Data-driven power battery safety risk assessment method and system thereof

Abstract

The present invention discloses a data-driven power battery safety risk assessment method and system, specifically relating to the technical field of battery safety monitoring, and is used to solve the problem of false alarms and missed detections caused by the confusion between the electrochemical relaxation effect and the characteristics of real fault signals in the existing fault detection methods during the static stage; through multi-source data acquisition based on voltage, temperature, and state of charge during the static stage, the extreme value point of the current slope is identified by weighted correction of the charge and discharge mode, the time-delay mapping relationship with the geometric shape mutation point of the voltage curve is established, abnormal inflection points are screened, and the fractal dimension attenuation characteristics are quantified; a dynamic threshold is generated by fusing the abnormal distribution density and the material damage index, and the composite criterion analysis is carried out by combining the frequency-domain energy offset characteristics extracted by time-frequency transformation and the thermal conduction time-delay phase difference, effectively stripping the relaxation interference signal and locking the hidden danger characteristics, realizing the full-dimensional safety state assessment from micro-defect evolution to macro-parameter abnormality, and improving the reliability of power battery fault detection under complex working conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of battery safety monitoring, and more specifically, to a data-driven power battery safety risk assessment method and system thereof. Background Art

[0002] The widespread application of power batteries in the field of electric vehicles has made their safety risk assessment a core concern of the industry; in the existing technology, safety monitoring of batteries in the static stage mainly relies on threshold judgment methods of parameters such as voltage and temperature, that is, by setting a fixed change rate threshold (such as voltage drop rate) to determine whether to trigger an abnormal alarm. This type of method assumes that parameter fluctuations in the static stage are only caused by self-discharge or environmental interference, and risk control can be achieved through smoothing filtering and static threshold rules; however, in complex environments and operating condition switching scenarios, the internal electrochemical relaxation effect of power batteries will dynamically evolve with factors such as temperature and historical load, resulting in a high similarity between the parameter change pattern in the static stage and the real fault signal, making it difficult to effectively distinguish the difference between the two.

[0003] The existing technology has not fully considered the dynamic offset characteristics of the electrochemical relaxation effect in complex environments and operating condition switching scenarios. Specifically, when the battery switches from high-load operation to a static state, its relaxation behaviors such as internal ion diffusion and interface charge redistribution are affected by the multiple coupling of temperature, charge state and aging degree, resulting in the evolution of relaxation signals and early fault signals (such as internal short circuit and lithium deposition) highly overlapping in the time and frequency domains. In scenarios such as high temperature and high charge state, normal relaxation can easily be misjudged as a fault, or real hidden dangers can be concealed, resulting in a safety blind spot where false alarms and missed detection risks coexist. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a data-driven power battery safety risk assessment method and system thereof to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] The data-driven power battery safety risk assessment method includes the following steps:

[0007] S1. Collect the voltage, temperature and charge state of the power battery in the static stage, and simultaneously obtain the last charge and discharge current value and charge and discharge mode;

[0008] S2, extracting the slope extreme value point of the last charge and discharge current value, weighting the slope extreme value point based on the charge and discharge mode, and identifying the geometric morphology mutation point timing of the voltage curve in the static stage;

[0009] S3. Establish the lag time mapping relationship between the slope extreme points and the geometric shape mutation points, and screen the abnormal inflection points with excessive time delays;

[0010] S4. Analyze the cumulative damage of historical working conditions to the microscopic defects of the electrode material based on the distribution density of abnormal inflection points and the fractal dimension decay characteristics of the voltage relaxation curve during the static stage, and generate a dynamic threshold interval in combination with the state of charge;

[0011] S5. Perform time-frequency transformation on the voltage during the static stage, extract the frequency-domain energy centroid offset and the time-domain fluctuation envelope, and extract the time-delay phase difference for the temperature during the static stage;

[0012] S6. Based on the judgment results of whether the time-domain fluctuation envelope is within the dynamic threshold interval and whether the time-delay phase difference is lower than the preset critical value, perform safety risk determination.

[0013] In a preferred embodiment, the voltage, temperature, and state of charge of the power battery during the static stage are collected, and the last charge and discharge current value and the charge and discharge mode are obtained synchronously, including:

[0014] S1-1. When the power battery enters the static stage, continuously collect the static voltage data at the first preset sampling frequency, continuously collect the static temperature data at the second preset sampling frequency, and obtain the state of charge data in real time through the battery management system;

[0015] S1-2. When the power battery ends the last charge and discharge, record the current value at the end of charge and discharge as the last charge and discharge current value, and obtain the charge and discharge mode classification identifier during the charge and discharge process. The charge and discharge mode classification identifier includes at least one of constant current charging, pulse discharging, and constant voltage charging;

[0016] S1-3. Align the timestamps of the static voltage data, static temperature data, and state of charge data with the timestamps of the last charge and discharge current value and the charge and discharge mode classification identifier to generate a time-synchronized static stage dataset.

[0017] In a preferred embodiment, extract the slope extreme points of the last charge and discharge current value, and perform weight correction on the slope extreme points based on the charge and discharge mode to identify the timing of the geometric shape mutation points of the voltage curve during the static stage, including:

[0018] S2-1. Perform mode adaptability screening on the slope change characteristics of the last charge and discharge current curve based on the charge and discharge mode classification identifier. Among them, the slope extreme points at the charge and discharge switching stage are retained in the pulse discharge mode, and the slope extreme points at the charging cut-off stage are retained in the constant current charging mode;

[0019] S2-2, dynamically weight the selected slope extreme points, adjust the weight coefficient of the extreme points in the pulse discharge mode according to the nonlinear relationship of the corresponding charge and discharge switching rate, and adjust the weight coefficient of the extreme points in the constant current charging mode according to the charging cut-off current gradient;

[0020] S2-3. Perform multi-scale curvature analysis on the geometric mutation points of the voltage curve in the static stage, extract the mutation points with abnormal curvature change direction and amplitude, and generate a time series of geometric mutation points.

[0021] In a preferred embodiment, a hysteresis time mapping relationship between the slope extreme point and the geometric morphology mutation point is established to screen abnormal inflection points with excessive hysteresis, including:

[0022] S3-1, matching the time difference between the weighted corrected slope extreme value time series and the geometric morphology mutation point time series, calculating the time difference between each slope extreme value point and the first subsequent geometric morphology mutation point, and generating the original time-lag data set;

[0023] S3-2, according to the charge and discharge mode classification identifier, the dynamic time lag matching window is set, the matching window in the pulse discharge mode is dynamically adjusted based on the inverse correlation between the charge and discharge switching rate and the relaxation rate, and the matching window in the constant current charging mode is dynamically adjusted based on the inverse correlation between the charge cut-off current gradient and the material stress release rate;

[0024] S3-3, filtering the original time-delay data in the dynamic time-delay matching window, removing the time difference data outside the window, and retaining the time-delay data in the window as the effective time-delay set;

[0025] S3-4. Based on the statistical distribution characteristics of the time-lag data of historical healthy batteries, the benchmark interval of the effective time-lag set is calculated, and the time-lag points that deviate from the benchmark interval are marked as abnormal inflection points.

[0026] In a preferred embodiment, according to the distribution density of abnormal inflection points and the fractal dimension attenuation characteristics of the voltage relaxation curve in the static stage, the cumulative damage of the microscopic defects of the electrode material caused by the historical working conditions is analyzed, and the dynamic threshold interval is generated in combination with the charge state, including:

[0027] S4-1. Count the frequency of abnormal inflection points within a unit time, and calculate the distribution density of abnormal inflection points based on the degree of concentration of abnormal inflection points on the voltage curve;

[0028] S4-2, calculating the fractal dimension decay rate of the voltage relaxation curve in the static stage, and extracting the average decrease rate of the fractal dimension over time as the fractal dimension decay characteristic;

[0029] S4-3. Quantify the cumulative damage degree of the historical working conditions to the microscopic defects of the electrode material based on the weighted fusion result of the abnormal inflection point distribution density and the fractal dimension decay characteristics;

[0030] S4-4. Compensate and correct the cumulative damage degree according to the state of charge to generate a dynamic threshold interval that changes non-linearly with the state of charge.

[0031] In a preferred embodiment, step S4-2 includes:

[0032] S4-2a. Calculate the fractal dimension sequence of the voltage relaxation curve in the rest stage. Calculate the fractal dimension values through multi-time window sliding to generate a decay sequence of the fractal dimension changing with time;

[0033] S4-2b. Based on the correlation between the stress release of the electrode material and the ion diffusion path, extract the dynamic coupling characteristics of the fractal dimension decline rate and the curvature change direction in the decay sequence. When the fractal dimension decline rate and the curvature change direction are in the same direction, it is marked as the fast decay section, and when they are in the opposite direction, it is marked as the slow decay section;

[0034] S4-2c. Statistically calculate the cumulative duration ratio of the fast decay section and the number of curvature jumps in the slow decay section to generate a quantization index of the fractal dimension decay characteristics.

[0035] In a preferred embodiment, perform time-frequency transformation on the voltage in the rest stage, extract the frequency-domain energy centroid offset and the time-domain fluctuation envelope, and extract the time-delay phase difference for the temperature in the rest stage, including:

[0036] S5-1. Perform wavelet packet transformation on the voltage data in the rest stage, divide the preset frequency bands and calculate the energy distribution of each preset frequency band, and extract the frequency-domain energy centroid offset as the difference between the centroid frequency at each moment and the reference frequency of the corresponding frequency band of the standard healthy battery;

[0037] S5-2. Filter the interference components of the frequency-domain energy centroid offset based on the dynamic threshold interval;

[0038] S5-3. Perform Hilbert transformation on the time-domain waveform of the voltage data to extract the time-domain fluctuation envelope as the average of the upper and lower envelopes of the voltage amplitude changing with time;

[0039] S5-4. Perform cross-correlation analysis on the temperature data sequence in the rest stage, calculate its time-delay phase difference relative to the envelope of the voltage curve, and the time-delay phase difference is the time offset corresponding to the peak of the cross-correlation function.

[0040] In a preferred embodiment, based on the judgment results of whether the time-domain fluctuation envelope is within the dynamic threshold interval and whether the time-delay phase difference is lower than the preset critical value, perform safety risk determination, including:

[0041] If the time-domain fluctuation envelope is within the dynamic threshold range and the time-delay phase difference is lower than the preset critical value, it is determined as the superimposed interference of the microscopic defects and relaxation effect of the electrode material; otherwise, match the frequency-domain energy centroid offset with the frequency-domain fingerprint library of the preset faults to trigger hierarchical alarms.

[0042] In a preferred embodiment, the construction logic of the frequency-domain fingerprint library of the preset faults is as follows: perform standardized charge and discharge tests on power batteries with known internal short-circuit faults and lithium plating fault types, collect voltage data during the static stage and extract the frequency-domain energy centroid offset characteristics. Among them, the internal short-circuit fault shows the characteristic of continuous positive offset of the energy centroid in the low-frequency band, and the lithium plating fault shows the characteristic of periodic fluctuation offset in the middle-frequency band; associate and map the characteristics of different fault modes with the energy distribution law of the corresponding frequency bands to form a frequency-domain fingerprint library with the fault type as the index and the frequency-domain energy centroid offset mode as the characteristic. The low-frequency band corresponds to the positive offset mode of the internal short-circuit fault, and the middle-frequency band corresponds to the periodic fluctuation mode of the lithium plating fault.

[0043] On the other hand, the present invention provides a data-driven power battery safety risk assessment system, including:

[0044] Static data acquisition module: collect the voltage, temperature and state of charge of the power battery during the static stage, and synchronously obtain the last charge and discharge current value and the charge and discharge mode;

[0045] Extreme value weight correction module: extract the slope extreme points of the last charge and discharge current value, correct the weights of the slope extreme points based on the charge and discharge mode, and identify the timing of the geometric shape mutation points of the voltage curve during the static stage;

[0046] Time-delay screening and analysis module: establish the lag time mapping relationship between the slope extreme points and the geometric shape mutation points, and screen the abnormal inflection points with excessive time delay;

[0047] Fractal dynamic threshold module: analyze the cumulative damage of the microscopic defects of the electrode material by historical working conditions according to the distribution density of the abnormal inflection points and the fractal dimension attenuation characteristics of the voltage relaxation curve during the static stage, and generate a dynamic threshold range in combination with the state of charge;

[0048] Time-frequency feature extraction module: perform time-frequency transformation on the voltage during the static stage, extract the frequency-domain energy centroid offset and the time-domain fluctuation envelope, and extract the time-delay phase difference of the temperature during the static stage;

[0049] Safety risk determination module: perform safety risk determination based on the judgment results of whether the time-domain fluctuation envelope is within the dynamic threshold range and whether the time-delay phase difference is lower than the preset critical value.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. Through the multi-dimensional signal collaborative analysis and dynamic threshold generation mechanism, the problem of confusion between relaxation effect and real fault signals in the safety risk assessment of power batteries during the static stage is effectively solved. By introducing a cross-scale correlation model of the evolution characteristics of microscopic defects in electrode materials and electrochemical dynamic responses, the cumulative damage of charge and discharge historical conditions to the internal structure of the battery is incorporated into the real-time risk assessment system, enabling the safety state criterion to accurately reflect the actual health of the battery. Through the fusion of multiple characteristics such as quantifying material damage by fractal dimension attenuation characteristics, capturing hidden faults by frequency-domain energy centroid shift, and identifying thermodynamic anomalies by time-delay phase difference, the discrimination sensitivity of fault signals under complex conditions is significantly improved, and stable and reliable detection performance can still be maintained in extreme scenarios such as high temperature and high state of charge;

[0052] 2. Based on the quantification results of material damage and the dynamic adjustment of the judgment boundary according to the real-time state of charge, the adaptive matching between the safety risk assessment model and the actual aging process of the battery is realized. Through weighted correction of charge and discharge modes, time-series analysis of geometric shape mutation points, and calculation of abnormal inflection point distribution density, a full-link monitoring logic from microscopic defect evolution to macroscopic parameter anomalies is constructed to ensure that early hidden dangers can be effectively identified before thermal runaway is triggered; The combination of time-frequency transformation and relaxation effect decoupling technology successfully strips the interference of normal electrochemical recovery process on monitoring signals, making the feature extraction of potential faults such as internal short circuit and lithium plating highly specific and enhancing the reliability of safety control throughout the life cycle of power batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a flowchart of the data-driven power battery safety risk assessment method of the present invention;

[0054] Figure 2 is a schematic structural diagram of the data-driven power battery safety risk assessment system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0056] Example 1: Figure 1 A data-driven power battery safety risk assessment method of the present invention is given, which includes the following steps:

[0057] S1. Collect the voltage, temperature and state of charge of the power battery during the static stage, and synchronously obtain the last charge and discharge current value and charge and discharge mode;

[0058] S2. Extract the slope extreme point of the last charge-discharge current value, correct the weight of the slope extreme point based on the charge-discharge mode, and identify the time sequence of the geometric shape mutation point of the voltage curve in the rest stage;

[0059] S3. Establish the lag time mapping relationship between the slope extreme point and the geometric shape mutation point, and screen out the abnormal inflection points with excessive time delay;

[0060] S4. Analyze the cumulative damage of historical working conditions to the microscopic defects of the electrode material according to the distribution density of the abnormal inflection points and the fractal dimension decay characteristics of the voltage relaxation curve in the rest stage, and generate a dynamic threshold interval in combination with the state of charge;

[0061] S5. Perform time-frequency transformation on the voltage in the rest stage, extract the frequency-domain energy center of gravity offset and the time-domain fluctuation envelope, and extract the time-delay phase difference of the temperature in the rest stage;

[0062] S6. Based on the judgment results of whether the time-domain fluctuation envelope is within the dynamic threshold interval and whether the time-delay phase difference is lower than the preset critical value, perform safety risk determination.

[0063] S1. Collect the voltage, temperature and state of charge of the power battery in the rest stage, and synchronously obtain the last charge-discharge current value and the charge-discharge mode, including:

[0064] S1-1. When the power battery enters the rest stage, continuously collect the rest voltage data through a voltage sensor at a first preset sampling frequency, continuously collect the rest temperature data through a temperature sensor at a second preset sampling frequency, and obtain the state of charge data in real time through the battery management system;

[0065] S1-2. When the power battery ends the last charge-discharge, record the current value at the end of the charge-discharge as the last charge-discharge current value, and obtain the charge-discharge mode classification identifier during the charge-discharge process. The charge-discharge mode classification identifier includes at least one of constant current charging, pulse discharging and constant voltage charging;

[0066] S1-3. Align the timestamps of the rest voltage data, the rest temperature data and the state of charge data with the timestamps of the last charge-discharge current value and the charge-discharge mode classification identifier to generate a time-synchronized rest stage data set.

[0067] When the power battery enters the rest stage, the rest voltage data is continuously collected by a voltage sensor at a first preset sampling frequency, where the first preset sampling frequency is set according to the battery type and application scenario. For example, for lithium-ion power batteries, the first preset sampling frequency is set to collect once per second; the rest temperature data is continuously collected by a temperature sensor at a second preset sampling frequency, and the second preset sampling frequency is set according to the sensitivity of temperature change. For example, for the high-temperature environment monitoring scenario, the second preset sampling frequency is set to collect once every 2 seconds; the state of charge data is obtained in real time through the battery management system. The battery management system calculates the state of charge by the voltage integration method or the open-circuit voltage method at a preset time interval and stores the state of charge data in association with the timestamps of the voltage data and the temperature data.

[0068] When the power battery ends the last charge and discharge, the current value at the end of the charge and discharge is recorded as the last charge and discharge current value, which specifically includes: triggering the current sensor to record the current value at the moment of disconnection when the charge and discharge circuit is disconnected, and marking this current value as the last charge and discharge current value; obtaining the charge and discharge mode classification identifier during the charge and discharge process, and the charge and discharge mode classification identifier is determined by parsing the charge and discharge control instructions of the battery management system. For example, when the continuous current fluctuation amplitude in the charge and discharge control instruction exceeds the preset fluctuation threshold, it is determined as the pulse discharge mode, and when the current instruction is a constant value, it is determined as the constant current charging or constant voltage charging mode, and the charge and discharge mode classification identifier is stored in association with the timestamp of the last charge and discharge current value.

[0069] Align the timestamps of the rest voltage data, the rest temperature data, and the state of charge data with the timestamps of the last charge and discharge current value and the charge and discharge mode classification identifier to generate a time-synchronized rest stage data set, which specifically includes: normalizing the timestamps of all data sources according to the system clock of the data acquisition device. For data with different sampling frequencies, the linear interpolation method is used to align the low-frequency sampled data with the high-frequency sampled data under the same time reference. For example, the temperature data sampled every 2 seconds is interpolated to generate equivalent data with a timestamp per second to make it consistent with the timestamp of the voltage data sampled per second; for the last charge and discharge current value and the charge and discharge mode classification identifier, the corresponding data entries are inserted on the normalized time axis according to their recording time points, and finally a synchronous data set including voltage, temperature, state of charge, the last charge and discharge current value, and the charge and discharge mode classification identifier is formed.

[0070] During the generation of the dataset in the rest stage, the settings of the first preset sampling frequency and the second preset sampling frequency are dynamically adjusted according to the battery type and application scenario. For example, for high-energy-density power batteries, the first preset sampling frequency is increased to 2 times per second to ensure that voltage transient changes are completely captured; for the analysis of the charge-discharge mode classification identifier, the preset fluctuation threshold in the charge-discharge control instruction is determined according to the percentage of the battery rated current. For example, when the current fluctuation amplitude exceeds 20% of the rated current and the duration is less than 1 second, it is determined as the pulse discharge mode; in the process of timestamp alignment, the specific implementation of the linear interpolation method includes: for the high-frequency timestamps between two adjacent low-frequency sampling points, the interpolation result is calculated according to the values of the front and back sampling points in proportion to time. For example, if the timestamps t1 and t3 of the temperature data correspond to the temperature values T1 and T3 respectively, then the interpolated temperature T2 at the moment t2 = (T1*(t3 - t2)+T3*(t2 - t1)) / (t3 - t1); for the insertion of the last charge-discharge current value, if its recording time point does not exactly match the existing timestamps, then add this data entry at the position of the closest high-frequency timestamp and mark it as the current value at the end of charge-discharge.

[0071] Through the above steps, each data item in the rest-stage dataset is strictly aligned in the time dimension, ensuring that the subsequent sequential correlation analysis of features such as slope extreme points and geometric shape mutation points has a consistent data basis; the clear analysis of the charge-discharge mode classification identifier provides a classification basis for the weight correction based on the charge-discharge mode in the subsequent steps; the selection of the linear interpolation method in the time synchronization process ensures the compatibility of data with different sampling frequencies and avoids feature extraction errors caused by time deviation. For example, misaligned timestamps may cause deviations in the calculation of the lag time mapping relationship between slope extreme points and geometric shape mutation points.

[0072] S2. Extract the slope extreme points of the last charge-discharge current value, correct the weights of the slope extreme points based on the charge-discharge mode, and identify the sequence of geometric shape mutation points of the voltage curve in the rest stage, including:

[0073] S2-1. Perform pattern adaptability screening on the slope change characteristics of the last charge-discharge current curve based on the charge-discharge mode classification identifier. Among them, the pulse discharge mode retains the slope extreme points in the charge-discharge switching stage, and the constant current charging mode retains the slope extreme points in the charging cut-off stage;

[0074] S2-2. Perform dynamic weight correction on the screened slope extreme points. The extreme points in the pulse discharge mode adjust the weight coefficient according to the non-linear relationship of their corresponding charge-discharge switching rate, and the extreme points in the constant current charging mode adjust the weight coefficient according to the charging cut-off current gradient;

[0075] S2-3. Perform multi-scale curvature analysis on the geometric shape mutation points of the voltage curve in the rest stage, extract the mutation points with abnormal curvature change directions and amplitudes, and generate a time series of geometric shape mutation points.

[0076] In the mode adaptability screening, when the charge-discharge mode classification identifier is the pulse discharge mode, extract the slope extreme points where the change rate of the current value exceeds twice the rated current change rate within a preset time window during the charge-discharge switching stage. The preset time window is dynamically adjusted according to the rated charge-discharge rate of the battery. For example, for a battery with a rated charge-discharge rate of 1C, the preset time window is set to 100 milliseconds. If it is detected that the current value drops from 50A to 0A within this window and the change rate exceeds 100A / s, record this moment as the slope extreme point. When the charge-discharge mode classification identifier is the constant current charging mode, extract the slope extreme point corresponding to the maximum change rate during the process of the current value dropping from a constant value to zero in the charging cut-off stage. Specifically, calculate the current difference between adjacent time points at intervals of 10 milliseconds during the current drop process, and select the time point with the largest current difference as the extreme point.

[0077] When performing dynamic weight correction on the screened slope extreme points, for the slope extreme points in the pulse discharge mode, calculate the weight coefficient according to the time length of the sudden drop of the current during the charge-discharge switching stage. The sudden drop time length is defined as the time required for the current to drop from 90% of the rated value to 10%. For example, when the sudden drop time is 50 milliseconds, the weight coefficient is set to 1.5, and when the sudden drop time is 200 milliseconds, the weight coefficient is set to 1.2. For the slope extreme points in the constant current charging mode, adjust the weight coefficient according to the absolute value of the current gradient in the charging cut-off stage. The absolute value of the current gradient is the maximum instantaneous change rate during the current drop process. For example, when the detected maximum instantaneous change rate is 5A per second decrease, the weight coefficient is set to 1.2, and when it is 2A per second decrease, it is set to 1.0.

[0078] When performing multi-scale curvature analysis on the geometric shape mutation points of the voltage curve in the rest stage, perform second-order difference calculation on the voltage data using sliding windows with different time spans. The time span is set according to the relaxation characteristics of the battery. For example, for a battery with a faster relaxation response, set a 1-second time window to detect local curvature, and for a battery with a slower relaxation, set a 5-second time window to detect the macroscopic curvature trend. Detect the reversal points of the curvature change direction and the events where the jump amplitude of the curvature value exceeds the preset curvature threshold at each scale. The curvature threshold is obtained based on the statistics of historical healthy battery data. For example, the curvature threshold is set to 0.1 in the 1-second time window and 0.25 in the 5-second time window. When performing timestamp intersection screening on the mutation points detected at different scales, only retain the mutation points that are identified in at least two adjacent scales. For example, if a certain time point is detected with a curvature direction reversal and an amplitude exceeding the threshold in both the 1-second and 2-second windows, it is determined as a valid mutation point.

[0079] Through the above steps, the selected slope extreme points are corrected by the mode adaptability weight to form a weighted extreme point sequence. For example, the extreme point with a sudden drop time of 50 milliseconds in the pulse discharge mode has a weight of 1.5, and the curvature mutation point detected in the 1-second time window in the corresponding static stage will be given priority in subsequent analysis; the mutation point time series sequence generated by the multi-scale curvature analysis is strictly aligned with the weighted extreme point sequence in the time dimension. For example, when the extreme point with a weight of 1.0 in the constant current charging mode is associated with the macro curvature mutation point detected in the 5-second time window, the standard time difference threshold in this mode is used for hysteresis matching.

[0080] S3. Establish the lag time mapping relationship between the slope extreme point and the geometric morphology mutation point, and screen the abnormal inflection points with excessive lag time, including:

[0081] S3-1, matching the time difference between the weighted corrected slope extreme value time series and the geometric morphology mutation point time series, calculating the time difference between each slope extreme value point and the first subsequent geometric morphology mutation point, and generating the original time-lag data set;

[0082] S3-2, according to the charge and discharge mode classification identifier, the dynamic time lag matching window is set, the matching window in the pulse discharge mode is dynamically adjusted based on the inverse correlation between the charge and discharge switching rate and the relaxation rate, and the matching window in the constant current charging mode is dynamically adjusted based on the inverse correlation between the charge cut-off current gradient and the material stress release rate;

[0083] S3-3, filtering the original time-delay data in the dynamic time-delay matching window, removing the time difference data outside the window, and retaining the time-delay data in the window as the effective time-delay set;

[0084] S3-4. Based on the statistical distribution characteristics of the time-lag data of historical healthy batteries, the benchmark interval of the effective time-lag set is calculated, and the time-lag points that deviate from the benchmark interval are marked as abnormal inflection points.

[0085] The weighted corrected slope extreme point time series and the geometric morphology mutation point time series are matched with each other in time difference, specifically including: taking the weighted corrected slope extreme point timestamp as the benchmark, searching for the first mutation point after each extreme point in the geometric morphology mutation point time series, and calculating the time difference between the two. For example, an extreme point appears at 10:05:30, and the first mutation point appears at 10:06:05, then the time difference is 35 seconds. The time difference between all extreme points and the corresponding mutation points constitutes the original time-lag data set.

[0086] When setting the dynamic time-delay matching window according to the charge-discharge mode classification identifier, for the pulse discharge mode, the charge-discharge switching rate is defined as the rate at which the current drops from the rated value to zero, and the relaxation rate is defined as the average rate of change of the voltage curvature during the rest stage. The time length of the dynamic time-delay matching window is inversely proportional to the charge-discharge switching rate and directly proportional to the relaxation rate. For example, when the charge-discharge switching rate is high, the window time is shortened; when the relaxation rate is high, the window time is extended. For the constant current charging mode, the charging cut-off current gradient is defined as the maximum instantaneous change rate of the current drop during the end stage of charging, and the material stress release rate is defined as the average rate of change of the voltage curve curvature during the rest stage. The time length of the dynamic time-delay matching window is inversely proportional to the charging cut-off current gradient and directly proportional to the material stress release rate. For example, when the charging cut-off current gradient is large, the window time is shortened; when the material stress release rate is high, the window time is extended.

[0087] When screening the original time-delay data within the dynamic time-delay matching window, for the time-delay data in the pulse discharge mode, only the data with a time difference less than or equal to the window time length is retained. For example, when the window time is set to 60 seconds, the data points with a time difference exceeding 60 seconds are excluded; for the time-delay data in the constant current charging mode, the data points with a time difference within the window time range are retained. For example, when the window time is set to 120 seconds, only the data with a time difference between 0 and 120 seconds is retained.

[0088] When calculating the reference interval based on the statistical distribution characteristics of the time-delay data of historical healthy batteries, it specifically includes: collecting the time-delay data of multiple groups of healthy batteries under the same charge-discharge mode, and statistically calculating the mean and standard deviation of their time differences. The reference interval is set as the mean plus or minus three times the standard deviation. For the pulse discharge mode, if the mean of the historical data is 50 seconds and the standard deviation is 10 seconds, the reference interval is 20 seconds to 80 seconds; for the constant current charging mode, if the mean is 100 seconds and the standard deviation is 20 seconds, the reference interval is 40 seconds to 160 seconds. Comparing the time difference data in the effective time-delay set with the reference interval corresponding to the mode, if the time difference is lower than the lower limit of the interval or higher than the upper limit of the interval, it is marked as an abnormal inflection point. For example, if a time-delay data in the pulse discharge mode is 15 seconds, which is lower than the lower limit of the reference interval of 20 seconds, it is determined as an abnormal inflection point; if a time-delay data in the constant current charging mode is 170 seconds, which is higher than the upper limit of the reference interval of 160 seconds, it is also determined as an abnormal inflection point.

[0089] Through the above steps, the setting of the dynamic time-delay matching window fully combines the charge-discharge behavior characteristics and the material response characteristics. For example, in the pulse discharge mode, a high switching rate corresponds to a short window time, reflecting the characteristic of the electrode quickly relaxing after high-load switching; the statistical determination of the reference interval avoids the subjective deviation of artificially setting thresholds. For example, the interval is delimited based on the standard deviation range of historical healthy data, improving the objectivity and accuracy of anomaly detection.

[0090] S4. Analyze the cumulative damage of microscopic defects in the electrode material under historical working conditions based on the distribution density of abnormal inflection points and the fractal dimension decay characteristics of the voltage relaxation curve during the rest stage, and generate a dynamic threshold interval in combination with the state of charge, including:

[0091] S4-1. Count the occurrence frequency of abnormal inflection points per unit time, and calculate the distribution density of abnormal inflection points in combination with the aggregation degree of the distribution of abnormal inflection points on the voltage curve;

[0092] S4-2. Calculate the fractal dimension decay rate of the voltage relaxation curve during the rest stage, and extract the average decline rate of the fractal dimension over time as the fractal dimension decay characteristic;

[0093] Step S4-2 includes:

[0094] S4-2a. Calculate the fractal dimension sequence of the voltage relaxation curve during the rest stage, calculate the fractal dimension value through multi-time window sliding, and generate a decay sequence of the fractal dimension over time;

[0095] S4-2b. Based on the correlation between the stress release of the electrode material and the ion diffusion path, extract the dynamic coupling characteristics of the fractal dimension decline rate and the curvature change direction in the decay sequence. When the fractal dimension decline rate and the curvature change direction are in the same direction, it is marked as the rapid decay section, and when they are in the opposite direction, it is marked as the slow decay section;

[0096] S4-2c. Count the cumulative duration ratio of the rapid decay section and the number of curvature jumps in the slow decay section, and generate a quantitative index of the fractal dimension decay characteristic.

[0097] S4-3. Quantify the cumulative damage degree of the historical working conditions to the microscopic defects of the electrode material based on the weighted fusion result of the abnormal inflection point distribution density and the fractal dimension decay characteristic;

[0098] S4-4. Compensate and correct the cumulative damage degree according to the state of charge, and generate a dynamic threshold interval that changes non-linearly with the state of charge.

[0099] When calculating the fractal dimension decay rate of the voltage relaxation curve during the rest stage, the specific implementation method of the fractal dimension sequence calculation is as follows: adopt the fractal box-counting method, perform grid coverage analysis on the voltage curve with different time windows, and the time window span is set according to the battery relaxation response time. For example, a 1-second time window is used for the initial relaxation stage to capture the rapid deformation characteristics, and a 5-second time window is used for the middle and late relaxation stages to analyze the macroscopic decay trend. The grid division rule is: the initial grid size is one-tenth of the time window length. For example, a 1-second window corresponds to an initial grid size of 0.1 second, and the grid size is gradually reduced to 1% second. Count the minimum number of grids required to cover the curve at each size, and fit the fractal dimension value through the least squares method to generate a sequence of the fractal dimension decreasing over time.

[0100] The calculation method of the fractal dimension decline rate is as follows: Take the difference between the fractal dimension values of two adjacent time windows and divide it by the window time interval. For example, with a 1-second time window interval, if the fractal dimension value at time t1 is 1.5 and at time t2 is 1.45, then the decline rate is (1.5 - 1.45) / 1 = 0.05 / s. The determination rule for the curvature change direction is: Calculate the second-order difference value of the voltage data at an interval of 0.1 seconds. The second-order difference is approximately the second calculation of the difference between three adjacent voltage points. If the average value of the second-order differences of three consecutive points is greater than zero, it is determined as a positive change (curvature convex upward), and if it is less than zero, it is determined as a negative change (curvature concave downward). When the fractal dimension decline rate exceeds 0.05 per second and the curvature change direction is negative, it is marked as a rapid decay segment, indicating that the crack propagation inside the electrode material hinders ion diffusion; when the fractal dimension decline rate is lower than 0.02 per second and the curvature change direction is positive, it is marked as a slow decay segment, indicating local stress release but no macroscopic damage is triggered.

[0101] When calculating the cumulative duration ratio of the rapid decay segments, calculate the ratio of the total time length of all rapid decay segments to the total duration of the rest stage. For example, if the total duration of the rest stage is 600 seconds and the cumulative duration of the rapid decay segments is 120 seconds, then the ratio is 20%. When counting the number of curvature jumps in the slow decay segments, detect the number of jump events where the curvature value exceeds twice the average curvature value of the adjacent region within the slow decay segment. For example, if the average curvature value within a 0.5-second window adjacent to a certain slow decay segment is 0.1, when the curvature value of 0.25 is detected, it is determined as one jump event, and if the cumulative number is 3 times, then record the number of jumps as 3 times.

[0102] The calculation method of the abnormal inflection point distribution density is as follows: Count the number of abnormal inflection points per unit time and multiply it by the spatial aggregation coefficient. The spatial aggregation coefficient is obtained by calculating the ratio of the average time interval between adjacent points of the abnormal inflection points on the voltage curve to the average time interval of a standard healthy battery under the same charge-discharge mode. For example, if the average interval of abnormal inflection points of a certain battery is 50 seconds and the average interval of a standard healthy battery is 100 seconds, then the aggregation coefficient is 50 / 100 = 0.5; if 10 abnormal inflection points are detected in a certain period, then the distribution density is 10 × 0.5 = 5.

[0103] The generation rule of the quantization index for the fractal dimension decay characteristic is as follows: the weight of the cumulative duration ratio in the rapid decay section is set to 0.6, and the weight of the number of curvature jumps in the slow decay section is 0.4. The weight ratio is determined according to the contribution degrees of the two types of indexes to material damage in historical data. Specifically, through the analysis of 100 groups of data of the same type of aged batteries, the correlation coefficient between the duration ratio in the rapid decay section and the capacity decay is 0.8, and the correlation coefficient of the number of jumps in the slow decay section is 0.5. The weights are allocated according to the correlation coefficient ratio as 0.6:0.4. For example, if the cumulative duration ratio is 20% and the number of jumps is 3, then the decay characteristic value is 0.2×0.6 + 3×0.4 = 1.32.

[0104] The weighted fusion method for the cumulative damage degree is: the distribution density and the fractal dimension decay characteristic index are weighted and summed according to the ratio of 7:3. This ratio is determined according to the influence weights of macroscopic abnormal events and microscopic structure evolution in the formation mechanism of microscopic defects of the electrode material. For example, the probability of lithium dendrite growth corresponding to the macroscopic abnormal inflection point is 70%, and the probability of electrode crack propagation corresponding to the microscopic fractal decay is 30%. Then the weight of the distribution density is 0.7, and the weight of the decay characteristic is 0.3. If the distribution density is 9.6 and the decay characteristic value is 1.32, then the cumulative damage degree = 9.6×0.7 + 1.32×0.3 = 7.18.

[0105] The specific implementation method of the state of charge compensation and correction is: calibrate the sensitivity curve of material damage under different states of charge through charge-discharge cycle experiments. The experimental conditions are an ambient temperature of 25°C ± 2°C and a charge-discharge rate of 1C. For example, the experiment shows that when the state of charge is higher than 80%, the sensitivity of the material to overcharge damage is reduced by 30%. Accordingly, the compensation coefficient is set to 0.7, and the calculation result of the damage degree is multiplied by this coefficient to relax the lower limit of the threshold; at the same time, when the state of charge is higher than 80%, the risk of lithium plating increases, and the upper limit of the threshold is multiplied by the coefficient 0.9 to shrink the interval. The generation logic of the dynamic threshold interval is: based on the threshold interval [50, 100] of the new battery under standard working conditions, which is obtained by statistically analyzing the static voltage fluctuation range of the new battery during 1C charge-discharge cycles; the normalization coefficient 10 is determined through historical data regression analysis. Specifically, a linear relationship curve between the damage degree and the threshold adjustment amount is fitted, and the reciprocal of the slope is taken as the normalization coefficient.

[0106] For example, if the compensated damage degree is 8, then the dynamic threshold interval = the reference interval × (damage degree / 10), that is, [50×0.8, 100×0.8] = [40, 80].

[0107] The quantization index of the fractal dimension decay characteristic fully reflects the dynamic evolution process of the microscopic defects of the electrode material. For example, a high proportion of the cumulative duration of the rapid decay segment indicates the continuous expansion of material cracks, and a large number of curvature jumps in the slow decay segment reflects the frequent occurrence of local stress concentration; the weighted fusion of the abnormal inflection point distribution density and the fractal decay characteristic effectively couples the correlation between macroscopic abnormal events and microscopic material damage; the state of charge compensation and correction dynamically adapts the threshold interval based on the electrochemical characteristics of the battery. For example, at a high state of charge, the ionic activity increases, the lower limit is relaxed to avoid misjudging normal relaxation as a fault, and at the same time, the upper limit is shrunk to improve the detection sensitivity to early lithium plating.

[0108] S5. Perform time-frequency transformation on the voltage during the rest stage, extract the frequency-domain energy center-of-gravity offset and the time-domain fluctuation envelope, and extract the time-delay phase difference for the temperature during the rest stage, including:

[0109] S5-1. Perform wavelet packet transformation on the voltage data during the rest stage, divide the preset frequency bands and calculate the energy distribution of each preset frequency band, and extract the frequency-domain energy center-of-gravity offset as the difference between the center-of-gravity frequency at each moment and the reference frequency of the corresponding frequency band of a standard healthy battery;

[0110] S5-2. Filter the interference components of the frequency-domain energy center-of-gravity offset based on the dynamic threshold interval;

[0111] S5-3. Perform Hilbert transformation on the time-domain waveform of the voltage data, and extract the time-domain fluctuation envelope as the average of the upper and lower envelopes of the voltage amplitude changing with time;

[0112] S5-4. Perform cross-correlation analysis on the temperature data sequence during the rest stage, calculate its time-delay phase difference relative to the envelope of the voltage curve, and the time-delay phase difference is the time offset corresponding to the peak of the cross-correlation function.

[0113] When performing wavelet packet transform on the voltage data in the static stage, the division of the preset frequency band is determined according to the battery fault sensitive frequency band. For example, for lithium-ion power batteries, the preset frequency band is divided into three frequency bands: 0.1 Hz - 1 Hz, 1 Hz - 10 Hz, and 10 Hz - 100 Hz. The energy distribution of each frequency band is obtained by calculating the sum of the squares of the signals of each node after wavelet packet decomposition. The extraction method of the frequency domain energy center of gravity offset is as follows: calculate the center of gravity frequency of the energy distribution of each frequency band. The center of gravity frequency is obtained through the weighted average formula, and the weight is the proportion of the energy of each sub-frequency band in the total energy. For example, within the frequency band of 0.1 Hz - 1 Hz, there are three sub-frequency bands with energy proportions of 30%, 50%, and 20% respectively, and the corresponding center frequencies are 0.5 Hz, 0.8 Hz, and 0.3 Hz. Then the center of gravity frequency = 0.5×0.3 + 0.8×0.5 + 0.3×0.2 = 0.61 Hz; compare the center of gravity frequency with the reference frequency of the corresponding frequency band of the standard healthy battery. The reference frequency is determined by collecting the average value of the center of gravity frequencies of 100 groups of healthy batteries under the same charge and discharge mode. For example, the reference frequency of the 0.1 Hz - 1 Hz frequency band of the healthy battery is 0.6 Hz. If the current center of gravity frequency is 0.61 Hz, the offset is +0.01 Hz.

[0114] When filtering the interference components of the frequency domain energy center of gravity offset based on the dynamic threshold interval, the upper and lower limits of the dynamic threshold interval are from the dynamic threshold interval generated in step S4. For example, for the frequency band of 0.1 Hz - 1 Hz, the dynamic threshold interval is [-0.05 Hz, +0.05 Hz]. If the current offset is +0.01 Hz, it is determined as an interference component and excluded; if the offset is +0.06 Hz, it is retained as an effective fault feature. During the filtering process, it is necessary to ensure that the time stamps of the frequency domain energy center of gravity offset and the dynamic threshold interval are aligned. For example, if the dynamic threshold interval is updated every minute, the frequency domain offset needs to be matched and filtered according to the average value within each minute.

[0115] When performing Hilbert transform on the time domain waveform of the voltage data, the specific steps are as follows: perform Hilbert transform on the voltage sampling sequence to generate an analytic signal, and extract the instantaneous amplitude of the analytic signal as the time domain fluctuation envelope. The upper envelope is the sequence of local maximum values of the instantaneous amplitude, and the lower envelope is the sequence of local minimum values. The average value of the upper and lower envelopes is the voltage fluctuation envelope.

[0116] For example, the voltage sampling values in a certain period are [3.2 V, 3.3 V, 3.1 V, 3.4 V]. After Hilbert transform, the upper envelope is [3.25 V, 3.35 V, 3.25 V, 3.45 V], and the lower envelope is [3.15 V, 3.25 V, 3.05 V, 3.35 V]. Then the average value of the fluctuation envelope is [(3.25 + 3.15) / 2, (3.35 + 3.25) / 2,...] = [3.2 V, 3.3 V,...].

[0117] When performing cross - correlation analysis on the temperature data sequence in the static stage, first align the time stamps of the temperature data and the voltage fluctuation envelope data, and resample both of them in seconds. Calculate the cross - correlation function of the temperature sequence and the envelope sequence, and search for the time offset corresponding to the maximum value of the cross - correlation function as the time - delay phase difference. For example, if the temperature data starts to rise at the time stamp t = 10s and the voltage envelope starts to rise at t = 12s, and the peak of the cross - correlation function appears at a position 2 seconds behind, then the time - delay phase difference is 2 seconds. The physical meaning of the time - delay phase difference is the time measure of the heat conduction lag effect. For example, the crack propagation of the electrode material leads to local heat generation, and there is a time delay for the heat to transfer to the temperature sensor.

[0118] The dynamic threshold filtering of the frequency - domain energy centroid offset effectively distinguishes normal relaxation fluctuations from real fault signals. For example, the offset caused by normal relaxation is within ±0.05Hz, while the internal short - circuit fault causes the offset to exceed + 0.05Hz; the mean value of the time - domain fluctuation envelope reflects the overall voltage fluctuation trend and avoids instantaneous noise interference; the time - delay phase difference quantifies the heat conduction delay characteristic. For example, the time - delay phase difference of a normal battery is less than 5 seconds, while under the lithium - plating fault, the time - delay phase difference exceeds 10 seconds.

[0119] S6. Based on the judgment results of whether the time - domain fluctuation envelope is within the dynamic threshold range and whether the time - delay phase difference is lower than the preset critical value, perform safety risk determination, including:

[0120] If the time - domain fluctuation envelope is within the dynamic threshold range and the time - delay phase difference is lower than the preset critical value, it is determined as the superposition interference of the microscopic defects of the electrode material and the relaxation effect; otherwise, match the frequency - domain energy centroid offset with the frequency - domain fingerprint library of the preset faults and trigger hierarchical alarms.

[0121] The construction logic of the frequency - domain fingerprint library of the preset faults is as follows: Perform standardized charge - discharge tests on power batteries with known internal short - circuit faults and lithium - plating fault types, collect voltage data in the static stage and extract the frequency - domain energy centroid offset characteristics. Among them, the internal short - circuit fault shows the characteristic of continuous positive offset of the energy centroid in the low - frequency band, and the lithium - plating fault shows the characteristic of periodic fluctuation offset in the middle - frequency band; Map the characteristics of different fault modes to the energy distribution law of the corresponding frequency bands to form a frequency - domain fingerprint library with the fault type as the index and the frequency - domain energy centroid offset mode as the characteristic. The low - frequency band corresponds to the positive offset mode of the internal short - circuit fault, and the middle - frequency band corresponds to the periodic fluctuation mode of the lithium - plating fault.

[0122] When making a safety risk determination based on the judgment results of whether the time-domain fluctuation envelope is within the dynamic threshold range and whether the time-delay phase difference is lower than the preset critical value, it is necessary to preprocess the time-domain fluctuation envelope and the dynamic threshold range to ensure dimensional consistency. The preprocessing method is as follows: Normalize the voltage amplitude of the time-domain fluctuation envelope according to the rated voltage range of the standard healthy battery. For example, if the voltage range of the standard healthy battery is 3.0V - 4.2V, then the envelope voltage value of 3.3V is normalized to (3.3 - 3.0) / (4.2 - 3.0) = 0.25; the dynamic threshold range is also normalized according to this range. For example, the dynamic threshold range [0.2, 0.3] represents the allowable normalized voltage fluctuation range.

[0123] The specific logic of the safety risk determination is as follows: If the normalized time-domain fluctuation envelope is within the dynamic threshold range and the time-delay phase difference is lower than the preset critical value, it is determined as the superposition interference of the microscopic defects and relaxation effects of the electrode material. For example, if the normalized envelope value is 0.25 and is within the range [0.2, 0.3], and the time-delay phase difference is 3 seconds (lower than the critical value of 5 seconds), it is determined as superposition interference; if the envelope value is 0.35 and exceeds the range or the time-delay phase difference is 6 seconds (exceeding the critical value of 5 seconds), the subsequent fault matching process is triggered.

[0124] The setting basis of the preset critical value is the maximum value of the time-delay phase difference in the historical healthy battery data plus twice the standard deviation. For example, if the average value of the time-delay phase difference of 100 groups of healthy batteries is 2 seconds and the standard deviation is 1 second, then the critical value = 2 + 2×1 = 4 seconds, rounded to 5 seconds.

[0125] The construction logic of the preset fault frequency-domain fingerprint library is as follows: Conduct charge and discharge tests on batteries with known internal short circuit and lithium plating faults, collect voltage data during the static stage, and extract the feature of the frequency-domain energy center of gravity offset. For example, the feature of the frequency-domain energy center of gravity offset for the internal short circuit fault in the 0.1Hz - 1Hz frequency band is a continuous positive offset of 0.1Hz - 0.2Hz, and the feature for the lithium plating fault in the 1Hz - 10Hz frequency band is a periodic fluctuation offset of ±0.05Hz; Associate the fault features with the frequency bands and store them as the frequency-domain fingerprint library. For example, the internal short circuit corresponds to the positive offset mode in the 0.1Hz - 1Hz frequency band, and the lithium plating corresponds to the periodic fluctuation mode in the 1Hz - 10Hz frequency band.

[0126] When matching the frequency-domain energy centroid offset with the frequency-domain fingerprint database, a similarity calculation method is adopted. For example, the cosine similarity between the current offset and the preset fault feature in the corresponding frequency band is calculated. If the similarity exceeds 0.8, it is determined that the match is successful. The classification alarm trigger rule is as follows: when the internal short circuit match is successful, a first-level alarm (immediate shutdown) is triggered; when the lithium plating match is successful, a second-level alarm (power-limited operation) is triggered. For example, if the similarity between the current offset and the internal short circuit feature in the 0.1 Hz - 1 Hz frequency band is 0.85, a first-level alarm is triggered; if the similarity between the current offset and the lithium plating feature in the 1 Hz - 10 Hz frequency band is 0.75, no alarm is triggered.

[0127] The specific implementation method for dimension unification during the preprocessing process is as follows: when normalizing the time-domain fluctuation envelope, the rated voltage range is dynamically adjusted according to the battery model. For example, for ternary lithium batteries, it is 2.8 V - 4.25 V, and for lithium iron phosphate batteries, it is 2.5 V - 3.65 V; the dynamic threshold interval is normalized synchronously in the same range. For example, the dynamic threshold interval [0.15, 0.3] for ternary lithium corresponds to the actual voltage range from 2.8 + 0.15×(4.25 - 2.8) = 3.01 V to 2.8 + 0.3×(4.25 - 2.8) = 3.36 V. The normalized envelope value is directly compared with this interval without secondary conversion.

[0128] The calibration experimental method for the preset critical value of the time-delay phase difference is as follows: at an ambient temperature of 25°C, perform standard charge and discharge cycle tests on the same batch of batteries, record the time-delay phase difference data of healthy batteries, calculate their mean and standard deviation, and set the critical value as the mean plus three times the standard deviation. For example, if the mean time-delay phase difference in 100 cycle tests is 3 seconds and the standard deviation is 0.5 seconds, then the critical value is 3 + 3×0.5 = 4.5 seconds, rounded up to 5 seconds. The construction of the fault frequency-domain fingerprint database needs to cover battery data with different aging degrees. For example, collect the frequency-domain energy centroid offsets of 50 groups of internally short-circuited batteries at different cycle numbers (100 - 1000 times), and extract the common frequency offset features as fingerprints.

[0129] Through the above steps, the safety risk determination logic realizes the collaborative analysis of multi-dimensional signals (time-domain envelope, time-delay phase difference, frequency-domain offset). For example, the superposition interference determination needs to simultaneously meet the conditions that the time-domain fluctuation is controlled (the envelope is within the threshold interval) and the heat conduction delay is normal (the time-delay phase difference is low), while the fault alarm depends on the accurate matching of the frequency-domain features. The normalization preprocessing eliminates the interference of dimension differences on the determination logic, and the multi-fault mode coverage of the frequency-domain fingerprint database improves the pertinence of the alarm.

[0130] It should be noted that the superposition interference of microscopic defects and relaxation effects in the electrode material refers to the abnormal local electrochemical activity caused by defects such as microscopic cracks and lattice distortion (microscopic defects) generated by the historical charge-discharge cycles of the electrode material during the battery rest stage, which, together with the normal relaxation behaviors such as ion redistribution inside the battery and interfacial charge balance (relaxation effects), lead to a composite fluctuation phenomenon with similar morphological characteristics but different causes in the voltage curve. For example, microscopic cracks cause a sudden voltage drop due to the hindered diffusion path of lithium ions, while the normal relaxation process also causes a slow voltage drop due to the reconstruction of the electric double layer. After the superposition, the voltage fluctuation amplitude and timing characteristics converge, making it impossible for the traditional threshold method to distinguish material damage from normal relaxation.

[0131] Example 2: Figure 2 The structural schematic diagram of the data-driven power battery safety risk assessment system of the present invention is given. The data-driven power battery safety risk assessment system includes:

[0132] Rest data acquisition module: It acquires the voltage, temperature, and state of charge of the power battery during the rest stage, and synchronously obtains the last charge-discharge current value and charge-discharge mode;

[0133] Extreme value weight correction module: It extracts the slope extreme value points of the last charge-discharge current value, corrects the weights of the slope extreme value points based on the charge-discharge mode, and identifies the timing of the geometric shape mutation points of the voltage curve during the rest stage;

[0134] Time delay screening and analysis module: It establishes the mapping relationship between the slope extreme value points and the time lag of the geometric shape mutation points, and screens the abnormal inflection points with excessive time lags;

[0135] Fractal dynamic threshold module: According to the distribution density of the abnormal inflection points and the fractal dimension attenuation characteristics of the voltage relaxation curve during the rest stage, it analyzes the cumulative damage of the microscopic defects of the electrode material under historical working conditions, and generates a dynamic threshold interval in combination with the state of charge;

[0136] Time-frequency feature extraction module: It performs time-frequency transformation on the voltage during the rest stage, extracts the frequency-domain energy center of gravity offset and the time-domain fluctuation envelope, and extracts the time lag phase difference for the temperature during the rest stage;

[0137] Safety risk determination module: Based on the judgment results of whether the time-domain fluctuation envelope is within the dynamic threshold interval and whether the time lag phase difference is lower than the preset critical value, it conducts safety risk determination.

[0138] All the above formulas are dimensionless and take their numerical values for calculation. The formulas are obtained by software simulation of a large amount of collected data to get a formula closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0139] It should be noted that the present invention can be deployed on the device itself to implement embedded applications, or can also run on a PC or other terminals with a user interface, so as to meet various hardware environments and usage requirements.

[0140] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains one or more sets of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0141] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, devices, and modules can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0142] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there can be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of the devices or modules can be in electrical, mechanical or other forms.

[0143] The module described as a separation component may or may not be physically separated. The component shown as a module may or may not be a physical module. It may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0144] In addition, in each embodiment of the present application, each functional module can be integrated into a processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0145] If the above functions are implemented in the form of software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs and other various media that can store program codes.

[0146] As described above, the above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0147] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A data-driven power battery safety risk assessment method, characterized in that: The steps include: S1. Collect the voltage, temperature and charge state of the power battery in the static stage, and simultaneously obtain the last charge and discharge current value and charge and discharge mode; S2, extracting the slope extreme value point of the last charge and discharge current value, weighting the slope extreme value point based on the charge and discharge mode, and identifying the geometric morphology mutation point timing of the voltage curve in the static stage; S3, establish the lag time mapping relationship between the slope extreme point and the geometric morphology mutation point, and screen the abnormal inflection point with excessive lag time; S4. According to the distribution density of abnormal inflection points and the fractal dimension attenuation characteristics of the voltage relaxation curve in the static stage, the cumulative damage of the microscopic defects of the electrode material caused by the historical working conditions is analyzed, and the dynamic threshold interval is generated in combination with the charge state; S5, performing time-frequency transformation on the voltage in the static stage, extracting the frequency domain energy center of gravity offset and the time domain fluctuation envelope, and extracting the time lag phase difference of the temperature in the static stage; S6. Based on the judgment result of whether the time domain fluctuation envelope is within the dynamic threshold range and whether the time lag phase difference is lower than the preset critical value, a safety risk judgment is performed.

2. The data-driven power battery safety risk assessment method according to claim 1 is characterized in that: Collect the voltage, temperature and state of charge of the power battery in the static stage, and simultaneously obtain the last charge and discharge current value and charge and discharge mode, including: S1-1. When the power battery enters the static stage, the static voltage data is continuously collected at a first preset sampling frequency, the static temperature data is continuously collected at a second preset sampling frequency, and the state of charge data is acquired in real time through the battery management system; S1-2. When the power battery finishes the last charge and discharge, the current value at the end of the charge and discharge is recorded as the last charge and discharge current value, and the charge and discharge mode classification identifier during the charge and discharge process is obtained, and the charge and discharge mode classification identifier includes at least one of constant current charging, pulse discharge and constant voltage charging; S1-3, aligning the timestamps of the static voltage data, static temperature data and state of charge data with the timestamps of the last charge and discharge current value and the charge and discharge mode classification identifier to generate a time-synchronized static phase data set.

3. The data-driven power battery safety risk assessment method according to claim 1, characterized in that: Extract the slope extreme point of the last charge and discharge current value, perform weight correction on the slope extreme point based on the charge and discharge mode, and identify the geometric mutation point timing of the voltage curve in the static stage, including: S2-1. Based on the charge and discharge mode classification identification, the slope change characteristics of the last charge and discharge current curve are screened for mode adaptability, wherein the pulse discharge mode retains the slope extreme point in the charge and discharge switching stage, and the constant current charge mode retains the slope extreme point in the charge cut-off stage; S2-2, dynamically weight the selected slope extreme points, adjust the weight coefficient of the extreme points in the pulse discharge mode according to the nonlinear relationship of the corresponding charge and discharge switching rate, and adjust the weight coefficient of the extreme points in the constant current charging mode according to the charging cut-off current gradient; S2-3. Perform multi-scale curvature analysis on the geometric mutation points of the voltage curve in the static stage, extract the mutation points with abnormal curvature change direction and amplitude, and generate a time series of geometric mutation points.

4. The data-driven power battery safety risk assessment method according to claim 1, characterized in that: Establish the lag time mapping relationship between the slope extreme point and the geometric morphology mutation point, and screen the abnormal inflection points with excessive lag time, including: S3-1, matching the time difference between the weighted corrected slope extreme value time series and the geometric morphology mutation point time series, calculating the time difference between each slope extreme value point and the first subsequent geometric morphology mutation point, and generating the original time-lag data set; S3-2, according to the charge and discharge mode classification identifier, the dynamic time lag matching window is set, the matching window in the pulse discharge mode is dynamically adjusted based on the inverse correlation between the charge and discharge switching rate and the relaxation rate, and the matching window in the constant current charging mode is dynamically adjusted based on the inverse correlation between the charge cut-off current gradient and the material stress release rate; S3-3, filtering the original time-delay data in the dynamic time-delay matching window, removing the time difference data outside the window, and retaining the time-delay data in the window as the effective time-delay set; S3-4. Based on the statistical distribution characteristics of the time-lag data of historical healthy batteries, the benchmark interval of the effective time-lag set is calculated, and the time-lag points that deviate from the benchmark interval are marked as abnormal inflection points.

5. The data-driven power battery safety risk assessment method according to claim 1, characterized in that: According to the distribution density of abnormal inflection points and the fractal dimension attenuation characteristics of the voltage relaxation curve in the static stage, the cumulative damage of the microscopic defects of the electrode material caused by historical working conditions is analyzed, and the dynamic threshold interval is generated in combination with the charge state, including: S4-1. Count the frequency of abnormal inflection points within a unit time, and calculate the distribution density of abnormal inflection points based on the degree of concentration of abnormal inflection points on the voltage curve; S4-2, calculating the fractal dimension decay rate of the voltage relaxation curve in the static stage, and extracting the average decrease rate of the fractal dimension over time as the fractal dimension decay characteristic; S4-3. Based on the weighted fusion results of the abnormal inflection point distribution density and the fractal dimension attenuation characteristics, the cumulative damage degree of the micro defects of the electrode material caused by the historical working conditions is quantified; S4-4. Compensate and correct the accumulated damage degree according to the state of charge, and generate a dynamic threshold range that changes nonlinearly with the state of charge.

6. The data-driven power battery safety risk assessment method according to claim 5 is characterized in that: Step S4-2 includes: S4-2a, performing fractal dimension sequence calculation on the voltage relaxation curve in the static stage, calculating the fractal dimension value by sliding multiple time windows, and generating a decay sequence of the fractal dimension changing with time; S4-2b, based on the correlation between stress release of electrode materials and ion diffusion path, the dynamic coupling characteristics of the fractal dimension decrease rate and the curvature change direction in the decay sequence are extracted, where the fractal dimension decrease rate and the curvature change direction are in the same direction as the rapid decay segment, and in the opposite direction as the slow decay segment; S4-2c. Count the cumulative duration of the fast decay segment and the number of curvature jumps in the slow decay segment to generate a quantitative index of the fractal dimension decay characteristic.

7. The data-driven power battery safety risk assessment method according to claim 1, characterized in that: Perform time-frequency transformation on the voltage in the static stage, extract the frequency domain energy center of gravity offset and the time domain fluctuation envelope, and extract the time lag phase difference of the temperature in the static stage, including: S5-1, perform wavelet packet transformation on the voltage data in the static stage, divide the preset frequency bands and calculate the energy distribution of each preset frequency band, and extract the frequency domain energy center of gravity offset as the difference between the center of gravity frequency at each moment and the reference frequency of the corresponding frequency band of the standard healthy battery; S5-2, filtering the interference component of the frequency domain energy center of gravity offset based on the dynamic threshold interval; S5-3, perform Hilbert transform on the voltage data time domain waveform, and extract the time domain fluctuation envelope as the average of the upper and lower envelopes of the voltage amplitude changing with time; S5-4. Perform cross-correlation analysis on the temperature data sequence in the static stage, and calculate the time-delay phase difference relative to the envelope of the voltage curve. The time-delay phase difference is the time offset corresponding to the peak of the cross-correlation function.

8. The data-driven power battery safety risk assessment method according to claim 1, characterized in that: Based on the judgment results of whether the time domain fluctuation envelope is within the dynamic threshold range and whether the time lag phase difference is lower than the preset critical value, a safety risk judgment is performed, including: If the time domain fluctuation envelope is within the dynamic threshold range and the time lag phase difference is lower than the preset critical value, it is judged as the superposition interference of the microscopic defects of the electrode material and the relaxation effect; otherwise, the frequency domain energy center of gravity offset is matched with the frequency domain fingerprint library of the preset fault to trigger a graded alarm.

9. The data-driven power battery safety risk assessment method according to claim 8, characterized in that: The logic for constructing the frequency domain fingerprint library of preset faults is as follows: perform standardized charge and discharge tests on power batteries with known internal short circuit faults and lithium plating fault types, collect voltage data in the static stage and extract the frequency domain energy center of gravity offset characteristics, where the internal short circuit fault presents a continuous positive offset characteristic of the energy center of gravity in the low frequency band, and the lithium plating fault presents a periodic fluctuation offset characteristic in the medium frequency band; associate and map the characteristics of different fault modes with the energy distribution law of the corresponding frequency band to form a frequency domain fingerprint library with the fault type as the index and the frequency domain energy center of gravity offset pattern as the characteristic, the low frequency band corresponds to the positive offset mode of the internal short circuit fault, and the medium frequency band corresponds to the periodic fluctuation mode of the lithium plating fault.

10. A data-driven power battery safety risk assessment system, used to implement the data-driven power battery safety risk assessment method according to any one of claims 1 to 9, characterized in that: include: Static data acquisition module: collects the voltage, temperature and charge state of the power battery during the static stage, and simultaneously obtains the last charge and discharge current value and charge and discharge mode; Extreme value weight correction module: extracts the slope extreme value point of the last charge and discharge current value, performs weight correction on the slope extreme value point based on the charge and discharge mode, and identifies the timing of the geometric mutation point of the voltage curve in the static stage; Time lag screening analysis module: establish the lag time mapping relationship between the slope extreme point and the geometric morphology mutation point, and screen the abnormal inflection points with excessive time lag; Fractal dynamic threshold module: Based on the distribution density of abnormal inflection points and the fractal dimension attenuation characteristics of the voltage relaxation curve in the static stage, the module analyzes the cumulative damage of the microscopic defects of the electrode material caused by historical working conditions, and generates a dynamic threshold interval in combination with the charge state; Time-frequency feature extraction module: performs time-frequency transformation on the voltage in the static stage, extracts the frequency domain energy center of gravity offset and the time domain fluctuation envelope, and extracts the time lag phase difference of the temperature in the static stage; Safety risk determination module: Safety risk determination is performed based on the judgment results of whether the time domain fluctuation envelope is within the dynamic threshold range and whether the time lag phase difference is lower than the preset critical value.

Images