Double-domain robust battery fault diagnosis method and device based on returned battery energy storage system inconsistency

Abstract

The invention discloses a dual-domain robust battery fault diagnosis method and device based on the inconsistency of a returned battery energy storage system. The method comprises the following steps: determining a test battery pack according to a returned battery in the returned battery energy storage system; performing validity screening and time correction on the initial electrical data of the tested battery pack to obtain corrected electrical data; determining a global correlation index and a consistency index of each test battery pack according to a rank pair corresponding to the corrected electrical data; and determining the fault state of each test battery pack and the returned battery energy storage system based on the indexes, a preset correlation threshold and a consistency threshold. According to the invention, the fault of the returned battery system can be diagnosed steadily, quickly and comprehensively, and data support is provided for in-service safe operation and decommissioning echelon utilization of the battery.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis in the operation of a retired lithium-ion battery energy storage system, and in particular to a dual-domain robust battery fault diagnosis method and device based on inconsistency of a retired battery energy storage system. BACKGROUND

[0002] As the core unit of an energy storage system, a lithium-ion battery pack directly affects the operational safety and economy of an energy storage power station. However, in actual operation, there are often consistency differences within the battery pack, and the voltage signal is disturbed by noise under long-term static or trickle conditions, which can cause misdiagnosis of traditional methods based on voltage-current correlation. At the same time, a single consistency analysis method is not sensitive enough to early degradation or trend abnormalities, and it is difficult to discover potential risks in a timely manner. These problems seriously restrict the safe use of retired batteries and the health management of in-service battery packs. SUMMARY

[0003] In view of this, the purpose of the present application is to provide a dual-domain robust battery fault diagnosis method and device based on inconsistency of a retired battery energy storage system, thereby achieving robust, rapid and comprehensive diagnosis of the fault of the retired battery system, and providing data support for the safe operation of the battery in service and the stepped utilization of the retired battery.

[0004] In a first aspect, the embodiments of the present application provide a dual-domain robust battery fault diagnosis method based on inconsistency of a retired battery energy storage system. Further, the method comprises: S102: determining a test battery pack according to a retired battery in the retired battery energy storage system; S104: performing validity screening and time correction on initial electrical data of the test battery pack to obtain corrected electrical data; S106: determining a global correlation index of each test battery pack based on a Kendall rank correlation algorithm and according to a rank pair corresponding to the corrected electrical data; S108: determining a consistency index of each test battery pack according to the corrected electrical data; and S110: determining a fault state of each test battery pack and the retired battery energy storage system based on the global correlation index, the consistency index, a preset correlation threshold and a preset consistency threshold.

[0005] Further, S102 comprises: the total number of retired batteries provided in the retired battery energy storage system is X; the total number of test battery packs is R, R >> X; and each test battery pack comprises one retired battery or a plurality of retired batteries connected in series.

[0006] Further, S104 is performed for each test battery pack; S104 includes: S104-2: the initial electrical data of the test battery pack is filtered for validity to obtain K valid segments T; S104-4: the voltage and current in the initial electrical data of the K valid segments T are respectively subjected to outlier removal, noise suppression, baseline drift correction and time synchronization alignment to obtain corrected electrical data, wherein the corrected electrical data includes: corrected voltage and corrected current .

[0007] Furthermore, S104-2 includes: S104-2-2: Preset maximum window length L max Minimum window length L min Maximum window step size S max and minimum window step size S min S104-2-4: Determine the local signal-to-noise ratio (SNR) based on the initial electrical data of the test battery pack and the pre-acquired reference noise variance of the returned batteries; S104-2-6: Based on the local signal-to-noise ratio (SNR) and the maximum window length... L max Minimum window length L min Maximum window step size S max and minimum window step size S min Determine the window length L and window step size S of the test battery pack; S104-2-8: Determine the segment of interest T' based on the window length L and window step size S, wherein the number of segments of interest T' for each test battery pack is K'; S104-2-10: Eliminate segments of interest T' that do not meet the preset current threshold range to obtain valid segments T, wherein the number of valid segments T for each test battery pack is K.

[0008] Further, S104-4 includes: S104-4-2: the voltage in the effective segment T and current Outlier removal was performed separately to obtain the removal voltage. and removal current S104-4-4: Based on the moving average filtering method, the rejection voltages are respectively... and the rejection current High-frequency noise suppression is performed to obtain the filter voltage. and filter current S104-4-6: The filtered voltages are respectively... and filtered current baseline drift processing to obtain baseline voltage and baseline current ; S104-4-8: determining an optimal alignment time according to the cross-correlation function between the baseline voltage and baseline current ; S104-4-9: time aligning the baseline voltage and baseline current according to the optimal alignment time, to obtain corrected voltage and corrected current .

[0009] Further, the number of sampling points within each of the effective segments T is n, and for each effective segment T, S106 is performed; S106 includes: S106-2: constructing a voltage change amount sequence based on the corrected voltage ; S106-4: constructing a current sequence based on the corrected current ; S106-6: determining a plurality of Kendall rank pairs according to the voltage change amount sequence and the current sequence; S106-8: classifying the Kendall rank pairs, and determining a local correlation index based on the classification result, wherein each group of test battery groups includes K effective segments T, and each effective segment T corresponds to one local correlation index; S106-10: performing weighted aggregation on the local correlation indexes to obtain a global correlation index , and each effective segment T corresponds to one global correlation index .

[0010] Further, S106-8 includes: calculating the local correlation index corresponding to any one effective segment T based on the following formula: ; wherein n is the number of sampling points within the effective segment T, and C and D are the classification results of the Kendall rank pairs.

[0011] Further, for each effective segment T, S108 is performed; S108 includes: S108-2: determining a consistency parameter of each effective segment T in each group of test battery groups according to the corrected voltage and corrected current , respectively, and the consistency parameter includes: equivalent internal resistance, polarization voltage amplitude, voltage change slope, model fitting residual, and steady-state variance; S108-4: determining a consistency feature vector according to the consistency parameter; S108-6: determining a consistency index D of each effective segment T in each of the test battery groups according to the consistency feature vector. ​​

[0012] Further, S110 comprises: S110-2: determining whether the failure state of the test battery group is a correlation failure according to the global correlation indicator of the test battery group and a preset correlation threshold; S110-4: determining whether the failure state of the test battery group is a consistency failure according to the consistency indicator of the test battery group and a preset consistency threshold; and S110-6: determining the overall failure state of the returned battery energy storage system according to the failure state of each of the test battery groups.

[0013] In a second aspect, an embodiment of the present application provides a dual-domain robust battery fault diagnosis device based on inconsistency of a returned battery energy storage system, comprising: a first diagnosis module configured to determine a test battery group according to returned batteries in the returned battery energy storage system; a second diagnosis module configured to perform validity screening and time correction on initial electrical data of the test battery group to obtain corrected electrical data; a third diagnosis module configured to determine a global correlation indicator of each of the test battery groups based on Kendall rank correlation algorithm according to a rank pair corresponding to the corrected electrical data; a fourth diagnosis module configured to determine a consistency indicator of each of the test battery groups according to the corrected electrical data; and a fifth diagnosis module configured to determine a failure state of each of the test battery groups and the returned battery energy storage system based on the global correlation indicator, the consistency indicator, a preset correlation threshold and a preset consistency threshold.

[0014] The beneficial effects of the embodiments of the present application are as follows: The present application discloses a dual-domain robust battery fault diagnosis method and device based on inconsistency of a returned battery energy storage system, comprising determining a test battery group according to returned batteries in the returned battery energy storage system; performing validity screening and time correction on initial electrical data of the test battery group to obtain corrected electrical data; determining a global correlation indicator and a consistency indicator of each of the test battery groups according to a rank pair corresponding to the corrected electrical data; and determining a failure state of each of the test battery groups and the returned battery energy storage system based on the above indicators, a preset correlation threshold and a preset consistency threshold. The present application can realize robust, fast and comprehensive diagnosis of the returned battery system failure, and provide data support for in-service safety operation and retired echelon utilization of batteries.

[0015] Other features and advantages of the present application will be illustrated in the following description, or can be known or determined without doubt from the description, or can be known by implementing the above-mentioned technologies of the present application.

[0016] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the following preferred embodiments are specifically described in detail below, and the accompanying drawings are as follows. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the specific embodiments or prior art of the present application, the drawings needed in the description of the specific embodiments or prior art will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0018] Figure 1 The flow chart of the first dual-domain robust battery fault diagnosis method based on the inconsistency of the returned battery energy storage system provided by the present application; Figure 2 The flow chart of the second dual-domain robust battery fault diagnosis method based on the inconsistency of the returned battery energy storage system provided by the present application; Figure 3 The flow chart of the third dual-domain robust battery fault diagnosis method based on the inconsistency of the returned battery energy storage system provided by the present application; Figure 4 The flow chart of the fourth dual-domain robust battery fault diagnosis method based on the inconsistency of the returned battery energy storage system provided by the present application; Figure 5 The flow chart of the fifth dual-domain robust battery fault diagnosis method based on the inconsistency of the returned battery energy storage system provided by the present application; Figure 6 The schematic diagram of the dual-domain robust battery fault diagnosis device based on the inconsistency of the returned battery energy storage system provided by the present application. DETAILED DESCRIPTION

[0019] The technical solutions of the present application will be described below in conjunction with the embodiments, obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] The present application relates to the fault diagnosis technology in the operation of the retired lithium ion battery energy storage system, in particular to the fault early warning and diagnosis method of the energy storage system under different discharge rates, charge and discharge depths, inconsistency and other conditions. The present application is applied in the scene of battery state diagnosis.

[0021] At present, the fault diagnosis of battery system has made certain progress in academic research and engineering application. The traditional battery management system usually sets threshold values of monitoring quantities such as voltage, current and temperature, to realize the alarm and protection of overvoltage, overcurrent, overtemperature and other explicit faults. However, this kind of method is too dependent on single criterion, and it is difficult to reveal the potential risks of battery operation under complex working conditions, which has the following shortcomings: Firstly, the diagnostic criteria are simplified and the fault coverage is incomplete. Existing methods are mostly based on static thresholds or empirical rules, which can mainly handle significant faults such as over-limit faults. However, for early degradation, slight internal short circuit, polarization anomaly and other implicit faults, there is often a lack of sensitive and reliable diagnostic mechanisms. This makes it difficult to detect potential hazards in the early stages, increasing the risk of serious accidents such as thermal runaway. Secondly, the electrochemical mechanism is complex and the operating conditions are variable. Lithium-ion batteries have strong non-linear characteristics during charging and discharging, and are influenced by electrochemical dynamics, diffusion effects, polarization characteristics and other factors. The response not only depends on the operating current, but also is affected by temperature fluctuations, aging levels and battery inconsistencies. Traditional methods are difficult to fully consider these complex mechanisms at the model level, resulting in insufficient diagnostic accuracy and generalization ability. Thirdly, measurement errors and environmental noise have a significant impact. The sampling system within the BMS inevitably has measurement bias and noise interference, especially when the battery pack is stationary or in a trickle state for a long time. Voltage signals are easily dominated by random noise, leading to misjudgment in correlation-based diagnostic methods. Existing technologies lack specialized handling of such conditions, reducing the robustness of the diagnosis. Finally, experimental verification conditions are limited and method comparisons are lacking. Existing research is mostly conducted under ideal conditions of normal temperature and new batteries, lacking systematic verification covering a wide temperature range, multiple aging levels and different consistency levels. At the same time, there is a lack of uniform quantitative comparison between methods, making the applicability of their diagnostic performance in complex environments unclear.

[0022] In summary, existing technologies have deficiencies in fault type coverage, complex environment adaptability, noise resistance and system verification, making it difficult to meet the needs of safe and stable operation of battery systems under various conditions.

[0023] To address the above problems, the present application proposes a dual-domain robust battery fault diagnosis method, which combines correlation domain analysis and consistency domain analysis. The correlation domain is analyzed through current gating and window filtering, only in the current excited segment, using Kendall rank correlation coefficient to measure the monotonicity of voltage-current relationship, which can avoid the interference of standing noise and sensitively capture the abnormality of battery operation trend. The consistency domain extracts multi-dimensional features such as internal resistance, polarization, slope and model residual in the excited segment, constructs a feature vector and uses Mahalanobis distance to quantify the deviation from the group distribution, revealing the consistency anomaly between batteries. Combining the correlation anomaly and consistency anomaly results, a comprehensive judgment mechanism is formed to monitor the trend deviation and group deviation simultaneously, improving the comprehensiveness and accuracy of the diagnosis.

[0024] The specific implementation steps are described in detail as follows: Example 1 As Figure 1 , Figure 2 , Figure 3As shown, it is a flow chart of a two-domain robust battery fault diagnosis method based on inconsistency of returned battery energy storage system, which comprises the following steps: S102: determining a test battery group according to the returned batteries in the returned battery energy storage system.

[0025] S102 comprises that the total number of returned batteries arranged in the returned battery energy storage system is X, the total number of test battery groups is R, and R>>X. Each of the test battery groups comprises one returned battery or a plurality of returned batteries connected in series or in parallel.

[0026] Specifically, the mathematical relationship between the maximum value of R and X is .

[0027] In the specific implementation process, the selection of the test battery group can be determined according to actual needs.

[0028] S104: performing validity screening and time correction on the initial electrical data of the test battery group to obtain corrected electrical data (such as Figure 4 as shown).

[0029] Specifically, the returned battery energy storage system comprises R test battery groups, r represents the rth test battery group, each test battery group comprises K valid segments T, k represents the kth valid segment T, S104 is performed on each test battery group, therefore, R×K sets of corrected electrical data, i.e. corrected voltage and corrected current , can be obtained through S104.

[0030] S104 comprises: S104-2: performing validity screening on the initial electrical data of the test battery group to obtain K valid segments T.

[0031] Firstly, the initial electrical data is collected, as shown in Table 1, and then for each test battery group, K valid segments T are determined.

[0032] Table 1

[0033] The operation (charging or discharging) data of the returned battery of the energy storage power station in Table 1 provides basic data support for subsequent fault diagnosis. The terminal voltage (of of Table 1) and the total voltage (of Table 1 ) are used to reflect the electrochemical state and group consistency of the battery. Table 1 also collects the current, which is used to characterize the excitation degree of the battery during the charging and discharging process, and is the reference quantity for subsequent current gating and correlation analysis. Table 1 records the sampling time to ensure the synchronization between multiple physical quantities, which facilitates subsequent sliding window division and time alignment.

[0034] Specifically, based on the dynamic adjustment of signal-to-noise ratio, the current gating and window screening are used to segment the collected voltage and current signals, and by setting the current amplitude threshold or the current change rate threshold, the excited time segment with sufficient information amount is screened out, which is called effective segment T. The window in the static or trickle state is removed to avoid the interference of random noise on the subsequent analysis results.

[0035] S104-2 includes: S104-2-2: preset window length maximum value L max , window length minimum value L min , window step maximum value S max , and window step minimum value S min .

[0036] S104-2-4: determining the local signal-to-noise ratio SNR according to the initial electrical data of the test battery pack and the reference noise variance of the returned battery obtained in advance.

[0037] S104-2-6: determining the window length L and the window step S of the test battery pack according to the local signal-to-noise ratio SNR, the window length maximum value L max , window length minimum value L min , window step maximum value S max , and window step minimum value S min .

[0038] S104-2-2, S104-2-4, S104-2-6 include: 1) calculating the local signal-to-noise ratio : Formula 1; wherein, is the noise of the test battery pack in the static state (not charging, not discharging, non-operating state) obtained in advance, is the short-time variance of the voltage signal obtained according to Table 1.

[0039] 2) calculating the window length L and the window step S.

[0040] In the present application, the sliding window length L and the step S are dynamically adjusted based on the signal-to-noise ratio (Signal-to-Noise Ratio, SNR), and the following is the adjustment measure: Formula 2; Equation 3 In a preset local signal-to-noise ratio SNR range, wherein , the window length L and the step size S decrease linearly with the increase of SNR , and take the minimum value when SNR SNR max L min and S min , and take the maximum value when SNR min L max and S max。

[0041] S104-2-8: Determine the concerned segment T' according to the window length L and the window step size S, wherein the number of the concerned segment T' of each test battery group is K'.

[0042] S104-2-10: Remove the concerned segment T' that does not meet the preset current threshold range to obtain the effective segment T, wherein the number of the effective segment T of each test battery group is K.

[0043] S104-2-8 and S104-2-10 include: 1) Determine the concerned segment T' according to the window length L and the window step size S (according to Table 1, L=2 and S=1).

[0044] 2) Calculate the current amplitude mean and the current standard deviation in each time window T', if the current amplitude mean is greater than the preset first threshold value, or the current standard deviation is greater than the preset second threshold value, then the T' is an effective window T, otherwise it is an invalid window, remove the T', and finally obtain K effective segments T.

[0045] Beneficial effects: Through the above screening process, T in the static or trickle state is removed, and only the window with sufficient current excitation and effective dynamic information in the signal is processed subsequently, thereby avoiding the misjudgment caused by the noise-dominant voltage signal in the long static or trickle state.

[0046] Each group of test battery groups corresponds to one , the of different test battery groups are different, and thus the L, S, and the number K of effective segments T of each group of test battery groups r are also different.

[0047] ​​​S104-4: performing outlier rejection, noise suppression, baseline drift correction and time synchronization alignment on the voltage and current in the initial electrical data in the K effective segments T respectively, to obtain modified electrical data, wherein the modified electrical data comprises: modified voltage and modified current .

[0048] Specifically, S104-4 includes: performing data preprocessing on the voltage signal in the effective segment T1...Tk, eliminating transient glitches by Hampel filtering, suppressing high-frequency noise based on a sliding average filtering method, eliminating baseline drift by using a differential correction method, and performing time alignment on the voltage and current signals based on a cross-correlation method, to obtain modified voltage and modified current , so as to improve the accuracy and stability of data analysis.

[0049] S104-4 includes: S104-4-2: performing outlier rejection on the voltage and current in the effective segment T respectively, to obtain rejected voltage and rejected current .

[0050] S104-4-2 is also called outlier rejection, that is, performing outlier detection on the voltage signal of T1...Tk, and eliminating transient glitch points and sampling outliers by Hampel filtering, so as to ensure the continuity and stability of the signal.

[0051] Specifically, in Table 1, L=2, S=1, the number of voltage data corresponding to each T is n (i.e. the number of sampling points), n corresponds to L, generally, n can be an integer multiple of L.

[0052] In a sliding window with a length of L , the median of the voltage in the window T is calculated M , and the absolute deviation of the median is set to MAD , and the rejection coefficient a is set, when the following formula is satisfied, it is determined that the point is an outlier, and the median is used to replace the current , to obtain , otherwise, is taken as for the subsequent steps.

[0053] Formula 4; The current is processed in the same way to obtain the rejected current .

[0054] S104-4-4: Based on the moving average filtering method, the filtering voltage and the filtering current are obtained by suppressing high-frequency noise. .

[0055] SS104-4-4 is also called noise suppression: for the voltage signal after outlier rejection , the voltage signal in the effective segment is smoothed based on the moving average filtering method to suppress high-frequency noise while maintaining the overall trend of the signal without distortion. Specifically, a sliding window with a length of L is moved point by point along the time sequence, and at each time point t , the voltage values of the time point and the surrounding several (j) sampling points are taken to perform arithmetic average to obtain the filtering voltage signal .

[0056] Equation 5; wherein, V1 ( t - j ) is the voltage value at time t - j .

[0057] The output filtering signal is smoother than the original signal, which can effectively weaken high-frequency random noise while maintaining the voltage change trend without distortion.

[0058] Based on the same method, the filtering current is obtained.

[0059] S104-4-6: The filtering voltage and the filtering current are respectively subjected to baseline drift processing to obtain the baseline voltage and the baseline current .

[0060] SS104-4-6 is also called baseline offset correction, that is, for the voltage signal after noise suppression , the voltage signal is subjected to baseline correction using the difference correction method to eliminate low-frequency drift caused by temperature drift, sensor offset or long-term cumulative error, so that the voltage change can truly reflect the dynamic response of the battery.

[0061] Under the sampling period Δ t , the voltage difference of adjacent sampling points is calculated: Equation 6;​ wherein represents the voltage variation of time t relative to the previous sampling point.

[0062] When the difference value is less than the set threshold ϵ , it is considered as a noise item and is set to zero, and the difference signal is further accumulated to recover the voltage signal after baseline drift.

[0063] Formula 7; wherein, is the initial reference voltage, is the voltage signal after baseline drift correction.

[0064] Based on the same method, the current is processed to obtain the baseline current .

[0065] S104-4-8: Determine the optimal alignment time between the baseline voltage and the baseline current , and time align the baseline voltage and the baseline current respectively according to the optimal alignment time , to obtain the corrected voltage and the corrected current .

[0066] S104-4-8 is also called time synchronization and alignment. In order to eliminate the time sequence misalignment of the battery pack voltage signal and the current signal caused by sampling delay or sampling frequency difference in the collection process, the pre-processed voltage signal V ( t ) and the current signal I ( t ) are time synchronized and aligned. Cross-correlation analysis method is used to correct the time delay between the signals, so as to ensure that the voltage change can be accurately corresponding to the current excitation.

[0067] S104-4-8 includes: 1) In an effective analysis window T, calculate the cross-correlation function of the voltage signal and the current signal: Formula 8; wherein, τ is the time delay estimated in advance according to experience, , which is determined according to the sampling period and the system delay range.

[0068] 2) In this embodiment, to achieve the optimal cross-correlation function value among all candidate delay values, the maximum value corresponding to the delay is taken As the optimal alignment amount: Equation 9; Where, if is less than the set tolerance threshold , it is determined that the signal has been basically synchronized, and no shift correction is needed; if is greater than , the signal needs to be shifted and corrected.

[0069] 3) Shift correction: Equation 10; Equation 11; Where, and are the aligned voltage signal and time signal, respectively.

[0070] Beneficial effects: Based on S104, lower-noise, baseline-stable, and time-aligned voltage and current data can be obtained, ensuring that the data relied on for subsequent correlation analysis and consistency analysis is more reliable, thereby significantly improving the accuracy and robustness of the diagnostic results.

[0071] S106: Based on the Kendall rank correlation algorithm, determine the global correlation index of each test battery group according to the rank pair corresponding to the corrected electrical data.

[0072] S106 and S108 are as shown in Figure 5 .

[0073] S106 includes: using the Kendall rank correlation coefficient to analyze the correlation domain of the monotonic relationship between the test battery group voltage variation V and the current I, and using the Kendall rank correlation coefficient to measure the monotonic relationship between the monomer voltage variation and the current within the effective segment T, thereby identifying the abnormality of the battery operation trend. When the correlation coefficient is lower than the preset threshold, it is determined that the battery has a correlation anomaly.

[0074] Specifically, the number of sampling points in each effective segment T is n, and for each effective segment T, S106-2 to S106-8 are executed.

[0075] S106 includes: S106-2: Based on the corrected voltage , construct a voltage variation sequence.

[0076] Specifically, the voltage of the n sampling points in any one effective segment T is constructed as a voltage variation sequence {Δ 1, Δ 2,…, Δ n} and any one of them is denoted by e.

[0077] S106-4: determining a current sequence based on the corrected current S106-5: constructing a voltage change sequence.

[0078] Specifically, the current of n sampling points within any one valid segment T is constructed as a current sequence 1, 2,…, n} and any one of them is denoted by f.

[0079] S106-6: determining a plurality of Kendall rank pairs according to the voltage change sequence and the current sequence.

[0080] Specifically, a total of Kendall rank pairs can be constructed.

[0081] Table 2

[0082] Table 3

[0083] Specifically, Table 2 and Table 3 provide simplified experimental data for constructing Kendall rank pairs for reference.

[0084] S106-8: classifying the Kendall rank pairs, and determining a local correlation index based on the classification result, wherein each test battery group includes K valid segments T, and each valid segment T corresponds to one local correlation index.

[0085] When the following conditions are met: Formula 12; that is, the rank order direction of the current and the voltage change is the same, it is determined (i.e., classified) that the rank pair is a consistent pair, and the number of all consistent pairs is denoted by C .

[0086] When the following conditions are met: Formula 13; that is, the rank order direction of the current and the voltage change is opposite, it is determined (i.e., classified) that the rank pair is an inconsistent pair, and the number of all inconsistent pairs is denoted by D .

[0087] When I e = I for Δ V e =Δ V f When this occurs, the pair is considered a level pair, recorded as B, and is not counted. C or D .

[0088] That is, B + C + D = .

[0089] Within each effective window, the voltage change Δ of a single cell is... ( t ) and battery pack current ( t Kendall's rank correlation coefficient was calculated to obtain the local correlation index. : Formula 14.

[0090] Where n is the number of sampling points within the effective fragment T, and C and D are the results of Kendall's rank pair classification.

[0091] Specifically, the decommissioned battery energy storage system includes R groups of test battery packs, where r represents the r-th test battery pack. Each test battery pack includes K effective segments T, where k represents the k-th effective segment T. Each effective segment T corresponds to a local correlation index. This yields R*K local correlation indices. R*K global correlation indicators .

[0092] S106-10: Incorporate local correlation indicators We perform weighted aggregation to obtain the global relevance index. Each valid fragment T corresponds to one global relevance index.

[0093] S106-10 includes: 1) Calculate the signal-to-noise ratio of the preset m effective segments T based on the following formula, where the total number of effective segments T in each test battery pack is K, and m≤K: ; 2) Calculate the global correlation index (i.e., the aggregate correlation index): Formula 15 ； Formula 16; in, express m The sum of the signal-to-noise ratios of a (preset number) of windows. This represents the normalized signal-to-noise ratio.

[0094] S108: Determine the consistency index for each group of test battery packs based on the corrected electrical data.

[0095] Specifically, S108 includes: extracting a multi-dimensional feature vector for each battery, including features such as equivalent internal resistance, polarization amplitude, voltage change slope, model fitting residual, and steady-state variance, and performing consistency domain analysis. Consistency domain analysis involves extracting a multi-dimensional feature vector for each battery within the effective segment, including features such as equivalent internal resistance, polarization amplitude, voltage change slope, model fitting residual, and steady-state variance. By calculating the Mahalanobis distance between each battery's feature vector and the group distribution, it is determined whether the battery deviates from the group consistency range. When the Mahalanobis distance exceeds a preset threshold and persists for multiple windows, the battery is determined to have a consistency anomaly.

[0096] For any valid segment of any test battery pack, extract the following features (there are R test battery packs in total, each test battery pack has K valid segments, each valid segment T contains n sampling points, the number of valid segments K is different for different test battery packs, each valid segment T corresponds to a Mahalanobis distance D, so we can obtain R*K Mahalanobis distances D (i.e., consistency index D)): S108 includes: S108-2: According to the corrected voltage and corrected current The consistency parameters for each effective segment T in each test battery pack are determined separately. The consistency parameters include: equivalent internal resistance, polarization voltage amplitude, voltage change slope, model fitting residual, and steady-state variance.

[0097] 1) Equivalent internal resistance R： During a step change in current Δ I At that time, the instantaneous voltage drop Δ was measured. V The calculation formula is: R Formula 17; 2) Polarization voltage amplitude P： Polarization voltage amplitude P From instantaneous voltage V tr With steady-state voltage V ss Determining the difference: 2-1) Instantaneous voltage V tr calculate: In the kth effective window interval Internal detection of current excitation events to determine the timing of the main event. t 0,k When satisfied , The threshold value is determined for the main event, and the moment is determined as the main event.

[0098] To avoid triggering glitches, delay (pre-set value) after the main event, and calculate the average of the voltage in the short window t tr Calculate the instantaneous voltage: Equation 18; 2-2) Steady-state voltage V ss Calculate: After buffering for at least t min , find the first t s,k that satisfies the voltage rate of change less than its threshold value and the voltage variance less than its threshold value , that is , while continuously maintaining a duration greater than T min . Then the steady-state voltage is averaged in Equation 19; 2-3) The amplitude of the polarization voltage at this time Equation 20; 3) Slope intensity index S : Calculate the rate of change of voltage with time in the window T, and take the median as the feature quantity: Equation 21; 4) Model fitting residual E： Calculate the predicted voltage based on the existing equivalent circuit model (the equivalent circuit model is the existing classical model, input current data, and get the predicted voltage value ) and the actual voltage , the residual is calculated as: Equation 22; 5) Steady-state variance Var : Calculate the variance of the voltage in the steady-state current stage: Equation 23; Where n represents the number of sampling points.

[0099] S108-4: Determine the consistency feature vector x according to the consistency parameter.

[0100] Construct the consistency feature vector of the rth group of test batteries: ​ Equation 24.

[0101] S108-6: Determine the consistency index D of each valid segment T in each group of test battery packs according to the consistency feature vector.

[0102] Based on the feature vector x of the consistency of the valid segment T of all test battery packs, the group mean in the kth valid window T of the rth group (i.e., each group) of test battery packs is calculated μ k and the covariance matrix Σ k : Note: L, step S, K of T in the valid segment of different test battery packs r are different, but because the difference will not be too large, if K of different test battery packs r is different, the extra T is removed, and K of all test battery packs r is kept consistent.

[0103] The group mean of the kth window of the entire returned battery energy storage system: Equation 25.

[0104] represents the group mean of all R test battery packs in the kth valid window.

[0105] The covariance matrix of the kth window of the entire returned battery energy storage system: Equation 26.

[0106] The Mahalanobis distance D (i.e., the consistency index) of the kth valid window T of the rth group of test battery packs.

[0107] Equation 27.

[0108] S110: Determine the fault state of each group of test battery packs and the returned battery energy storage system based on the global correlation index, the consistency index of each group of test battery packs, the preset correlation threshold, and the preset consistency threshold.

[0109] S110 includes: S110-2: Determine whether the fault state of the test battery pack is a correlation fault according to the global correlation index of the test battery pack and the preset correlation threshold.

[0110] Specifically, there are K valid segments T in each group of test battery packs, and each valid segment T corresponds to 1 correlation index For each group of test battery packs: compare the aggregated (global) correlation index with the preset correlation threshold Comparison is made when and when the condition is met in consecutive p preset number (p < K) of sliding windows (effective segments) T, it is determined that the test battery group r has a correlation trend abnormality.

[0111] S110-4: Determine whether the failure state of the test battery group is a consistency failure according to the consistency index of the test battery group and a preset consistency threshold.

[0112] Specifically, there are K effective segments T in each test battery group, each effective segment T corresponds to 1 consistency index D, and for each test battery group: when D th D th is a preset consistency threshold) and the condition is met in consecutive p preset number of effective segments T, it is determined that the test battery group r has a consistency abnormality.

[0113] S110-6: Determine the overall failure state of the returned battery energy storage system according to the failure state of each test battery group.

[0114] Specifically, if there is a consistency failure and the number of test battery groups with correlation failure is greater than a preset threshold A, the overall failure state of the returned battery energy storage system is considered to be a first-level failure; if there is a consistency failure and the number of test battery groups with correlation failure is between A and B, the overall failure state of the returned battery energy storage system is considered to be a second-level failure, and the overall failure state of the returned battery energy storage system can be classified according to actual needs.

[0115] Beneficial effects: S110 adopts correlation domain and consistency domain joint modeling, calculates a comprehensive risk score, and completes failure determination according to a classification threshold; after S110, the fault diagnosis method of the application further includes: the diagnosis result is output to the EMS by the BMS, triggering emergency control such as power limiting / flow limiting, cooling setting down-regulation, and abnormal cell bypass / isolation, and closed-loop recording and tracing. Through the dual-domain fusion and BMS-EMS cooperative control, the fault detection reliability is improved and the false alarm risk is reduced, ensuring the safe operation and controllable scheduling of the battery group under complex working conditions.

[0116] The beneficial effects of the embodiments of the application are as follows: 1. The application can effectively solve the problem of misdiagnosis caused by static working condition noise, and take into account both trend abnormality and group deviation, achieving rapid and accurate diagnosis of the battery group, and providing reliable support for the screening and utilization of retired batteries and the safe operation of in-service batteries.​​​

[0117] 2. The method effectively avoids the interference of static / resting noise through current gating and robust statistical methods, improving the stability of diagnosis. It can not only find early trend abnormalities, but also identify group consistency problems, solving the defects of single method coverage. The method does not depend on ideal conditions and can maintain good performance in different SOC intervals, aging levels and multi-temperature environments. The required signals are voltage and current, which are both conventional BMS collected quantities. The algorithm has low complexity and can be directly integrated into existing systems. The method framework can be coordinated with temperature monitoring, capacity estimation, health prediction and other modules, and has the potential for further application expansion.

[0118] 3. The method can effectively overcome the problems of single criterion, insufficient robustness and limited applicability in the prior art, and realize robust, fast and comprehensive diagnosis of battery system faults, providing reliable protection for the safe operation and retired gradient utilization of batteries.

[0119] 4. The application proposes a double-domain fusion diagnosis method combining Kendall rank correlation and Mahalanobis distance, which realizes robust fault detection of the battery pack in complex operating environments. The technical process includes multiple key steps: first, based on the current gating strategy, the operating data is filtered to select the excited segments with identification information, avoiding the interference of static noise on the correlation analysis. Then, the Kendall rank correlation coefficient is used to represent the monotonicity of the voltage-current relationship in the effective segment to identify correlation abnormalities. At the same time, the consistency features such as internal resistance, polarization, slope and residual error under the excited segment are extracted to construct a feature vector and use Mahalanobis distance to quantify the consistency deviation between batteries to identify consistency abnormalities. Finally, the correlation abnormalities and consistency abnormalities are fused to determine the robust diagnosis of the battery pack health state.

[0120] 5. The application proposes a double-domain robust battery fault diagnosis method for improving the fault recognition accuracy and robustness of lithium-ion battery packs in complex working conditions. The method uses current gating and window filtering to only diagnose in the excited segment, avoiding the interference of static or trickle noise on the correlation; in the correlation domain, the Kendall rank correlation coefficient is used to measure the monotonicity of the voltage-current relationship to identify battery operation trend abnormalities; in the consistency domain, multi-dimensional features such as internal resistance, polarization, slope and residual error are extracted to construct a feature vector and use Mahalanobis distance to measure the deviation from the group distribution, thereby identifying consistency abnormalities. Finally, the correlation abnormalities and consistency abnormalities are fused to determine the fast and robust diagnosis of the battery pack fault. The method only relies on conventional monitoring quantities such as voltage and current, has low computational complexity, and can effectively improve the safety and reliability of the battery system, and provide support for the rapid screening and gradient utilization of retired batteries.

[0121] 6、The application can realize robust, rapid and comprehensive diagnosis of the returned battery system failure, and provide data support for the safe operation of the battery in service and the retired step utilization.

[0122] Embodiment two The embodiment provides a dual-domain robust battery fault diagnosis device based on inconsistency of a returned battery energy storage system, which comprises the following: A first diagnosis module is configured to determine a test battery group according to a returned battery in the returned battery energy storage system. A second diagnosis module is configured to perform validity screening and time correction on initial electrical data of the test battery group, to obtain corrected electrical data. A third diagnosis module is configured to determine a global correlation index of each test battery group according to a rank pair corresponding to the corrected electrical data based on a Kendall rank correlation algorithm. A fourth diagnosis module is configured to determine a consistency index of each test battery group according to the corrected electrical data. A fifth diagnosis module is configured to determine a fault state of each test battery group and the returned battery energy storage system based on the global correlation index, the consistency index, a preset correlation threshold and a preset consistency threshold.

[0123] The dual-domain robust battery fault diagnosis device based on inconsistency of a returned battery energy storage system provided by the embodiment of the application has the same implementation principle and technical effects as the foregoing dual-domain robust battery fault diagnosis method based on inconsistency of a returned battery energy storage system, and for brevity of description, the part not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiment.

[0124] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application.

Claims

1. A dual-domain robust battery fault diagnosis method based on inconsistencies in decommissioned battery energy storage systems, characterized in that, The method includes: S102: Determine the test battery pack based on the decommissioned batteries in the decommissioned battery energy storage system; S104: Perform validity screening and time correction on the initial electrical data of the test battery pack to obtain corrected electrical data; S106: Based on Kendall's rank correlation algorithm, determine the global correlation index of each test battery pack according to the rank pairs corresponding to the corrected electrical data; S108: Determine the consistency index of each group of the test battery packs based on the corrected electrical data; S110: Determine the fault status of each test battery pack and the decommissioned battery energy storage system based on the global correlation index, the consistency index, the preset correlation threshold, and the preset consistency threshold.

2. The dual-domain robust battery fault diagnosis method based on the inconsistency of decommissioned battery energy storage systems according to claim 1, characterized in that, S102 includes: The total number of decommissioned batteries installed in the decommissioned battery energy storage system is X; The total number of the test battery packs is R, where R >> X; Each of the test battery packs includes one decommissioned battery or multiple decommissioned batteries connected in succession.

3. The dual-domain robust battery fault diagnosis method based on the inconsistency of decommissioned battery energy storage systems according to claim 2, characterized in that, Perform S104 for each test battery pack; S104 includes: S104-2: Validate the initial electrical data of the test battery pack to obtain K valid segments T; S104-4: The voltage and current in the initial electrical data of the K valid segments T are subjected to outlier removal, noise suppression, baseline drift correction, and time synchronization alignment to obtain corrected electrical data. The corrected electrical data includes: corrected voltage... and corrected current .

4. The dual-domain robust battery fault diagnosis method based on the inconsistency of decommissioned battery energy storage systems according to claim 3, characterized in that, S104-2 includes: S104-2-2: Preset maximum window length L max Minimum window length L min Maximum window step size S max and minimum window step size S min ; S104-2-4: Determine the local signal-to-noise ratio (SNR) based on the initial electrical data of the test battery pack and the pre-acquired reference noise variance of the decommissioned battery; S104-2-6: Based on the local signal-to-noise ratio (SNR) and the maximum window length... L max Minimum window length L min Maximum window step size S max and minimum window step size S min Determine the window length L and window step size S of the test battery pack; S104-2-8: Determine the segment of interest T' according to the window length L and the window step size S, wherein the number of segments of interest T' for each test battery pack is K'; S104-2-10: Eliminate the segments of interest T' that do not meet the preset current threshold range to obtain the valid segments T, wherein the number of valid segments T for each test battery pack is K.

5. The dual-domain robust battery fault diagnosis method based on the inconsistency of decommissioned battery energy storage systems according to claim 4, characterized in that, S104-4 includes: S104-4-2: The voltage in the effective segment T and current Outlier removal was performed separately to obtain the removal voltage. and removal current ; S104-4-4: Based on the moving average filtering method, the rejection voltages are respectively... and the rejection current High-frequency noise suppression is performed to obtain the filter voltage. and filter current ; S104-4-6: The filtered voltages are respectively... and filter current Baseline drift processing is performed to obtain the baseline voltage. and baseline current ; S104-4-8: According to the baseline voltage and baseline current The cross-correlation function between them determines the optimal alignment time. And according to the optimal alignment time For baseline voltage and baseline current Time alignment was performed separately to obtain the corrected voltage. and corrected current .

6. The dual-domain robust battery fault diagnosis method based on the inconsistency of decommissioned battery energy storage systems according to claim 5, characterized in that, The number of sampling points in each valid segment T is n. For each valid segment T, S106 is executed. S106 includes: S106-2: Based on the corrected voltage Construct a voltage change sequence; S106-4: Based on the corrected current Construct current sequences; S106-6: Determine multiple Kendall rank pairs based on the voltage change sequence and the current sequence; S106-8: Classify the Kendall rank pairs and determine the local correlation index based on the classification results. Each test battery pack includes K effective segments T, and each effective segment T corresponds to one local correlation index. S106-10: Incorporate local correlation indicators Perform weighted aggregation to obtain the global relevance index. Each valid segment T corresponds to one global relevance index. .

7. The dual-domain robust battery fault diagnosis method based on the inconsistency of decommissioned battery energy storage systems according to claim 6, characterized in that, S106-8 includes: The local correlation index corresponding to any valid segment T is calculated based on the following formula: ； Where n is the number of sampling points within the effective fragment T, and C and D are the results of Kendall's rank pair classification.

8. The dual-domain robust battery fault diagnosis method based on the inconsistency of decommissioned battery energy storage systems according to claim 7, characterized in that, For each valid segment T, execute S108; S108 includes: S108-2: According to the corrected voltage and corrected current The consistency parameters for each effective segment T in each test battery pack are determined separately. The consistency parameters include: equivalent internal resistance, polarization voltage amplitude, voltage change slope, model fitting residual, and steady-state variance. S108-4: Determine the consistency feature vector based on the consistency parameters; S108-6: Determine the consistency index D for each valid segment T in each group of the test battery packs based on the consistency feature vector.

9. The dual-domain robust battery fault diagnosis method based on the inconsistency of decommissioned battery energy storage systems according to claim 8, characterized in that, S110 includes: S110-2: Determine whether the fault state of the test battery pack is a correlated fault based on the global correlation index of the test battery pack and the preset correlation threshold; S110-4: Determine whether the fault state of the test battery pack is a consistency fault based on the consistency index of the test battery pack and the preset consistency threshold; S110-6: Determine the overall fault status of the decommissioned battery energy storage system based on the fault status of each group of test battery packs.

10. A dual-domain robust battery fault diagnosis device based on the inconsistency of a decommissioned battery energy storage system, characterized in that, The device includes: The first diagnostic module is used to determine the test battery pack based on the decommissioned batteries in the decommissioned battery energy storage system. The second diagnostic module is used to perform validity screening and time correction on the initial electrical data of the test battery pack to obtain corrected electrical data. The third diagnostic module is used to determine the global correlation index of each test battery pack based on the Kendall rank correlation algorithm and the rank pairs corresponding to the corrected electrical data. The fourth diagnostic module is used to determine the consistency index of each group of the test battery packs based on the corrected electrical data. The fifth diagnostic module is used to determine the fault status of each test battery pack and the decommissioned battery energy storage system based on the global correlation index, the consistency index, the preset correlation threshold, and the preset consistency threshold.

Images