lithium-ion power batteries. By combining ensemble empirical mode decomposition (EEMD) and an autoregressive time series model (AR), fault features are extracted from voltage signals collected by lithium-ion power battery sensors, thereby improving the accuracy of lithium-ion power battery fault diagnosis.

[0005] A method for extracting fault features of lithium-ion power batteries based on EEMD-AR. The specific process is as follows:

[0006] Step 1. Collect the voltage signals of lithium-ion power batteries under different working conditions respectively, and evenly add normally distributed white noise ni (t) Then, the voltage signal is decomposed by EMD to obtain the IMF component;

[0007] Step 2. Perform AR modeling on the normalized IMF components;

[0008] Step 3. Find the AR model of the ith IMF component and The average and Find the eigenvector of the i-th IMF component;

[0009] Step 4. Calculate the Mahalanobis distance d between the ith IMF component and the eigenvector of the lithium-ion power battery voltage signal j,i ;

[0010] Step 5. Choose appropriate weighting parameters a1, a2, ..., a n ,satisfy Calculate the comprehensive judgment distance d of lithium-ion power batteries j ;

[0011] Step 6. Compare d j The size of , takes the state corresponding to the minimum comprehensive discrimination distance as the identification type of the lithium-ion power battery voltage signal, so that the working state and fault type of the lithium-ion power battery can be judged, and the recognition accuracy can be calculated.

[0012] Furthermore, in step 1, the voltage signals of the lithium-ion power battery under different working conditions are respectively collected, and the normally distributed white noise n is uniformly added. i (t) After that, the voltage signal is decomposed by EMD to obtain the IMF component, which includes the following steps:

[0013] 1.1 Find all the extreme points of the original signal f(x), and then convert all the maximum points M of f(x) into max and the minimum point M min Connect them, record them as upper envelope and lower envelope, find the average value of their sum m1(x), and the difference r1(x) between f(x) and m1(x);

[0014] 1.2 Determine whether r1(x) meets the IMF decomposition conditions. If so, the decomposed IMF component IMF1 is obtained. If not, it needs to be processed until it meets the IMF decomposition conditions.

[0015] 1.3 Remove IMF1 from the f(x) signal and get the difference between the two as shown in formula (1):

[0016] r1(x)=f(x)-IMF1 (1)

[0017] 1.4 Take r1(x) as the initial value and repeat the above steps to get the second component of IMF IMF2. And so on to get r i (x)(i=1,2,...,n).

[0018] Furthermore, in step 2, performing AR modeling on the normalized IMF components includes the following steps:

[0019] 2.1 The standard deviation of each IMF is normalized to ensure that IMFs at different scales have similar energy distribution;

[0020] 2.2 For each normalized IMF component c i (t) Establish the autoregressive model AR as shown in formula (2):

[0021]

[0022] Furthermore, obtaining the eigenvector of the i-th IMF component in step 3 includes the following steps:

[0023] 3.1 Obtaining the autoregressive parameters in the AR model for each IMF component and calculate the variance The calculation formula is shown in formula (3):

[0024]

[0025] 3.2 Find the i-th IMF component in the AR model and The average and

[0026] 3.3 The eigenvector of the i-th IMF component is shown in formula (4):

[0027]

[0028] Further, in step 4, the Mahalanobis distance d between the ith IMF component and the eigenvector of the lithium-ion power battery voltage signal is calculated. j,i The calculation formula is shown in formula (5):

[0029]

[0030] Furthermore, in step 5, appropriate weighting parameters a1, a2, ..., a are selected. n ,satisfy Calculate the comprehensive judgment distance d of lithium-ion power batteries j The formula is shown in formula (6):

[0031]

[0032] Furthermore, in step 6, d is compared j The size of , taking the state corresponding to the minimum comprehensive discrimination distance as the identification type of the lithium-ion power battery voltage signal, so as to judge the working state and fault type of the lithium-ion power battery, and calculating the recognition accuracy includes the following steps:

[0033] 6.1 Distance between two populations;

[0034] 6.2 Multi-population distance discrimination;

[0035] 6.3 Calculate and evaluate the recognition accuracy.

[0036] The advantages of the present invention over the prior art are: the state of a lithium-ion power battery when a fault occurs is complex, the sensor cannot directly detect the specific source of the fault, and the fault diagnosis method currently used for lithium-ion power batteries has a low accuracy rate, so a method for extracting fault features of lithium-ion power batteries based on EEMD-AR is proposed. The method first adds the collected voltage signal of the lithium-ion power battery to the normally distributed white noise and then performs EEMD decomposition to decompose it into several intrinsic mode functions (IMFs), then AR models the IMF components and extracts the autoregressive parameters and variances therein, obtains the Mahalanobis distance between each IMF component and the eigenvector, uses the Mahalanobis distance discriminant method to identify fault features, and finally calculates and evaluates the recognition accuracy. The method improves the accuracy of fault feature extraction for lithium-ion power batteries to a certain extent, can effectively identify the specific source of the fault, and improves the safety of lithium-ion power battery applications. Description of the drawings:

[0037] Figure 1 This is the algorithm flow chart of EMD decomposition method;

[0038] Figure 2 It is the flow chart of EEMD algorithm;

[0039] Figure 3 This is a flowchart for extracting fault features of lithium-ion power batteries based on the EEMD-AR method; Specific implementation method:

[0040] In order to explain the purpose, technical scheme and features of the present invention more clearly, the present invention is further explained below with reference to the accompanying drawings. Figure 1 This is the flow chart of the EMD decomposition algorithm. Figure 2 This is the EEMD algorithm flow chart. Figure 3FIG. 1 is a flowchart of a lithium-ion power battery fault feature extraction program based on the EEMD-AR method. It should be understood that the specific implementation described herein is to explain the present invention, but not to limit the present invention.

[0041] The EEMD-AR-based lithium-ion power battery fault feature extraction method described in this embodiment is performed according to the following steps:

[0042] Step 1: Collect the voltage signals of lithium-ion power batteries under different working conditions respectively, and evenly add normally distributed white noise n i (t) Then, the voltage signal is decomposed by EMD to obtain the IMF component;

[0043] Find all the extreme points of the original signal f(x), and then convert all the maximum points of f(x) into max and the minimum point M min Connect them, record them as upper envelope and lower envelope, find the average value of the sum of the two m1(x), and the difference r1(x) between f(x) and m1(x) as shown in equations (7) and (8):

[0044]

[0045] r1(x)=f(x)-m1(x) (8)

[0046] If r1(x) satisfies the decomposition condition of IMF, then r1(x) is the first component c1(t) of f(x). If it does not meet the decomposition condition, r1(x) is used as the initial data again and the above steps are repeated to obtain equation (9):

[0047] r 11 (x) = r1(x) - m 11 (x) (9)

[0048] In the formula, m 11 (x) is the average value of the sum of the upper envelope and the lower envelope of r1(x), and then r is determined. 11 (x) Whether it meets the IMF decomposition conditions. If it still does not meet the conditions, it needs to be reprocessed multiple times to obtain formula (10):

[0049] r 1k (x) = r 1(k-1 )(x)-m 1k (x) (10)

[0050] Make m 1k (x) satisfies the IMF decomposition condition, denoted as IMF1 = r 1k (x)

[0051] After removing IMF1 from f(x), we get formula (11):

[0052] r1(x)=f(x)-IMF1 (11)

[0053] Then take r1(x) as the initial value and repeat the above steps to obtain the second component of IMF IMF2. Similarly, we can obtain formula (12):

[0054] r i (x) = r i-1 (x)-IMF i ,i=1,2,...,n (12)

[0055] If there is no IMF component in the signal, the condition shown in formula (13) must be met:

[0056]

[0057] In the formula, m 1k (x) is the average envelope obtained in the current cycle calculation, m 1(k-1) (x) is the average envelope obtained in the last cycle calculation, and the SD value should be between 0.2 and 0.3. Therefore, the original data consists of the sum of the IMF component and the final residual as shown in formula (14):

[0058]

[0059] Finally, the original signal f(x) is decomposed into n IMF components from high frequency to low frequency.

[0060] Step 2: Perform AR modeling on the normalized IMF components;

[0061] The standard deviation of each IMF is normalized to ensure that IMFs at different scales have similar energy distributions. For each normalized IMF component, an autoregressive model AR is established as shown in formula (15):

[0062]

[0063] In the formula, n is the order; is the autoregressive model parameter; n i (t) is the residual of the model, which has a mean of 0 and a variance of A normally distributed white noise sequence.

[0064] The accuracy of the AR model is affected by the order of the AR model. The minimum prediction error criterion (FPE) is used to determine the accuracy of the AR model, as shown in formula (16):

[0065]

[0066] Where N is the length of the signal series, P is the AR model order, is the residual variance, P starts to increase from 1, and when FPE(P) reaches the minimum value, the corresponding P is the optimal order.

[0067] Step 3: Find the AR model of the ith IMF component and The average and Find the eigenvector of the i-th IMF component;

[0068] Obtain the autoregressive parameters in the AR model for each IMF component and calculate the variance The calculation formula is shown in formula (17):

[0069]

[0070] Find the i-th IMF component in the AR model and The average and

[0071] The eigenvector formula of the i-th IMF component is shown in formula (18):

[0072]

[0073] Step 4: Calculate the Mahalanobis distance d between the ith IMF component and the eigenvector of the lithium-ion power battery voltage signal j,i ;

[0074] d j,i The calculation formula is shown in formula (19):

[0075]

[0076] Step 5: Select appropriate weighting parameters a1, a2, ..., a n ,satisfy Calculate the comprehensive judgment distance d of lithium-ion power batteries j ;

[0077] Calculate the comprehensive judgment distance d of lithium-ion power batteries j The formula is shown in formula (20):

[0078]

[0079] Step 6. Compare d jThe size of , takes the state corresponding to the minimum comprehensive discrimination distance as the identification type of the lithium-ion power battery voltage signal, so that the working state and fault type of the lithium-ion power battery can be judged, and the recognition accuracy can be calculated.

[0080] Distance discrimination between two populations. Suppose there are two p-dimensional populations G1 and G2, their mean vectors are μ1 and μ2 respectively, their covariance matrices are Σ1 and Σ2 respectively, x is a quantity to be discriminated, and the judgment rule shown in formula (21) is used to judge which population x belongs to:

[0081]

[0082] Multi-population distance discrimination. Suppose there are k p-dimensional populations G1, G2, …, G k , whose mean vectors are μ1,μ2,…,μ k , the covariance matrices are Σ1,Σ2,…,Σ k , x is the measurement to be measured, and the judgment rule shown in formula (22) is used to determine which population x belongs to:

[0083] x∈G i ,like

[0084] Due to the unknown nature of the overall distribution, the accuracy of the discrimination method is estimated by back-substituting the discrimination results. The evaluation recognition accuracy is calculated as follows:

[0085]

[0086] Where n 12 It means that in the population with sample size n1, there are n 12 The samples are judged as population 2, n 21 It means that in the population with sample size n2, there are n 21 The samples are judged to be population one.

Claims

1. A method for extracting fault features of lithium-ion power batteries based on EEMD-AR, characterized in that: The method comprises the following steps: Step 1: Collect the voltage signals of lithium-ion power batteries under different working conditions respectively, and evenly add normally distributed white noise n i (t) Then, the voltage signal is decomposed by EMD to obtain the IMF component; In the step 1, the voltage signal after adding white noise is expressed as follows after EMD decomposition: Step 2: Perform AR modeling on the normalized IMF components; In the step 2: using the following formula to establish an autoregressive model AR for each IMF component; Step 3: Find the i-th IMF component in the AR model and The average and Find the eigenvector of the i-th IMF component; In step 3: find the autoregressive parameters in the AR model of each IMF component and calculate the variance The eigenvector formula of the i-th IMF component is: Step 4: Calculate the Mahalanobis distance d between the ith IMF component and the eigenvector of the lithium-ion power battery voltage signal j,i ; In the step 4: d j,i The calculation formula is: Step 5: Select appropriate weighting parameters a1, a2, ..., a n ,satisfy Calculate the comprehensive discrimination distance of lithium-ion power batteries; Step 6: Compare d j The size of , takes the state corresponding to the minimum comprehensive discrimination distance as the identification type of the lithium-ion power battery voltage signal, so that the working state and fault type of the lithium-ion power battery can be judged, and the recognition accuracy can be calculated.

Images