A method for detecting abnormal cells in a lithium battery pack based on probability distribution

Abstract

The application provides a kind of abnormal monomer detection method in lithium battery pack based on probability distribution, by the charging voltage data of monomer unified statistical feature extraction is carried out, and subsequent cluster analysis is carried out to corresponding statistical value, avoid the existing processing mode of monomer voltage frame by frame acquisition, significantly reduce the cost of calculation.The Gaussian mixture model used in the method belongs to unsupervised clustering, compared with the K-means clustering method used in the prior art, it can more smoothly fit different distribution forms and has higher accuracy.The method of the application is more suitable for the scale of hundreds of monomers in the current vehicle battery pack, the obtained statistical value distribution is relatively concentrated, easier to process, so as to simplify the fitting classification and provide faster processing speed and higher calculation efficiency.

Description

Technical Field

[0001] This invention belongs to the field of lithium battery fault diagnosis technology, specifically relating to a method for detecting abnormal cells within a lithium battery pack based on probability distribution. Background Technology

[0002] Currently, due to individual differences in manufacturing and subsequent use, individual lithium batteries that make up automotive power battery packs often exhibit inconsistencies in internal resistance, capacity, and performance. Furthermore, as service time and charge-discharge cycles accumulate, the inconsistencies among individual cells in the battery pack will further intensify, leading to reduced battery life, performance degradation, and various safety issues. In severe cases, it may even cause thermal runaway or fire risks. Therefore, it is necessary to detect abnormal cells in the battery pack as early as possible and eliminate potential safety hazards in advance.

[0003] Because lithium batteries in real-world vehicles exhibit the characteristics of a typical nonlinear cell system, their system state changes in real time. However, the amount of data that can be collected for detecting abnormal cells is very limited. Furthermore, existing battery management systems (BMS) are limited by computing power and cannot detect anomalies by scanning the internal structure of the lithium battery or measuring its internal resistance. Many existing technologies are difficult to apply in real-world vehicles. For example, existing patents CN111007401A, CN115099282A, CN109165687A, and CN111090050A utilize battery test sample data combined with corresponding machine learning algorithms to detect abnormal cells. However, these solutions are highly dependent on the quantity and quality of the test data, failing to meet the real-time and accuracy requirements of real-world applications, and placing significant computational and storage burdens on the BMS. Therefore, providing an accurate and practical solution for detecting abnormal cells within lithium battery packs that effectively utilizes limited data and computing resources is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0004] In view of this, and to address the technical problems existing in this field, the present invention provides a method for detecting abnormal cells within a lithium battery pack based on probability distribution, specifically including the following steps:

[0005] Step 1: For each individual battery cell in the actual vehicle's lithium battery pack, use sensors to collect charging voltage data during charging.

[0006] Step 2: Based on SOC, each charging process is divided into multiple charging segments. The charging voltage data of each battery cell in the same charging segment are statistically analyzed, and the following six statistical values ​​of each battery cell in the charging segment and time series are calculated: the average value, standard deviation, third-order normalized moment (i.e., skewness, kurtosis), unconditional maximum likelihood of autoregression process, and kurtosis of sample absolute Fourier transform spectrum.

[0007] Step 3: Establish a Gaussian mixture model p(x|θ) to describe the probability density of the distribution of each statistical value: where x represents the observation of the statistical sample, θ represents the distribution of the result of x, which is composed of k clusters of Gaussian components; the number of clusters k is determined based on the AIC and BIC criteria to determine the optimal value, and the parameters of θ are obtained iteratively through the expectation-maximization method (EM method);

[0008] Extract the statistical values ​​generated by each cell in any charging cycle, and obtain the specific distribution parameters through the expectation-maximization (EM) method iteratively: calculate the observed value x. i The probability from the Gaussian component of the k-th cluster makes the probability likelihood The logarithm reaches its maximum, and iterative calculations bring the algorithm to converge, ultimately yielding the Gaussian distributions θ. k and its corresponding weight α k ;

[0009] Using the Gaussian mixture model obtained according to the above algorithm, cluster analysis is performed on the statistical values ​​generated by the battery cells, and the log-likelihood of the statistical values ​​with respect to the Gaussian mixture model is calculated; if a cell shows obvious outliers, it is considered that the cell may be abnormal.

[0010] Step 4: For any statistical value, establish horizontal and vertical coordinate axes based on the magnitude of the statistical value and the log-likelihood of the statistical value calculated in Step 3, and determine the corresponding coordinates of each battery cell; take the coordinates of the cell with the largest log-likelihood value as the origin, and calculate the Euclidean distance between the other cells and the origin.

[0011] Step 5: For each individual cell, plot the Euclidean distance change curve during vehicle operation, set a sliding window and calculate the integral area of ​​the curve within the sliding window range; determine whether the integral area exceeds the set threshold, if so, identify the corresponding battery cell as an abnormal cell, otherwise identify it as a normal cell.

[0012] Furthermore, in step one, the charging voltage data of each battery cell is extracted and calculated using Python's Tsfresh utility library.

[0013] Furthermore, the average value of the voltage difference between subsequent adjacent time series calculated in step two is specifically calculated based on the following formula:

[0014]

[0015] In the formula, n is the number of subsequent time series frames;

[0016] The unconditional maximum likelihood X of the autoregressive process t The calculation is based on the following formula:

[0017]

[0018] In the formula, ε is the autoregressive coefficient. t Let k be the random disturbance term, and k be the maximum lag order of the autoregressive process.

[0019] Furthermore, the Gaussian mixture model established in step three takes the following form:

[0020]

[0021] In the formula, K is the number of Gaussian components, or clusters, in the model. For Gaussian distribution θ k The probability density, θ k Follows the mean μ k The variance is σ 2 k Gaussian distribution, θ k The corresponding weight α k satisfy:

[0022]

[0023] For the observation x of sample i i The likelihood of it belonging to the k-th Gaussian component is defined as:

[0024] p(z i =k∣θ)

[0025] In the formula, z i This indicates the Gaussian distribution model to which the sample belongs;

[0026] The likelihood of sample i can be obtained by multiplying the likelihoods of all sub-models:

[0027]

[0028] The corresponding log-likelihood is:

[0029]

[0030] Furthermore, the clustering process of the EM maximum likelihood estimation algorithm specifically includes the following expectation step (E-step) and maximization step (M-step):

[0031] First, the observed value x is calculated using the expectation step. i The probability of coming from the k-th Gaussian component:

[0032]

[0033] Then, through a maximization step, the following parameters are iteratively calculated to maximize the log-likelihood LL(θ):

[0034]

[0035]

[0036]

[0037] Where N is the number of observation samples, and t is the time step of the iteration;

[0038] In the above process, the optimal number of clusters k is determined within the range of 1-5 based on the following AIC and BIC criteria:

[0039] AIC = -2ln(L) + 2k

[0040] BIC = -2ln(L) + ln(n)·k

[0041] In the formula, L is the likelihood function and n is the number of samples.

[0042] Furthermore, in step three, based on the number of outliers, battery cells with three or more outliers in the statistical values ​​are specifically identified as abnormal cells, battery cells with two outliers in the statistical values ​​are identified as cells with risks, and battery cells with one or fewer outliers in the statistical values ​​are identified as normal cells.

[0043] The probability distribution-based abnormal cell detection method for lithium battery packs provided by this invention extracts unified statistical features from the charging voltage data of individual cells and performs subsequent cluster analysis on the corresponding statistical values. This avoids the existing processing method of collecting individual cell voltages frame by frame, significantly reducing computational costs. The Gaussian mixture model used in the method is an unsupervised clustering method, which, compared to the K-means clustering method used in the prior art, can more smoothly fit different distribution forms and has higher accuracy. The method of this invention is more suitable for the scale of hundreds of cells in current automotive battery packs, resulting in a relatively concentrated distribution of statistical values ​​that is easier to process, thereby simplifying the fitting and classification process and providing faster processing speed and higher computational efficiency. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the method provided by the present invention;

[0045] Figure 2 This is a schematic diagram illustrating the identification of anomalous monomers by clustering the kurtosis of the absolute Fourier transform spectrum of statistical values ​​in an example based on the present invention.

[0046] Figure 3 This is a schematic diagram illustrating the identification of abnormal and normal single entities through sliding window processing in an example based on the present invention. Detailed Implementation

[0047] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] The present invention provides a method for detecting abnormal cells within a lithium battery pack based on probability distribution, such as... Figure 1 As shown, the specific steps include:

[0049] Step 1: For each individual battery cell in the actual vehicle's lithium battery pack, use sensors to collect charging voltage data during charging.

[0050] Step 2: Based on SOC, each charging process is divided into multiple charging segments. The charging voltage data of each battery cell in the same charging segment are statistically analyzed, and the following six statistical values ​​of each battery cell in the charging segment and time series are calculated: the mean change of voltage difference between subsequent adjacent time series, the standard deviation, the third normalized moment (i.e., skewness), the kurtosis, the unconditional maximum likelihood of the autoregressive process, and the kurtosis of the sample absolute Fourier transform spectrum.

[0051] Step 3: Establish a Gaussian mixture model p(x|θ) to describe the probability density of the distribution of each statistical value: where x represents the observation of the statistical sample, θ represents the distribution of the result of x, which is composed of k clusters of Gaussian components; the number of clusters k is determined based on the AIC and BIC criteria to determine the optimal value, and the parameters of θ are obtained iteratively through the expectation-maximization method (EM method);

[0052] Extract the statistical values ​​generated by each cell in any charging cycle, and obtain the specific distribution parameters through the expectation-maximization (EM) method iteratively: calculate the observed value x. i The probability from the Gaussian component of the k-th cluster makes the probability likelihood The logarithm reaches its maximum, and iterative calculations bring the algorithm to converge, ultimately yielding the Gaussian distributions θ. k and its corresponding weight α k ;

[0053] Using the Gaussian mixture model obtained according to the above algorithm, cluster analysis is performed on the statistical values ​​generated by the battery cells, and the log-likelihood of the statistical values ​​with respect to the Gaussian mixture model is calculated; if a cell shows obvious outliers, it is considered that the cell may be abnormal.

[0054] Step 4: For any statistical value, establish horizontal and vertical coordinate axes based on the magnitude of the statistical value and the log-likelihood of the statistical value calculated in Step 3, and determine the corresponding coordinates of each battery cell; take the coordinates of the cell with the largest log-likelihood value as the origin, and calculate the Euclidean distance between the other cells and the origin.

[0055] Step 5: For each individual cell, plot the Euclidean distance change curve during vehicle operation, set a sliding window and calculate the integral area of ​​the curve within the sliding window range; determine whether the integral area exceeds the set threshold, if so, identify the corresponding battery cell as an abnormal cell, otherwise identify it as a normal cell.

[0056] In a preferred embodiment of the present invention, the charging voltage data of each battery cell in step one is specifically extracted and calculated using the Tsfresh tool library of Python.

[0057] In a preferred embodiment of the present invention, the average value of the voltage difference between subsequent adjacent time series calculated in step two is specifically calculated based on the following formula:

[0058]

[0059] In the formula, n is the number of subsequent time series frames;

[0060] The unconditional maximum likelihood X of the autoregressive process t The calculation is based on the following formula:

[0061]

[0062] In the formula, ε is the autoregressive coefficient. t Let k be the random disturbance term, and k be the maximum lag order of the autoregressive process.

[0063] In a preferred embodiment of the present invention, the Gaussian mixture model established in step three specifically takes the following form:

[0064]

[0065] In the formula, K is the number of Gaussian components, or clusters, in the model. For Gaussian distribution θ k The probability density, θ k Follows the mean μ k The variance is σ 2k Gaussian distribution, θ k The corresponding weight α k satisfy:

[0066]

[0067] For the observation x of sample i i The likelihood of it belonging to the k-th Gaussian component is defined as:

[0068] p(z i =k∣θ)

[0069] In the formula, z i This indicates the Gaussian distribution model to which the sample belongs;

[0070] The likelihood of sample i can be obtained by multiplying the likelihoods of all sub-models:

[0071]

[0072] The corresponding log-likelihood is:

[0073]

[0074] In a preferred embodiment of the present invention, the clustering process of performing the EM maximum likelihood estimation algorithm specifically includes the following expectation step (E-step) and maximization step (M-step):

[0075] E-step: Calculate the observation x i The probability of coming from the k-th Gaussian component:

[0076]

[0077] M-step: The following parameters are calculated iteratively to maximize the log-likelihood LL(θ) by maximizing the following steps:

[0078]

[0079]

[0080]

[0081] Where N is the number of observation samples, and t is the time step of the iteration;

[0082] In the above process, the optimal number of clusters k is determined within the range of 1-5 based on the following AIC and BIC criteria:

[0083] AIC = -2ln(L) + 2k

[0084] BIC = -2ln(L) + ln(n)·k

[0085] In the formula, L is the likelihood function and n is the number of samples.

[0086] In a preferred embodiment of the present invention, in step three, based on the number of outliers, battery cells with three or more outliers in statistical values ​​are specifically identified as abnormal cells, battery cells with two outliers in statistical values ​​are identified as cells with risks, and battery cells with one or fewer outliers in statistical values ​​are identified as normal cells.

[0087] Figure 2 The diagram illustrates outlier detection based on the statistical value `value__fft_aggregated__aggtype_"kurtosis"`. A significant outlier is observed in the graph, with its log-likelihood significantly lower than other monomers. Comparison with actual accident reports confirms this outlier as the actual occurrence of an accident. Further verification using other statistical values ​​shows that the probability distributions of the remaining five values ​​all point to this severe outlier, demonstrating the effectiveness of this invention.

[0088] Since in-vehicle systems cannot make judgments based on images in practical applications, this invention converts the recognition results of Gaussian mixture models into simple threshold judgments. If an outlier appears in the individual statistical values, the point will be significantly far away from other normal individual points. Therefore, the Euclidean distance between the coordinates of each individual point and the origin can be used to effectively reflect the degree of outlier and anomaly.

[0089] In an example of this invention, by extracting the Euclidean distances between each individual cell involved in the accident and the origin cell, a curve showing the change of the Euclidean distance between the individual cells during vehicle operation is plotted. A sliding window is established, and the integral area of ​​the distance curve within the sliding window over the charging cycle is calculated. The width of the sliding window is generally determined to be 5. Within this window, if no abnormality occurs in the individual cell, its integral area will fluctuate at a low level, unlikely to exceed 0.5. A threshold of 0.5 can be set; if the integral area of ​​any individual cell exceeds 0.5, it is considered an abnormal individual and an alarm is triggered. Figure 3 As shown, the left side represents the integral area within the Euclidean distance sliding window of the abnormal individual, while the right side represents a randomly selected normal individual from the same vehicle. The sliding window width is 5. It can be seen that the curve for the accident individual rapidly climbs to 1 or higher before the accident, reaching its peak at a certain point, while the curve for the normal individual never exceeds 0.5, hovering below 0.2.

[0090] It should be understood that the sequence number of each step in the embodiments of the present invention does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0091] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting abnormal cells within a lithium battery pack based on probability distribution, characterized in that: Specifically, the following steps are included: Step 1: For each individual battery cell in the actual vehicle's lithium battery pack, use sensors to collect charging voltage data during charging. Step 2: Based on SOC, each charging process is divided into multiple charging segments. The charging voltage data of each battery cell in the same charging segment are statistically analyzed, and the following six statistical values ​​of each battery cell in the charging segment and time series are calculated: the average value, standard deviation, third-order normalized moment (i.e., skewness, kurtosis), unconditional maximum likelihood of autoregression process, and kurtosis of sample absolute Fourier transform spectrum. Step 3: Establish a Gaussian mixture model p(x|θ) to describe the probability density of the distribution of each statistical value: where x represents the observation of the statistical sample, θ represents the distribution of the result of x, which is composed of k clusters of Gaussian components; the number of clusters k is determined based on the AIC and BIC criteria to determine the optimal value, and the parameters of θ are obtained iteratively through the expectation-maximization method (EM method); Extract the statistical values ​​generated by each cell in any charging cycle, and obtain the specific distribution parameters through the expectation-maximization (EM) method iteratively: calculate the observed value x. i The probability from the Gaussian component of the k-th cluster makes the probability likelihood The logarithm reaches its maximum, and iterative calculations bring the algorithm to converge, ultimately yielding the Gaussian distributions θ. k and its corresponding weight α k ; Using the Gaussian mixture model obtained according to the above algorithm, cluster analysis is performed on the statistical values ​​generated by the battery cells, and the log-likelihood of the statistical values ​​with respect to the Gaussian mixture model is calculated; if a cell shows obvious outliers, it is considered that the cell may be abnormal. Step 4: For any statistical value, establish horizontal and vertical coordinate axes based on the magnitude of the statistical value and the log-likelihood of the statistical value calculated in Step 3, and determine the corresponding coordinates of each battery cell; take the coordinates of the cell with the largest log-likelihood value as the origin, and calculate the Euclidean distance between the other cells and the origin. Step 5: For each individual cell, plot the Euclidean distance change curve during vehicle operation, set a sliding window and calculate the integral area of ​​the curve within the sliding window range; determine whether the integral area exceeds the set threshold, if so, identify the corresponding battery cell as an abnormal cell, otherwise identify it as a normal cell.

2. The method as described in claim 1, characterized in that: In step one, the charging voltage data of each battery cell is extracted and calculated using Python's Tsfresh utility library.

3. The method as described in claim 1, characterized in that: The average voltage difference between subsequent adjacent time series calculated in step two is specifically calculated based on the following formula: In the formula, n is the number of subsequent time series frames; The unconditional maximum likelihood X of the autoregressive process t The calculation is based on the following formula: In the formula, ε is the autoregressive coefficient. t Let k be the random disturbance term, and k be the maximum lag order of the autoregressive process.

4. The method as described in claim 1, characterized in that: The Gaussian mixture model established in step three takes the following form: In the formula, K is the number of Gaussian components, or clusters, in the model. The Gaussian distribution θ that sample x follows k The probability density, θ k Follows the mean μ k The variance is σ 2 k Gaussian distribution, θ k The corresponding weight α k satisfy: For the observation x of sample i i The likelihood of it belonging to the k-th Gaussian component is defined as: p(z i =k∣θ) In the formula, z i This indicates the Gaussian distribution model to which the sample belongs; The likelihood of sample i can be obtained by multiplying the likelihoods of all sub-models: The corresponding log-likelihood is:

5. The method as described in claim 4, characterized in that: The clustering process of the EM maximum likelihood estimation algorithm specifically includes the following expectation step (E-step) and maximization step (M-step): First, the observed value x is calculated using the expectation step. i The probability of coming from the k-th Gaussian component: Then, through a maximization step, the following parameters are iteratively calculated to maximize the log-likelihood LL(θ): Where N is the number of observation samples, and t is the time step of the iteration; In the above process, the optimal number of clusters k is determined within the range of 1-5 based on the following AIC and BIC criteria: AIC = -2ln(L) + 2k BIC = -2ln(L) + ln(n)·k In the formula, L is the likelihood function and n is the number of samples.

6. The method as described in claim 1, characterized in that: In step three, based on the number of outliers, battery cells with three or more outliers are identified as abnormal cells, battery cells with two outliers are identified as normal cells with potential risks, and battery cells with one or fewer outliers are identified as normal cells.

Images