Battery fault early warning and diagnosis method, system and device and storage medium

Abstract

The invention relates to the technical field of battery detection, and particularly provides a battery fault early warning and diagnosis method, which comprises the following steps: acquiring operation parameters of a battery pack in real time, processing the operation parameters based on a preset judgment condition, generating a corresponding preprocessing signal according to a noise environment state, and extracting spatial-temporal characteristics to perform weak signal enhancement processing, so as to obtain a battery fault early warning and diagnosis result. Generating an enhanced feature set; based on the enhanced feature set, performing space-time fusion calculation according to a preset weight relation to obtain a fault energy accumulation value; dynamically correcting a fault judgment threshold according to the health state of the battery and the real-time environment parameters; and when the fault energy accumulation value exceeds the fault judgment threshold after dynamic correction, outputting a graded early warning signal. By capturing weak fault features of the battery pack and combining spatial domain feature extraction of temperature difference and voltage distribution and a signal enhancement technology, the detection sensitivity of early hidden faults is improved, and in a lithium battery safety early warning scene, the fault detection time is shortened, the false alarm rate is reduced and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of battery detection, and in particular to a battery fault early warning and diagnosis method and system. Background Art

[0002] Lithium-ion batteries, as complex, nonlinear systems that are sensitive to both the environment and noise, are subject to numerous fault types, subtle fault characteristics, and significant individual variability. Fault diagnosis faces the dual challenges of high misdiagnosis and missed diagnosis rates, making accurate fault detection and location extremely challenging. Furthermore, lithium-ion battery failures can develop rapidly. For example, an internal short circuit can cause the battery to rapidly release energy and generate heat, ultimately leading to combustion and explosion. To prevent these catastrophic events, a battery management system (BMS) must be capable of detecting and identifying faults early, shortening diagnostic time and providing proactive safety warnings to minimize the risks associated with battery failure.

[0003] Most existing fault diagnosis algorithms use a threshold-based diagnostic approach. When the abnormal characteristic value of a battery caused by a fault exceeds a preset threshold, the fault diagnosis system issues a fault alarm. In practice, traditional diagnostic algorithms often adopt a conservatively high threshold setting to avoid misdiagnosis caused by environmental noise. However, this also increases the risk of missing early faults due to their hidden characteristics. If the threshold is lowered to reduce the fault omission rate, the lower threshold will be frequently hit by abnormal characteristic values ​​caused by environmental noise, resulting in frequent misdiagnosis and affecting the normal operation of the battery system. Therefore, traditional threshold-based fault diagnosis methods cannot resolve the contradiction between misdiagnosis and missed diagnosis.

[0004] To this end, the present invention provides a battery fault early warning and diagnosis method to solve the above problems. Summary of the Invention

[0005] In order to overcome the deficiencies of the prior art, the present invention provides a battery fault early warning and diagnosis method and system to solve the problems in the prior art.

[0006] One embodiment of the present invention provides a battery failure early warning and diagnosis method, comprising the following steps: Acquire operating parameters of the battery pack in real time and perform branching processing on the operating parameters based on a preset environmental noise intensity determination condition; if the environment is determined to be a strong noise environment, perform noise stripping processing on the operating parameters to generate a preprocessed signal; if the environment is determined to be a non-strong noise environment, use the operating parameters as the preprocessed signal; Extracting spatiotemporal features from the preprocessed signal, the extracted spatiotemporal features including signal energy features of a preset target frequency band, battery pack cell temperature difference distribution features, and cell voltage distribution features; performing weak signal enhancement processing on the extracted spatiotemporal features to generate an enhanced feature set; Based on the spatiotemporal features in the enhanced feature set, a spatiotemporal fusion calculation is performed according to a preset weight relationship to obtain a fault energy accumulation value; Dynamically adjust the fault judgment threshold based on battery health status and real-time environmental parameters; When the fault energy accumulation value exceeds the fault determination threshold value after dynamic correction, a graded warning signal is output.

[0007] In one embodiment, the step of acquiring the operating parameters of the battery pack in real time and performing branch processing on the operating parameters based on a preset environmental noise intensity determination condition specifically includes: Perform a primary determination based on the historical noise intensity records of the battery pack's environment. If the average noise intensity over the past preset continuous time period is lower than the preset safety noise threshold, the environment is directly determined to be a non-strong noise environment. Otherwise, a second-level judgment is executed to perform spectrum analysis on the currently acquired operating parameters and extract the energy of noise components above the preset cutoff frequency. If the energy of the noise component exceeds the dynamic noise threshold, it is determined to be a strong noise environment. Only when the secondary determination output is a strong noise environment, a noise stripping process is performed to generate the pre-processed signal.

[0008] In one embodiment, the weak signal enhancement process includes: a) High-frequency signal enhancement path: Perform frequency domain decomposition on the signal energy characteristics of the preset target frequency band, and extract the frequency band component with a preset bandwidth and a characteristic frequency determined by the real-time battery status as the center frequency point; Calculating a dynamic gain value based on a ratio of the signal amplitude of the frequency band component to the ambient noise amplitude and applying the dynamic gain value to the frequency band component to obtain a high-frequency enhancement component; b) Low-frequency trend maintenance path: Perform sliding average filtering on the temperature difference distribution characteristics and voltage distribution characteristics of the battery pack cells, where the filter window length is inversely proportional to the battery charge and discharge rate, to obtain low-frequency enhancement features; c) Cross-scale fusion: The high-frequency enhancement component and the low-frequency enhancement feature are weightedly superimposed according to a preset fusion coefficient to generate the enhancement feature set.

[0009] In one embodiment, in the step of performing spatiotemporal fusion calculation based on the spatiotemporal features in the enhanced feature set according to a preset weight relationship to obtain the fault energy accumulation value, the preset weight relationship is a dynamic weight relationship, and the dynamic weight relationship is dynamically adjusted according to the real-time working status parameters of the battery pack, wherein: The real-time working status parameter includes at least one of the ambient temperature, battery charge and discharge rate, and battery state of charge; The dynamic weight relationship includes a time domain feature weight, a first spatial domain feature weight, and a second spatial domain feature weight, wherein the time domain feature is a signal energy feature, the first spatial domain feature is a battery pack cell temperature difference distribution feature, and the second spatial domain feature is a battery pack cell voltage distribution feature; the dynamic adjustment of the dynamic weight relationship satisfies: When the ambient temperature is lower than a temperature threshold, increasing the weight of the first spatial domain feature; When the charge and discharge rate is higher than the rate threshold, increasing the weight of the time domain feature; The second spatial domain feature weight remains within a preset threshold range.

[0010] In one embodiment, in the step of performing spatiotemporal fusion calculation based on the spatiotemporal features in the enhanced feature set according to a preset weight relationship to obtain the fault energy accumulation value, the fault energy accumulation value is calculated by: The temporal and spatial features in the enhanced feature set are mapped to the characteristic values ​​and reference values ​​of the fault energy calculation formula, and the fault energy cumulative value is calculated using the following formula; in, For the moment a target characteristic value of the battery pack after weak signal enhancement, wherein the target characteristic value is selected from at least one of a signal energy characteristic, a battery pack single cell temperature difference distribution characteristic, and a battery pack single cell voltage distribution characteristic; is the normal operating condition reference value corresponding to the target characteristic value; is the characteristic sampling interval; To calculate the cumulative duration; Moreover, the selection and correspondence of the target characteristic value The assignment of is associated and adapted with the dynamic weight relationship, specifically satisfying: When the ambient temperature is lower than the temperature threshold, the battery pack temperature difference distribution characteristics are preferentially selected as Calculate the weight ratio corresponding to the increase in the normal baseline value of the single temperature difference; When the charge and discharge rate is higher than the rate threshold, the signal energy feature is preferentially selected as Calculate the weight ratio corresponding to the normal baseline value of the increased signal energy.

[0011] In one embodiment, the setting of the normal operating condition reference value further includes: Collect historical operating data of the battery pack and pre-process it to obtain standardized feature data applied to the machine model as input data; The standardized feature data obtained after preprocessing is used to train a battery individual health state model using a machine learning algorithm to obtain a trained battery individual health state model. The battery individual health state model receives the operating parameters of the battery pack in real time and outputs a personalized normal operating condition reference interval that dynamically adapts to the battery life cycle; wherein the personalized normal operating condition reference interval is dynamically updated with the number of battery cycles and the degree of aging; When the health status dispersion of single cells in the same batch of battery packs is monitored to exceed a preset threshold, the adaptive correction of the personalized normal operating condition reference interval is triggered. The correction process is combined with real-time working status parameters to perform multi-dimensional dynamic adjustment.

[0012] In one embodiment, after the step of performing spatiotemporal fusion calculation based on the spatiotemporal features in the enhanced feature set according to a preset weight relationship to obtain the fault energy accumulation value, the following steps are also included: Obtain historical fault data and real-time working status parameters, combine them with the calculation results of the fault energy accumulation value, perform data standardization preprocessing, and generate a multivariate time series feature data set; Based on the standardized multivariate time series feature data set, a multivariate time series prediction model is constructed using a machine learning algorithm. The multivariate time series prediction model is used to predict the fault energy growth rate in the future period; The predicted fault energy growth rate is compared with the dynamically corrected fault judgment threshold change gradient in real time. When the predicted fault energy growth rate exceeds the product of the preset safety factor and the fault judgment threshold change gradient, the advance warning mechanism is triggered; According to the relative deviation between the predicted fault energy growth rate and the change gradient of the fault judgment threshold, multiple warning levels are divided, and the remaining safe operation time is calculated based on the current fault energy accumulation value, the dynamically corrected fault judgment threshold and the predicted fault energy growth rate.

[0013] This application also relates to a battery failure early warning and diagnosis system, comprising: A data acquisition module, configured to acquire operating parameters of the battery pack in real time, perform branch processing on the operating parameters based on a preset environmental noise intensity determination condition, and generate a preprocessing signal; a noise stripping module, configured to perform noise stripping processing on the operating parameters and generate a preprocessing signal when determining that the environment is a strong noise environment; A feature extraction module is used to extract spatiotemporal features from the preprocessed signal, wherein the extracted spatiotemporal features include signal energy features of a preset target frequency band, battery pack cell temperature difference distribution features, and cell voltage distribution features; and perform weak signal enhancement processing on the extracted spatiotemporal features to generate an enhanced feature set; A fusion calculation module is used to perform spatiotemporal fusion calculation based on the spatiotemporal features in the enhanced feature set according to a preset weight relationship to obtain a fault energy accumulation value; Dynamic correction module, used to dynamically correct the fault judgment threshold according to the battery health status and real-time environmental parameters; The early warning output module is used to output a graded early warning signal when the accumulated value of the fault energy exceeds the fault judgment threshold value after dynamic correction.

[0014] The present application also relates to a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned battery fault warning and diagnosis method are implemented.

[0015] The present application also relates to a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned battery fault early warning and diagnosis method are implemented.

[0016] The battery fault warning and diagnosis method and system provided in the above embodiments have the following beneficial effects: 1. Through environmental noise adaptive branch processing, interference signals are removed and weak fault characteristics of battery packs are accurately captured under strong noise. The spatial feature extraction and signal enhancement technology of temperature difference and voltage distribution are combined to improve the detection sensitivity of early hidden faults. Through the spatiotemporal fusion calculation of fault energy accumulation value, the fault deterioration trend is quantified and the risk trajectory is predicted. Finally, based on the battery health status and environmental parameters, the threshold is dynamically corrected to achieve the accurate output of graded warning signals. This fundamentally solves the technical difficulties of traditional methods in high misdiagnosis rate, high missed diagnosis rate and response lag. In lithium battery safety warning scenarios, it shortens the fault detection time and reduces the false alarm rate.

[0017] 2. In one embodiment, by introducing a dynamic weight relationship, the weights of the time domain features, the first spatial domain features, and the second spatial domain features are dynamically adjusted according to the real-time working status parameters of the battery pack (ambient temperature, battery charge and discharge rate, battery state of charge, etc.), effectively solving the problem that the traditional fixed weight diagnosis method cannot accurately match the importance of fault features under different working conditions. In a low-temperature environment, increasing the weight of the temperature difference distribution feature of the battery pack cells can capture potential faults caused by uneven temperature in advance; under high charge and discharge rate conditions, increasing the weight of the signal energy feature can accurately identify abnormal energy fluctuations under high load. This dynamic weight mechanism breaks the "one-size-fits-all" limitation of the traditional threshold diagnosis method, avoids misdiagnosis and missed diagnosis due to environmental changes or differences in working conditions, makes the fault diagnosis results more in line with the actual operating status of the battery, and significantly improves the accuracy and reliability of fault diagnosis.

[0018] 3. In one of the embodiments, by deeply integrating the enhanced feature set with the fault energy calculation formula, the weak abnormal features in the battery operation process are converted into quantifiable and accumulative fault energy values, solving the problem that it is difficult to effectively evaluate early hidden fault features in traditional threshold diagnosis methods. Combined with the dynamic weight relationship, appropriate target features are selected under different working conditions for fault energy accumulation calculation, so that the fault energy accumulation value can truly reflect the fault development trend. Compared with traditional diagnostic methods, this calculation method has achieved a leap from "static threshold judgment" to "dynamic energy accumulation analysis", which can accurately capture feature changes in the early stage of the fault, greatly shorten the fault diagnosis time, and reduce the misdiagnosis rate and missed diagnosis rate caused by feature misjudgment. It provides a more scientific and accurate quantitative basis for early fault warning of lithium batteries, and effectively improves the safety and reliability of battery management systems.

[0019] 4. In one embodiment, historical battery pack operating data is collected to train an individual health status model, generating a personalized normal operating condition reference interval that dynamically adapts to the battery's aging level. This overcomes the bottleneck of traditional fixed thresholds that are difficult to match individual battery differences. This reference interval is combined with dynamic weighting rules (such as prioritizing cell temperature difference characteristics at low temperatures) to achieve a coordinated correction of the reference value and feature weight. When the cell health status dispersion exceeds the threshold, the reference interval is adjusted in multiple dimensions based on real-time operating status parameters (ambient temperature, charge and discharge rate). This ensures improved temperature difference characteristic fault tolerance under low temperature conditions and enhanced voltage anomaly sensitivity under high rate conditions, thereby reducing the false positive rate.

[0020] 5. In one embodiment, a multivariate time-series prediction model is constructed based on the accumulated fault energy value (incorporating dynamically selected feature values ​​and weights), achieving a transition from real-time diagnosis to trend prediction. This model compares the predicted fault energy growth rate with the gradient change of the dynamically corrected threshold in real time. When the predicted value exceeds the safety threshold, a graded advance warning is triggered. Simultaneously, a personalized baseline interval correction mechanism is combined to accurately output the remaining safe operating time. This method converts the accumulated fault energy value into a quantifiable risk trajectory, providing a sufficient window for handling critical faults such as thermal runaway, effectively improving the proactiveness and reliability of battery system safety warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0022] Figure 1A flowchart of a battery fault warning and diagnosis method provided by an embodiment of the present invention; Figure 2 A block diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0024] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture. If the specific posture changes, the directional indications will also change accordingly.

[0025] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited to "first" and "second" may explicitly or implicitly include at least one of such features. In addition, if "and / or" or "and / or" appears in the full text, its meaning includes three parallel solutions. Taking "A and / or B" as an example, it includes solution A, solution B, or solutions that satisfy both A and B. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in this field to implement. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0026] Reference Figure 1 One embodiment of the present invention provides a battery fault early warning and diagnosis method, comprising the following steps: S10, acquiring operating parameters of the battery pack in real time, and performing branch processing on the operating parameters based on a preset environmental noise intensity determination condition; If it is determined to be a strong noise environment, noise stripping processing is performed on the operating parameters to generate a preprocessing signal; if it is determined to be a non-strong noise environment, the operating parameters are used as the preprocessing signal; S20, extracting spatiotemporal features from the preprocessed signal, the extracted spatiotemporal features including signal energy features of a preset target frequency band, battery pack cell temperature difference distribution features, and cell voltage distribution features; performing weak signal enhancement processing on the extracted spatiotemporal features to generate an enhanced feature set; S30, based on the spatiotemporal features in the enhanced feature set, performing spatiotemporal fusion calculation according to a preset weight relationship to obtain a fault energy accumulation value; S40, dynamically modifying the fault determination threshold according to the battery health status and real-time environmental parameters; S50: When the fault energy accumulation value exceeds the fault determination threshold value after dynamic correction, a graded warning signal is output.

[0027] In this embodiment, interference signals are stripped away through environmental noise-adaptive branch processing, and weak fault characteristics of the battery pack are accurately captured under strong noise. The spatial feature extraction and signal enhancement technology of temperature difference and voltage distribution are combined to improve the detection sensitivity of early hidden faults. The spatiotemporal fusion calculation of the fault energy accumulation value is used to quantify the fault deterioration trend and predict the risk trajectory. Finally, the threshold is dynamically corrected based on the battery health status and environmental parameters to achieve accurate output of graded warning signals, which fundamentally solves the technical difficulties of traditional methods in high misdiagnosis rate, high missed diagnosis rate and response lag. In the lithium battery safety warning scenario, the fault detection time is shortened and the false alarm rate is reduced.

[0028] As described in step S10 above, the battery management system (BMS) collects real-time operating parameters of the battery pack, including internal state parameters such as the battery pack's current signal, cell voltage, and temperature. Based on preset environmental noise intensity criteria (e.g., voltage fluctuation standard deviation > 50mV or temperature sampling noise > 2°C), the operating parameters are processed in a branching manner. If a strong noise environment is determined (e.g., the current surge at vehicle startup or sensor noise in a low-temperature environment), the operating parameters are subjected to noise removal using a wavelet transform threshold denoising algorithm, preserving valid signal features and generating a preprocessed signal. If a non-strong noise environment is determined (e.g., normal operating conditions during steady driving), the operating parameters are directly used as the preprocessed signal. This noise branching mechanism prevents signal distortion in strong noise environments that could lead to misjudgment of faults, thereby improving data reliability.

[0029] As described in step S20 above, performing spatiotemporal feature extraction on the preprocessed signal includes: Time domain features: Calculate the signal energy characteristics of a preset target frequency band (e.g., 5-100 Hz) (avoiding low-frequency noise <5 Hz) to capture the weak electrical signal fluctuations associated with the chemical reactions inside the battery. Spatial features: Extract the temperature difference distribution characteristics of the battery pack cells (such as the proportion of areas with a temperature difference between cells greater than 5°C) and the voltage distribution characteristics of the cells (such as the number of cells with a voltage difference greater than 100mV).

[0030] The empirical mode decomposition (EMD) combined with the adaptive threshold method is used to enhance the weak signals of spatiotemporal features, suppress background noise interference, and generate an enhanced feature set including energy features, temperature difference features, and voltage features.

[0031] Through the fusion of spatiotemporal features and weak signal enhancement, the hidden characteristics of early battery failures (such as voltage fluctuations caused by micro-short circuits and local overheating temperature differences) are converted into recognizable explicit features.

[0032] As described in step S30 above, a spatiotemporal fusion calculation is performed based on the spatiotemporal features in the enhanced feature set according to preset weightings (e.g., a 40% weight for time-domain energy features, a 35% weight for temperature difference distribution features, and a 25% weight for voltage distribution features). This weighting is dynamically adjusted based on the battery's health status (e.g., an aging battery's weighting for temperature difference features is increased to 45%). The accumulated fault energy is then calculated through weighted summation to quantify the energy release level of the battery's internal fault.

[0033] Through the spatiotemporal fusion of dynamic weights, the one-sidedness of single feature diagnosis is solved and the collaborative characterization of multi-dimensional fault features is achieved.

[0034] As described in step S40 above, the fault determination threshold is dynamically adjusted based on the battery's state of health (SOH) and real-time environmental parameters (e.g., ambient temperature of -20°C, charge / discharge rate of 2C). When the SOH is less than 80%, the initial threshold is lowered by 15% to increase the fault sensitivity of aging batteries. When the ambient temperature is less than 0°C, the temperature-dependent threshold is adjusted in conjunction with the Arrhenius equation to compensate for the impact of low temperatures on battery performance.

[0035] This breaks the limitations of traditional fixed thresholds, allowing the fault judgment criteria to adapt to the actual health status and operating conditions of the battery in real time, reducing the missed diagnosis rate.

[0036] As described in step S50 above, when the accumulated fault energy exceeds the dynamically corrected fault determination threshold, a three-level hierarchical warning mechanism is triggered: Level 1 warning (energy accumulation value > 1.2 × threshold): The vehicle's battery management system (BMS) pushes a fault prompt to the instrument panel and adjusts the upper limit of the charge and discharge current; Level 2 warning (energy accumulation value > 1.5 × threshold): The battery thermal management system is linked to start the liquid cooling cycle and limit the charge and discharge power; Level 3 warning (energy accumulation value > 2.0 × threshold): locates abnormal cells based on voltage distribution characteristics (for example, the voltage of cell No. 37 deviates from the mean by 25%), and simultaneously triggers the main circuit power outage.

[0037] Through graded early warning, quantitative management of fault risks is achieved, differentiated handling strategies are provided for faults of different severity, and system safety is improved.

[0038] In one feasible embodiment, assume that a certain brand of electric vehicle is equipped with an NCM811 power battery pack (200 series-parallel structure) and is driving in a cold area with a temperature of -10°C. The battery management system (BMS) performs a fault warning through the following process: Step S10: Real-time collection of battery pack operating parameters reveals that the voltage fluctuation standard deviation reaches 80mV (exceeding the strong noise threshold of 50mV), which indicates a low-temperature and strong noise environment. Wavelet transform is used to remove noise from the voltage signal to retain the true voltage change trend.

[0039] Step S20: Extract the signal energy characteristics of the target frequency band (20-80 Hz), and find that the energy value is 30% higher than that under normal operating conditions. At the same time, it is detected that in the single-cell temperature difference distribution, the temperature difference between cells 35 and 40 reaches 7°C (exceeding the 5°C threshold). The weak temperature difference characteristics are enhanced using the EMD algorithm.

[0040] Step S30: Calculate the fault energy accumulation value based on the preset weights (time domain energy 40%, temperature difference 35%, voltage 25%). The current value is 0.78. ( is the initial threshold).

[0041] Step S40: Since the ambient temperature is -10°C and the battery SOH is 75%, the dynamic correction threshold is 0.9 (The initial threshold is lowered by 10%).

[0042] Step S50: When the vehicle climbs the slope and the charge / discharge rate rises to 2.5C, the fault energy accumulation value rises rapidly to 1.1×0.9 =0.99 , triggering a level one warning: the BMS reduces the charging current from 150A to 100A, and displays "Risk of local battery overheating, it is recommended to reduce the load" on the instrument panel.

[0043] Through the full-process fault diagnosis mechanism, an early warning is triggered when the battery experiences early local overheating. The response time is shorter than that of traditional fixed threshold solutions, avoiding missed faults due to noise interference in low-temperature environments, and ensuring the battery safety of electric vehicles under extreme working conditions.

[0044] In one embodiment, in step S10, the operating parameters of the battery pack are acquired in real time, and the operating parameters are subjected to branch processing based on a preset environmental noise intensity determination condition, specifically including: S11. Perform a primary determination based on the historical noise intensity records of the battery pack's environment. If the average noise intensity over a preset continuous period of time is lower than a preset safety noise threshold, the environment is directly determined to be a non-strong noise environment. S12. Otherwise, a secondary determination is performed to perform spectrum analysis on the currently acquired operating parameters and extract the energy of noise components above the preset cutoff frequency. If the energy of the noise components exceeds the dynamic noise threshold, the environment is determined to be a strong noise environment. Only when the secondary determination output is a strong noise environment, the noise stripping process is performed to generate the pre-processed signal.

[0045] This embodiment utilizes a dual-cascaded decision mechanism. First, a time-series historical noise intensity assessment is used to pre-determine the ambient noise status, avoiding misjudgments due to transient interference. Second, a frequency-band scanning noise determination based on operating parameters is performed to accurately capture high-frequency, strong noise components. Finally, noise stripping is initiated only when the secondary verification confirms a strong noise environment, enabling on-demand allocation of computing resources. This fundamentally avoids the issues of wasted computing power and residual interference caused by traditional single-stage determination, providing lightweight, real-time decision support for fault diagnosis in high-noise scenarios for power batteries.

[0046] As described in step S11 above, based on the battery pack's historical operation database, the system retrieves the environmental noise intensity records (such as voltage fluctuation standard deviation and temperature sampling noise) for a preset continuous period (e.g., one hour) in real time and calculates the average noise intensity. If this average value is lower than the preset safety noise threshold (e.g., voltage fluctuation standard deviation <50mV), the system directly determines that the noise level is not high, and the secondary determination process is skipped.

[0047] Through statistical analysis of historical noise, rapid screening can be achieved, redundant calculations in low-noise environments can be avoided, the time required for single noise determination can be shortened, and the real-time performance of the system can be improved.

[0048] As described in step S12 above, when the primary determination fails (e.g., the average noise intensity is ≥ 50 mV), it is necessary to dynamically correct the noise threshold to accurately identify the strong noise environment.

[0049] The significance of dynamically correcting the noise threshold is: The noise energy (high-frequency noise component) of the battery is affected by two core factors: Charge and discharge rate (C): During high-rate charge and discharge, the current density inside the battery changes dramatically, causing electrode polarization and intensified side reactions, resulting in a surge in noise energy.

[0050] Ambient temperature: Under low temperature conditions, the viscosity of the electrolyte increases and the ion migration speed slows down → "local polarization" is prone to occur at the electrode interface → the noise energy increases significantly.

[0051] If a fixed threshold (such as ) Determine a strong noise environment, which may exist: At high magnification / low temperature, the actual noise energy is much higher than → It will be misjudged as "strong noise" and over-trigger the noise stripping (causing loss of useful signal). At low magnification / high temperature, the actual noise energy is much lower than → “Strong noise” will be missed (leading to false alarms in fault diagnosis).

[0052] Therefore, a fast Fourier transform (FFT) spectrum analysis is performed on the current operating parameters (voltage and temperature signals) to extract the energy of high-frequency noise components above the preset cutoff frequency (such as 100 Hz). The dynamic noise threshold is adaptively adjusted according to the real-time operating parameters and the noise threshold is dynamically corrected using the following formula: Dynamic noise threshold = basic threshold × charge and discharge rate correction term × ambient temperature correction term.

[0053] Specifically: If the energy of the high-frequency noise component exceeds the dynamic threshold, it is determined to be a strong noise environment, triggering the wavelet transform noise stripping process.

[0054] Spectral analysis is used to distinguish between mechanical vibration noise (low frequency) and electrochemical reaction noise (high frequency). The threshold is dynamically corrected using the temperature-rate dual parameter. The threshold is automatically lowered in a low-temperature environment of -10°C to avoid missing pulse noise during cold start.

[0055] In a feasible embodiment, assume that a certain brand of electric vehicle is equipped with an NCM811 power battery pack (200 series-parallel structure) and is driving in a cold area with a temperature of -10°C: Initial judgment stage: The noise records of the battery pack for the past hour were retrieved, and it was found that the standard deviation of voltage fluctuation reached 75mV during the low-temperature startup phase, but dropped to 40mV during stable driving. The overall average noise intensity was 55mV (≥50mV safety threshold). The primary judgment failed and entered the secondary judgment.

[0056] Secondary determination stage: Perform FFT analysis on the current operating parameters and extract the energy of high-frequency noise components above 100Hz as 36m At this time, the charge and discharge rate is 2.0C, the ambient temperature is -10℃, and the dynamic noise threshold is calculated as: Because 36m > , it is determined to be a strong noise environment, triggering the noise stripping process, and using the wavelet transform threshold denoising algorithm to perform noise stripping processing on the operating parameters.

[0057] In one embodiment, the weak signal enhancement process in step S20 includes: a) High-frequency signal enhancement path: Perform frequency domain decomposition on the signal energy characteristics of the preset target frequency band, and extract the frequency band component with a preset bandwidth and a characteristic frequency determined by the real-time battery status as the center frequency point; Calculating a dynamic gain value based on a ratio of the signal amplitude of the frequency band component to the ambient noise amplitude and applying the dynamic gain value to the frequency band component to obtain a high-frequency enhancement component; b) Low-frequency trend maintenance path: Perform sliding average filtering on the temperature difference distribution characteristics and voltage distribution characteristics of the battery pack cells, where the filter window length is inversely proportional to the battery charge and discharge rate, to obtain low-frequency enhancement features; c) Cross-scale fusion: The high-frequency enhancement component and the low-frequency enhancement feature are weightedly superimposed according to a preset fusion coefficient to generate the enhancement feature set.

[0058] In this embodiment, a cross-scale dual-path enhancement mechanism is adopted to dynamically lock the characteristic frequency band based on the real-time status of the battery to achieve targeted amplification of high-frequency fault signals; combined with dynamic adaptive filtering of the low-frequency feature sliding window, the slow degradation trend characteristics of the battery are fully preserved; and finally, an enhanced feature set is output through weighted fusion, which breaks through the noise interference limit while maintaining the identifiability of all-scale fault modes.

[0059] Specifically: a) High-frequency signal enhancement path (to solve the problem of environmental noise masking high-frequency fault signals).

[0060] The Fast Fourier Transform (FFT) converts the time-domain signal into a frequency-domain energy distribution, focusing on a preset target frequency band (e.g., 10-100 Hz) that is strongly correlated with battery failures. This frequency band typically contains fault signatures such as poor electrode contact and electromagnetic noise generated by side reactions. This separates the high-frequency fault signatures in the mixed signal from the time-domain ambient noise, preventing interference from low-frequency noise such as ambient vibration and motor electromagnetic interference.

[0061] Based on real-time battery state parameters (such as SOC, SOH, and temperature), the characteristic frequency is dynamically calculated using electrochemical models (such as the PNGV equivalent circuit model and the porous electrode theoretical model) combined with these real-time battery state parameters (SOC, SOH, and temperature). For example, the PNGV model derives the relationship between electrode polarization impedance and frequency. As the SEI thickens due to battery aging (decreased SOH), the characteristic frequency of the polarization impedance will drift from 50 Hz to 30 Hz. This model is used to calculate the fault-sensitive frequency under the current operating conditions in real time, ensuring the accuracy of frequency band component extraction. Alternatively, a machine learning algorithm (such as random forest or long short-term memory (LSTM)) can be used to construct a battery state-characteristic frequency mapping model using historical battery data (fault characteristic frequencies at different SOC, SOH, and temperatures) as a training set. The current battery state parameters (e.g., SOC = 85%, SOH = 78%, and temperature 25°C) are input in real time, and the pre-trained model outputs the characteristic frequency under the corresponding operating conditions (e.g., 42 Hz). The algorithm automatically learns the correlation between battery state changes and characteristic frequency drift, eliminating the need for manual parameter adjustment. Centered on this frequency, the frequency band components of a preset bandwidth (e.g., ±5Hz) are intercepted to accurately lock in fault-sensitive signals. This overcomes the limitations of traditional fixed-band analysis and dynamically adapts to characteristic frequency offsets caused by changes in battery status to avoid missed fault signals. Unlike traditional methods that fix characteristic frequencies (which can easily miss faults due to changes in battery status), this solution dynamically calculates fault-sensitive frequencies under current operating conditions through electrochemical models (e.g., PNGV) or machine learning algorithms (e.g., LSTM). For example, when the battery SOH drops from 90% to 70%, the characteristic frequency drifts from 50Hz to 35Hz. The model can capture this change in real time, ensuring that the frequency band component extraction covers the actual fault signal.

[0062] The ratio of the fault signal amplitude to the ambient noise amplitude within the target frequency band is calculated in real time. If the ratio is greater than 1 (e.g., 3mV fault signal vs. 2mV noise), the signal is amplified proportionally (e.g., gain = 10 × 1.5 = 15 times). If the ratio is less than 1, the signal is maintained or suppressed. Dynamic gain adjustment enables targeted enhancement of the actual fault signal. This avoids the problem of synchronous noise amplification caused by fixed gain, allowing weak fault signals to be highlighted even in high-noise environments (such as during motor startup).

[0063] b) Low-frequency trend maintenance path (solves the problem of operating condition fluctuations masking low-frequency fault signals).

[0064] A sliding average filter algorithm is used to smooth low-frequency, slowly varying signals such as cell temperature differences (normal temperature differences <0.5°C) and cell voltage differences (normal voltage differences <50mV). This algorithm calculates the average of N consecutive sampling points to suppress transient fluctuations during the charge and discharge process (such as voltage spikes during fast charging). It also preserves slowly changing trends in temperature and voltage (such as a 0.3°C temperature difference increase over 10 minutes) and filters out high-frequency noise interference under normal operating conditions.

[0065] Dynamically adjust the filter window duration according to the charge and discharge rate (Formula: , The benchmark duration is 60 seconds. For example, at 3C high-rate charging and discharging, the window duration is shortened to 20 seconds to quickly respond to transient fluctuations; at 0.5C low-rate charging and discharging, it is extended to 120 seconds to avoid oversmoothing the true trend. This addresses the shortcomings of traditional fixed-window filtering: excessively long windows at high rates can lead to trend lag, while excessively short windows at low rates can cause residual noise, achieving dynamic adaptation of the filtering effect to the operating conditions.

[0066] c) Cross-scale fusion (solving the problem of high-frequency / low-frequency signal separation).

[0067] Fusion weights (e.g., 0.6 for high frequency and 0.4 for low frequency) are preset based on the battery fault type (e.g., thermal runaway, overcharge, micro-short circuit), or dynamically adjusted based on real-time diagnostic needs. High-frequency enhanced fault signatures (e.g., 45mV high-frequency noise) are weighted and merged with low-frequency trend signals (e.g., 0.8°C temperature differential increase) to generate an enhanced feature set containing full-scale fault information. This overcomes the one-sidedness of single-frequency analysis and integrates high-frequency early fault signatures with low-frequency fault development trends, providing complete and accurate input data for subsequent fault energy calculations.

[0068] In one embodiment, in step S30, the preset weight relationship is a dynamic weight relationship, and the dynamic weight relationship is dynamically adjusted according to the real-time working state parameters of the battery pack, wherein: The real-time working status parameter includes at least one of the ambient temperature, battery charge and discharge rate, and battery state of charge; The dynamic weight relationship includes a time domain feature weight, a first spatial domain feature weight, and a second spatial domain feature weight, wherein the time domain feature is a signal energy feature, the first spatial domain feature is a battery pack cell temperature difference distribution feature, and the second spatial domain feature is a battery pack cell voltage distribution feature; the dynamic adjustment of the dynamic weight relationship satisfies: When the ambient temperature is lower than a temperature threshold, increasing the weight of the first spatial domain feature; When the charge and discharge rate is higher than the rate threshold, increasing the weight of the time domain feature; The second spatial domain feature weight remains within a preset threshold range.

[0069] In this embodiment, by introducing a dynamic weight relationship, the weights of the time domain features, the first spatial domain features, and the second spatial domain features are dynamically adjusted based on the real-time operating status parameters of the battery pack (ambient temperature, battery charge and discharge rate, battery state of charge, etc.), effectively solving the problem that the traditional fixed-weight diagnosis method cannot accurately match the importance of fault features under different operating conditions. In low-temperature environments, increasing the weight of the temperature difference distribution feature of the battery pack cells can capture potential faults caused by uneven temperature in advance; under high charge and discharge rate conditions, increasing the weight of the signal energy feature can accurately identify abnormal energy fluctuations under high load. This dynamic weight mechanism breaks the "one-size-fits-all" limitation of the traditional threshold diagnosis method, avoids misdiagnosis and missed diagnosis due to environmental changes or differences in operating conditions, makes the fault diagnosis results more consistent with the actual operating status of the battery, and significantly improves the accuracy and reliability of fault diagnosis.

[0070] Specifically, real-time working status parameters affect weight distribution in the following ways: Ambient temperature: reflects the thermal environment in which the battery pack operates. In a low temperature environment, the chemical reaction rate inside the battery decreases, which can easily lead to local temperature abnormalities and affect battery life.

[0071] Charge and discharge rate: reflects the size of the battery's charge and discharge current. High-rate charge and discharge will intensify the internal polarization of the battery, generating more heat and voltage fluctuations.

[0072] Battery State of Charge (SOC): Indicates the current battery charge. The internal resistance and thermal characteristics of the battery are different in different SOC ranges, and the fault manifestation characteristics are different.

[0073] For dynamic weight adjustment strategy: (1) Time domain feature weight (signal energy feature).

[0074] Technical issue: During high-rate charging and discharging (such as fast charging), the internal reaction of the battery is intense, and the energy characteristics (time domain characteristics) of voltage / temperature fluctuations can better reflect the changes in battery status.

[0075] Adjustment logic: When the charge / discharge rate exceeds the rate threshold (e.g., 2C), the time domain feature weight is automatically increased. By increasing the proportion of signal energy features, the sensitivity to transient faults (e.g., lithium deposition and poor contact) at high rates is enhanced.

[0076] Mathematical expression: Time domain feature weight = basic weight × (1 + × (charge and discharge rate - rate threshold)), is the gain coefficient.

[0077] (2) The first spatial feature weight (single-unit temperature difference distribution feature).

[0078] Technical issue: In a low temperature environment (such as <0°C), the temperature difference between each cell in the battery pack is aggravated, which can easily cause local overcooling or thermal runaway.

[0079] Adjustment logic: When the ambient temperature falls below a threshold (e.g., 5°C), the weight of the first spatial domain feature is automatically increased. This amplifies the impact of individual temperature difference distribution features and focuses on monitoring temperature imbalances at low temperatures.

[0080] Mathematical expression: The first spatial feature weight = basic weight × (1 + × (temperature threshold - ambient temperature), is the gain coefficient.

[0081] (3) Second spatial feature weight (single cell voltage distribution feature).

[0082] Technical issue: Cell voltage difference is a key indicator of battery pack consistency, but excessive focus may lead to neglect of other fault characteristics.

[0083] Adjustment logic: The weight of the second spatial domain feature is limited to a preset threshold range (such as 0.3-0.5) to ensure its stable monitoring of battery pack consistency while avoiding excessively high weights that mask other fault signals.

[0084] Mathematical expression: The second spatial domain feature weight = min (max (basic weight × f (SOC), lower limit value), upper limit value), where f (SOC) is the adjustment function related to SOC.

[0085] Dynamic weight synergy mechanism: Dynamically adjust the weights of each feature through real-time working status parameters to achieve multi-dimensional and accurate monitoring of battery failures: Low-temperature scenarios: Increase the weight of the single-unit temperature difference distribution feature to prioritize identifying local performance degradation caused by low temperatures.

[0086] High-magnification scenario: Increase the weight of signal energy features and focus on capturing transient anomalies under high current.

[0087] Full operating range: stabilizes the cell voltage distribution characteristic weights and continuously monitors the battery pack consistency.

[0088] Compared with the traditional fixed weight method, this dynamic weight allocation mechanism can more flexibly adapt to the real-time working status of the battery pack, significantly improve the recognition of fault characteristics under different working conditions, and provide more reliable feature input for the subsequent fault type determination in step S40.

[0089] In one embodiment, in step S30, the fault energy accumulation value is calculated as follows: The temporal and spatial features in the enhanced feature set are mapped to the characteristic values ​​and reference values ​​of the fault energy calculation formula, and the fault energy cumulative value is calculated using the following formula; in, For the moment a target characteristic value of the battery pack after weak signal enhancement, wherein the target characteristic value is selected from at least one of a signal energy characteristic, a battery pack single cell temperature difference distribution characteristic, and a battery pack single cell voltage distribution characteristic; is the normal operating condition reference value corresponding to the target characteristic value; is the characteristic sampling interval; To calculate the cumulative duration; Moreover, the selection and correspondence of the target characteristic value The assignment of is associated and adapted with the dynamic weight relationship, specifically satisfying: When the ambient temperature is lower than the temperature threshold, the battery pack temperature difference distribution characteristics are preferentially selected as Calculate the weight ratio corresponding to the increase in the normal baseline value of the single temperature difference; When the charge and discharge rate is higher than the rate threshold, the signal energy feature is preferentially selected as Calculate the weight ratio corresponding to the normal baseline value of the increased signal energy.

[0090] In this embodiment, by deeply integrating the enhanced feature set with the fault energy calculation formula, the weak abnormal features in the battery operation process are converted into quantifiable and accumulative fault energy values, solving the problem that it is difficult to effectively evaluate the early hidden fault features in the traditional threshold diagnosis method. Combined with the dynamic weight relationship, appropriate target features are selected under different working conditions for fault energy accumulation calculation, so that the fault energy accumulation value can truly reflect the fault development trend. Compared with traditional diagnostic methods, this calculation method has achieved a leap from "static threshold judgment" to "dynamic energy accumulation analysis", which can accurately capture feature changes in the early stage of the fault, greatly shorten the fault diagnosis time, and reduce the misdiagnosis rate and missed diagnosis rate caused by feature misjudgment. It provides a more scientific and accurate quantitative basis for early fault warning of lithium batteries, effectively improving the safety and reliability of the battery management system.

[0091] Specifically, the degree to which battery characteristics deviate from the normal state is quantified in the form of "energy accumulation". The greater the deviation and the longer the duration, the higher the fault energy, which intuitively reflects the severity and development trend of the fault.

[0092] Adaptation rules of target features and benchmark values, target features Selection and normal operating conditions benchmark value The assignment of is linked to the dynamic weight relationship. The specific adaptation logic is: (1) When the ambient temperature is lower than the threshold (e.g. <5°C).

[0093] In low-temperature environments, the temperature difference distribution characteristics of battery cells are more sensitive to faults (such as temperature difference abnormalities caused by local cold starts). Therefore: Prioritize the single temperature difference distribution feature as the target feature (such as the maximum temperature difference between cells); Dynamic Adjustment : Increase the weight of the "normal benchmark value of single-unit temperature difference" (such as relaxing the upper limit of normal temperature difference from 0.5℃ to 0.8℃ to compensate for natural temperature fluctuations under low temperatures).

[0094] This prevents normal temperature difference fluctuations in low-temperature environments from being misjudged as faults, while strengthening the identification of real temperature difference faults (such as coolant blockage).

[0095] (2) When the charge and discharge rate is higher than the threshold (e.g. >2C).

[0096] During high-rate charge and discharge, signal energy characteristics (such as voltage fluctuation energy) can better reflect battery internal polarization, lithium deposition and other faults. Therefore: Prioritize signal energy features as target features (such as the signal energy value of the target frequency band); Dynamic Adjustment : Increase the weight of "normal signal energy benchmark value" (such as increasing the normal energy limit from 30 Increased to 50 , adapted to normal energy fluctuations at high rates).

[0097] This is to avoid misjudgment caused by normal energy fluctuations at high magnification, while accurately capturing high-magnification-specific faults (such as energy mutations caused by diaphragm puncture).

[0098] Collaborative mechanism for energy accumulation calculation: Through "dynamic feature selection + benchmark value adaptation", the fault energy accumulation calculation is realized: Adaptive working conditions: Focusing on temperature difference in low-temperature scenarios and energy in high-rate scenarios, solving the one-sidedness of traditional single-feature calculations; Dynamic compensation of baseline value: adjust the normal baseline according to the working conditions to avoid misjudgment caused by environmental / working condition fluctuations; Energy quantification is intuitive: Deviations are accumulated in an integral form to convert “instantaneous fault characteristics” into “traceable energy indicators” to facilitate subsequent fault threshold determination (step S40).

[0099] In one embodiment, the setting of the normal operating condition reference value further includes: Collect historical operating data of the battery pack and pre-process it to obtain standardized feature data applied to the machine model as input data; The standardized feature data obtained after preprocessing is used to train a battery individual health state model using a machine learning algorithm to obtain a trained battery individual health state model. The battery individual health state model receives the operating parameters of the battery pack in real time and outputs a personalized normal operating condition reference interval that dynamically adapts to the battery life cycle; wherein the personalized normal operating condition reference interval is dynamically updated with the number of battery cycles and the degree of aging; When the health status dispersion of single cells in the same batch of battery packs is monitored to exceed a preset threshold, the adaptive correction of the personalized normal operating condition reference interval is triggered. The correction process is combined with real-time working status parameters to perform multi-dimensional dynamic adjustment.

[0100] In this embodiment, historical battery pack operating data is collected to train an individual health status model, generating a personalized normal operating condition reference interval that dynamically adapts to the battery's aging process. This overcomes the bottleneck of traditional fixed thresholds that struggle to adapt to individual battery differences. This reference interval, combined with dynamic weighting rules (for example, prioritizing cell temperature difference characteristics at low temperatures), enables coordinated correction of reference values ​​and feature weights. When the cell health status dispersion exceeds a threshold, the reference interval is adjusted across multiple dimensions based on real-time operating status parameters (ambient temperature, charge and discharge rate). This ensures improved temperature difference characteristic tolerance at low temperatures and enhanced voltage anomaly sensitivity at high rate conditions, thereby reducing the probability of false positives.

[0101] Specifically, the setting of the normal operating condition benchmark value is achieved through "historical data driven + model dynamic adaptation + discreteness trigger correction". The specific logic is as follows: (1) Historical data preprocessing: constructing standardized input.

[0102] Collect historical operating data of the battery pack throughout its life cycle (covering parameters such as voltage, temperature, charge and discharge rate under different operating conditions and health status) and preprocess it through the following steps: Outlier filtering: Eliminate abnormal data caused by sensor failure and extreme working conditions (such as short circuit protection); Normalization: Mapping features of different dimensions (such as voltage 0-5V, temperature -20-60°C) to the same scale (such as 0-1); Time series feature construction: Extract feature change trends within a sliding window (such as the standard deviation of voltage fluctuation within 5 minutes).

[0103] This solves the problems of "dimensional heterogeneity, uneven distribution, and noise" in the original data, generates high-quality standardized feature data, and provides reliable input for machine learning model training.

[0104] (2) Machine learning training: dynamically adapted health status model.

[0105] The preprocessed standardized feature data is fed into a machine learning algorithm (such as LSTM, random forest) to train the battery individual health status model: Model input: real-time operating parameters (voltage, temperature, charge and discharge rate, number of cycles, etc.); Model output: personalized normal operating condition reference range (such as the normal voltage fluctuation range and temperature difference range of a single battery); Dynamic update mechanism: The model adjusts the reference range in real time according to the number of battery cycles and aging degree (SOH decay) (for example, after 1000 cycles, the upper limit of normal voltage fluctuation is relaxed from 20mV to 30mV).

[0106] It breaks through the defect of traditional "fixed benchmark value" and enables the normal operating condition benchmark to dynamically adapt with battery aging and cycle number, solving the problem of "misjudgment / missed judgment caused by the same benchmark for new and aged batteries".

[0107] (3) Discreteness trigger correction: multi-dimensional dynamic adjustment.

[0108] Real-time calculation of the health status dispersion of cells in the same batch of battery packs (such as the standard deviation of each cell's SOH). When the dispersion exceeds the preset threshold (such as the SOH standard deviation > 10%): Trigger condition: Determining that the battery pack consistency has deteriorated (e.g., some cells have aged prematurely); Correction mechanism: Combined with real-time operating status parameters (ambient temperature, charge and discharge rate), the normal operating condition benchmark interval is adjusted in multiple dimensions: In low-temperature environments, the single-cell temperature difference reference range is narrowed (enhanced identification of temperature difference anomalies); During high-rate charge and discharge, the voltage fluctuation reference range is expanded (tolerating normal fluctuations at high rates).

[0109] To solve the problem of "individual benchmark failure after battery pack consistency deteriorates", discreteness monitoring is used to trigger adaptive correction, so that the benchmark interval is adapted to individual aging and compatible with batch differences and real-time operating conditions.

[0110] Through the entire process of "historical data preprocessing → machine learning modeling → discreteness trigger correction", a dynamic, personalized, and self-adaptive normal operating condition benchmark system has been built: Historical data-driven approach ensures the benchmark matches the battery's full life cycle characteristics; The machine learning model enables dynamic updates of the benchmark as it ages and cycles; Discreteness triggers correction to solve the benchmark failure problem after batch consistency deteriorates.

[0111] This scheme is the fault energy accumulation calculation ( The assignment of the normal operating condition provides an accurate and dynamic benchmark for normal operating conditions, enabling fault diagnosis to adapt to individual battery differences and be compatible with complex operating condition fluctuations, fundamentally improving the accuracy and robustness of fault warnings.

[0112] In one embodiment, after step S30, the following steps are further included: Obtain historical fault data and real-time working status parameters, combine them with the calculation results of the fault energy accumulation value, perform data standardization preprocessing, and generate a multivariate time series feature data set; Based on the standardized multivariate time series feature data set, a multivariate time series prediction model is constructed using a machine learning algorithm. The multivariate time series prediction model is used to predict the fault energy growth rate in the future period; The predicted fault energy growth rate is compared with the dynamically corrected fault judgment threshold change gradient in real time. When the predicted fault energy growth rate exceeds the product of the preset safety factor and the fault judgment threshold change gradient, the advance warning mechanism is triggered; According to the relative deviation between the predicted fault energy growth rate and the change gradient of the fault judgment threshold, multiple warning levels are divided, and the remaining safe operation time is calculated based on the current fault energy accumulation value, the dynamically corrected fault judgment threshold and the predicted fault energy growth rate.

[0113] In this embodiment, a multivariate time-series prediction model is constructed based on the accumulated fault energy value (incorporating dynamically selected feature values ​​and weights), achieving a leap from real-time diagnosis to trend prediction. By comparing the predicted fault energy growth rate with the gradient of a dynamically corrected threshold in real time, a graded advance warning is triggered when the predicted value exceeds the safety threshold. Simultaneously, a personalized baseline interval correction mechanism is incorporated to accurately output the remaining safe operating time. This method converts the accumulated fault energy value into a quantifiable risk trajectory, providing a sufficient window for handling critical faults such as thermal runaway, effectively improving the proactiveness and reliability of battery system safety warnings.

[0114] Specifically, after step S30, the additional advance warning mechanism is implemented through "multivariate time series analysis + dynamic threshold comparison + remaining life prediction". The specific logic is as follows: (1) Construction of multivariate time series features.

[0115] Data collection: Integrate historical fault data (such as overcharging and thermal runaway cases), real-time operating status parameters (temperature, charge and discharge rate, SOC), and the accumulated fault energy value calculated in step S30; Standardization preprocessing: Map heterogeneous data (such as temperature -20~60℃, fault energy 0~1000J) to a unified scale (such as [-1,1]); Time series feature generation: Extract feature change trends (such as fault energy growth rate and temperature gradient) based on a sliding window (such as 5 minutes) to construct a multivariate time series dataset.

[0116] By converting "static fault indicators" into "dynamic timing features", the temporal correlation of fault evolution is captured, providing structured input for machine learning predictions.

[0117] (2) Prediction of fault energy growth rate.

[0118] Model construction: Using time series prediction algorithms such as LSTM and Transformer, we train a multivariate time series prediction model based on a standardized data set; Model output: Predict the fault energy growth rate in the future period (e.g. 30 minutes) ; Dynamic adaptation: Multivariable time series prediction model parameters are updated in real time as battery age (SOH decreases) and operating conditions change (temperature fluctuations) to ensure prediction accuracy.

[0119] By sensing the potential change trend of fault energy in advance, the hysteresis problem of traditional "threshold-triggered warning" can be solved.

[0120] (3) Dynamic threshold comparison and advance warning triggering Dynamic threshold gradient: Based on the dynamic weight relationship, the gradient of the fault judgment threshold is calculated in real time (e.g. the rate of threshold drop accelerates at low temperatures); Safety factor product: preset safety factor (As in 1.2), define the safety boundary as · ; Warning trigger condition: When the predicted fault energy growth rate > · When the advance warning is triggered.

[0121] Through the dynamic comparison of "predicted growth rate vs threshold change gradient", early warning can be given before the fault energy approaches the threshold, reserving a time window for handling.

[0122] (4) Multi-level warning level classification and remaining life calculation.

[0123] i) Warning level classification: ii) Calculation of remaining safe operating time: in, is the current fault judgment threshold, It is the current fault energy accumulation value.

[0124] "Visualization of risk levels" is achieved through graded early warning, supporting differentiated response strategies; remaining life prediction quantifies safety boundaries and provides accurate time reference for maintenance decisions.

[0125] Through the entire chain of "time series feature extraction → growth rate prediction → dynamic threshold comparison → risk quantification", an advanced, hierarchical, and quantifiable early warning system has been established: Multivariable time series models capture the evolution of faults, breaking through the limitations of traditional single-variable early warning systems; The dynamic threshold comparison mechanism adapts to complex working conditions and battery aging to avoid false positives and false negatives. Remaining life calculation provides a time window for preventive maintenance and reduces safety risks.

[0126] This solution works in deep collaboration with the aforementioned dynamic weight relationship and fault energy accumulation calculation to form a complete diagnostic chain from "feature enhancement → fault quantification → risk prediction", significantly improving the safety and reliability of the battery system.

[0127] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0128] In one embodiment, a battery fault warning and diagnosis system is provided. The battery fault warning and diagnosis system corresponds to the battery fault warning and diagnosis method in the above embodiment. The battery fault warning and diagnosis system includes: A data acquisition module, configured to acquire operating parameters of the battery pack in real time, perform branch processing on the operating parameters based on a preset environmental noise intensity determination condition, and generate a preprocessing signal; a noise stripping module, configured to perform noise stripping processing on the operating parameters and generate a preprocessing signal when determining that the environment is a strong noise environment; A feature extraction module is used to extract spatiotemporal features from the preprocessed signal, wherein the extracted spatiotemporal features include signal energy features of a preset target frequency band, battery pack cell temperature difference distribution features, and cell voltage distribution features; and perform weak signal enhancement processing on the extracted spatiotemporal features to generate an enhanced feature set; A fusion calculation module is used to perform spatiotemporal fusion calculation based on the spatiotemporal features in the enhanced feature set according to a preset weight relationship to obtain a fault energy accumulation value; Dynamic correction module, used to dynamically correct the fault judgment threshold according to the battery health status and real-time environmental parameters; The early warning output module is used to output a graded early warning signal when the accumulated value of the fault energy exceeds the fault judgment threshold value after dynamic correction.

[0129] The specific definition of a battery fault warning and diagnostic system can be found in the definition of a battery fault warning and diagnostic method above and will not be repeated here. Each module in the above-mentioned battery fault warning and diagnostic system can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor of the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0130] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 2 The computer device includes a processor, a memory, a network interface, and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used for data storage, data processing, data analysis, etc. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a battery fault early warning and diagnosis method is implemented.

[0131] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a battery fault warning and diagnosis method is implemented.

[0132] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a battery failure early warning and diagnosis method is implemented.

[0133] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0134] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0135] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A battery failure early warning and diagnosis method, characterized in that: The steps include: Acquire operating parameters of the battery pack in real time and perform branching processing on the operating parameters based on a preset environmental noise intensity determination condition; if the environment is determined to be a strong noise environment, perform noise stripping processing on the operating parameters to generate a preprocessed signal; if the environment is determined to be a non-strong noise environment, use the operating parameters as the preprocessed signal; Extracting spatiotemporal features from the preprocessed signal, the extracted spatiotemporal features including signal energy features of a preset target frequency band, battery pack cell temperature difference distribution features, and cell voltage distribution features; performing weak signal enhancement processing on the extracted spatiotemporal features to generate an enhanced feature set; Based on the spatiotemporal features in the enhanced feature set, a spatiotemporal fusion calculation is performed according to a preset weight relationship to obtain a fault energy accumulation value; Dynamically adjust the fault judgment threshold based on battery health status and real-time environmental parameters; When the fault energy accumulation value exceeds the fault determination threshold value after dynamic correction, a graded warning signal is output.

2. The battery failure warning and diagnosis method according to claim 1, characterized in that: The step of acquiring the operating parameters of the battery pack in real time and performing branch processing on the operating parameters based on a preset environmental noise intensity determination condition specifically includes: Perform a primary determination based on the historical noise intensity records of the battery pack's environment. If the average noise intensity over the past preset continuous time period is lower than the preset safety noise threshold, the environment is directly determined to be a non-strong noise environment. Otherwise, a second-level judgment is executed to perform spectrum analysis on the currently acquired operating parameters and extract the energy of noise components above the preset cutoff frequency. If the energy of the noise component exceeds the dynamic noise threshold, it is determined to be a strong noise environment. Only when the secondary determination output is a strong noise environment, a noise stripping process is performed to generate the pre-processed signal.

3. The battery failure warning and diagnosis method according to claim 1, characterized in that: The weak signal enhancement processing includes: a) High-frequency signal enhancement path: Perform frequency domain decomposition on the signal energy characteristics of the preset target frequency band, and extract the frequency band component with a preset bandwidth and a characteristic frequency determined by the real-time battery status as the center frequency point; Calculating a dynamic gain value based on a ratio of the signal amplitude of the frequency band component to the ambient noise amplitude and applying the dynamic gain value to the frequency band component to obtain a high-frequency enhancement component; b) Low-frequency trend maintenance path: Perform sliding average filtering on the temperature difference distribution characteristics and voltage distribution characteristics of the battery pack cells, where the filter window length is inversely proportional to the battery charge and discharge rate, to obtain low-frequency enhancement features; c) Cross-scale fusion: The high-frequency enhancement component and the low-frequency enhancement feature are weightedly superimposed according to a preset fusion coefficient to generate the enhancement feature set.

4. The battery failure warning and diagnosis method according to claim 1, wherein: In the step of performing spatiotemporal fusion calculation based on the spatiotemporal features in the enhanced feature set according to a preset weight relationship to obtain the fault energy accumulation value, the preset weight relationship is a dynamic weight relationship, and the dynamic weight relationship is dynamically adjusted according to the real-time working status parameters of the battery pack, wherein: The real-time working status parameter includes at least one of the ambient temperature, battery charge and discharge rate, and battery state of charge; The dynamic weight relationship includes a time domain feature weight, a first spatial domain feature weight, and a second spatial domain feature weight, wherein the time domain feature is a signal energy feature, the first spatial domain feature is a battery pack cell temperature difference distribution feature, and the second spatial domain feature is a battery pack cell voltage distribution feature; the dynamic adjustment of the dynamic weight relationship satisfies: When the ambient temperature is lower than a temperature threshold, increasing the weight of the first spatial domain feature; When the charge and discharge rate is higher than the rate threshold, increasing the weight of the time domain feature; The second spatial domain feature weight remains within a preset threshold range.

5. The battery failure warning and diagnosis method according to claim 4, characterized in that: In the step of performing spatiotemporal fusion calculation based on the spatiotemporal features in the enhanced feature set according to a preset weight relationship to obtain the fault energy accumulation value, the calculation method of the fault energy accumulation value is: The temporal and spatial features in the enhanced feature set are mapped to the characteristic values ​​and reference values ​​of the fault energy calculation formula, and the fault energy cumulative value is calculated using the following formula; in, For the moment a target characteristic value of the battery pack after weak signal enhancement, wherein the target characteristic value is selected from at least one of a signal energy characteristic, a battery pack single cell temperature difference distribution characteristic, and a battery pack single cell voltage distribution characteristic; is the normal operating condition reference value corresponding to the target characteristic value; is the characteristic sampling interval; To calculate the cumulative duration; Moreover, the selection and correspondence of the target characteristic value The assignment of is associated and adapted with the dynamic weight relationship, specifically satisfying: When the ambient temperature is lower than the temperature threshold, the battery pack temperature difference distribution characteristics are preferentially selected as Calculate the weight ratio corresponding to the increase in the normal baseline value of the single temperature difference; When the charge and discharge rate is higher than the rate threshold, the signal energy feature is preferentially selected as Calculate the weight ratio corresponding to the normal baseline value of the increased signal energy.

6. The battery failure warning and diagnosis method according to claim 5, characterized in that: The setting of the normal operating condition reference value also includes: Collect historical operating data of the battery pack and pre-process it to obtain standardized feature data applied to the machine model as input data; The standardized feature data obtained after preprocessing is used to train a battery individual health state model using a machine learning algorithm to obtain a trained battery individual health state model. The battery individual health state model receives the operating parameters of the battery pack in real time and outputs a personalized normal operating condition reference interval that dynamically adapts to the battery life cycle; wherein the personalized normal operating condition reference interval is dynamically updated with the number of battery cycles and the degree of aging; When the health status dispersion of single cells in the same batch of battery packs is monitored to exceed a preset threshold, the adaptive correction of the personalized normal operating condition reference interval is triggered. The correction process is combined with real-time working status parameters to perform multi-dimensional dynamic adjustment.

7. The battery failure warning and diagnosis method according to claim 5, characterized in that: After the step of performing spatiotemporal fusion calculation based on the spatiotemporal features in the enhanced feature set according to a preset weight relationship to obtain the fault energy accumulation value, the following steps are also included: Obtain historical fault data and real-time working status parameters, combine them with the calculation results of the fault energy accumulation value, perform data standardization preprocessing, and generate a multivariate time series feature data set; Based on the standardized multivariate time series feature data set, a multivariate time series prediction model is constructed using a machine learning algorithm. The multivariate time series prediction model is used to predict the fault energy growth rate in the future period; The predicted fault energy growth rate is compared with the dynamically corrected fault judgment threshold change gradient in real time. When the predicted fault energy growth rate exceeds the product of the preset safety factor and the fault judgment threshold change gradient, the advance warning mechanism is triggered; According to the relative deviation between the predicted fault energy growth rate and the change gradient of the fault judgment threshold, multiple warning levels are divided, and the remaining safe operation time is calculated based on the current fault energy accumulation value, the dynamically corrected fault judgment threshold and the predicted fault energy growth rate.

8. A battery fault warning and diagnosis system, used to implement the steps of a battery fault warning and diagnosis method according to any one of claims 1 to 7, characterized in that: include: A data acquisition module, configured to acquire operating parameters of the battery pack in real time, perform branch processing on the operating parameters based on a preset environmental noise intensity determination condition, and generate a preprocessing signal; a noise stripping module, configured to perform noise stripping processing on the operating parameters and generate a preprocessing signal when determining that the environment is a strong noise environment; A feature extraction module is used to extract spatiotemporal features from the preprocessed signal, wherein the extracted spatiotemporal features include signal energy features of a preset target frequency band, battery pack cell temperature difference distribution features, and cell voltage distribution features; Perform weak signal enhancement processing on the extracted spatiotemporal features to generate an enhanced feature set; A fusion calculation module is used to perform spatiotemporal fusion calculation based on the spatiotemporal features in the enhanced feature set according to a preset weight relationship to obtain a fault energy accumulation value; Dynamic correction module, used to dynamically correct the fault judgment threshold according to the battery health status and real-time environmental parameters; The early warning output module is used to output a graded early warning signal when the accumulated value of the fault energy exceeds the fault judgment threshold value after dynamic correction.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the battery fault early warning and diagnosis method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the battery fault early warning and diagnosis method according to any one of claims 1 to 7 are implemented.

Images