Battery pack multi-fault diagnosis method and system based on signal decomposition and entropy feature fusion

Abstract

The invention relates to the technical field of battery management, in particular to a battery pack multi-fault diagnosis method and system based on signal decomposition and entropy feature fusion. The method comprises the following steps: acquiring voltage data of each single battery in a battery pack in real time, and performing segmentation processing on voltage time sequence data by adopting a sliding window mechanism based on a Fibonacci sequence; adaptively optimizing key parameters of variational mode decomposition by using a Schrodinger optimization algorithm, and performing adaptive decomposition on the voltage signal in each window to obtain a plurality of intrinsic mode functions; calculating the Shannon entropy value of each intrinsic mode function, preferably selecting a preset number of representative mode components according to the Shannon entropy characteristic contribution degree, and carrying out superposition reconstruction to obtain a reconstructed voltage signal; extracting the Shannon entropy, the spectrum sparsity index and the time domain stability index of the reconstructed voltage signal to form a three-dimensional fault feature vector; and identifying a connection fault, a short circuit fault and a sensor fault through a multi-level threshold judgment strategy based on the three-dimensional fault feature vector.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery management, in particular to a battery pack multi-fault diagnosis method and system based on signal decomposition and entropy feature fusion. BACKGROUND

[0002] In actual use, the battery pack is prone to multiple faults under complex working conditions, such as loose connection, internal or external short circuit, sensor failure, etc. If these faults cannot be detected and distinguished in time, they may cause serious safety accidents such as thermal runaway, seriously threatening the safe operation of electric vehicles.

[0003] In the prior art, the absolute threshold of voltage and current is used for alarm, which is easily affected by the initial inconsistency and aging degree of the battery pack, resulting in false alarm or missed alarm. In addition, some algorithms based on dynamic time warping or complex models have large calculation amount, are difficult to implement online diagnosis on the battery management system with limited computing resources, and have single feature extraction, which is difficult to uniquely identify the specific fault type.

[0004] Therefore, there is an urgent need for a diagnosis method that can online, quickly and accurately identify multiple fault types and is robust to battery pack inconsistency. SUMMARY

[0005] In view of the above problems, the present disclosure provides a battery pack multi-fault diagnosis method and system based on signal decomposition and entropy feature fusion to overcome the above problems or at least partially solve the above problems, aiming to realize fast and accurate identification of connection faults, short circuit faults and sensor faults, improve the diagnosis reliability under real battery pack working conditions, and be suitable for online application of battery management system (BMS).

[0006] The purpose of the present application can be achieved by the following technical solutions: The first aspect of the present application provides a battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion, comprising: Real-time acquisition of voltage data of each single battery in the battery pack, and segmenting the voltage time series data using a sliding window mechanism based on the Fibonacci sequence; Using the Schrodinger optimization algorithm to adaptively optimize the key parameters of variational mode decomposition, the voltage signal in each window is adaptively decomposed to obtain multiple intrinsic mode functions; Calculate the Shannon entropy value of each intrinsic mode function, and select a preset number of representative modal components according to the Shannon entropy feature contribution degree, and superimpose and reconstruct to obtain a reconstructed voltage signal; Extract the Shannon entropy, spectral sparsity index and time domain stability index of the reconstructed voltage signal to form a three-dimensional fault feature vector; Based on the three-dimensional fault feature vector, a multi-level threshold judgment strategy is used to identify connection faults, short circuit faults and sensor faults.

[0007] Further, The key parameters of the variational modal decomposition are adaptively optimized using the Schrödinger optimization algorithm, including: Initialize a population, and each individual represents a set of modal numbers and penalty coefficients of variational modal decomposition; Take the sum of the Shannon entropies of all intrinsic modal functions as the fitness function, and iteratively update the parameter combination through the Schrödinger optimization algorithm; Output the optimal parameter combination that minimizes the fitness function, which is used for variational modal decomposition of the voltage signal.

[0008] Further, The fitness function is defined as: ; ; Wherein, is the energy probability distribution of the jth IMF component at point i, is the jth intrinsic modal function, is the number of intervals divided by the signal amplitude range.

[0009] Further, The calculation of the spectral sparsity index includes: Perform a fast Fourier transform on the reconstructed voltage signal to obtain the power spectrum ; Normalize the power spectrum to a probability distribution, denoted as: ; Calculate the Shannon entropy of the probability distribution as the spectral sparsity index, denoted as: .

[0010] Further, The calculation of the time-domain stability index includes: Segment the reconstructed voltage signal and calculate the root mean square value of each segment as the segmented energy sequence , Wherein, ; Perform linear regression on the segmented energy sequence to obtain the absolute value of the slope and the residual sum of squares ; Calculate the time-domain stability index based on the absolute value of the slope and the residual sum of squares, denoted as: .

[0011] Further, The multi-level threshold judgment strategy comprises: Firstly, it is judged whether the Shannon entropy of each single battery exceeds a normal threshold value, and if so, it is determined that there is an abnormality; For the abnormal battery, it is judged whether the spectral sparsity index and the time domain stability index thereof simultaneously fall below a sensor failure threshold value, and if so, it is determined as sensor failure; If it is not sensor failure, a dynamic reference is set based on the average value of the spectral sparsity index and the time domain stability index of the normal single battery, and by comparing the deviation degree of the abnormal battery feature from the dynamic reference, short-circuit failure and connection failure are distinguished.

[0012] The second aspect of the technical scheme of the present application provides a battery pack multi-fault diagnosis system based on signal decomposition and entropy feature fusion, comprising: A data acquisition module is configured to acquire voltage data of each single battery in the battery pack in real time; A signal preprocessing module is configured to segmentally process the voltage time series data by using a sliding window mechanism based on a Fibonacci sequence; An adaptive decomposition module is configured to optimize variational mode decomposition parameters by using a Schrodinger optimization algorithm, and decompose the voltage signal; A signal reconstruction module is configured to select and reconstruct an intrinsic mode function based on Shannon entropy; A feature extraction module is configured to extract Shannon entropy, spectral sparsity index and time domain stability index; A fault diagnosis module is configured to identify fault types based on a three-dimensional fault feature vector and a multi-level threshold judgment strategy.

[0013] Further, In a battery pack multi-fault diagnosis system based on signal decomposition and entropy feature fusion, the adaptive decomposition module comprises a parameter optimization unit and a signal decomposition unit, wherein the parameter optimization unit uses a Schrodinger optimization algorithm to minimize the sum of Shannon entropy as an objective function, and the signal decomposition unit performs variational mode decomposition using the optimized parameters.

[0014] In a battery pack multi-fault diagnosis system based on signal decomposition and entropy feature fusion, the fault diagnosis module comprises an abnormality detection unit, a sensor failure identification unit and a fault distinguishing unit, which sequentially perform multi-level judgment.

[0015] The third aspect of the technical scheme of the present application provides a computer readable storage medium having instructions stored thereon, which, when executed by one or more processors, cause the processors to perform the battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion as described in the first aspect.

[0016] The technical scheme provided by the present application can bring the following beneficial effects: 1、 The method provided by the present application diagnoses according to the mutation of the dynamic characteristics of the signal caused by the fault, rather than the static absolute value of the voltage, and thus has natural robustness to the inconsistency inherent in the battery pack due to different initial capacities and aging degrees, effectively avoiding false positives.

[0017] 2、 The adaptive window sampling strategy based on the Fibonacci sequence is proposed to balance calculation and detection, solving the problem of low calculation efficiency caused by the fixed step length of the traditional sliding window method.

[0018] 3、 The Schrodinger optimization algorithm is used to adaptively optimize the variational modal decomposition parameters, avoiding the complexity of manual parameter adjustment and improving the accuracy and stability of signal decomposition.

[0019] 4、 Without relying on complex sequence matching algorithms such as dynamic time warping, the calculation efficiency is high, and it is suitable for BMS embedded deployment.

[0020] 5、 By constructing a three-dimensional feature of Shannon entropy, spectral sparsity index and time domain stability index, and extracting fault features from randomness, frequency domain and time domain stability, the limitations of traditional methods relying on single features or thresholds are broken through, and the identification of different fault types from anomaly detection is realized.

[0021] The above description is only a summary of the technical scheme of the present disclosure, in order to more clearly understand the technical means of the present disclosure, the content of the specification can be implemented, and in order to make the above and other purposes, features and advantages of the present disclosure more obvious and easy to understand, the following specific embodiments of the present disclosure are described. BRIEF DESCRIPTION OF DRAWINGS

[0022] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for purposes of illustrating the preferred embodiments only and are not to be construed as limiting the present disclosure. Moreover, in the accompanying drawings, like reference numerals designate like parts throughout the several views. In the drawings: Figure 1 is a step flowchart of a battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion provided by an embodiment of the present specification; Figure 2 is a flowchart of a battery pack multi-fault diagnosis method provided by an embodiment of the present specification; Figure 3 is an IMF decomposition schematic diagram under normal voltage provided by an embodiment of the present specification; wherein, Figure 3 (a) in the figure is a curve graph of the change of voltage value with sampling points when the original signal is normal; Figure 3 (b) in the figure is a curve graph of the change of amplitude with sampling points when the Shannon entropy is 4.8526;Figure 3 (c) is a graph showing the change in amplitude with sampling points when the Shannon entropy is 2.1812; Figure 3 The graph in (d) shows the amplitude variation with sampling points when the Shannon entropy is 2.0377. Figure 3 (e) is the curve showing the change in amplitude with sampling points when the Shannon entropy is 2.1570; Figure 3 In the middle (f), the amplitude varies with the sampling points when the Shannon entropy is 1.9230; Figure 4 This is a schematic diagram of IMF decomposition under short-circuit fault provided in the embodiments of this specification; wherein, Figure 4 (a) is a graph showing the voltage value as a function of sampling points during a short-circuit fault; Figure 4 (b) is a graph showing the change in amplitude with sampling points when the Shannon entropy is 5.3216; Figure 4 (c) is the curve showing the change in amplitude with sampling points when the Shannon entropy is 3.7866; Figure 4 The graph in (d) shows the change in amplitude with sampling points when the Shannon entropy is 1.7888. Figure 4 (e) is the curve showing the change in amplitude with sampling points when the Shannon entropy is 3.1366; Figure 5 This is a schematic diagram of IMF decomposition under connectivity failure provided in the embodiments of this specification; wherein, Figure 5 (a) shows the voltage value as a function of sampling points when there is a connection fault; Figure 5 (b) is a graph showing the change in amplitude with sampling points when the Shannon entropy is 5.3604. Figure 5 (c) is a graph showing the change in amplitude with sampling points when the Shannon entropy is 5.2056; Figure 5 The graph in (d) shows the change in amplitude with sampling points when the Shannon entropy is 4.2779. Figure 5 (e) is the curve showing the change in amplitude with sampling points when the Shannon entropy is 4.2146; Figure 5 The graph in (f) shows the change in amplitude with sampling points when the Shannon entropy is 3.9594. Figure 6 This is a schematic diagram of IMF decomposition under sensor failure provided in the embodiments of this specification; wherein, Figure 6 (a) shows the voltage value as a function of sampling points when the sensor fails; Figure 6 (b) is a graph showing the change in amplitude with sampling points when the Shannon entropy is 5.1630. Figure 6 (c) is the curve showing the change in amplitude with sampling points when the Shannon entropy is 5.0903; Figure 6 The graph in (d) shows the change in amplitude with sampling points when the Shannon entropy is 4.7592. Figure 6 (e) is the curve showing the change in amplitude with sampling points when the Shannon entropy is 4.8636; Figure 7 is a shannon entropy curve schematic diagram provided by an embodiment of the specification; Figure 8 is a structural schematic diagram of a battery pack multi-fault diagnosis system based on signal decomposition and entropy feature fusion provided by an embodiment of the specification. DETAILED DESCRIPTION

[0023] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art. The technical solutions provided by the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0024] The embodiments of the present application provide a battery pack multi-fault diagnosis method and system based on signal decomposition and entropy feature fusion.

[0025] In a first aspect of the technical solutions of the present application, a battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion is provided, comprising the following steps: S101, real-time acquisition of voltage data of each single battery in the battery pack, and segment processing of the voltage time series data using a sliding window mechanism based on the Fibonacci sequence; S102, using the Schrodinger optimization algorithm (SGA) to adaptively optimize the key parameters of the variational mode decomposition (VMD), and adaptively decomposing the voltage signal in each window to obtain a plurality of intrinsic mode functions; S103, calculating the Shannon entropy value of each intrinsic mode function, and selecting a preset number of representative modal components according to the Shannon entropy feature contribution degree, and superimposing and reconstructing to obtain a reconstructed voltage signal; S104, extracting the Shannon entropy, spectral sparsity index and time domain stability index of the reconstructed voltage signal to form a three-dimensional fault feature vector; S105, based on the three-dimensional fault feature vector, identifying the connection fault, short circuit fault and sensor fault through a multi-level threshold judgment strategy.

[0026] Based on the S101 step, signal acquisition and preprocessing is performed, specifically as follows: During battery pack operation, the voltage data of each individual cell is collected in real time by the Battery Management System (BMS). To address the low computational efficiency caused by the fixed step size of traditional sliding window methods, this invention proposes an adaptive window sampling strategy based on the Fibonacci sequence (subsequence: [1, 2, 3, 5, 8, 13, 21, 34]) to balance computation and detection. First, the standard deviation of the voltage data within the current sliding window is calculated. ,Will Standard deviation of historical window voltage Compare them; when Less than When the voltage is not significantly abnormal, a low-density sampling mode ([8, 13, 21, 34]) is used to reduce the computational load; when Greater than When a voltage anomaly is detected, a high-density sampling mode ([1, 2, 3, 5]) is used to quickly capture fault characteristics. In this invention, the selection of the Fibonacci sequence is based on the golden ratio property of the sequence, and the ratio of adjacent step lengths approaches the golden ratio φ≈1.618. This mathematical property ensures that when the sampling density needs to be adjusted due to changes in the system state, the change ratio of the step length is naturally reasonable, avoiding abrupt resolution changes, ensuring the continuity of signal analysis, and reducing computational efficiency.

[0027] Let the window size be W and the sliding step size be s, then the window overlap rate is... This invention proposes to optimize window overlap using the following formula: Optimization objective = min( ) in The time-varying step size is selected from the Fibonacci sequence.

[0028] In this invention, the selection of the Fibonacci sequence is based on the golden ratio property of the sequence. The ratio of adjacent step lengths approaches the golden ratio φ≈1.618. This mathematical property ensures that when the system state changes and the sampling density needs to be adjusted, the change ratio of the step length is naturally reasonable, avoiding sudden changes in resolution, ensuring the continuity of signal analysis, and improving computational efficiency.

[0029] Based on step S102, SGA-VMD adaptive optimization and decomposition are performed. This step aims to adaptively decompose the non-stationary battery voltage signal into a series of quasi-stationary intrinsic mode functions (IMFs) with a center frequency. Addressing the problems of traditional swarm intelligence optimization algorithms easily getting trapped in local extrema and exhibiting poor convergence speed when optimizing VMD parameters, this invention proposes using the Schrödinger optimization algorithm (SGA) to adaptively optimize the VMD parameters (number of modes k, penalty coefficient α). Specifically: (1) Initialize SGA parameters: Initialize the position of all search agents (particles), each particle represents a (k, α) combination; set the population size and maximum number of iterations, define the optimization variable dimension m=2 and range, calculate the fitness of each initial particle, and record the global optimal solution.

[0030] (2) Definition of fitness function: The sum of minimum Shannon entropies is defined as the fitness function to minimize the uncertainty of the signal components after decomposition. The fitness function is as follows: ; ; in, Let i be the energy probability distribution of the j-th IMF component at point i. Let j be the j-th eigenmode function. It is the number of intervals into which the signal amplitude range is divided.

[0031] (3) Optimize the search process: First, for each individual in the population, perform VMD decomposition and calculate its fitness value; where an individual is a ( , ) parameter pairs.

[0032] Then, the global search capability of the SGA algorithm is used to update the position of individuals in the population; optionally, a random restart strategy is introduced (to reinitialize individuals with a small probability) to enhance the ability to escape local optima.

[0033] Finally, repeat the above steps until the maximum number of iterations is reached.

[0034] This step adaptively decomposes the non-stationary battery voltage signal into a series of quasi-stationary intrinsic mode functions (IMFs) with a center frequency, where each IMF represents a feature in the original voltage signal, and therefore these IMFs can be used for fault diagnosis.

[0035] The target number of IMF in VMD Penalty coefficient It is a key parameter that plays a decisive role in the signal decomposition effect, directly affecting the amplitude, frequency and contribution of the IMF; Directly control the number of IMFs obtained from the decomposition: An excessively large value will lead to over-decomposition and introduce pseudo-modalities, while an excessively small value will result in under-decomposition, preventing the latent features in the time series from being fully extracted; penalty weights It also has a significant impact on decomposition performance, and its relationship with the bandwidth of each modal component is inverse; Too low a value can easily cause modal aliasing, interfering with the identification of effective features; too high a value can suppress aliasing, but may cause the loss of local detail information.

[0036] This step is aimed at the problem that traditional swarm intelligence optimization algorithms (such as grey wolf optimization algorithm GWO, northern goshawk optimization algorithm NGO, etc.) are prone to fall into local extremum and have poor convergence speed when optimizing VMD parameters. This paper proposes to use Schrödinger optimizer (SGA) to adaptively optimize VMD parameters (a, b, c) , );SGA algorithm uses wave-particle duality, combines probability exploration and deterministic use of "twin update mechanism", and realizes global exploration and local fine search. In addition, the algorithm itself has fewer parameters, is not sensitive to the initial population setting, and uses adaptive switching threshold to dynamically adjust the exploration and development ratio, taking into account the global search ability and convergence speed, which can effectively deal with optimization problems in complex search space; when dealing with battery voltage signals affected by noise and working condition changes, it can show more stable optimization performance than traditional algorithms, ensuring the consistency of VMD decomposition effect, thereby improving the reliability of the entire diagnosis system.

[0037] For step S103, this step is based on Shannon entropy IMF optimization and reconstruction, which includes: For each IMF component obtained by decomposition, calculate its Shannon entropy value SE, and the calculation method is consistent with the formula used in step S102 for SGA fitness evaluation, thereby reducing the calculation complexity. According to its characteristic contribution (from low to high top three), the representative signal containing fault features is selected, and the selected IMF component is superimposed to reconstruct a new voltage signal. The lower the Shannon entropy value, the more regular and orderly the amplitude sequence of the component, the clearer the specific physical mode (such as periodic oscillation, steady-state deviation) caused by the fault, and the less the random noise interference.

[0038] This step can effectively remove signal noise and redundancy on the one hand, and enhance the adaptability of the method to battery inconsistency on the other hand. The inherent inconsistency and slow aging process of the battery pack are usually reflected in the low-frequency trend or overall baseline fluctuation of the signal. By reconstructing the IMF selected by Shannon entropy, these low-frequency and trend components dominated by inconsistency and aging can be effectively filtered out, so that the diagnosis method relies more on the dynamic feature mutation triggered by the fault in a specific frequency band, rather than the absolute static value of the voltage, thereby significantly reducing the dependence on the initial state and aging degree of the battery.

[0039] Based on step S104, the Shannon entropy, spectral sparsity index and time domain stability index of the reconstructed voltage signal are extracted, which includes: First, Shannon entropy SE is taken as the first dimension feature of the fault, which is used to quantify the uncertainty or randomness of the signal in signal processing; the higher the entropy value, the more chaotic and unpredictable the signal is; the lower the entropy value, the more orderly and regular the signal is.

[0040] For the reconstructed voltage signal, the Shannon entropy in each window is calculated by using a sliding window, and then the entropy value curve changing with time is generated by sliding the window.

[0041] Under normal conditions, the battery voltage signal shows a smooth and orderly change rule, and the Shannon entropy value maintains a low level with slight fluctuations, reflecting the inherent certainty and stability of the system. When a fault occurs, whether it is a voltage drop caused by a short circuit fault, intermittent oscillation caused by a connection fault, or random noise introduced by a sensor fault, it will inject significant unpredictable components into the voltage signal, disrupting the original regularity and causing the Shannon entropy value to rise significantly.

[0042] Specifically, as shown in Figure 3 , Figure 4 , Figure 5 , Figure 6 , wherein the curve reflects the change of the voltage value with the sampling point; and the curves of IMF1 to IMF5 reflect the change of the amplitude with the sampling point; wherein, Figure 3 (a) in the original signal, the curve of the voltage value with the sampling point, the abscissa is the sampling point, and the ordinate is the voltage; Figure 3 (b) in the curve of the amplitude with the sampling point when the Shannon entropy is 4.8526, the abscissa is the sampling point, and the ordinate is the amplitude; Figure 3 (c) in the curve of the amplitude with the sampling point when the Shannon entropy is 2.1812, the abscissa is the sampling point, and the ordinate is the amplitude; Figure 3 (d) in the curve of the amplitude with the sampling point when the Shannon entropy is 2.0377, the abscissa is the sampling point, and the ordinate is the amplitude; Figure 3 (e) in the curve of the amplitude with the sampling point when the Shannon entropy is 2.1570, the abscissa is the sampling point, and the ordinate is the amplitude; Figure 3 (f) in the curve of the amplitude with the sampling point when the Shannon entropy is 1.9230, the abscissa is the sampling point, and the ordinate is the amplitude; wherein, Figure 4 (a) in the curve of the voltage value with the sampling point when a short circuit fault occurs, the abscissa is the sampling point, and the ordinate is the voltage; Figure 4 (b) in the curve of the amplitude with the sampling point when the Shannon entropy is 5.3216, the abscissa is the sampling point, and the ordinate is the amplitude; Figure 4 (c) in the curve of the amplitude with the sampling point when the Shannon entropy is 3.7866, the abscissa is the sampling point, and the ordinate is the amplitude;Figure 4 (d) represents the curve showing the change in amplitude with sampling points when the Shannon entropy is 1.7888. The horizontal axis represents the sampling points, and the vertical axis represents the amplitude. Figure 4 (e) represents the curve showing the change in amplitude with sampling points when the Shannon entropy is 3.1366. The horizontal axis represents the sampling points, and the vertical axis represents the amplitude. in, Figure 5 In the middle (a), the curve of voltage value changing with sampling points when there is a connection fault is shown. The horizontal axis is the sampling point and the vertical axis is the voltage. Figure 5 (b) represents the curve of amplitude variation with sampling points when Shannon entropy is 5.3604, with the horizontal axis representing sampling points and the vertical axis representing amplitude. Figure 5 (c) represents the curve of amplitude change with sampling points when Shannon entropy is 5.2056, with the horizontal axis representing sampling points and the vertical axis representing amplitude; Figure 5 (d) represents the curve of amplitude change with sampling points when Shannon entropy is 4.2779, with the horizontal axis representing sampling points and the vertical axis representing amplitude; Figure 5 (e) represents the curve showing the change in amplitude with sampling points when the Shannon entropy is 4.2146. The horizontal axis represents the sampling points, and the vertical axis represents the amplitude. Figure 5 (f) represents the curve of amplitude change with sampling points when Shannon entropy is 3.9594, with the horizontal axis representing sampling points and the vertical axis representing amplitude; in, Figure 6 In the middle (a), the curve of voltage value changing with sampling points when the sensor fails is shown. The horizontal axis is the sampling point and the vertical axis is the voltage. Figure 6 (b) represents the curve of amplitude variation with sampling points when Shannon entropy is 5.1630, with the horizontal axis representing sampling points and the vertical axis representing amplitude; Figure 6 (c) represents the curve of amplitude change with sampling points when Shannon entropy is 5.0903, with the horizontal axis representing sampling points and the vertical axis representing amplitude; Figure 6 (d) represents the curve showing the change in amplitude with sampling points when the Shannon entropy is 4.7592. The horizontal axis represents the sampling points, and the vertical axis represents the amplitude. Figure 6 (e) represents the curve showing the change in amplitude with sampling points when the Shannon entropy is 4.8636. The horizontal axis represents the sampling points, and the vertical axis represents the amplitude. like Figure 4 As shown, under a short-circuit fault, the entropy value jumps to a high level with slight fluctuations; as Figure 5 As shown, under connection failure, the entropy value increases moderately and is accompanied by continuous fluctuations; as Figure 6 As shown, the entropy value typically reaches its highest level under sensor failure. Therefore, this invention uses the Shannon entropy distribution for anomaly detection, and its significantly increased value is direct evidence that the system has transitioned from an ordered state to a disordered or abnormal state.

[0043] Secondly, the fault fingerprint with strong distinguishability is constituted by two indexes of spectral sparsity index (SSI) and time stability index (TSI) together with Shannon entropy, and the specific identification of connection fault, short circuit fault and sensor fault is realized by analyzing the distribution mode in the feature space; wherein, the SSI is used to represent the prominent degree of fault characteristic frequency components in the spectrum.

[0044] The calculation of the spectral sparsity index includes: The fast Fourier transform is performed on the reconstructed voltage signal to obtain a power spectrum ; The power spectrum is normalized into a probability distribution, which is expressed as: ; wherein, and are the one-sided power spectral densities at the discrete frequency points and after the fast Fourier transform (FFT) is performed on the reconstructed voltage signal; is the number of frequency points; The Shannon entropy of the probability distribution is calculated as the spectral sparsity index, which is expressed as: .

[0045] The greater the SSI value is, the more concentrated the spectrum is, and the fault energy is concentrated on a few specific frequency components, such as the short circuit point formed by the short circuit fault, which forms a strong and continuous disturbance source, and the energy is highly concentrated, corresponding to a high SSI value; the smaller the SSI value is, the more dispersed the spectrum is, and the fault energy is uniformly distributed in a wide frequency band, and there is no prominent characteristic frequency, corresponding to random noise, so the SSI of the sensor fault is extremely low.

[0046] Then, the time stability index (TSI) is calculated, which quantifies the stability degree of the signal in the time domain by analyzing the trend stability of the segmented energy sequence of the reconstructed voltage signal, to distinguish whether the fault forms a stable new state or a continuous unstable disturbance.

[0047] The calculation of the time stability index includes: The reconstructed voltage signal is segmented, and for the segment signal , the root mean square value RMS thereof is calculated as the energy level of the segment, and the amplitude and fluctuation of the signal are considered to obtain a segmented energy sequence ; wherein, ; L is the length (number of sampling points) of each segment of signal; the reconstructed long voltage signal is uniformly or non-overlappingly divided into a plurality of segments, and each segment contains L consecutive sampling points; represents the amplitude of the i-th sampling point in the k-th segment of signal; Linear regression is performed on the segmented energy sequence, denoted as , to obtain the absolute value of the slope and the residual sum of squares ; The time-domain stability index is calculated based on the absolute value of the slope and the residual sum of squares, denoted as: , wherein is not 0, and the meaning is to prevent from being zero and being unable to calculate.

[0048] The formula combines the strength of the trend change of the energy sequence (measured by the absolute value of the slope ) and the strength of the random fluctuation deviating from the trend (measured by the residual sum of squares ).

[0049] High TSI value: indicates that after the fault occurs, the system enters and maintains a new, relatively stable state, and the energy profile of the signal changes smoothly; corresponding to a short-circuit fault, the voltage will stabilize at a new, lower platform, forming a new state that is continuous and stable, so the TSI value is high; Medium TSI value: indicates that the fault state exists, but has certain volatility and is not completely stable. The voltage drop corresponding to the connection fault is continuous, but the change in arc or contact resistance accompanying it will introduce fluctuations, causing the energy sequence to change to some extent, so the TSI is moderate; Low TSI value: indicates that the fault signal itself is extremely unstable and has not formed any stable new pattern, and the energy profile fluctuates dramatically; corresponding to the random jump or drift produced by the sensor fault, the signal form is completely unstable and has no continuity, so the TSI is extremely low.

[0050] Based on the step S105, the multi-level threshold judgment strategy includes: First, determine whether the Shannon entropy of each single battery exceeds the normal threshold value, and if it does, determine that there is an anomaly; For the abnormal battery, determine whether the spectral sparsity index and the time-domain stability index are simultaneously lower than the sensor fault threshold value, and if so, determine that it is a sensor fault; If it is not a sensor fault, set a dynamic reference based on the average of the spectral sparsity index and the time-domain stability index of the normal single battery, and distinguish between short-circuit faults and connection faults by comparing the deviation of the abnormal battery characteristics from the dynamic reference.

[0051] In order to verify the effect of the present application, the following experiment is designed: To establish the quantitative threshold of SSI and TSI for fault classification in the application, a battery pack model composed of 5 single batteries in series is built in MATLAB or Simulink environment to simulate different faults; considering the inherent inconsistency of the battery in actual use, the initial SOC (between 90% and 100%) and initial temperature (between 24℃ and 26℃) of the 5 single bodies are set to different values and 0.5C constant current charging experiment is carried out.

[0052] The specific steps are: A 5mΩ contact resistance Rcf is connected in parallel between No. 1 battery and No. 2 battery through switch S1 to simulate connection fault, when connection fault occurs in the battery pack, switch S1 is opened, when the battery pack works normally, switch S1 is closed. A short circuit fault is simulated on No. 3 battery, a 500Ω short circuit resistance Risc is connected in series with switch S2, and then connected in parallel to the two ends of the battery to simulate short circuit fault. When the battery has short circuit fault, switch S2 is closed, when the battery works normally, switch S2 is opened. Sensor fault is simulated on No. 5 battery by adding fault injection module to trigger fault signal. Multiple experiments are carried out for each fault type to obtain fault data, the collected data are processed according to the diagnostic method described in the application, and finally a three-dimensional feature vector [SE, SSI, TSI] is obtained for each data sample; the three-dimensional feature vectors of all samples are plotted in the feature space, the statistical distribution of the features under different fault types is analyzed, and the classification threshold is set to realize the unique and specific identification of connection fault, short circuit fault and sensor fault.

[0053] Based on this, the application adopts a classification method based on rule threshold to realize multi-fault diagnosis and identification. By setting reasonable threshold, three types of faults can be accurately detected and diagnosed.

[0054] By checking whether the Shannon entropy (SE) of each single battery exceeds the normal threshold (TH_se_normal), it is judged whether the system has abnormality, if the SE of all single bodies does not exceed the threshold, it is determined that the battery pack is in normal state, otherwise it is determined that there is abnormality. After detecting the abnormality, sensor fault screening is carried out immediately, if the spectral sparsity index (SSI) and time domain stability index (TSI) of a single body are both lower than the set extremely low threshold (TH_sensor), it is determined as sensor fault.

[0055] For abnormal conditions of non-sensor fault, the system calculates the average value (SSI_avg, TSI_avg) of SSI and TSI of the remaining normal single bodies after excluding the identified fault single body to establish dynamic normal benchmark.

[0056] Finally, based on this benchmark to distinguish short circuit and connection fault: if the SSI and TSI of a certain abnormal monomer are significantly higher than the normal benchmark (more than 2 times), it is determined as a short circuit fault, because the short circuit will cause strong and unique disturbance; if the characteristics of the abnormal monomer do not deviate significantly from the normal benchmark, it is determined as a connection fault, because the influence of the connection fault (such as loosening) is relatively mild and will affect the adjacent monomers.

[0057] Among them, the person skilled in the art can set the normal state threshold according to the actual battery type, system working condition and diagnostic accuracy requirement, and the present application does not limit this.

[0058] The second aspect of the technical scheme of the present application provides a battery pack multi-fault diagnosis system 800 based on signal decomposition and entropy feature fusion, comprising: A data acquisition module 801 is configured to acquire voltage data of each monomer battery in the battery pack in real time. A signal preprocessing module 802 is configured to segment the voltage time series data by using a sliding window mechanism based on the Fibonacci sequence. An adaptive decomposition module 803 is configured to optimize the variational mode decomposition parameters by using the Schrodinger optimization algorithm, and decompose the voltage signal. A signal reconstruction module 804 is configured to select the intrinsic mode function based on the Shannon entropy feature contribution degree and reconstruct it. A feature extraction module 805 is configured to extract the Shannon entropy, the spectral sparsity index and the time domain stability index. A fault diagnosis module 806 is configured to identify the fault type based on the three-dimensional fault feature vector and the multi-level threshold value judgment strategy.

[0059] Further, In a battery pack multi-fault diagnosis system based on signal decomposition and entropy feature fusion, the adaptive decomposition module comprises a parameter optimization unit and a signal decomposition unit, wherein the parameter optimization unit uses the Schrodinger optimization algorithm to minimize the sum of the Shannon entropy as the objective function, and the signal decomposition unit performs variational mode decomposition using the optimized parameters.

[0060] In a battery pack multi-fault diagnosis system based on signal decomposition and entropy feature fusion, the fault diagnosis module comprises an anomaly detection unit, a sensor fault identification unit and a fault distinguishing unit, which sequentially perform multi-level judgment.

[0061] The third aspect of the technical scheme of the present application provides a computer readable storage medium having instructions stored thereon, which, when executed by one or more processors, cause the processor to perform the battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion as described in the first aspect.

[0062] The embodiment can divide the method into functional modules according to the method examples described above. For example, each functional module can be used, or two or more functions can be integrated into one processing module. The integrated module can be implemented in the form of hardware. It should be noted that the division of modules in the embodiment is illustrative, and is only a logical functional division. In actual implementation, another division mode can be used.

[0063] When each functional module is divided according to each function, the method can include a data acquisition module, a signal preprocessing module, an adaptive decomposition module, a signal reconstruction module, a feature extraction module, a fault diagnosis module, and the like. It should be noted that all related contents of each step involved in the method embodiment can be referred to the function description of the corresponding functional module, and will not be described here.

[0064] The embodiment also provides a computer readable storage medium, which includes but is not limited to a disk memory, a CD-ROM, an optical memory, and the like, and stores computer program codes. When the computer program codes run on a computer, the computer program codes make the computer execute the related method steps to implement the battery pack multi-fault diagnosis method and system based on signal decomposition and entropy feature fusion provided in the embodiment.

[0065] The embodiment also provides a computer program product, which makes the computer execute the related steps to implement the battery pack multi-fault diagnosis method and system based on signal decomposition and entropy feature fusion provided in the embodiment when the computer program product runs on the computer. The beneficial effects of the embodiment can be referred to the beneficial effects of the corresponding method provided above, and will not be described here.

[0066] Through the description of the above embodiments, those skilled in the art can understand that, for the convenience and brevity of description, only the division of the above functional modules is used as an example. In actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.

[0067] In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the division of the apparatus embodiments is only a logical function division, and there can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another apparatus, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different apparatuses can be indirect couplings or communication connections through some interfaces, apparatuses or units, and can be in electrical, mechanical or other forms.

[0068] It should be noted that, in the description of the disclosure, the relative positional relationship indicated by the terms such as "upper", "lower", "front", "back", "left" and "right" is based on the positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the indicated position or element must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as a limitation of the disclosure.

[0069] The above merely illustrates the embodiments of the present disclosure, and is not intended to limit the present disclosure. The present disclosure can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present disclosure shall be included in the scope of the claims of the present disclosure.

Claims

1. A battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion, characterized in that, The method comprises the following steps: Real-time acquisition of voltage data of each single battery in the battery pack, and segmented processing of the voltage time series data by using a sliding window mechanism based on the Fibonacci sequence; Adaptive optimization of key parameters of variational mode decomposition by using the Schrödinger optimization algorithm, adaptive decomposition of the voltage signal in each window, and obtaining of a plurality of intrinsic mode functions; Calculation of the Shannon entropy value of each intrinsic mode function, selection of a preset number of representative modal components according to the Shannon entropy characteristic contribution, superposition and reconstruction to obtain a reconstructed voltage signal; Extraction of the Shannon entropy, spectral sparsity index and time domain stability index of the reconstructed voltage signal to form a three-dimensional fault feature vector; Based on the three-dimensional fault feature vector, a multi-level threshold judgment strategy is used to identify connection faults, short circuit faults and sensor faults.

2. The battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion according to claim 1, characterized in that: The adaptive optimization of key parameters of variational mode decomposition by using the Schrödinger optimization algorithm comprises: Initialization of a population, each individual representing a group of modal numbers and penalty coefficients of variational mode decomposition; Minimization of the sum of the Shannon entropies of all intrinsic mode functions as the fitness function, and iterative updating of the parameter combination by using the Schrödinger optimization algorithm; Output of the optimal parameter combination that minimizes the fitness function, for variational mode decomposition of the voltage signal.

3. The battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion according to claim 2, characterized in that: The fitness function is defined as: ； ； wherein, is the energy probability distribution of the jth IMF component at the i point, is the jth intrinsic mode function, is the number of intervals dividing the signal amplitude range.

4. The battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion according to claim 1, characterized in that: The calculation of the spectral sparsity index comprises: performing a fast Fourier transform on the reconstructed voltage signal to obtain a power spectrum ; Normalization of the power spectrum into a probability distribution, denoted as: ； Calculation of the Shannon entropy of the probability distribution as the spectral sparsity index, denoted as: 。 5. The battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion according to claim 1, characterized in that: The calculation of the time domain stability index comprises: segmenting the reconstructed voltage signal and calculating a root mean square value of each segment as a sequence of segment energies , wherein ; Linear regression on the segmented energy sequence to get the slope absolute value and the residual sum of squares ; Calculation of the time domain stability index based on the absolute value of the slope and the residual sum of squares, is represented as: .

6. The battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion according to claim 1, characterized in that: The multi-level threshold judgment strategy comprises: Firstly, it is judged whether the Shannon entropy of each single battery exceeds the normal threshold value, and if it exceeds, it is determined that there is an abnormality; For the abnormal battery, it is judged whether the spectral sparsity index and the time domain stability index simultaneously fall below the sensor fault threshold value, and if so, it is determined as a sensor fault; If it is not a sensor fault, a dynamic reference is set based on the average values of the spectral sparsity index and the time domain stability index of the normal single batteries, and the deviation degree of the abnormal battery features from the dynamic reference is compared to distinguish the short circuit fault and the connection fault.

7. A battery pack multi-fault diagnosis system based on signal decomposition and entropy feature fusion, characterized in that, The method comprises the following steps: A data acquisition module is used to real-time acquisition of voltage data of each single battery in the battery pack; A signal preprocessing module is used to segmented processing of the voltage time series data by using a sliding window mechanism based on the Fibonacci sequence; An adaptive decomposition module is configured to optimize variational modal decomposition parameters using a Schrödinger optimization algorithm and decompose the voltage signal; A signal reconstruction module is configured to select an eigenmode function based on a Shannon entropy feature contribution degree and reconstruct the signal; A feature extraction module is configured to extract a Shannon entropy, a spectral sparsity index, and a time-domain stability index; A fault diagnosis module is configured to identify a fault type based on a three-dimensional fault feature vector and a multi-level threshold value judgment strategy.

8. The battery pack multi-fault diagnosis system based on signal decomposition and entropy feature fusion according to claim 7, characterized in that The adaptive decomposition module comprises a parameter optimization unit and a signal decomposition unit, wherein the parameter optimization unit uses a Schrödinger optimization algorithm to minimize the sum of Shannon entropies as an objective function, and the signal decomposition unit performs variational modal decomposition using the optimized parameters.

9. The battery pack multi-fault diagnosis system based on signal decomposition and entropy feature fusion according to claim 7, characterized in that, The fault diagnosis module comprises an anomaly detection unit, a sensor fault identification unit, and a fault differentiation unit, which sequentially perform multi-level judgments.

10. A computer-readable storage medium, characterized in that, An apparatus has instructions stored thereon that, when executed by one or more processors, cause the processors to perform the battery pack multi-fault diagnosis method based on signal decomposition and entropy feature fusion according to any one of claims 1-6.

Images