A method for progressive fault diagnosis and early warning of electric vehicle battery systems

Abstract

This invention discloses a progressive fault diagnosis and early warning method for lithium-ion battery systems in electric vehicles, used to assess the safety status of battery packs. The method mainly consists of three parts: the first part performs symplectic geometric mode decomposition on the original battery voltage of the charge-discharge segment of the battery system within each sliding window of appropriate length to obtain the static voltage components reflecting the battery state; the second part extracts the correlation coefficient and difference coefficient between the static voltage components of each individual cell and the static voltage components of the "average individual cell"; the third part inputs the correlation coefficient and difference coefficient features into a connectivity-based outlier factor algorithm combined with a specific threshold to identify abnormal individual cells. This method can perfectly decompose the voltage curve to obtain the static components, avoid the influence of mode aliasing on the results, and accurately identify abnormal individual cells.

Description

Technical Field

[0001] This invention belongs to the field of energy storage technology, and in particular relates to a data-driven method for fault diagnosis and early warning of lithium-ion battery systems for electric vehicles. Background Technology

[0002] Lithium-ion batteries have been widely used in electric vehicles due to their numerous advantages. However, due to differences in raw materials and production processes, inconsistencies exist among individual cells in a battery system. These individual cells are located at different temperatures and positions, and coupled with the complex operating conditions of electric vehicles and user misuse, initial inconsistencies increase, potentially leading to a series of reliability issues and even malfunctions, severely impacting the safe operation of electric vehicles. Battery system failures originate from the battery itself and electrical connection components. Sudden failures occur rapidly with limited pre-failure data, making prediction difficult; while progressive failures occur slowly with traceable evolutionary processes. Utilizing big data mining techniques can accurately grasp the potential patterns in battery systems, identify outliers and anomalies, and allow more time for early fault-tolerant control, preventing further deterioration and effectively protecting public safety and property. Therefore, data-driven research on fault diagnosis and early warning for lithium-ion power battery systems is of great significance for expanding diagnostic technologies, accelerating the iterative upgrading of the automotive industry chain, and ensuring the safety of vehicle operation.

[0003] In recent years, with the rapid development of artificial intelligence and machine learning, as well as the advancements in cloud computing and hardware, data-driven methods, including machine learning and statistical methods, have been widely applied in the field of electric vehicle battery system fault diagnosis. Examples include Kalman filtering models, variational mode decomposition methods, wavelet decomposition methods, local outlier factor algorithms, and isolated forest algorithms. However, existing methods for battery system fault diagnosis and early warning using a combination of signal decomposition and machine learning algorithms suffer from problems such as mode aliasing and difficulty in parameter selection when decomposing noisy voltage signals. Therefore, there is an urgent need to apply efficient signal decomposition-based methods to obtain voltage characterization signal components and combine them with machine learning algorithms to achieve early fault diagnosis and early warning for battery systems. Summary of the Invention

[0004] This invention provides a data-driven method for progressive fault diagnosis and early warning of lithium-ion battery systems for electric vehicles, with the aim of accurately identifying abnormal individual battery cells in advance.

[0005] 1. Fault Diagnosis and Early Warning Process

[0006] This invention proposes a battery system fault diagnosis strategy based on signal decomposition and anomaly detection algorithms. The method first performs symplectic geometric mode decomposition on the original charging or discharging voltage signals, decomposing them into intrinsic mode functions (IMFs). These IMFs reflect the internal state of individual cells or the response to external stimuli in different frequency bands. Then, features are extracted from the static components, including the correlation coefficient and difference coefficient between each individual cell voltage and the "average individual cell" voltage, and these are standardized using Z-scores to convert them into a unified metric. Finally, a connectivity-based outlier factor (COF) algorithm, which offers good stability and high accuracy, is used to identify faulty individual cells. The fault diagnosis process is attached. Figure 1 As shown, the specific steps are as follows:

[0007] Step 1: In a length of Within a sliding window (SW), symplectic geometric mode decomposition is performed on the charging or discharging voltage segment, and the static components in the IMFs are selected to form a new voltage matrix. :

[0008] (1.1)

[0009] In the formula, This represents the number of individual cells in the battery system.

[0010] Step 2: Assuming in Each individual cell The mean and variance of the static voltage component are respectively and So, single cell battery The correlation coefficient between the "average single cell" and the "average single cell" can be calculated using formula (1.2):

[0011] (1.2)

[0012] (1.3)

[0013] In the formula, and They are The mean and variance. In a battery system, the load configuration significantly affects the trend of terminal voltage variation. To identify more patterns in the static voltage component of each individual cell, a detrending fluctuation process will be used. This process eliminates trend voltage by measuring the difference between the static voltage component of each individual cell at each sampling point and the static voltage component of the "average individual cell". Individual cell The coefficient of difference can be calculated according to equation (1.4):

[0014] (1.4)

[0015] And use formula (1.5) to apply it to each The calculated result and Perform Z-scores standardization:

[0016] (1.5)

[0017] In the formula, and These are the mean and standard deviation of the correlation coefficient at each sampling point, respectively.

[0018] Step 3: The autocorrelation coefficient and cross-correlation coefficient features at each location are combined to form a new feature vector, and the features of each feature vector are calculated. Individual cell value:

[0019] (1.6)

[0020] if Exceeding the threshold This indicates that the first [unit / item] in the battery system Individual cells in A malfunction has occurred.

[0021] (1.7)

[0022] In the formula, and They represent the first Individual cells in No abnormalities or abnormalities exist.

[0023] 2. Battery system operating data of the vehicle under study

[0024] For the vehicle with progressive fault No. 1, operational data of the battery system from 23:32:43 on June 17, 2019 to 10:54:30 on July 15, 2019 were selected, with a voltage sampling interval of 30 seconds. (See attached...) Figure 2(a) It can be observed that the selected data includes 15 charge or discharge cycles. Faulty cell #83 exhibits a significant outlier around the 2600th sampling point, and this outlier becomes increasingly pronounced as the electric vehicle battery system operates. The BMS triggers an alarm at the 3722nd sampling point. It is noteworthy that the current variation during the charging and discharging process of an electric vehicle is a complex and highly strategic process. During the charging phase, a multi-stage constant current fast charging strategy is employed. When the battery charge is low, the charger provides a fixed high current to quickly charge the battery. As the battery charge increases, the current gradually decreases to avoid overcharging and battery damage. During the discharging process, the current output is adjusted according to the required power and driving conditions. For example, during highway driving or acceleration, the battery needs to provide high power, and the current output increases accordingly. The current variation of vehicle No.2 during the first discharge and charging cycle is shown in the attached figure. Figure 2 As shown in (d), the current values ​​during the charging and discharging stages are negative and positive, respectively, according to the data acquisition protocol. It can be seen from the figure that the current distribution during charging is relatively simple, while the current distribution during discharging is complex. Furthermore, since the voltage signal of an electric vehicle battery system is a typical time series, it has different time-domain and frequency-domain waveforms, mainly including two different modes: static and dynamic. The static part reflects the battery state, while the dynamic part is the transient response of the battery system to current changes. Therefore, from the attached... Figure 2 (b) and 2(c) visually demonstrate that the voltage data curve during the charging phase is smoother than that during the discharging phase in the first charge and discharge cycle of the selected data.

[0025] For vehicle No. 2 with progressive fault, operational data of the battery system from 10:13:22 on May 23, 2019 to 16:53:00 on May 26, 2019 were selected, with a voltage sampling interval of 20 seconds. (See attached data.) Figure 3 (a) shows the voltage curves for the selected time period. It can be seen that the selected data includes four charge and discharge cycles. Before the BMS alarm, the maximum and minimum voltages of all individual cells did not exceed the cutoff voltage. Cells #3 and #82 consistently had lower voltages. Due to data noise and circuit balancing during discharge, this small deviation could not be clearly detected by the BMS until the last charging segment, when the BMS issued an alarm at sampling point 4386. Analysis of the background alarm data shows that the battery system inconsistency alarm continued until the end of charging. The voltage data curves during the alarm phase are magnified, as shown in the attached figure. Figure 3 As shown in (b), the voltage curves of individual cells #3 and #82 show obvious outlier phenomena.

[0026] For vehicle No. 3 (normal vehicle), operational data of the battery system from 12:02:00 on June 8, 2022 to 22:10:00 on June 9, 2022 was selected, with a voltage sampling interval of 20 seconds. No alarms were issued by the BMS during the operation period. The voltage curve of the battery system during the selected time period is attached. Figure 4 As shown, it contains 7 charge or discharge cycles.

[0027] 3. Discussion of Fault Diagnosis and Early Warning Results

[0028] 3.1 Comparison of different methods for decomposing noisy voltage signals

[0029] Empirical Mode Decomposition (EMD) is a mature signal decomposition method that uses extremum point envelope interpolation to iteratively select single-component signals that meet certain conditions. It has been widely applied in various fault diagnosis fields. The static components of the original voltage signal of a normal vehicle decomposed using EMD and SGMD are shown in the attached figure. Figure 5 As shown in (b) and 5(c), it can be observed that the static component signals obtained by both decomposition methods are similar to the contours of the original voltage signal. However, the signal components decomposed by EMD exhibit mode aliasing, as indicated by the red arrow. This is because the EMD method first needs to determine the local extrema of the signal during the decomposition process, and then connects all the local maxima and minima with cubic splines to form upper and lower envelopes. The mean curve is then obtained from the upper and lower envelopes. During the process of obtaining the envelope, individual prominent noise points in the signal can affect the selection of extrema, resulting in uneven distribution of extrema. Consequently, the obtained envelope is a combination of the local envelope of prominent noise points and the envelope of the true signal. The mean calculated from this envelope, and the IMFs components selected after filtering, contain the inherent modes of the signal and prominent noise points, or contain the inherent modes of adjacent characteristic time scales, thus producing mode aliasing. In contrast, the SGMD method can effectively decompose the original voltage signal into static and dynamic components that meet different analytical needs, which is something that the EMD method cannot perfectly achieve.

[0030] 3.2 Method for determining the length of the sliding window

[0031] In the fault diagnosis of electric vehicle battery systems, the battery switch (SW) can be used to achieve online data monitoring. However, the length of the SW has a significant impact on the accuracy of the fault detection algorithm. A longer SW improves fault detection performance but also increases computational cost and reduces efficiency. Therefore, determining an SW of appropriate length is essential for battery system fault diagnosis. Considering that this chapter focuses on extracting the correlation coefficients of individual cell voltage curves for battery system fault diagnosis, the SW length is often set relatively large. To balance computation time, an evaluation function is defined... To determine the length of SW, The higher the value, the greater the difference between normal and abnormal monomers, making it easier to detect abnormal monomers. The calculation formula is as follows:

[0032] (1.8)

[0033] In the formula, For length equal to The difference between the COF fraction of an abnormal single cell and the maximum COF fraction of a normal single cell under SW conditions. For calculating time.

[0034] In order to calculate The appropriate length of the SW is determined by the length of the SW, and the battery system voltage data of the vehicle with progressive fault No.1 is substituted into the evaluation function. The calculated values ​​are shown in the attached figure. Figure 6 As shown in the figure, the analysis results indicate that when the length of SW is equal to 50, the difference between the COF score of the #83 abnormal cell and the highest COF score of other normal cells is the largest. At this time, the evaluation function... The value also reaches its maximum, indicating that the anomaly detection model is most effective at distinguishing between normal and abnormal single cells under this condition. Therefore, the length of SW can be set to 50. In addition, the SW movement step size can be set to 1 to improve the monitoring effect.

[0035] 3.3 Fault Diagnosis and Early Warning Results

[0036] Under normal circumstances, the voltage variation trends of normal cells in the same battery system are roughly the same. When a battery system malfunctions, the voltage data curve of the faulty cell will be significantly different from that of the normal cell. However, the number of faulty cells in a battery system is usually small, and the abnormal voltage data of a few faulty cells will not affect the overall trend of the battery system. Therefore, the proposed method uses the average value of the static voltage components of all cells in the SW as a reference, defined as the "average cell" voltage static component. Subsequently, the correlation coefficient between the static voltage components of each cell in the SW and the "average cell" voltage static component is calculated, as well as the difference coefficient of each cell's static voltage component after eliminating the "average cell" voltage static component. The standardized characteristics of the voltage data of the first discharge cycle of vehicle No.1 are attached. Figure 7As shown, there are a total of 782 z-scores (SWs). The graph reveals that the standardized correlation coefficient and variance coefficient of cell #83 are significantly different from other normal cells, and the standardized correlation coefficient of cell #83 is greater than the standardized variance coefficient. According to the z-scores standardization rule, if the standardized correlation coefficient is greater than 3 times the standard deviation, these standardized coefficients can be considered abnormal. The black dashed line in the graph represents the range of 3 times the standard deviation (due to the relatively small standard deviation, a supplementary section is provided). Figure 7 The vertical axis is quite large (intuitively resembling a straight line). It can be seen that the standardized correlation coefficients and dissimilarity coefficients of some normal individual cells significantly exceed the normal thresholds. Relying solely on z-scores standardization rules leads to an extremely high false alarm rate. To reduce the false alarm rate and achieve efficient diagnosis of faulty individual cells, a fault diagnosis algorithm based on connectivity outlier factors is proposed. A two-dimensional feature matrix is ​​constructed using the extracted correlation coefficient features, and then the fault diagnosis algorithm is applied to identify each feature point in the matrix.

[0037] After applying the COF fault diagnosis algorithm to the two-dimensional feature matrix composed of standardized correlation coefficients and standardized difference coefficients, a COF score for each individual cell is generated at each SW. To quickly identify abnormal individual cells, a suitable threshold needs to be set. This invention uses a trial-and-error method to select the fault alarm threshold. The highest COF score of normal individual cells in all discharge and charge cycles of the selected vehicle battery system was calculated. To accurately identify faulty individual cells and ensure that normal individual cells do not generate false alarms, the alarm threshold for an SW length of 50 can be set to 7. Figure 8 Figures (a) and (b) show the fault diagnosis results of the No.1 vehicle battery system during the first discharge or charge. As can be seen from the figures, the COF score of cell #83 was higher than the fault alarm threshold in both the discharge cycle and the charge cycle. Cell #83 was first detected as abnormal at the 110th SW in the discharge cycle, which corresponds to the 149th sampling point. The sampling time was 11:52:00 on June 23, 2019. The BMS issued an alarm at the 3722nd sampling point, at 15:42:30 on July 15, 2019. It can be seen that this method can provide early warning of faulty cells 3573 sampling points before the BMS issues an alarm, with the fault alarm time being about 22 days in advance.

[0038] The fault diagnosis results for the first discharge and charge cycle of the vehicle battery system No. 2 are attached. Figure 9As shown, in the first discharge cycle, cells #82 and #3 were flagged as abnormal at sampling point 574 (SW), corresponding to sampling point 623, with a sampling time of 05:37:00 on May 24, 2019. Meanwhile, the BMS issued an alarm at sampling point 4836, with an alarm time of 10:56:31 on May 26, 2019. The comparison shows that the proposed method issues an alarm 2 days and 5 hours earlier than the BMS, providing the driver with ample time for repairs. (Appendix) Figure 9 (b) shows the fault diagnosis and warning results of the first charging cycle data. It can be seen from the figure that at the first SW, the COF scores of cells #82 and #3 exceeded the set threshold.

[0039] The fault diagnosis results for the battery system of vehicle No. 3 during the first discharge and charge cycle are attached. Figure 10 As shown in the figure, it can be seen that the COF score of all individual cells exceeds the safety threshold, and the diagnostic results indicate that they are all normal individual cells.

[0040] In summary, for vehicles No.1 and No.2 with progressive failures, the electric vehicle battery system fault diagnosis and early warning method of the present invention can accurately identify abnormal cells and detect early minor faults in individual cells up to 22 days in advance of the BMS alarm system. Therefore, the fault diagnosis and early warning method is effective and feasible.

[0041] 3.4 Comparison of Fault Diagnosis Results Before and After Charging Cycle Data Decomposition

[0042] During electric vehicle operation, the battery system generates varying currents based on driving power demands, causing fluctuations in the voltage of individual battery cells. This voltage data contains significant noise, including static components reflecting battery state and dynamic components reflecting transient current changes. However, during battery charging, a multi-stage constant-current fast charging mode is employed, resulting in stable voltage changes in individual battery cells, with the voltage data almost equal to the static components reflecting battery state. (Appendix) Figure 11 The comparison of fault diagnosis results before and after using the SGMD method to decompose charging cycle data is presented. The graph shows that the COF score of individual cells is significantly higher without the SGMD method than after its application, and the COF score curve of faulty cells differs significantly from that of normal cells. Considering the voltage characteristics during the charging cycle, the alarm threshold before decomposition is set to 7. Before using the SGMD method, an anomaly was detected at the 28th switching frequency (SW), while after using the SGMD method, an anomaly was detected at the 10th SW. This demonstrates that using the SGMD method to decompose the charging cycle data allows for earlier fault detection.

[0043] This invention addresses the problems of mode mixing and parameter selection difficulties in the decomposition of noisy voltage signals by existing methods. It proposes a battery system fault diagnosis method based on symplectic geometric mode decomposition and local outlier factors of connectivity. The specific contributions are as follows: (1) An evaluation function is constructed to select the length of the SW in order to decompose the voltage data curve. The evaluation function takes into account the difference between normal and abnormal single cells and the calculation time. (2) Voltage segments are selected in the charge and discharge cycle stages using SWs of appropriate length. The selected voltage segments are decomposed using the symplectic geometric mode decomposition method and static components reflecting the battery state are extracted. (3) Considering that the number of faulty single cells in the battery system is small and the abnormal voltage data of individual faulty single cells will not affect the overall trend of the battery system, the correlation coefficient and difference coefficient between the static voltage component of each single cell and the static voltage component of the "average single cell" are calculated. The two coefficients reflecting the fault characteristics are standardized to eliminate the difference between the dimensions. (4) The two-dimensional feature matrix of the coefficients is input into the COF algorithm and combined with a specific threshold to realize the accurate identification and early warning of progressively faulty single cells in the electric vehicle battery system.

[0044] The methods in the embodiments of the present invention are implemented using electronic devices; therefore, it is necessary to describe the relevant electronic devices. For this purpose, embodiments of the present invention provide an electronic device comprising: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor can invoke logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0045] Furthermore, when the logical instructions in at least one of the aforementioned memories can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0046] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0047] The accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of the present invention. Based on this understanding, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, or sometimes in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0048] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading this application specification, they can still modify or make equivalent substitutions to the specific implementation of the present invention, but these modifications or changes do not depart from the protection scope of the pending claims of the present invention. Attached Figure Description

[0049] Figure 1 This is a flowchart for fault diagnosis and early warning.

[0050] Figure 2 Data curves for the battery system of vehicle No.1: (a) Voltage curve during the selected time period; (b) Voltage curve for the first discharge cycle; (c) Voltage curve for the first charge cycle; (d) Current for the first cycle.

[0051] Figure 3 The voltage curves for the battery system of vehicle No.2 are as follows: (a) Voltage curves during the selected time period; (b) A partial enlarged view of the voltage curves during the alarm phase.

[0052] Figure 4 This is the voltage curve of the battery system for vehicle No. 3.

[0053] Figure 5 The discharge cycle voltage curves were decomposed using different methods: (a) raw voltage data; (b) static component decomposed using SGMD; (c) static component decomposed using EMD.

[0054] Figure 6 The evaluation function values ​​under different SW are: (a) the COF score of the abnormal cell and the highest value of other normal cells; (b) the evaluation function value.

[0055] Figure 7 Standardized characteristics of all SWs for vehicle No.1 during the first discharge cycle: (a) standardized correlation coefficient; (b) standardized difference coefficient.

[0056] Figure 8 The fault diagnosis results for the battery system of vehicle No.1 are as follows: (a) first discharge cycle; (b) first charge cycle.

[0057] Figure 9 The fault diagnosis results for the battery system of vehicle No.2 are as follows: (a) first discharge cycle; (b) first charge cycle.

[0058] Figure 10 The fault diagnosis results for the battery system of vehicle No.3 are as follows: (a) first discharge cycle; (b) first charge cycle.

[0059] Figure 11 Comparison of fault diagnosis before and after decomposition of charging cycle data: (a) using SGMD decomposition; (b) not using SGMD decomposition.

Claims

1. A method for progressive fault diagnosis and early warning of lithium-ion battery systems for electric vehicles, characterized in that, The method includes the following steps; S1, obtain the static component of the voltage curve, and use symplectic geometric mode decomposition to decompose the voltage data curve of the battery system charge and discharge cycle into intrinsic mode functions, and select the static component that reflects the internal state of the individual cell. S2, extract features characterizing the fault, extract features of the static components within each sliding window, including the correlation coefficient and difference coefficient between the voltage of each individual cell and the voltage of the "average individual cell", and convert them into a uniform metric by Z-scores standardization. S3, diagnosis and early warning of abnormal single cells, uses a connectivity-based outlier factor algorithm with good qualitative and high accuracy and a specific threshold to determine the faulty single cells; Geometric mode decomposition method for obtaining static components of voltage curves Includes the following steps; Trajectory matrix construction: For a given one-dimensional signal According to Takens' embedding theorem, a one-dimensional signal can be reconstructed into a multi-dimensional signal. The trajectory matrix can be obtained by reconstructing a given one-dimensional voltage signal. ； in, and These are the embedding dimension and the latency, respectively. Delay time Set to 1, embedding dimension Determined by the following formula; ； in, For the length of a one-dimensional signal data, The sampling frequency is typically 0.1 Hz, 0.2 Hz, and 0.3 Hz in electric vehicle battery systems. The maximum dominant peak frequency in the signal power spectral density; Symmetric geometric matrix transformation: based on the trajectory matrix Reconstructing the Hamiltonian matrix ; ； in, Given a symmetric matrix, in order to... To satisfy SGST, define another Hamiltonian matrix. Then the orthogonal symplectic matrix It can be constructed as; ； Among them, matrix It possesses the properties of a symplectic matrix and is an orthogonal symplectic matrix, thus preserving the Hamiltonian matrix during matrix transformations. Spatial phase structure; matrix It is an upper triangular matrix; it can be transformed using the Schmidt orthogonalization method. Let the upper triangular matrix be... eigenvalues , while symmetric matrix The eigenvalues ​​will be equal to The eigenvalues ​​of the matrix are then... The eigenvalues ​​are arranged in descending order as follows: And denote the matrix The eigenvectors corresponding to the eigenvalues ​​are To simplify the process, the orthogonal symplectic matrix in the above formula is... Householder matrix can be used Instead, the matrix is ​​obtained by applying Hamiltonian... Obtained by Schur decomposition; reconstructed initial single-component matrix Through the Householder matrix Seek; ； Then reconstruct the matrix It can be represented as; ； Diagonal averaging; reducing the dimension to... initial component matrix Diagonal averaging is performed; for ease of calculation, a matrix is ​​set up. The elements are ,definition; ； For matrix Perform diagonal averaging; ； The matrix will be reconstructed according to the above formula. Transform into A length of Reconstructed components The original one-dimensional time series can be represented as The sum of the reconstructed components; ； Similar component recombination is necessary because the components derived from SGMD decomposition are not always independent and may contain components with the same features or frequencies. Therefore, it is required to reconstruct the matrix. Similarity analysis is performed to merge similar components; since the eigenvectors are sorted in descending order during symplectic geometric decomposition, the matrix... The components at the front of the signal include one-dimensional signals. The main frequency components; from the components Begin by combining the remaining portions with... Perform similarity analysis and pair components with high similarity with other components. The symplectic geometric components are obtained by merging. ; then, the components will be formed The components from the matrix Delete, and obtain the residual signal. Then the normalized mean square error (NMSE) of the signal decomposition is: ； in, Represents the residual components. It is the number of iterations; the above process is repeated continuously, and the iteration stops when NMSE is less than the set threshold. The final result can be expressed as: ； in, for Number of components; The static component of the voltage curve is obtained using the aforementioned symplectic geometric mode decomposition method; in a length of Within a sliding window, symplectic geometric mode decomposition is performed on the charging or discharging voltage segment, and static components are selected to form a new voltage matrix. ; ； In the formula, This represents the intrinsic mode components obtained after symplectic geometric mode decomposition of the voltage of each battery cell. This refers to the number of individual cells in the battery system. Extracting features that characterize a fault includes the following steps; Assuming in Each individual cell The mean and variance of the static voltage component are respectively and So, single cell battery The correlation coefficient between "average single cell" and "average single cell" can be calculated as follows: ； ； In the formula, and They are The mean and variance. In a battery system, the load configuration significantly affects the trend of terminal voltage variation. To identify more patterns in the static voltage component of each individual cell, a detrending fluctuation process will be used. This process eliminates trend voltage by measuring the difference between the static voltage component of each individual cell at each sampling point and the static voltage component of the "average individual cell". Individual cell The coefficient of difference can be calculated according to the following formula; ； And use the following formula for each The calculated result and Perform Z-scores standardization; ； In the formula, and These are the mean and standard deviation of the correlation coefficient at each sampling point, respectively. The diagnosis and early warning of abnormal individual cells include the following steps; Will The autocorrelation coefficient and cross-correlation coefficient features at each location are combined to form a new feature vector, and the features of each feature vector are calculated. Individual cell value; ； if Exceeding the threshold This indicates that the first [unit / item] in the battery system Individual cells in A malfunction occurred at this location; ； In the formula, and They represent the first Individual cells in No abnormalities or abnormalities exist.

2. A computer storage medium, characterized in that, The storage medium stores program data, which, when executed, implements the method as described in any one of claims 1.

Images