A method and device for detecting battery cell failure in an energy storage system
Through the successive variational mode model and box graph local outlier detection, combined with sliding window slope analysis, the complexity and time-long problems in energy storage battery fault diagnosis are solved, and fast and accurate battery cell fault detection is achieved. Early warning of failures has improved the stability and safety of the energy storage system.
Patent Information
- Application Number
- CN202311832549.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-27
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2043-12-27
AI Technical Summary
The prior art is complex and time-consuming to diagnose energy storage battery faults, and has failed to effectively solve the risks of battery aging and thermal runaway, and lacks engineering practicality.
The data decomposition is performed using a successive variational modal model, combined with the box graph local outlier detection and sliding window slope analysis, to achieve rapid and accurate diagnosis of battery cell failure.
It improves the speed, accuracy and universality of battery cell failure detection, and can warn of potential failures 55 minutes in advance, reduce operating costs, and improve the stability of energy storage systems.
Smart Images

Figure CN117554844B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of energy storage battery fault detection, and in particular to a battery cell fault detection method and device for an energy storage system. Background Art
[0002] The safety and thermal runaway issues of energy storage batteries seriously affect the development of the electrochemical energy storage industry. Due to improper use behavior, differences in factory processes and other factors, serious electrical and mechanical abuse will occur. These abuses will accelerate battery aging and rapidly reduce the service life. With the increase in the inconsistency of the monomers, it is easy to trigger the risk of thermal runaway. Therefore, power battery fault diagnosis and safety warning play an important role in saving people's lives and property safety, which can ensure the safe and stable operation of the energy storage system, improve system efficiency and reduce operating costs.
[0003] Currently, most neural networks or derivatives of neural networks are used for battery fault diagnosis and safety warning. However, the optimization of neural network models is complex, and a deep neural network has disadvantages such as long optimization time, without considering the practicality of the project. Summary of the invention
[0004] The present invention provides a method and device for detecting battery cell faults in an energy storage system, which can effectively diagnose battery cell faults in the energy storage system and improve the rapidity, accuracy and universality of battery cell fault detection.
[0005] In order to solve the above technical problems, an embodiment of the present invention provides a method and device for detecting battery cell failure in an energy storage system, including:
[0006] Obtain the thermal runaway data of each battery cell in the energy storage system under operating conditions and obtain the full life cycle data;
[0007] The whole life cycle data is pre-processed in a cleaning state to obtain a number of charging and discharging fragment data;
[0008] Based on the pre-built successive variational modal model, by adjusting the model parameters, the data of each charging and discharging segment is subjected to successive variational modal decomposition to obtain the maximum voltage dynamic fluctuation component and voltage dynamic fluctuation component corresponding to each charging and discharging segment data;
[0009] Based on the selected outlier threshold, each maximum voltage dynamic fluctuation component is subjected to local outlier detection in the box plot to obtain the abnormal battery cell detection result;
[0010] According to the selected first sliding window and the preset slope condition, the internal fault diagnosis of the battery is performed on each voltage dynamic fluctuation component to obtain the internal fault detection result of the battery cell.
[0011] By implementing the embodiment of the present invention, data of thermal runaway of each battery cell in the energy storage system under operating conditions are obtained to obtain full life cycle data; the full life cycle data is preprocessed for cleaning state to obtain a plurality of charge and discharge segment data; based on a pre-constructed successive variational modal model, each charge and discharge segment data is subjected to successive variational modal decomposition by adjusting model parameters to obtain the maximum voltage dynamic fluctuation component and voltage dynamic fluctuation component corresponding to each charge and discharge segment data; based on the selected outlier threshold, each maximum voltage dynamic fluctuation component is subjected to box plot local outlier detection to obtain abnormal battery cell detection results; according to the selected first sliding window and the preset slope condition, each voltage dynamic fluctuation component is subjected to battery internal fault diagnosis to obtain battery cell internal fault detection results. By performing state cleaning preprocessing and successive variational mode decomposition on the full life cycle data under operating conditions, and using box plot local outlier detection and battery internal fault diagnosis, the abnormal faults and internal faults of the battery cells of the energy storage system can be effectively diagnosed. This method can be applied to various operating conditions and can quickly and accurately diagnose battery cells of different models. It is verified through the data of thermal runaway of the energy storage system, which further effectively illustrates the effectiveness of battery cell fault detection and improves the speed, accuracy and universality of battery cell fault detection.
[0012] As a preferred solution, the whole life cycle data is pre-processed to clean the state to obtain several charging and discharging fragment data, specifically:
[0013] Use the preset data cleaning method to process the null values and abnormal values in the full life cycle data to obtain the cleaned full life cycle data;
[0014] Based on the positive and negative current, the monotonically increasing characteristics of the state of charge and the data point continuity condition, the full life cycle data after cleaning is divided into charge and discharge states to obtain the data of each charge and discharge segment; among which, the data point continuity condition is that the number of continuous data points is not less than the preset number value.
[0015] As a preferred solution, the pre-built successive variational modal model is specifically:
[0016] By introducing quadratic penalty terms and Lagrange multiplier terms, the constrained optimization problem of the successive empirical mode structure is converted into an unconstrained optimization problem, and the pre-constructed successive variational mode model is obtained, which is:
[0017]
[0018] Among them, u k (t) is the single-component AM / FM signal obtained by the kth decomposition, ω k is the center frequency of the single-component AM / FM signal obtained by the kth decomposition, ω is the frequency parameter, ξ is the balance parameter, is the Wiener filtering result of the current residual of the saddle point after the modal component Fourier transform, Find the frequency domain value of the modal component time series after the saddle point and Fourier transformation, Find the frequency domain value of the original input signal after the saddle point and the first Fourier transform, Find the frequency domain value of the original input signal after the saddle point and the kth Fourier transform, is the filter, γ is the Lagrange multiplication operator, k is the total number of modes, ∞ is the value added by iterative optimization, is the frequency response of the ith filter, is the frequency response of the original signal after Fourier transform, is the frequency response of the Lagrange multiplication operator.
[0019] As a preferred solution, adjust the model parameters, specifically:
[0020] The input in the pre-built successive variational modal model is set to the single cell voltage sequence of the battery, the original data compactness index is set to the preset index value, the Gaussian noise is set to 0, the standard tolerance of convergence is set to the preset tolerance value, and the stop type is set to 1.
[0021] As a preferred solution, based on the selected outlier threshold, each maximum voltage dynamic fluctuation component is subjected to local outlier detection in a box plot to obtain abnormal battery cell detection results, specifically:
[0022] According to each maximum voltage dynamic fluctuation component, a maximum dynamic voltage fluctuation component matrix is obtained;
[0023] Calculate a mean matrix and a standard deviation matrix based on the selected second sliding window and the maximum dynamic voltage fluctuation component matrix;
[0024] According to the mean matrix and standard deviation matrix, the outlier threshold is selected to obtain the outlier points. The formula is:
[0025] ο=(U>μ+W*σ)|(U<μ-W*σ)
[0026] Among them, o is the outlier, W is the outlier threshold, μ is the mean matrix, σ is the standard deviation matrix, and U is the maximum dynamic voltage fluctuation component matrix;
[0027] Draw outliers and box plots, and based on the box plots, judge abnormal data to obtain abnormal battery cell detection results.
[0028] As a preferred solution, according to the selected first sliding window and the preset slope condition, the internal fault diagnosis of the battery is performed on each voltage dynamic fluctuation component to obtain the internal fault detection result of the battery cell, specifically:
[0029] The dynamic fluctuation components of each voltage are dimensionlessly processed to obtain the absolute value of the mean difference of each voltage;
[0030] According to the absolute value of each voltage mean difference, sliding is performed based on the selected first sliding window, the kurtosis factor is calculated, and the slope between the highest peak and the lowest peak in the current window is calculated to obtain the slope of the current window;
[0031] If the slope of the current window meets the first slope condition, it is determined that the current battery cell has an internal short circuit fault; wherein the first slope condition is specifically:
[0032] fault1=k c <0|((k c ≤0)&(0≤k n <1))
[0033] Among them, k c is the slope of the current window, k n is the slope of the next window, fault1 is an internal short circuit fault;
[0034] If the slope of the current window meets the second slope condition, it is determined that the current battery cell is an open circuit fault; wherein the second slope condition is specifically:
[0035] fault2=(0≤k c <1)|((0≤k c <1)&k n <0)
[0036] Among them, fault2 is an open circuit fault;
[0037] The battery cells corresponding to the internal short circuit faults and open circuit faults are counted to obtain the internal fault detection results of the battery cells.
[0038] As a preferred solution, before counting the battery cells corresponding to the internal short circuit faults and the open circuit faults and obtaining the internal fault detection results of the battery cells, the method further includes:
[0039] When the sliding of the selected first sliding window ends, the number of short-circuit faults and open-circuit faults that meet the preset slope condition is counted, a sliding total window of the upper limit of the threshold of fault processing is selected, the slope of the window is calculated based on the sliding total window, and the internal short-circuit fault and open-circuit fault of the corresponding battery cell are determined according to the preset slope condition and the slope of the window.
[0040] In order to solve the same technical problem, an embodiment of the present invention further provides a battery cell fault detection device for an energy storage system, comprising: a data acquisition module, a data preprocessing module, a modal decomposition module, an abnormal battery cell detection module and a battery internal fault diagnosis module;
[0041] The data acquisition module is used to obtain the thermal runaway data of each battery cell in the energy storage system under the operating conditions to obtain the full life cycle data;
[0042] The data preprocessing module is used to perform cleaning state preprocessing on the full life cycle data to obtain a number of charging and discharging fragment data;
[0043] The modal decomposition module is used to perform successive variation modal decomposition on each charging and discharging segment data based on a pre-built successive variation modal model by adjusting model parameters to obtain the maximum voltage dynamic fluctuation component and voltage dynamic fluctuation component corresponding to each charging and discharging segment data;
[0044] The abnormal battery cell detection module is used to perform local outlier detection of each maximum voltage dynamic fluctuation component on the box plot based on the selected outlier threshold, and obtain the abnormal battery cell detection result;
[0045] The battery internal fault diagnosis module is used to perform battery internal fault diagnosis on each voltage dynamic fluctuation component according to the selected first sliding window and the preset slope condition, and obtain the battery single body internal fault detection result.
[0046] In order to solve the same technical problem, an embodiment of the present invention further provides a computer device, including a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, a battery cell fault detection method of an energy storage system is implemented.
[0047] In order to solve the same technical problem, an embodiment of the present invention further provides a computer-readable storage medium storing a computer program, which implements a battery cell fault detection method for an energy storage system when the computer program is executed by a processor.
[0048] Compared with the prior art, the present invention has the advantages of:
[0049] (1) By adjusting the model parameters of the pre-built successive variational modal model, the abnormal battery cell detection results are obtained, which solves the difficulties of variational modal decomposition (VMD) such as slow convergence speed and high calculation time, reduces the uncertainty of the number of modes and parameters that need to be set in advance, and can accurately locate the location of the abnormal cell, which can play a certain early warning role for the inconsistency of the cell.
[0050] (2) The generated slope of the sliding window is used to realize the internal fault detection of the single cell. For the internal short circuit fault and open circuit fault, the slope is calculated and statistical methods are used. The potential internal short circuit fault and open circuit fault are detected by the number of times the slope of the current sliding window and the next sliding window changes. The warning time can be advanced by up to 55 minutes, providing enterprises and users with sufficient time to discover and deal with the faulty single cells in the energy storage system before thermal runaway occurs, saving lives and property while improving the stability of the electrochemical energy storage system.
[0051] (3) The battery cell fault detection method is realized by using successive variational mode decomposition, local outlier detection of box plots and battery internal fault diagnosis. It can effectively diagnose the battery cell faults of energy storage systems. It has a wide range of applications. While reducing the amount of calculation, it has the advantages of speed, accuracy and universality, and can be better applied to engineering practice.
[0052] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 : A schematic diagram of a flow chart of an embodiment of a method for detecting a battery cell failure in an energy storage system provided by the present invention;
[0054] Figure 2 : A simplified flow chart of an embodiment of a battery cell fault detection method for an energy storage system provided by the present invention;
[0055] Figure 3 : A schematic diagram of the maximum voltage fluctuation component and local outlier detection of a battery cell in an embodiment of a battery cell fault detection method for an energy storage system provided by the present invention;
[0056] Figure 4 : A schematic diagram of applying thermal runaway data of an energy storage system to internal fault detection in an embodiment of a battery cell fault detection method for an energy storage system provided by the present invention;
[0057] Figure 5 : A structural schematic diagram of an embodiment of a battery cell fault detection device for an energy storage system provided by the present invention. DETAILED DESCRIPTION
[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0059] Embodiment 1
[0060] Please refer to Figure 1 , is a schematic diagram of a flow chart of a battery cell fault detection method for an energy storage system provided by an embodiment of the present invention, wherein the simplified flow chart is as follows: Figure 2 The battery cell fault detection method includes steps 101 to 105, each of which is as follows:
[0061] Step 101: Acquire thermal runaway data of each battery cell in the energy storage system under operating conditions to obtain full life cycle data.
[0062] In this embodiment, the data of thermal runaway of the battery under operating conditions is obtained from the data platform of the energy storage system. As an example, the data reset index is used to read the data such as the charge and discharge current, SOC value, etc. obtained from the platform, and some of the battery data obtained is shown in Table 1 below:
[0063] Table 1 Part of the battery data obtained
[0064]
[0065] It should be noted that cells 1 to 97 are battery cells of the energy storage system. The numbers in the table represent the voltage values sampled at each moment, in V. The insulation resistance is the resistance of the entire energy storage system at each sampling moment, in KΩ. Battery SOC, or state of charge, is used to reflect the remaining capacity of the battery. Its value is defined as the ratio of the remaining capacity to the battery capacity, usually expressed as a percentage.
[0066] Step 102: Perform cleaning state preprocessing on the full life cycle data to obtain a number of charge and discharge segment data.
[0067] In this embodiment, the data of the entire life cycle is obtained through the data platform of the energy storage system. Due to the huge amount of data in the entire life cycle of the battery, the data of the entire life cycle needs to be divided into small sample charging and discharging fragment data, and the data of the entire life cycle of the extracted energy storage system battery operation is cleaned and the state division is preprocessed to obtain the charging and discharging fragment data, which includes charging fragment data and discharging fragment data. By dividing the data of the entire life cycle of the power battery into small sample data of charging and discharging, the necessary data support is provided for the subsequent detection of abnormal monomers and the diagnosis of potential internal short circuit / open circuit faults.
[0068] Optionally, step 102 is specifically as follows: using a preset data cleaning method to process data null values and abnormal values in the full life cycle data to obtain the cleaned full life cycle data; based on the positive and negative current, the monotonically increasing characteristics of the state of charge and the data point continuity condition, the cleaned full life cycle data is divided into charge and discharge states to obtain each charge and discharge segment data; wherein, the data point continuity condition is that the number of continuous data points is not less than a preset number value.
[0069] In this embodiment, the data cleaning methods such as removal or difference are first used to process the empty values and abnormal values of the data in the whole life cycle, and then according to the positive and negative current and the monotonically increasing characteristics of the state of charge SOC, the data is divided into charging and discharging segments to obtain the data of the charging segment and the discharging segment under the condition of ensuring the continuity of the data points. The data point continuity condition is that the number of continuous data points is not less than the preset number value, such as: the data points are at least 10 consecutive. As an example, the positive and negative current, whether the SOC monotonically increases is a null value, and it is ensured that the data points are at least 10 consecutive. Finally, all data information is extracted by recording the index values of the starting point and the end point to obtain the charging and discharging segment data.
[0070] Step 103: Based on the pre-constructed successive variational modal model, by adjusting the model parameters, the data of each charging and discharging segment are subjected to successive variational modal decomposition to obtain the maximum voltage dynamic fluctuation component and the voltage dynamic fluctuation component corresponding to the data of each charging and discharging segment.
[0071] In this embodiment, the data of the charge and discharge segments are decomposed using successive variational mode decomposition (SVMD) to obtain the largest dynamic voltage fluctuation component. The maximum voltage fluctuation component is used, which has strong adaptability and significantly improves the accuracy of abnormal monomer detection while reducing modal aliasing. The parameters of the successive variational mode are adjusted to avoid overfitting or underfitting due to the increase or decrease of the parameters affecting the convergence speed.
[0072] Optional, pre-built successive variational modal models, specifically:
[0073] By introducing quadratic penalty terms and Lagrange multiplier terms, the constrained optimization problem of the successive empirical mode structure is converted into an unconstrained optimization problem, and the pre-constructed successive variational mode model is obtained, which is:
[0074]
[0075] Among them, u k (t) is the single-component AM / FM signal obtained by the kth decomposition, ω k is the center frequency of the single-component AM / FM signal obtained by the kth decomposition, that is, the iteration value, ω is the frequency parameter, ξ is the balance parameter, is the Wiener filtering result of the current residual of the saddle point after the modal component Fourier transform, Find the frequency domain value of the modal component time series after the saddle point and Fourier transformation, Find the frequency domain value of the original input signal after the saddle point and the first Fourier transform, Find the frequency domain value of the original input signal after the saddle point and the kth Fourier transform, is the filter, γ is the Lagrange multiplication operator, k is the total number of modes, and ∞ is the value increased by iterative optimization, starting from 1 to ∞. is the frequency response of the ith filter, is the frequency response of the original signal after Fourier transform, is the frequency response of the Lagrange multiplication operator.
[0076] In this embodiment, the maximum voltage dynamic fluctuation component is obtained by applying successive empirical mode decomposition. Based on the successive empirical mode structure, a constrained optimization problem is mainly transformed into an unconstrained optimization problem to construct a successive variational modal model. The constrained optimization problem has the following formula:
[0077]
[0078] Subject to:u k (t)+f r (t) = f signal
[0079] Where: δ(t) is the Dirac function, t is the time variable, j is the imaginary part of the complex number, represents the gradient calculation process, u k (t) is the single-component AM / FM signal obtained by the k-th decomposition; is the residual signal; u i (t) is the modal component obtained by decomposition; f u (t) is the unprocessed part of the input signal; ω k is the center frequency of the single-component AM / FM signal obtained by the kth decomposition; * represents convolution; ξ is the balance parameter; β k (t) and β i (t) are filters The impulse response of signal is the original signal.
[0080] For the constrained optimization problem, by introducing the quadratic penalty term and the Lagrange multiplier term, a constrained optimization problem is converted into an unconstrained optimization problem. The conversion formula of the unconstrained optimal problem is as follows:
[0081]
[0082] Optionally, adjust the model parameters, specifically:
[0083] The input in the pre-built successive variational modal model is set to the single cell voltage sequence of the battery, the original data compactness index is set to the preset index value, the Gaussian noise is set to 0, the standard tolerance of convergence is set to the preset tolerance value, and the stop type is set to 1.
[0084] In this embodiment, the model parameters of the successive variation mode are adjusted, and the inputs in the successive variation are set to the single cell voltage sequence of the battery, the original data compactness index is set to a preset index value (such as: 3000), the Gaussian noise tau is set to 0, the convergence standard tolerance to l is set to a preset tolerance value (such as: 1e-7), and the stop type is set to 1. The setting of this parameter is debugged according to manual experience, which can better avoid the influence of the convergence speed due to the increase or decrease of the parameters, resulting in overfitting or underfitting. By adjusting the parameters of the model, the possibility of over / underfitting is reduced and the accuracy of the model decomposition is improved.
[0085] Step 104: Based on the selected outlier threshold, perform box plot local outlier detection on each maximum voltage dynamic fluctuation component to obtain abnormal battery cell detection results.
[0086] In this embodiment, the maximum voltage fluctuation component of each monomer sequence is obtained through successive variational mode decomposition, and the box plot and outlier detection are drawn through the maximum voltage fluctuation component. According to the average value and median of the maximum voltage dynamic fluctuation component, the appropriate threshold is automatically selected, and the outliers and box plots are drawn, so as to determine the abnormal monomers and the number according to the pattern of the outliers, and obtain the abnormal battery monomer detection result, that is, the abnormal battery monomer detection result includes abnormal battery monomers and the number of abnormal battery monomers. It should be noted that since the abnormal monomers include multiple abnormal situations such as monomer inconsistency, internal short circuit fault and open circuit fault, internal short circuit fault and open circuit fault diagnosis are further performed in step 105.
[0087] Optionally, step 104 specifically includes step 1041 to step 1044, each step is specifically as follows:
[0088] Step 1041: Obtain a maximum dynamic voltage fluctuation component matrix according to each maximum voltage dynamic fluctuation component.
[0089] In this embodiment, the maximum voltage dynamic fluctuation component is obtained through successive variational modal decomposition, and the maximum dynamic voltage fluctuation component matrix is as follows:
[0090]
[0091] in, is the maximum dynamic voltage fluctuation component of the nth monomer, and U is the maximum dynamic voltage fluctuation component matrix.
[0092] Step 1042: Calculate the mean matrix and the standard deviation matrix based on the selected second sliding window and the maximum dynamic voltage fluctuation component matrix.
[0093] In this embodiment, by setting the second sliding window to 10, the mean and standard deviation of the maximum dynamic fluctuation component are calculated as follows:
[0094]
[0095] In the formula, is the mean value of the maximum dynamic fluctuation component of the nth voltage sequence, μ is the mean matrix of the maximum dynamic fluctuation component, is the standard deviation of the maximum dynamic fluctuation component of the nth voltage sequence, and σ is the standard deviation matrix of the maximum dynamic fluctuation component.
[0096] Step 1043: According to the mean matrix and the standard deviation matrix, select the outlier threshold to obtain the outlier. The formula is:
[0097] ο=(U>μ+W*σ)|(U<μ-W*σ)
[0098] Among them, o is the outlier, W is the outlier threshold, μ is the mean matrix, σ is the standard deviation matrix, and U is the maximum dynamic voltage fluctuation component matrix.
[0099] In this embodiment, a box-and-whisker outlier plot of the battery pack generation parameters is drawn, and outliers are drawn using an improved local outlier algorithm of the box-and-whisker plot. The threshold selection formula for the outliers is as follows:
[0100] ο=(U>μ+W*σ)|(U<μ-W*σ)
[0101] Where o is an outlier, and W is the defined outlier threshold, which is set according to the 3λ principle.
[0102] Step 1044: draw outliers and box plots, and based on the box plots, determine abnormal data to obtain abnormal battery cell detection results.
[0103] In this embodiment, the maximum voltage fluctuation component of the battery cell and the local outlier detection schematic diagram are as follows: Figure 3 As shown, draw a box plot outlier to find abnormal monomers:
[0104] R IQ =Q 3 -Q 1
[0105] B L =Q 1 -1.5×R IQ
[0106] B U =Q 3 +1.5×R IQ
[0107] In the formula, Q 3 With Q 1 are the upper and lower quartiles, respectively; R IQ is the interquartile range of the data; using R IQ Calculate the upper and lower whiskers B with the upper and lower quartiles L With B U Used to determine whether the data is abnormal.
[0108] Step 105: According to the selected first sliding window and the preset slope condition, the internal fault diagnosis of the battery is performed on each voltage dynamic fluctuation component to obtain the internal fault detection result of the battery cell.
[0109] In this embodiment, in order to detect the battery cell sequence of internal short circuit fault and open circuit fault, the thermal runaway data of the energy storage system is applied to the internal fault detection schematic diagram, such as Figure 4 As shown, the internal fault is an internal short circuit fault or an open circuit fault. After processing in step 103, the voltage dynamic fluctuation component is obtained, and then the absolute value of the characteristic difference is obtained through dimensionless processing of the kurtosis. Then, a suitable sliding window is selected, and the slopes of the highest peak and the lowest peak in the sliding window are calculated until the sliding ends. The number of times that potential internal short circuit faults and open circuit faults that meet the slope condition occur is counted, and 5% (the upper limit of the threshold for fault processing) of the total sliding window is selected as the threshold to detect the internal short circuit fault or open circuit fault single cell, and the battery single cell internal fault detection result is obtained, including the battery single cell sequence of the internal short circuit fault and the open circuit fault.
[0110] Optionally, step 105 specifically includes step 1051 to step 1056, and each step is specifically as follows:
[0111] Step 1051: perform dimensionless processing on each voltage dynamic fluctuation component to obtain the absolute value of each voltage mean difference.
[0112] In this embodiment, the absolute value of the characteristic difference (the absolute value of the voltage mean difference) is obtained through dimensionless processing of the kurtosis, and the formula is as follows:
[0113]
[0114] In the formula, is the voltage sequence v of the jth battery cell j The dimensionless eigenvalue of |v j | represents the amplitude of the signal, p(v j) represents the probability density distribution of the signal value, l,m>0 represents different Minkowski constants, and M represents the total number of cell voltage sequences. It should be noted that the voltage dynamic fluctuation component of a battery cell represents the voltage sequence value of the battery cell.
[0115] Step 1052: According to the absolute values of the mean differences of the voltages, the first sliding window is selected for sliding, the kurtosis factor is calculated, and the slope between the highest peak and the lowest peak in the current window is calculated to obtain the slope of the current window.
[0116] In this embodiment, the kurtosis factor is calculated as the ratio of the third-order center distance to the fourth power of the standard deviation:
[0117]
[0118] In the formula, KI y is the kurtosis of the difference y, y is the difference between the single cell voltage sequence value in a sliding window and the average value of all single cell voltage sequences in the window, that is, the absolute value of the characteristic difference, d is the standard deviation, and ε=3,η=2.
[0119] Calculate the slope between the highest peak and the lowest peak in the sliding window using the following formula:
[0120] K=[k 1 … k n ]
[0121] Where K is the calculated sliding window slope matrix, k n is the slope between the highest peak and the lowest peak in the nth sliding window.
[0122] Step 1053: If the slope of the current window satisfies the first slope condition, it is determined that the current battery cell is an internal short circuit fault; wherein the first slope condition is specifically:
[0123] fault1=k c <0|((k c ≤0)&(0≤k n <1))
[0124] Among them, k c is the slope of the current window, k n is the slope of the next window, and fault1 is an internal short circuit fault.
[0125] In this embodiment, the preset slope condition includes a first slope condition and a second slope condition. The internal short circuit fault fault1 is determined according to the slope k and the first slope condition, and the open circuit fault fault2 is determined according to the slope k and the second slope condition.
[0126] Step 1054: If the slope of the current window satisfies the second slope condition, it is determined that the current battery cell is an open circuit fault; wherein the second slope condition is specifically:
[0127] fault2=(0≤k c <1)|((0≤k c <1)&k n <0)
[0128] Among them, fault2 is an open circuit fault.
[0129] Step 1055: When the sliding of the selected first sliding window ends, the number of short circuit faults and open circuit faults that meet the preset slope condition is counted, the total sliding window of the upper limit of the threshold for fault processing is selected, the slope of the window is calculated based on the total sliding window, and the internal short circuit fault and open circuit fault of the corresponding battery cell are determined according to the preset slope condition and the slope of the window.
[0130] In this embodiment, after the processing of step 103, the voltage dynamic fluctuation component is obtained, and then the absolute value of the characteristic difference is obtained through dimensionless processing of the kurtosis, and then a suitable sliding window (first sliding window) is selected, and the slopes of the highest peak and the lowest peak in the sliding window are calculated until the sliding ends, and the number of times that potential internal short circuit faults and open circuit faults that meet the slope condition occur is counted, and 5% (the upper limit of the threshold for fault processing) of the total sliding window is selected as the threshold, and the internal short circuit fault or open circuit fault single cell is detected, and the battery single cell internal fault detection result is obtained by statistics, including the battery single cell sequence of internal short circuit fault and open circuit fault.
[0131] Step 1056: Count the battery cells that are determined to have internal short-circuit faults and open-circuit faults, and obtain the battery cell internal fault detection results.
[0132] In this embodiment, the slope generated by the sliding window is used to diagnose the short circuit fault / open circuit fault in the battery:
[0133] Internal short circuit fault: fault1=k c <0|((k c ≤0)&(0≤k n <1))
[0134] Open circuit fault: fault2=(0≤k c <1)|((0≤k c <1)&k n <0)
[0135] Whether a fault occurs is determined by comparing the number of potential faults that meet the slope with the threshold. If the number of internal short circuit faults is greater than the number of open circuit faults, the monomer sequence of the number of internal short circuit faults is counted. If the two are equal, the number of internal short circuit faults and open circuit faults are counted.
[0136] In implementing the embodiment of the present invention, the maximum voltage fluctuation component of each monomer is calculated by successive variational mode decomposition (SVMD decomposition), and abnormal monomers are detected by box plots and outliers. Secondly, the absolute value of the voltage mean difference is obtained by successive variational mode decomposition, and the slope formed by the highest peak and the lowest peak of the current window and the next window is calculated and counted to detect potential internal short circuit / open circuit faults. Finally, the fault detection method proposed in the present invention is verified by the operating data of thermal runaway of the single battery of the energy storage system, and an early warning can be given 55 minutes before the thermal runaway occurs. The results show that the method has reliability and engineering application value.
[0137] By using different types of energy storage system data for verification, and verifying on data of different energy storage system models and working conditions, through parameter input or model adjustment, after simulation, the internal short circuit / open circuit fault detection is relatively high. Using different working conditions, energy storage system models and power battery models, the accuracy and effectiveness of the battery single cell fault detection method of the present invention are verified, and it can be well transferred to engineering applications.
[0138] By implementing the embodiment of the present invention, data of thermal runaway of each battery cell in the energy storage system under operating conditions are obtained to obtain full life cycle data; the full life cycle data is preprocessed for cleaning state to obtain a plurality of charge and discharge segment data; based on a pre-constructed successive variational modal model, each charge and discharge segment data is subjected to successive variational modal decomposition by adjusting model parameters to obtain the maximum voltage dynamic fluctuation component and voltage dynamic fluctuation component corresponding to each charge and discharge segment data; based on the selected outlier threshold, each maximum voltage dynamic fluctuation component is subjected to box plot local outlier detection to obtain abnormal battery cell detection results; according to the selected first sliding window and the preset slope condition, each voltage dynamic fluctuation component is subjected to battery internal fault diagnosis to obtain battery cell internal fault detection results. By performing state cleaning preprocessing and successive variational mode decomposition on the full life cycle data under operating conditions, and using box plot local outlier detection and battery internal fault diagnosis, the abnormal faults and internal faults of the battery cells of the energy storage system can be effectively diagnosed. This method can be applied to various operating conditions and can quickly and accurately diagnose battery cells of different models. It is verified through the data of thermal runaway of the energy storage system, which further effectively illustrates the effectiveness of battery cell fault detection and improves the speed, accuracy and universality of battery cell fault detection.
[0139] Compared with the prior art, the present invention has the advantages of:
[0140] (1) By adjusting the model parameters of the pre-built successive variational modal model, the abnormal battery cell detection results are obtained, which solves the difficulties of variational modal decomposition (VMD) such as slow convergence speed and high calculation time, reduces the uncertainty of the number of modes and parameters that need to be set in advance, and can accurately locate the location of the abnormal cell, which can play a certain early warning role for the inconsistency of the cell.
[0141] (2) The generated slope of the sliding window is used to realize the internal fault detection of the single cell. For the internal short circuit fault and open circuit fault, the slope is calculated and statistical methods are used. The potential internal short circuit fault and open circuit fault are detected by the number of times the slope of the current sliding window and the next sliding window changes. The warning time can be advanced by up to 55 minutes, providing enterprises and users with sufficient time to discover and deal with the faulty single cells in the energy storage system before thermal runaway occurs, saving lives and property while improving the stability of the electrochemical energy storage system.
[0142] (3) The battery cell fault detection method is realized by using successive variational mode decomposition, local outlier detection of box plots and battery internal fault diagnosis. It can effectively diagnose the battery cell faults of energy storage systems. It has a wide range of applications. While reducing the amount of calculation, it has the advantages of speed, accuracy and universality, and can be better applied to engineering practice.
[0143] Embodiment 2
[0144] Accordingly, see Figure 5 , Figure 5 1 is a schematic diagram of the structure of a second embodiment of a battery cell fault detection device for an energy storage system provided by the present invention. Figure 5 As shown, a data acquisition module 501, a data preprocessing module 502, a modal decomposition module 503, an abnormal battery cell detection module 504 and a battery internal fault diagnosis module 505;
[0145] The data acquisition module 501 is used to acquire the data of thermal runaway of each battery cell in the energy storage system under the operating conditions to obtain the full life cycle data;
[0146] The data preprocessing module 502 is used to perform cleaning state preprocessing on the full life cycle data to obtain a number of charge and discharge segment data;
[0147] The modal decomposition module 503 is used to perform a successive variation modal decomposition on each charging and discharging segment data based on a pre-built successive variation modal model by adjusting the model parameters, so as to obtain the maximum voltage dynamic fluctuation component and the voltage dynamic fluctuation component corresponding to each charging and discharging segment data;
[0148] The abnormal battery cell detection module 504 is used to perform local outlier detection of the box plot on each maximum voltage dynamic fluctuation component based on the selected outlier threshold value to obtain an abnormal battery cell detection result;
[0149] The battery internal fault diagnosis module 505 is used to perform battery internal fault diagnosis on each voltage dynamic fluctuation component according to the selected first sliding window and the preset slope condition, and obtain the battery cell internal fault detection result.
[0150] By implementing the embodiment of the present invention, data of thermal runaway of each battery cell in the energy storage system under operating conditions are obtained to obtain full life cycle data; the full life cycle data is preprocessed for cleaning state to obtain a plurality of charge and discharge segment data; based on a pre-constructed successive variational modal model, each charge and discharge segment data is subjected to successive variational modal decomposition by adjusting model parameters to obtain the maximum voltage dynamic fluctuation component and voltage dynamic fluctuation component corresponding to each charge and discharge segment data; based on the selected outlier threshold, each maximum voltage dynamic fluctuation component is subjected to box plot local outlier detection to obtain abnormal battery cell detection results; according to the selected first sliding window and the preset slope condition, each voltage dynamic fluctuation component is subjected to battery internal fault diagnosis to obtain battery cell internal fault detection results. By performing state cleaning preprocessing and successive variational mode decomposition on the full life cycle data under operating conditions, and using box plot local outlier detection and battery internal fault diagnosis, the abnormal faults and internal faults of the battery cells of the energy storage system can be effectively diagnosed. This method can be applied to various operating conditions and can quickly and accurately diagnose battery cells of different models. It is verified through the data of thermal runaway of the energy storage system, which further effectively illustrates the effectiveness of battery cell fault detection and improves the speed, accuracy and universality of battery cell fault detection.
[0151] The above-mentioned battery cell fault detection device for an energy storage system can implement a battery cell fault detection method for an energy storage system of the above-mentioned method embodiment. The options in the above-mentioned method embodiment are also applicable to this embodiment and will not be described in detail here. The rest of the contents of the embodiment of this application can refer to the contents of the above-mentioned method embodiment and will not be described in detail in this embodiment.
[0152] In addition, an embodiment of the present application further provides a computer device, which includes a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the steps in any of the above method embodiments are implemented.
[0153] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.
[0154] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for detecting battery cell failure in an energy storage system, characterized in that: include: Obtain the thermal runaway data of each battery cell in the energy storage system under operating conditions and obtain the full life cycle data; The whole life cycle data is pre-processed in a cleaning state to obtain a plurality of charge and discharge segment data; Based on the pre-built successive variational modal model, by adjusting the model parameters, the data of each charging and discharging segment are subjected to successive variational modal decomposition to obtain the maximum voltage dynamic fluctuation component and the voltage dynamic fluctuation component corresponding to each charging and discharging segment data; Based on the selected outlier threshold, each of the maximum voltage dynamic fluctuation components is subjected to a box plot local outlier detection to obtain an abnormal battery cell detection result; According to the selected first sliding window and the preset slope condition, the battery internal fault diagnosis is performed on each of the voltage dynamic fluctuation components to obtain the battery monomer internal fault detection result; wherein, according to the selected first sliding window and the preset slope condition, the battery internal fault diagnosis is performed on each of the voltage dynamic fluctuation components to obtain the battery monomer internal fault detection result, specifically: The dynamic fluctuation components of the voltages are dimensionally processed to obtain the absolute values of the mean differences of the voltages; wherein the dimensionless processing is performed using the following formula: in, is the voltage sequence v of the jth battery cell j The dimensionless eigenvalue of |v j | represents the amplitude of the signal, p(v j ) represents the probability density distribution of the signal value, l,m>0 represents different Minkowski constants, and M represents the total number of monomer voltage sequences; According to the absolute value of the mean difference of each voltage, sliding is performed based on the selected first sliding window, the kurtosis factor is calculated, and the slope between the highest peak and the lowest peak in the current window is calculated to obtain the slope of the current window; wherein the kurtosis factor is calculated by the following formula: Among them, KI y is the kurtosis of the difference y, y is the difference between the single-cell voltage sequence value in a sliding window and the average value of all single-cell voltage sequences in the window, that is, the absolute value of the characteristic difference, d is the standard deviation, and ε=3,η=2; If the slope of the current window satisfies the first slope condition, it is determined that the current battery cell has an internal short circuit fault; wherein the first slope condition is specifically: fault1=k c <0|((k c ≤0)&(0≤k n <1)) Among them, k c is the slope of the current window, k n is the slope of the next window, fault1 is the internal short circuit fault; If the slope of the current window satisfies the second slope condition, it is determined that the current battery cell is an open circuit fault; wherein the second slope condition is specifically: fault2=(0≤k c <1)|((0≤k c <1)&k n <0) Wherein, fault2 is the open circuit fault; The battery cells that are determined to correspond to the internal short-circuit fault and the open-circuit fault are counted to obtain the internal fault detection result of the battery cells.
2. The method for detecting battery cell failure of an energy storage system according to claim 1, characterized in that: The whole life cycle data is pre-processed for cleaning status to obtain a number of charge and discharge segment data, specifically: Using a preset data cleaning method to process null values and abnormal values in the full life cycle data to obtain cleaned full life cycle data; Based on the positive and negative current, the monotonically increasing characteristics of the state of charge and the data point continuity condition, the full life cycle data after cleaning is divided into charge and discharge states to obtain the data of each charge and discharge segment; wherein, the data point continuity condition is that the number of continuous data points is not less than a preset number value.
3. The method for detecting battery cell failure of an energy storage system according to claim 1, characterized in that: The pre-built successive variational modal model is specifically: The constrained optimization problem of the successive empirical mode structure is converted into an unconstrained optimization problem by introducing a quadratic penalty term and a Lagrange multiplier term, and the pre-constructed successive variational modal model is obtained, specifically: Among them, u k is the single-component AM / FM signal obtained by the kth decomposition, ω k is the center frequency of the single-component AM / FM signal obtained by the kth decomposition, ω is the frequency parameter, ξ is the balance parameter, is the Wiener filtering result of the current residual of the saddle point after the modal component Fourier transform, Find the frequency domain value of the modal component time series after the saddle point and Fourier transformation, Find the frequency domain value of the original input signal after the saddle point and the first Fourier transform, Find the frequency domain value of the original input signal after the saddle point and the kth Fourier transform, is the frequency response of the kth filter, γ is the Lagrange multiplication operator, k is the total number of modes, ∞ is the value added by iterative optimization, is the frequency response of the ith filter, is the frequency response of the original signal after Fourier transform, is the frequency response of the Lagrange multiplication operator.
4. The method for detecting battery cell failure of an energy storage system according to claim 1, characterized in that: The adjustment model parameters are specifically: The input in the pre-built successive variational modal model is set to the single cell voltage sequence of the battery, the original data compactness index is set to a preset index value, the Gaussian noise is set to 0, the standard tolerance of convergence is set to a preset tolerance value, and the stop type is set to 1.
5. The method for detecting battery cell failure of an energy storage system according to claim 1, characterized in that: Based on the selected outlier threshold, each of the maximum voltage dynamic fluctuation components is subjected to a box plot local outlier detection to obtain an abnormal battery cell detection result, specifically: According to each of the maximum voltage dynamic fluctuation components, a maximum dynamic voltage fluctuation component matrix is obtained; Calculate a mean matrix and a standard deviation matrix based on the selected second sliding window and the maximum dynamic voltage fluctuation component matrix; According to the mean matrix and the standard deviation matrix, the outlier threshold is selected to obtain the outlier. The formula is: ο=(U>μ+W*σ)|(U<μ-W*σ) Wherein, o is the outlier, W is the outlier threshold, μ is the mean matrix, σ is the standard deviation matrix, and U is the maximum dynamic voltage fluctuation component matrix; The outliers and box plots are drawn, and based on the box plots, abnormal data is determined to obtain the abnormal battery cell detection result.
6. The method for detecting battery cell failure of an energy storage system according to claim 1, characterized in that: Before counting the battery cells determined to have the internal short circuit fault and the open circuit fault, and obtaining the battery cell internal fault detection result, the method further includes: When the sliding of the selected first sliding window ends, the number of short-circuit faults and open-circuit faults that meet the preset slope condition is counted, a sliding total window of an upper threshold value of fault processing is selected, the slope of the window is calculated based on the sliding total window, and the internal short-circuit fault and the open-circuit fault of the corresponding battery cell are determined according to the preset slope condition and the slope of the window.
7. A battery cell fault detection device for an energy storage system, characterized in that: include: Data acquisition module, data preprocessing module, modal decomposition module, abnormal battery cell detection module and battery internal fault diagnosis module; Wherein, the data acquisition module is used to acquire the data of thermal runaway of each battery cell in the energy storage system under the operating conditions to obtain the full life cycle data; The data preprocessing module is used to perform cleaning state preprocessing on the full life cycle data to obtain a plurality of charge and discharge segment data; The modal decomposition module is used to perform a successive variation modal decomposition on each of the charging and discharging segment data based on a pre-built successive variation modal model by adjusting model parameters to obtain a maximum voltage dynamic fluctuation component and a voltage dynamic fluctuation component corresponding to each of the charging and discharging segment data; The abnormal battery cell detection module is used to perform box plot local outlier detection on each of the maximum voltage dynamic fluctuation components based on the selected outlier threshold to obtain an abnormal battery cell detection result; The battery internal fault diagnosis module is used to perform battery internal fault diagnosis on each of the voltage dynamic fluctuation components according to the selected first sliding window and the preset slope condition, and obtain the battery monomer internal fault detection result; wherein, the battery internal fault diagnosis is performed on each of the voltage dynamic fluctuation components according to the selected first sliding window and the preset slope condition, and the battery monomer internal fault detection result is obtained, specifically: The dynamic fluctuation components of the voltages are dimensionally processed to obtain the absolute values of the mean differences of the voltages; wherein the dimensionless processing is performed using the following formula: in, is the voltage sequence v of the jth battery cell j The dimensionless eigenvalue of |v j | represents the amplitude of the signal, p(v j ) represents the probability density distribution of the signal value, l,m>0 represents different Minkowski constants, and M represents the total number of monomer voltage sequences; According to the absolute value of the mean difference of each voltage, sliding is performed based on the selected first sliding window, the kurtosis factor is calculated, and the slope between the highest peak and the lowest peak in the current window is calculated to obtain the slope of the current window; wherein the kurtosis factor is calculated by the following formula: Among them, KI y is the kurtosis of the difference y, y is the difference between the single-cell voltage sequence value in a sliding window and the average value of all single-cell voltage sequences in the window, that is, the absolute value of the characteristic difference, d is the standard deviation, and ε=3,η=2; If the slope of the current window satisfies the first slope condition, it is determined that the current battery cell has an internal short circuit fault; wherein the first slope condition is specifically: fault1=k c <0|((k c ≤0)&(0≤k n <1)) Among them, k c is the slope of the current window, k n is the slope of the next window, fault1 is the internal short circuit fault; If the slope of the current window satisfies the second slope condition, it is determined that the current battery cell is an open circuit fault; wherein the second slope condition is specifically: fault2=(0≤k c <1)|((0≤k c <1)&k n <0) Wherein, fault2 is the open circuit fault; The battery cells that are determined to correspond to the internal short-circuit fault and the open-circuit fault are counted to obtain the internal fault detection result of the battery cells.
8. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the method for detecting battery cell failure of an energy storage system according to any one of claims 1 to 6 is implemented.
9. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, implements the battery cell fault detection method for the energy storage system according to any one of claims 1 to 6.
