A battery system fault detection method and system

Abstract

This invention discloses a battery system fault detection method and system, relating to the field of battery fault detection technology. The method includes the following steps: collecting performance data during the stable operation phase of the battery and generating a degradation evolution baseline trajectory; collecting real-time performance data during the continuous operation phase of the battery and generating the current degradation evolution trajectory; performing trend decoupling analysis on the current degradation evolution trajectory using a variational mode decomposition algorithm and determining its consistency with the degradation evolution baseline trajectory to obtain potential abnormal evolution behavior; analyzing the potential abnormal evolution behavior using a deviation algorithm to obtain anomalous evolution characteristics and anomalous evolution trends; determining the existence of a fault based on the anomalous evolution characteristics, determining the fault evolution stage based on the anomalous evolution trend, and obtaining the fault detection result based on the fault existence and fault evolution stage. This invention generates fault detection results through decoupling analysis, thereby achieving fault detection in battery systems.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery fault detection, in particular to a battery system fault detection method and system. BACKGROUND

[0002] In practical application scenarios such as new energy vehicles and energy storage power stations, the health status of a lithium ion battery system directly determines the operation safety and service life of the equipment. Under long-term vehicle working conditions, the battery pack will exhibit nonlinear long-term degradation characteristics due to internal factors such as electrode material aging, electrolyte decomposition, and SEI film growth. Meanwhile, influenced by external factors such as driving behavior fluctuations, environmental temperature changes, and charging rate differences, the degradation trajectories of core parameters such as voltage and capacity inevitably mix short-term working condition disturbances and potential fault signals. Therefore, by comparing the consistency of the real-time degradation characteristics of the battery with the health benchmark, the abnormal evolution trend can be accurately identified, which is the core technical direction of early warning of battery faults and is also the key to ensuring the reliable operation of the battery system.

[0003] The traditional battery system fault detection method usually uses a current degradation evolution trajectory direct comparison method. This method first constructs a degradation evolution benchmark trajectory containing parameters such as voltage and capacity based on long-term operation data of a healthy battery, and clearly defines the normal evolution law of the healthy battery. Then, the current degradation evolution trajectory of the battery to be detected is collected in real time, and the overall similarity between the current degradation evolution trajectory and the degradation evolution benchmark trajectory is calculated directly. The consistency level of the battery to be detected with the health status is judged by a pre-set threshold, and then abnormal conditions are identified.

[0004] However, the current degradation evolution trajectory used in the current degradation evolution trajectory direct comparison method is a mixture of three types of characteristics: change direction, change speed, and change persistence. These three types of characteristics are not decoupled specifically, resulting in that the comparison object of the consistency judgment contains instantaneous characteristics of short-term disturbances and mutation characteristics of abnormal trends, which ultimately causes deviation in the consistency judgment result and affects the accuracy and reliability of fault detection. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a battery system fault detection method and system to solve the problems in the background art.

[0006] To achieve the above purpose, the present application is implemented by the following technical scheme: a battery system fault detection method and system, comprising the following steps: Step S1: collecting performance data in a stable running phase of the battery and generating a degradation evolution benchmark trajectory; collecting real-time performance data in a continuous running phase of the battery and generating a current degradation evolution trajectory; Step S2: Use the variational mode decomposition algorithm to perform trend decoupling analysis on the current degradation evolution trajectory, and make a consistency judgment with the degradation evolution baseline trajectory to obtain potential abnormal evolution behavior; Step S3: Analyze the potential anomalous evolutionary behavior using the deviation algorithm to obtain anomalous evolutionary characteristics and anomalous evolutionary trends; Step S4: Based on the anomalous evolution characteristics, determine the existence of potential fault risks; based on the abnormal evolution trend, determine the fault evolution stage of potential fault risks; and based on the existence of faults and the fault evolution stage, obtain fault detection results.

[0007] Preferably, generating a baseline trajectory for degradation and evolution includes: After collecting performance data arranged by timestamps and preprocessing it, a set of standardized core performance data time series is obtained. ,in Represents the performance data at time r. The normalized value; The construction of the degradation and evolution baseline trajectory specifically selects sub-intervals representing historical stable operating stages from the standardized core performance data time series. and its corresponding parameter subsequence P( By performing curve fitting based on physical mechanisms on the parameter subsequence, a mathematical expression characterizing the inherent healthy aging pattern is extracted. The mathematical form of the degradation evolution baseline trajectory function is: ; in, is the baseline trajectory function for degradation and evolution; a is the coefficient describing the initial nonlinear change; b is the initial rate of change parameter; c is the slope of the long-term linear change; d is the fitting constant; t is the normalized discrete-time index. This set of fitting parameters {a,b,c,d} and the degradation evolution baseline trajectory function they determine Together, they constitute the baseline trajectory B of degradation and evolution; Current Degradation and Evolutionary Trajectory It is directly defined by the standardized core performance data time series P(T); it adopts the same data preprocessing process, time normalization starting point, and performance data standardization and extraction method as when constructing the degradation and evolution baseline trajectory B.

[0008] Preferably, the trend decoupling analysis of the current degradation evolution trajectory using the variational mode decomposition algorithm includes: The variational mode decomposition method is used to adaptively decompose the signal of the current degenerate evolution trajectory. The variational mode decomposition is achieved by solving a constrained variational problem, and the mathematical expression is: ; in, It is the k-th eigenmode function component; It is the center frequency of the k-th component; It is the Dirac function; j is the imaginary unit; It is a time partial derivative operator; This represents the convolution operation; The L2 norm is used to measure bandwidth; f is the signal of the current degenerate evolution trajectory of the input; k is the index of the intrinsic mode function components; By solving this constrained variational problem, a set of eigenmode function components is obtained { }

[0009] Preferably, consistency assessment with the baseline trajectory of degenerative evolution reveals potential anomalous evolutionary behaviors, including: After completing the variational mode decomposition, the eigenmode function components obtained from the decomposition are... (i) Used for consistency analysis; A set of comparable reference sequences at the same time point is generated based on the degradation and evolution baseline trajectory B. The reference sequence represents the ideal evolutionary behavior that should be exhibited under fault-free conditions; the intrinsic mode function components are calculated. (i) and reference sequence Consistency coefficient in terms of direction of change, rate of change, and duration of change; Consistency coefficient of change direction Calculated using Pearson correlation coefficient; rate of change consistency coefficient. The correlation coefficient of the instantaneous frequency difference sequence of the components is used for evaluation; the consistency coefficient of change persistence is also assessed. Evaluation is conducted by analyzing the attenuation characteristics of the component amplitude; Based on the consistency analysis results, if a certain intrinsic mode function component (i) Consistency coefficient of change direction Consistency coefficient of change rate Or the coefficient of consistency of change persistence If any one or more indicators in the intrinsic mode function are consistently below or above their preset normal threshold range, then the intrinsic mode function component is determined to carry abnormal evolutionary behavior, and it is decoupled from the normal aging trend and marked as potential abnormal evolutionary behavior.

[0010] Preferably, the deviation algorithm is used to analyze potential anomalous evolutionary behavior, and the resulting anomalous evolutionary features include: Anomalous modal components are the eigenmode functions that are marked as potential anomalous evolutionary behaviors; Anomalous evolution characteristics are quantified by the deviation of anomalous intensity, and the instantaneous intensity index based on anomalous modal components is calculated. Compared with the baseline intensity index determined by the baseline trajectory of degradation evolution Comparison; Anomalous evolutionary characteristics Further, by calculating the average value of the standardized anomaly intensity deviation within the observation window, it becomes a comprehensive measure of the severity of the anomaly over a period of time: ; in, It represents anomalous evolutionary characteristics; T is the length of the selected analysis time window; This is the starting point of the analysis.

[0011] Preferably, the abnormal evolutionary trends include: Abnormal evolution trend It is captured by calculating the rate of change of the anomalous intensity of the anomalous modal components. The mathematical expression is the rate of change of the anomalous evolution feature in the time dimension, as follows: ; in, Indicates an abnormal evolutionary trend; It is the time derivative operator.

[0012] Preferably, the presence of potential fault risks determined based on anomalous evolution characteristics includes: The basis for determining the existence of a fault is its anomalous evolution characteristics. Does it continuously exceed the preset anomaly detection threshold for Q consecutive analysis periods? If satisfied > If the continuity condition is not met, then the battery system is determined to be faulty.

[0013] Preferably, the failure evolution stages that determine potential failure risks based on abnormal evolution trends include: After confirming the existence of the fault, based on the abnormal evolution trend To quantify the dynamic evolution of the fault, thereby determining its evolution stage, if ≤ If so, the fault is determined to be in its early stage, characterized by a slow increase in the degree of abnormality; if < ≤ If so, the fault is determined to have entered the development stage, indicating that degradation has accelerated significantly; if > If this is the case, the fault is considered to have reached a critical stage, indicating that performance may soon undergo a rapid decline; among them, and The empirical threshold is obtained by performing regression analysis on degradation process data from known failure cases.

[0014] Preferably, the fault detection results obtained based on the fault presence and fault evolution stage include: After the fault determination is completed, a structural hierarchy orientation analysis is further performed; Obtain anomalous modal components that are marked as potentially anomalous evolutionary behaviors. (t); simultaneously, the time series of observation parameters for each module m are directly collected. ; By calculating abnormal mode components (t) and the time series of observation parameters for each module The correlation coefficient between the positioning To quantify the degree of linear correlation between them and locate the correlation coefficient. The calculation formula is defined as follows: ; in, It is the positioning correlation coefficient of the m-th module to be determined; (t) is the anomalous mode component, that is, the value of the labeled anomalous intrinsic mode function component at discrete time point t; The sampled values ​​of the observation parameters of the m-th module at the same time point t; It is the average value of the abnormal mode components at all N time points; is the average value of the time series of the observation parameters of the m-th module within the same time window; N is the total number of sampling points within the selected analysis time window; t is the time index; By comparing the positioning correlation coefficients of all modules The size of the correlation coefficient is used to determine the module with the largest correlation coefficient value as the most likely source of failure that causes system-level anomalies, thereby completing the structural hierarchy orientation. The final output is a structured fault detection result, which integrates fault presence, fault evolution stage, and structural hierarchy.

[0015] A battery system fault detection system includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0016] Beneficial effects This invention provides a battery system fault detection method, which involves machine learning and deep learning technologies, and has the following beneficial effects: (1) By employing variational mode decomposition for trend decoupling analysis, the nonlinear and non-stationary characteristics of battery degradation data can be accurately adapted, decomposing the complex current degradation evolution trajectory into corresponding pure-state intrinsic mode components, thus achieving accurate separation of the three types of features. This process effectively eliminates the interference of short-term operating condition disturbances on consistency judgment, making the normal degradation features used for subsequent comparison with the baseline trajectory purer, avoiding judgment bias caused by non-target signals from the source, and significantly improving the accuracy of consistency judgment.

[0017] (2) By introducing a parameter optimization mechanism, the method of determining variational mode decomposition parameters is improved, which solves the drawback of traditional variational mode decomposition relying on experience to set the number of modes K and the penalty factor α. By adaptively determining the optimal K value and combining it with the optimized α value, the parameters are accurately matched, so that the single-component modes decomposed by variational mode decomposition can more clearly characterize the core features of battery degradation, namely the monotonicity of the direction of change, the gradient of the rate of change, and the stability of the duration of change. This greatly enhances the trend recognition accuracy of single-component modes and provides a highly reliable data source for the subsequent extraction of anomalous evolution features.

[0018] (3) By adopting deviation quantification to identify anomalous evolution features, the method is more targeted and accurate. The relative difference between the decoupled pure-state features and the baseline trajectory is calculated, thereby realizing the quantitative characterization of anomalous evolution features. This indicator is a dimensionless parameter and is not affected by factors such as battery type and capacity specifications. It can accurately capture explicit anomalies such as sudden changes in the rate of change, as well as implicit anomalies such as gradual changes in the duration of change. It effectively distinguishes between the deterioration of true consistency and the false inconsistency caused by disturbances, and significantly improves the reliability and generalization ability of anomaly identification. Attached Figure Description

[0019] Figure 1 This is a flowchart of a battery system fault detection method proposed in this invention.

[0020] Figure 2 This invention provides a hierarchical diagram of potential abnormal evolution behavior obtained from a battery system fault detection method.

[0021] Figure 3 This is a hierarchical diagram of the fault detection results obtained by the battery system fault detection method proposed in this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figures 1-3 This invention provides a technical solution: a battery system fault detection method. Specifically, the following battery system fault detection method is provided; please refer to [link / reference]. Figure 1 The method includes the following steps: Step S1: Collect performance data during the stable operation phase of the battery and generate a baseline trajectory for degradation evolution; collect real-time performance data during the continuous operation phase of the battery and generate the current degradation evolution trajectory.

[0024] To establish an analytical benchmark that can distinguish between normal aging and abnormal degradation of battery systems, this step first standardizes and preprocesses the collected performance data, then uses the performance data from the historical stable operation phase after commissioning to construct a long-term degradation evolution benchmark trajectory, and on this basis, continuously characterizes the real-time performance data after entering normal operation to generate the current degradation evolution trajectory that can be directly used for comparative analysis.

[0025] Data preprocessing is a crucial step in ensuring the consistency and accuracy of subsequent analysis, and mainly includes time alignment, data cleaning, and normalization. First, performance data arranged by timestamps is collected using smart sensors deployed in the battery system. These smart sensors possess self-calibration, anti-interference, and digital output capabilities. The collected performance data includes, but is not limited to, total voltage. (t), total current (t) and representative temperature (t). After the intelligent sensor amplifies, filters, and performs analog-to-digital conversion on the original signal, it outputs a standardized digital signal. Then, the performance data is resampled and aligned with the time series at a uniform sampling time interval Δt to form a regular time point sequence. Subsequently, the output data of the smart sensors is cleaned using physical constraints or statistical methods to remove noise and transient interference. Module-level data (such as average voltage / temperature of each module) is collected simultaneously to prepare for constructing its baseline curve.

[0026] The purpose of data normalization is to eliminate the influence of dimensions and map performance data of different orders of magnitude to a comparable scale, thereby focusing on their long-term evolution trends. For directly measured voltage, current, and temperature parameters, normalization is performed separately: Voltage normalized value... ,in This refers to the rated voltage and the normalized value of the current. ,in This is the maximum permissible operating current; temperature normalized value. ,in The selected reference temperature (such as the average temperature under rated operating conditions).

[0027] More importantly, to characterize the overall degradation state of the battery system, it is necessary to extract and normalize the derived performance indicators from the performance data that can comprehensively reflect the degree of aging. Among them, the available capacity retention rate of the battery system under the standard charge-discharge cycle condition is one of the key indicators, which is usually obtained by analyzing the complete charge or discharge curve and directly expressed as a percentage relative to the initial rated capacity, that is, it is already in a normalized form. Another key indicator is the relative change rate of the DC internal resistance of the battery system , which can be calculated through the voltage-current response under specific pulse conditions and normalized to , where is the reference internal resistance measured at the initial operation stage, is the DC internal resistance calculated at time . After the above preprocessing, a set of clean and comparable standardized core performance data time series is obtained, where represents the normalized value of the r-th performance data at time .

[0028] To construct the degradation evolution reference trajectory, a sub-interval representing the historical stable operation stage is specifically selected from the time series of the standardized core performance data (where <N) and its corresponding parameter subsequence P( ). By performing curve fitting based on physical mechanisms on this parameter subsequence, a mathematical expression characterizing the inherent healthy aging law can be extracted. An empirical model widely applicable to describe the decline or growth behavior of key parameters such as the capacity retention rate and internal resistance growth rate of lithium-ion batteries during long-term operation is a mixed model including exponential initial change and linear long-term change. Taking the capacity retention rate as an example, the mathematical form of the degradation evolution reference trajectory function is: ; where, is the degradation evolution reference trajectory function, representing the fitted value of the performance parameter of the degradation evolution reference trajectory at any time t; a is the coefficient describing the initial non-linear change; b is the initial change rate parameter; c is the long-term linear change slope; d is the fitting constant to adjust the curve reference position; t is the normalized discrete time index, taking the equivalent standard cycle number of the battery. The model parameters a, b, c, and d are obtained through the historical data subset ( , P( The nonlinear least squares fitting of the curve yields a distinct physical meaning: a and b (usually b < 0) together describe the rapidly changing nonlinear components in the initial commissioning phase due to side reactions (such as SEI film growth); c (usually c < 0) describes the relatively constant linear slope that dominates long-term aging; d is a fitting constant used to adjust the baseline position of the curve. This set of fitting parameters {a, b, c, d} and the baseline trajectory function of degradation evolution they determine... Together, they constitute the degradation evolution baseline trajectory B, which quantifies the expected path of core performance evolution over time under fault-free conditions.

[0029] It should be noted that the aforementioned degradation evolution baseline trajectory function The aim is to characterize the healthy aging patterns of batteries throughout their main effective lifespan. The exponential term... The text describes the initial nonlinear decay caused by processes such as the growth of the solid electrolyte interface film; linear terms are also described. This describes the capacity decay phase dominated by linearity in the intermediate stage; the parameter c < 0 guarantees that the capacity decay is within the observation and fitting time range t ∈ [ , Within the historical stable operating phase (i.e., the period of stable operation), the function can accurately track the decay trend. When time t is extremely large, this degradation evolution baseline trajectory function... Mathematically, it tends towards negative infinity, but this does not represent its physical predictive significance. In practical battery systems, the battery reaches the end of its lifespan before its performance parameters approach their physical lower limit (e.g., 0%), and this degradation evolution baseline trajectory function... The application phase has ended. Therefore, the functional form is physically reasonable and valid within its applicable timescale.

[0030] In addition to obtaining the baseline degradation evolution trajectory, a comparable real-time current degradation evolution trajectory is needed to dynamically assess the current and near-term operational health status. Current Degradation Evolution Trajectory It is directly defined by the complete, identically preprocessed parameter sequence P(T). Specifically, the current degenerative evolution trajectory It is an ordered set of points {( ,p( ))|n=0,1,...,N}, where p( The selected data (such as capacity retention) comprehensively reflects the health status and represents real-time performance data. To ensure effective comparison with the degradation evolution baseline trajectory and to place both within the same analytical framework, the current degradation evolution trajectory employs the same data preprocessing procedures, time normalization starting point, and performance data standardization and extraction methods as the baseline trajectory. In this way, the baseline trajectory B provides a reference for how aging should occur, while the current degradation evolution trajectory... This reflects the actual aging process in real time, and the two are strictly aligned in terms of time scale and energy dimension, laying a directly comparable data foundation for trend decoupling and consistency analysis in subsequent steps.

[0031] Step S2: Use variational mode decomposition algorithm to perform trend decoupling analysis on the current degradation evolution trajectory, and make consistency judgment with the degradation evolution baseline trajectory to obtain potential abnormal evolution behavior.

[0032] This step, based on the current degradation evolution trajectory established in step S1, aims to perform trend decoupling analysis on the current degradation evolution trajectory through variational mode decomposition and compare its consistency with the baseline degradation evolution trajectory to distinguish between normal aging changes and abnormal fault-induced changes. This step first employs the variational mode decomposition method, utilizing a parameter optimization strategy to obtain the optimal number of modes K and the penalty factor. An adaptive decomposition is performed on the signal of the current degenerate evolution trajectory. Variational mode decomposition is achieved by solving a constrained variational problem, the mathematical expression of which is: ; in, It is the k-th eigenmode function component; It is the center frequency of the k-th component; The Dirac function is used to construct analytic signals; j is the imaginary unit. It is a time partial derivative operator; This represents the convolution operation; The L2 norm is used to measure bandwidth; f is the signal of the current degenerate evolution trajectory of the input; k is the index of the intrinsic mode function component. The optimized parameter K is used to specify the number of modes to be decomposed, i.e., the number of intrinsic mode function components. The bandwidth constraint, acting as a penalty factor, controls the component to ensure that the decomposition result accurately captures different frequency components in the signal. By solving this variational problem, a set of eigenmode function components can be obtained { } and the corresponding center frequency { }

[0033] It should be noted that the center frequency The center frequency is a key intermediate parameter in the decomposition process, representing the frequency domain energy centroid of each component. Iterative updates ensure that each intrinsic mode function (IMF) component has a compact bandwidth around its center frequency, effectively separating variations at different scales in the signal. Although the center frequency itself does not directly participate in subsequent consistency calculations, it indirectly improves the quality of the IMF components by optimizing their frequency domain positioning, providing clean and reliable input data for consistency analysis. This means that the center frequency's role is primarily in the decomposition stage, ensuring the accuracy of the decomposition, while the consistency analysis is based on high-quality IMF components.

[0034] After completing the variational mode decomposition, the intrinsic mode function components obtained from the decomposition are analyzed based on the trajectory of the performance parameters. (i) For consistency analysis. To this end, a set of comparable reference sequences at the same time point must first be generated based on the degradation evolution baseline trajectory B. For low-frequency modes that characterize long-term aging trends, the reference sequence It can be directly obtained through the baseline trajectory function of degradation evolution. The data are obtained by sampling on the time series. For other modes characterizing specific periodic fluctuations or noise, their reference sequences may be close to zero or a constant value, depending on the ideal behavior defined by the degradation evolution baseline trajectory. These reference sequences represent the ideal evolutionary behavior that should be exhibited under fault-free conditions. Subsequently, the eigenmode function components are calculated. (i) and reference sequence Consistency coefficient in terms of direction of change, speed of change, and duration of change.

[0035] Consistency analysis is based on three specific indicators: consistency coefficient of direction of change. The consistency coefficient of the rate of change was assessed using the Pearson correlation coefficient. The consistency coefficient of change persistence is evaluated through instantaneous frequency difference assessment. The evaluation is based on amplitude decay characteristics. Three specific indicators are used to quantitatively compare the similarity between the current degradation trajectory components and the baseline degradation trajectory components using mathematical formulas.

[0036] Consistency coefficient of change direction The Pearson correlation coefficient is used for calculation, and the formula is as follows: ; in, This represents the value of the k-th eigenmode function component of the current degradation trajectory after variational mode decomposition at time i; From the baseline trajectory function of degeneration and evolution The value of the generated reference sequence at time i; and These are the components of the intrinsic mode function. and reference sequence The mean; N is the total number of sampling points. The consistency coefficient in this direction of change. It measures the degree of linear correlation between the current degradation trajectory component and the baseline degradation trajectory component. The closer the value is to 1, the higher the consistency of the direction of change. When there is an abnormal deviation, the value decreases significantly.

[0037] Consistency coefficient of change rate The evaluation is performed by calculating the correlation coefficient of the instantaneous frequency difference sequence of the components. First, the instantaneous frequency is extracted from each eigenmode function component. (i), for example, obtained through the Hilbert transform. ,in It is the phase function of the component. Then, the difference sequence of instantaneous frequencies is calculated. Instantaneous frequency difference sequence corresponding to the degradation and evolution baseline trajectory sequence (i), where the instantaneous frequency difference sequence (i) is obtained by referring to the difference sequence of instantaneous frequencies. That is, through the smoothed degradation evolution baseline trajectory sequence (i) Obtained by approximation using the Hilbert transform. (The coefficient of consistency in rate of change is missing.) The formula is: ; in, It is the current intrinsic mode function component The instantaneous frequency difference sequence; It corresponds to the reference component. The reference instantaneous frequency difference sequence; and These are instantaneous frequency difference sequences (i) and the reference instantaneous frequency difference sequence The mean of the difference sequence is N-1, where N-1 is the length of the difference sequence. This rate of change consistency coefficient... The consistency of the rate of change is quantified. A value close to 1 indicates that the rate of change pattern matches the benchmark, while a large deviation indicates an anomaly.

[0038] Consistency coefficient of change The attenuation characteristics of the component amplitudes are evaluated by combining the correlation coefficient of the amplitude sequence with the attenuation trend. Let... The current component amplitude is obtained from the Hilbert transform envelope. This is the baseline amplitude. The coefficient for consistency of variation persistence. The formula is: ; in, It is the current intrinsic mode function component The amplitude sequence; It corresponds to the reference component. The reference amplitude sequence; and These are amplitude sequences and reference amplitude sequence The mean of the variation. The coefficient of consistency of this variation. Capture the persistence pattern of amplitude decay. A high value indicates a consistent decay pattern, while a low value indicates a persistence anomaly, such as premature decay or persistent oscillation.

[0039] To improve the accuracy of trend decoupling and avoid mode aliasing, this step introduces a parameter optimization strategy aimed at adaptively determining the optimal number of modes K and penalty factor for variational mode decomposition (VMD). This ensures that normal aging trends, abnormal evolution characteristics, and noise in the signal can be effectively separated into different intrinsic mode functions. Parameter optimization is achieved based on indices such as orthogonality, energy ratio, and variational energy entropy.

[0040] Orthogonality index The formula for identifying split components is as follows: ; in, (i) is the value of the k-th intrinsic mode function component at the i-th sampling point; n is the number of sampling points. An excessively large value of this index indicates that there is mode aliasing or splitting between components.

[0041] Energy percentage The formula used to distinguish between effective components and noise components is: ; in, It is the energy of the k-th component. E is the total signal energy. Components with excessively small ratios It is considered noise.

[0042] Variational energy entropy Used to optimize the secondary penalty factor Its calculation expression is: ; in, It represents the energy percentage of the kth component; It is the variational energy entropy of the k-th component. The value corresponding to the minimum VEE value This is the optimal value. Given the number of modes K, by calculating the variational energy entropy (VEE) corresponding to different penalty factor α values ​​and finding its minimum, the penalty factor α that minimizes the variational energy entropy VEE is determined as the optimal parameter under the current number of modes K. In actual optimization algorithms (such as gradient descent or grid search), this minimization process is achieved iteratively, with the following stopping condition: iteration stops when the adjustment step size of the penalty factor α tends to stabilize; at the same time, to prevent infinite loops, a maximum number of iterations is set as a safety guarantee for the entire search process.

[0043] By iteratively calculating these metrics, the optimal number of modes K and the penalty factor can be adaptively determined. These optimized parameters are directly fed back into the variational mode decomposition formula for re-decomposition, thereby improving the robustness and accuracy of the decomposition and enabling trend decoupling to better capture subtle anomalies. To adaptively determine the optimal number of modes K and penalty factor α for variational mode decomposition (VMD), a two-level optimization strategy based on index evaluation is adopted. The specific process is as follows: First, candidate ranges for the number of modes K and penalty factor α are defined to form a parameter search space. Then, for each candidate number of modes K, a search is performed within the range of penalty factor α, with the goal of finding the penalty factor α value that optimizes the overall evaluation index of the current decomposition. Specifically, for each set of parameters (K, α) to be evaluated, VMD decomposition is performed to obtain K intrinsic mode function components, and then the energy of each component and the orthogonality index between components are calculated simultaneously. Based on the calculated energy proportions of each component Furthermore, the variational energy entropy characterizing the sparsity of the energy distribution is obtained. Next, the penalty factor α that minimizes the variational energy entropy is selected as the locally optimal penalty factor for the current K value. (K), this step aims to obtain the purest and most concentrated mode decomposition results. This is achieved by calculating the local optimal penalty factor corresponding to the number of modes K for all candidate modes. After optimizing (K), the final decision-making stage weighs the orthogonality index of the decomposition results corresponding to the modality number K values ​​of each mode. (Measures the degree of modal aliasing) and the energy percentage of the top few primary modes. The sum (ensuring the main signal element is effectively extracted) is used to determine the globally optimal number of modes. Thus, the optimal parameter combination ( , ( It was determined that this combination can effectively separate different trend components while minimizing over-decomposition and mode aliasing.

[0044] Finally, based on the consistency analysis results, if a certain intrinsic mode function component (i) Consistency coefficient of change direction Consistency coefficient of change rate Or the coefficient of consistency of change persistence If any one or more indicators in the intrinsic mode function (IMF) consistently fall below (or exceed) its preset normal threshold range, then the IMF component is determined to carry abnormal evolutionary behavior, and is decoupled from the normal aging trend, marked as a potential abnormal evolutionary behavior. This process ensures early identification of abnormal changes and provides a high-precision trend separation basis for subsequent fault detection.

[0045] It should be noted that the preset normal threshold range is determined through statistical learning of the performance data accumulated by the battery system under long-term historical healthy operating conditions. Specifically, firstly, a large amount of battery performance degradation data under known fault-free operating conditions is collected, and variational mode decomposition is performed on it according to the method described in this invention to extract the consistency coefficients of the change direction, change rate, and change persistence of each intrinsic mode function component under different time windows; subsequently, these coefficient values ​​are statistically analyzed, and the percentile method is used to define the normal fluctuation boundary by taking the 5% to 95% percentile.

[0046] Step S3: Use the deviation algorithm to analyze the potential abnormal evolutionary behavior to obtain anomalous evolutionary characteristics and anomalous evolutionary trends.

[0047] This step, building upon the decoupled potential anomalous evolutionary behaviors identified in step S2, aims to quantify these behaviors using a deviation algorithm to extract their static severity and dynamic rate of change, thereby obtaining anomalous evolutionary characteristics and trends. This step focuses on deriving two key indicators from the anomalous modal components already labeled as potential anomalous evolutionary behaviors: anomalous evolutionary characteristics quantify the severity of the anomalous behavior through anomalous intensity deviation, and anomalous evolutionary trends capture the dynamic evolution speed of the anomalous behavior through the rate of change of anomalous intensity. These two key indicators are calculated based on the anomalous modal components themselves and can effectively identify the intensity and development momentum of anomalous behaviors.

[0048] First, when step S2 determines that the intrinsic mode function component carries abnormal evolutionary behavior and decouples it from the normal aging trend, marking it as potential abnormal evolutionary behavior, the intrinsic mode function is the abnormal mode component of this step.

[0049] Anomalous evolution characteristics are quantified by the deviation of anomalous intensity. This deviation reflects the degree to which the anomalous intensity carried by the anomalous modal component deviates from the background fluctuation intensity during normal operation. Its calculation is based on the instantaneous intensity index of the anomalous modal component. Compared with the baseline intensity index determined by the baseline trajectory of degradation evolution Comparison of the instantaneous intensity exponents of anomalous modal components. Defined as the square of the amplitude of this component, the mathematical expression is: ; in, It is the instantaneous intensity exponent of the anomalous mode component at time t, obtained by analyzing the anomalous mode component. (t) can be obtained directly by squaring it; (t) represents the anomalous modal component, derived from the potential anomalous evolutionary behavior in step S2, i.e., the labeled anomalous eigenmode function component. This instantaneous intensity index... The intensity of the potential anomalous evolutionary behavior at time t was calculated.

[0050] Benchmark Strength Index Then, by constructing the degradation evolution baseline trajectory function in step S1 By sampling at the same time point and calculating the square, the mathematical expression is: ; in, It is the baseline intensity index at time t during normal aging, obtained by analyzing the baseline trajectory function of degradation evolution. Calculate the value at the corresponding time point t and then square it. It is the degradation evolution baseline trajectory function obtained by fitting historical stable operating data in step S1. This baseline intensity index... This represents the energy evolution path that should occur under fault-free conditions.

[0051] Anomalous evolutionary characteristics Further, by calculating the average value of the standardized anomaly intensity deviation within the observation window, it becomes a comprehensive measure of the severity of the anomaly over a period of time: ; in, It is an anomalous evolutionary characteristic, which is a dimensionless quantity; T is the length of the selected analysis time window; This involves analyzing the initial time step; the standardization strength deviation of the integration operation at each time step. A cumulative average is performed. This standardized intensity deviation normalizes the absolute deviation of the anomalous intensity to the expected baseline intensity level, thus making the anomalous evolution characteristics... It can eliminate the influence of changes in overall intensity levels across different periods, directly reflecting the relative magnitude of the anomaly's severity. Anomalous evolution characteristics. The larger the value, the more significant the deviation of the intensity of the anomalous modal component from the normal baseline, and the more pronounced the anomalous evolution characteristics.

[0052] After calculating the anomalous evolution characteristics, the abnormal evolution trend This is captured by calculating the rate of change of anomalous intensity of anomalous modal components. The rate of change of anomalous intensity describes the rate of change of the anomalous evolutionary feature itself over time, reflecting whether the potential anomalous evolutionary behavior is intensifying, weakening, or remaining stable. Its mathematical expression is the rate of change of the anomalous evolutionary feature over the time dimension, as follows: ; in, Indicates an abnormal evolutionary trend; This is the time derivative operator. This formula calculates the rate of change of the standardized anomaly intensity deviation over time. In practical numerical calculations, it can be approximated by calculating the difference in this deviation between consecutive time points. Anomaly evolution trend. It can identify the dynamic characteristics of potential abnormal evolutionary behavior. Positive values ​​indicate that the severity of the abnormality is intensifying, while negative values ​​indicate that the severity of the abnormality is mitigating. The absolute value reflects the rate of change. For example, rapid positive changes may correspond to the initial stage of a sudden failure, while slow negative changes may indicate that the abnormality is gradually recovering.

[0053] Through the above calculations, anomalous evolutionary features were extracted from potential abnormal evolutionary behaviors. and abnormal evolution trend These two indicators quantify abnormal behavior from the perspectives of static intensity and dynamic change, respectively, providing a core data foundation for fault risk assessment and hierarchical location in subsequent steps.

[0054] Step S4: Based on the anomalous evolution characteristics, determine the existence of potential fault risks; based on the abnormal evolution trend, determine the fault evolution stage of potential fault risks; and based on the existence of faults and the fault evolution stage, obtain fault detection results.

[0055] This step uses the anomalous evolution features extracted in step S3. and abnormal evolution trend Using this as the core input, the system completes the final determination and precise location of fault risks, and outputs a fault detection result that includes the presence of faults, the stage of fault evolution, and the structural hierarchy.

[0056] First, the basis for determining the existence of a fault is the anomalous evolution characteristics. Does it continuously exceed the preset anomaly detection threshold for Q consecutive analysis periods? The anomaly detection threshold It is not arbitrarily set, but determined through statistical methods based on health status data accumulated over a long period of operation. Specifically, it is calculated from historical stable operating phases to determine the normal operating status. The values ​​are subjected to distribution analysis, and the 95th or 99th percentile is used as a threshold to ensure reliable capture of statistically significant deviations. If the following conditions are met... > If the continuity condition is not met, then the battery system is determined to be faulty.

[0057] After confirming the existence of the fault, the abnormal evolution trend calculated in step S3 is used as a basis. This is used to quantify the dynamic evolution of faults, thereby determining their evolution stages. The determination of the evolution stage is based on the abnormal evolution trend. This is achieved by comparing the thresholds with a set of thresholds determined based on degradation analysis of historical failure cases. Specifically, if... ≤ If so, the fault is determined to be in its early stage, characterized by a slow increase in the degree of abnormality; if < ≤ If so, the fault is determined to have entered the development stage, indicating that degradation has accelerated significantly; if > If this is the case, the fault has reached a critical stage, indicating that performance may soon deteriorate rapidly. and These are empirical thresholds obtained through regression analysis of degradation process data from known failure cases. They transform abstract evolutionary trends into specific, calculable criteria.

[0058] To achieve precise fault location, this step, after fault determination, further performs structural hierarchy orientation analysis. The core principle is that abnormal evolutionary behavior at the system level must be triggered by a fault in a lower-level module; therefore, the system-level abnormal signal and the directly observed parameters of the faulty module should have the strongest temporal synchronization. Specifically, the abnormal mode components marked as potential abnormal evolutionary behaviors in step S2 are first acquired. (t), which represents the core anomaly pattern extracted from the overall performance data. Simultaneously, the time series of observation parameters for each module m are directly collected. For example, the average voltage or average temperature of the module are raw signals that are directly monitored in the battery system and do not require complex preprocessing.

[0059] Next, the components of the anomalous intrinsic mode function are calculated. (t) and the time series of observation parameters for each module The correlation coefficient between the positioning This location correlation coefficient is used to quantify the degree of linear correlation between them. The calculation formula is defined as follows: ; in, It is the positioning correlation coefficient of the m-th module to be determined; (t) is the abnormal mode component, which comes from the variational mode decomposition result of step S2; The sampled values ​​of the observed parameters (such as voltage) of the m-th module at the same time point t; It is the average value of the abnormal mode components at all N time points; is the average value of the observation sequence of the m-th module within the same time window; N is the total number of sampling points within the selected analysis time window; t is the time index. The calculated location correlation coefficient... It is a dimensionless number between -1 and 1. The closer its value is to 1, the higher the synchronization between the module's operational fluctuations and system-level abnormal modes.

[0060] It should be noted that all performance data and module observation data to be analyzed are timestamped using the same clock source. Since the sampling intervals for VMD decomposition processing and the original data acquisition may differ, interpolation methods are needed to unify the data sequences to the same time series. Specifically, a baseline time series can be selected (e.g., the sampling time points of performance parameters), and then linear interpolation or spline interpolation methods can be used to unify the observation parameter sequences of each module. Resampling to this reference time point ensures that at any calculation time t, the components... (t) and Both represent the state at the same physical moment. After this synchronization process, both can be used to calculate the positioning correlation coefficient based on the same time index t.

[0061] By comparing the positioning correlation coefficients of all modules The size of the correlation coefficient is used to determine the module with the largest correlation coefficient value as the most likely source of failure that could cause system-level anomalies, thereby completing the structural hierarchy pointing.

[0062] Finally, by synthesizing all the above analyses, a structured fault detection result is output. This result integrates: fault presence (whether an alarm is triggered), fault evolution stage (early / development / critical stage), and structural hierarchy (e.g., "Module M3"). This comprehensive output allows the detection conclusion to go beyond a simple "fault" alarm, providing a quantitative basis for advanced management functions such as preventative maintenance, load adjustment, and system life prediction, thereby significantly enhancing the engineering practical value of the method.

[0063] This technical solution collects performance data from battery stable and continuous operation phases, preprocesses it to construct a degradation baseline trajectory and the current degradation trajectory, employs parameter-optimized variational mode decomposition to adaptively separate three types of pure-state features: direction of change, velocity, and persistence, and combines consistency judgment to screen potential anomalies. A deviation algorithm quantifies anomalous evolution characteristics and abnormal trends, ultimately achieving full-dimensional identification of fault presence, evolution stage, and structural hierarchy. This solution solves the judgment bias problem caused by trajectory aliasing in traditional methods, improves the accuracy and generalization ability of anomaly identification, and provides reliable technical support for early warning and maintenance of battery system faults.

[0064] The present invention also protects a battery system fault detection system, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0066] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

Claims

1. A method for detecting battery system faults, characterized in that, Includes the following steps: Step S1: Collect performance data during the stable operation phase of the battery and generate a baseline trajectory for degradation evolution; collect real-time performance data during the continuous operation phase of the battery and generate the current degradation evolution trajectory. Step S2: Use the variational mode decomposition algorithm to perform trend decoupling analysis on the current degradation evolution trajectory, and make a consistency judgment with the degradation evolution baseline trajectory to obtain potential abnormal evolution behavior; Step S3: Analyze the potential anomalous evolutionary behavior using the deviation algorithm to obtain anomalous evolutionary characteristics and anomalous evolutionary trends; Step S4: Based on the anomalous evolution characteristics, determine the existence of potential fault risks; based on the abnormal evolution trend, determine the fault evolution stage of potential fault risks; and based on the existence of faults and the fault evolution stage, obtain fault detection results.

2. The battery system fault detection method according to claim 1, characterized in that, Generate a baseline trajectory for degradation and evolution, including: After collecting performance data arranged by timestamps and preprocessing it, a set of standardized core performance data time series is obtained. ,in Represents the performance data at time r. The normalized value; The construction of the degradation and evolution baseline trajectory specifically selects sub-intervals representing historical stable operating stages from the standardized core performance data time series. and its corresponding parameter subsequence P( By performing curve fitting based on physical mechanisms on the parameter subsequence, a mathematical expression characterizing the inherent healthy aging pattern is extracted. The mathematical form of the degradation evolution baseline trajectory function is: ； in, is the baseline trajectory function for degradation and evolution; a is the coefficient describing the initial nonlinear change; b is the initial rate of change parameter; c is the slope of the long-term linear change; d is the fitting constant; t is the normalized discrete-time index. This set of fitting parameters {a,b,c,d} and the degradation evolution baseline trajectory function they determine Together, they constitute the baseline trajectory B of degradation and evolution; Current Degradation and Evolutionary Trajectory It is directly defined by the standardized core performance data time series P(T); it adopts the same data preprocessing process, time normalization starting point, and performance data standardization and extraction method as when constructing the degradation and evolution baseline trajectory B.

3. The battery system fault detection method according to claim 2, characterized in that, The variational mode decomposition algorithm is used to perform trend decoupling analysis on the current degradation evolution trajectory, including: The variational mode decomposition method is used to adaptively decompose the signal of the current degenerate evolution trajectory. The variational mode decomposition is achieved by solving a constrained variational problem, and the mathematical expression is: ； in, It is the k-th eigenmode function component; It is the center frequency of the k-th component; It is the Dirac function; j is the imaginary unit; It is a time partial derivative operator; This represents the convolution operation; The L2 norm is used to measure bandwidth; f is the signal of the current degenerate evolution trajectory of the input; k is the index of the intrinsic mode function components; By solving the constrained variational problem, a set of eigenmode function components { } 4. The battery system fault detection method according to claim 3, characterized in that, Consistency assessment with the baseline trajectory of degenerative evolution reveals potential anomalous evolutionary behaviors, including: After completing the variational mode decomposition, the eigenmode function components obtained from the decomposition are... (i) Used for consistency analysis; A set of comparable reference sequences at the same time point is generated based on the degradation and evolution baseline trajectory B. The reference sequence represents the ideal evolutionary behavior that should be exhibited under fault-free conditions; the intrinsic mode function components are calculated. (i) and reference sequence Consistency coefficient in terms of direction of change, rate of change, and duration of change; Consistency coefficient of change direction Calculated using Pearson correlation coefficient; rate of change consistency coefficient. The correlation coefficient of the instantaneous frequency difference sequence of the components is used for evaluation; the consistency coefficient of change persistence is also assessed. Evaluation is conducted by analyzing the attenuation characteristics of the component amplitude; Based on the consistency analysis results, if a certain intrinsic mode function component (i) Consistency coefficient of change direction Consistency coefficient of change rate Or the coefficient of consistency of change persistence If any one or more indicators in the intrinsic mode function are consistently below or above their preset normal threshold range, then the intrinsic mode function component is determined to carry abnormal evolutionary behavior, and it is decoupled from the normal aging trend and marked as potential abnormal evolutionary behavior.

5. A battery system fault detection method according to claim 4, characterized in that, The deviation algorithm is used to analyze potential anomalous evolutionary behavior, resulting in anomalous evolutionary features, including: Anomalous modal components are the eigenmode functions that are marked as potential anomalous evolutionary behaviors; Anomalous evolution characteristics are quantified by the deviation of anomalous intensity, and the instantaneous intensity index based on anomalous modal components is calculated. Compared with the baseline intensity index determined by the baseline trajectory of degradation evolution Comparison; Anomalous evolutionary characteristics Further, by calculating the average value of the standardized anomaly intensity deviation within the observation window, it becomes a comprehensive measure of the severity of the anomaly over a period of time: ； in, It represents anomalous evolutionary characteristics; T is the length of the selected analysis time window; This is the starting point of the analysis.

6. A battery system fault detection method according to claim 5, characterized in that, Abnormal evolutionary trends were observed, including: Abnormal evolution trend It is captured by calculating the rate of change of the anomalous intensity of the anomalous modal components. The mathematical expression is the rate of change of the anomalous evolution feature in the time dimension, as follows: ； in, Indicates an abnormal evolutionary trend; It is the time derivative operator.

7. A battery system fault detection method according to claim 6, characterized in that, The existence of potential fault risks is determined based on anomalous evolution characteristics, including: The basis for determining the existence of a fault is its anomalous evolution characteristics. Does it continuously exceed the preset anomaly detection threshold for Q consecutive analysis periods? If satisfied > If the continuity condition is not met, then the battery system is determined to be faulty.

8. A battery system fault detection method according to claim 7, characterized in that, Based on the judgment of abnormal evolution trends, the failure evolution stages of potential failure risks are identified, including: After confirming the existence of the fault, based on the abnormal evolution trend To quantify the dynamic evolution of the fault, thereby determining its evolution stage, if ≤ If so, the fault is determined to be in its early stage, characterized by a slow increase in the degree of abnormality; if < ≤ If so, the fault is determined to have entered the development stage, indicating that degradation has accelerated significantly; if > If this is the case, the fault is considered to have reached a critical stage, indicating that performance may soon undergo a rapid decline; among them, and The empirical threshold is obtained by performing regression analysis on degradation process data from known failure cases.

9. A battery system fault detection method according to claim 8, characterized in that, The fault detection results are obtained based on the existence and evolution stages of the fault, including: After the fault determination is completed, a structural hierarchy orientation analysis is further performed; Obtain anomalous modal components that are marked as potentially anomalous evolutionary behaviors. (t); simultaneously, the time series of observation parameters for each module m are directly collected. ; By calculating abnormal mode components (t) and the time series of observation parameters for each module The correlation coefficient between the positioning To quantify the degree of linear correlation between them and locate the correlation coefficient. The calculation formula is defined as follows: ； in, It is the positioning correlation coefficient of the m-th module to be determined; (t) is the anomalous mode component, that is, the value of the labeled anomalous intrinsic mode function component at discrete time point t; The sampled values ​​of the observation parameters of the m-th module at the same time point t; It is the average value of the abnormal mode components at all N time points; is the average value of the time series of the observation parameters of the m-th module within the same time window; N is the total number of sampling points within the selected analysis time window; t is the time index; By comparing the positioning correlation coefficients of all modules The size of the correlation coefficient is used to determine the module with the largest correlation coefficient value as the most likely source of failure that causes system-level anomalies, thereby completing the structural hierarchy orientation. The final output is a structured fault detection result, which integrates fault presence, fault evolution stage, and structural hierarchy.

10. A battery system fault detection system, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-9.

Images