An intelligent fault detection method, device and storage medium for a battery pack

Abstract

The application discloses an intelligent fault detection method and device for a battery pack and a storage medium, relates to the technical field of battery detection, and comprises the following steps: a multi-parameter signal set is established, and state change analysis of the battery pack is performed to identify an abnormal coupling mode; after the abnormal coupling mode is configured to be abnormally concerned, real-time fault judgment criteria are established; historical data of a health state of the battery pack are read, real-time monitoring data sets are acquired, fault abnormality is identified, and a first fault abnormality identification result is established; the real-time fault judgment criteria are synchronized to a joint identification model for joint simulation residual error analysis to establish a second fault abnormality identification result, interactive verification is performed in combination with the first fault abnormality identification result, and fault abnormality is reported. The application solves the technical problems of low fault detection precision of the battery pack, inability to realize multi-parameter correlation analysis and fault early warning in the prior art, and achieves the technical effects of improving fault identification accuracy and realizing early warning of the fault of the battery pack.

Description

Technical Field

[0001] This invention relates to the field of battery testing technology, and specifically to an intelligent fault detection method, device and storage medium for battery packs. Background Technology

[0002] During long-term operation, the performance of battery packs is affected by a variety of factors, including voltage, current, temperature, electrolyte density, and plate deformation, with complex dynamic coupling relationships between different parameters. Traditional fault detection methods typically rely on single electrical parameters or static thresholds for judgment, which are insufficient to reflect the true electrochemical state inside the battery, leading to delayed fault identification or a high false alarm rate. Furthermore, the lack of comprehensive analysis and modeling of multi-dimensional monitoring data makes it impossible to accurately assess the battery pack's operating status and predict fault trends in advance. Summary of the Invention

[0003] This application provides an intelligent fault detection method, device, and storage medium for battery packs, which addresses the technical problems of low fault detection accuracy, inability to perform multi-parameter correlation analysis, and early fault warning in existing battery pack technologies.

[0004] In view of the above problems, this application provides an intelligent fault detection method, device and storage medium for battery packs.

[0005] A first aspect of this application provides an intelligent fault detection method for a battery pack, the method comprising:

[0006] Simultaneously collect data on voltage, current, temperature, electrolyte density, and plate deformation of the battery pack to establish a multi-parameter signal set. Input this multi-parameter signal set into a multi-parameter dynamic coupling analysis model to perform state change analysis of the battery pack and identify abnormal coupling patterns. Configure abnormal attention based on the abnormal coupling patterns, dynamically adjust the fault threshold curve of the battery pack, and establish a real-time fault judgment standard. Read historical data on the battery pack's health status and obtain a real-time monitoring dataset. Based on the real-time fault judgment standard, the historical data, and the real-time monitoring dataset, identify fault anomalies and establish a first fault anomaly identification result. Synchronize the real-time fault judgment standard to a joint identification model, and use the joint identification model to perform joint simulation residual analysis to establish a second fault anomaly identification result. The joint identification model is an electrochemical-electrical joint identification model. Interactively verify the first and second fault anomaly identification results and report the fault anomaly.

[0007] A second aspect of this application provides an electronic device, including: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the intelligent fault detection method for a battery pack provided in this application.

[0008] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements an intelligent fault detection method for a battery pack provided in this application.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] This application simultaneously collects voltage, current, temperature, electrolyte density, and plate deformation data of the battery pack to establish a multi-parameter signal set. The multi-parameter signal set is input into a multi-parameter dynamic coupling analysis model to perform state change analysis of the battery pack and identify abnormal coupling modes. Based on the abnormal coupling modes, abnormal attention is configured, and the fault threshold curve of the battery pack is dynamically adjusted to establish a real-time fault judgment standard. Historical data on the health status of the battery pack is read, and a real-time monitoring dataset is obtained. Based on the real-time fault judgment standard, the historical data, and the real-time monitoring dataset, fault anomaly identification is performed to establish a first fault anomaly identification result. The real-time fault judgment standard is synchronized to a joint identification model, and joint simulation residual analysis is performed using the joint identification model to establish a second fault anomaly identification result. The joint identification model is an electrochemical-electrical joint identification model. The first and second fault anomaly identification results are interactively verified, and a fault anomaly is reported. This invention addresses the technical problems of low accuracy in battery pack fault detection and the inability to perform multi-parameter correlation analysis and early fault warning in existing technologies. By constructing a multi-parameter dynamic coupling analysis model and combining it with electrochemical-electrical joint simulation residual analysis, it achieves dynamic coupling identification of multi-dimensional parameters and fault correlation judgment, thereby improving the accuracy of fault identification and enabling early warning of battery pack faults. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic flowchart of an intelligent fault detection method for a battery pack provided in an embodiment of this application;

[0013] Figure 2 This is a schematic diagram of the structure of an exemplary electronic device of this application.

[0014] Explanation of reference numerals in the attached drawings: Bus 300, Receiver 301, Processor 302, Transmitter 303, Memory 304, Bus Interface 305. Detailed Implementation

[0015] This application provides an intelligent fault detection method, device, and storage medium for battery packs. It addresses the technical problems of low fault detection accuracy and inability to achieve multi-parameter correlation analysis and early fault warning in existing battery pack technologies. By constructing a multi-parameter dynamic coupling analysis model and combining it with electrochemical-electrical joint simulation residual analysis, it achieves dynamic coupling identification of multi-dimensional parameters and fault correlation judgment, thereby improving the accuracy of fault identification and enabling early warning of battery pack faults.

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0017] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0018] Example 1, as Figure 1 As shown, this application provides an intelligent fault detection method for battery packs, the method comprising:

[0019] Step S100: Synchronously collect data on the voltage, current, temperature, electrolyte density, and plate deformation of the battery pack to establish a multi-parameter signal set.

[0020] In this embodiment, high-precision sensors and signal acquisition channels are deployed at various preset locations within the battery pack to monitor operating parameters such as voltage, current, temperature, electrolyte density, and plate deformation in real time. Specifically, voltage is obtained by connecting the sampling terminals to the two electrodes of a single battery cell to capture the instantaneous potential difference; current is captured using shunt measurement or Hall effect sensing to capture the current flowing in the charging and discharging circuit; temperature is obtained by attaching thermocouples or thermistors to the battery casing and its ends to acquire surface temperature changes; electrolyte density is obtained by sensing changes in the refractive index or conductivity of the electrolyte using density sensing elements to obtain real-time density values; and plate deformation is monitored using displacement sensing elements to detect the spacing between plates or minute deformations of the casing. All these signals are synchronously sampled under the same time reference and, after unified timestamp calibration and formatting, form a multi-parameter signal containing data on voltage, current, temperature, electrolyte density, and plate deformation.

[0021] Step S200: Input the multi-parameter signal set into the multi-parameter dynamic coupling analysis model, perform state change analysis of the battery pack, and identify abnormal coupling modes.

[0022] In this embodiment, when inputting a multi-parameter signal set into a multi-parameter dynamic coupling analysis model to perform state change analysis of the battery pack, the multi-parameter signal set is first normalized and time-synchronized in the preprocessing layer of the multi-parameter dynamic coupling analysis model to construct a multi-dimensional state vector reflecting the operating characteristics of the battery pack. Subsequently, the multi-dimensional state vector is passed to a hierarchical coupling network containing a feature decoupling layer, a temporal coupling layer, and an energy mapping layer to sequentially complete the extraction of inter-parameter correlation features, analysis of temporal coupling relationships, and energy distribution mapping. Finally, based on the output results of the hierarchical coupling network, abnormal coupling patterns are identified.

[0023] Furthermore, in the method provided in the application embodiments, inputting the multi-parameter signal set into a multi-parameter dynamic coupling analysis model to perform state change analysis of the battery pack and identify abnormal coupling modes, further includes:

[0024] The multi-parameter signal set is normalized and time-synchronized preprocessed through the preprocessing layer of the multi-parameter dynamic coupling analysis model to establish a multi-dimensional state vector; the multi-dimensional state vector is sent to the hierarchical coupling network of the multi-parameter dynamic coupling analysis model, the hierarchical coupling network including a feature decoupling layer, a temporal coupling layer, and an energy mapping layer; and abnormal coupling pattern recognition is performed according to the hierarchical coupling network.

[0025] In this embodiment, the acquired multi-parameter signal set is first input into the preprocessing layer of the multi-parameter dynamic coupling analysis model. The monitored parameters are normalized to ensure that data such as voltage, current, temperature, electrolyte density, and electrode deformation are within a unified dimensional range, avoiding calculation deviations caused by differences in the amplitude of a single parameter. Simultaneously, time synchronization processing ensures that different signals are aligned with the same time base, guaranteeing consistent data timing. The processed data are then combined to form a multi-dimensional state vector.

[0026] Next, the multidimensional state vector is sent to the hierarchical coupling network of the multi-parameter dynamic coupling analysis model to perform abnormal coupling pattern recognition. The hierarchical coupling network includes a feature decoupling layer, a temporal coupling layer, and an energy mapping layer. During abnormal coupling pattern recognition, the feature decoupling layer first performs covariance decomposition on the multidimensional state vector to obtain the cross-correlation matrix between each monitoring parameter, and extracts strongly coupled parameter pairs to establish a strongly coupled parameter subset, while calculating the coupling weights of each parameter pair. Then, the temporal coupling layer performs temporal coherence spectrum calculation based on a sliding time window on the strongly coupled parameter subset to obtain a temporal coupling strength sequence, and performs weighted superposition according to the coupling weights to construct a global temporal coupling matrix reflecting the characteristics of parameter changes over time. Next, the energy mapping layer maps the global temporal coupling matrix to the energy distribution space through Fourier transform to form an energy distribution matrix. Finally, coupling pattern clustering is performed based on the energy distribution matrix to establish a coupling pattern set, and anomaly detection is performed to output the abnormal coupling patterns.

[0027] Furthermore, in the method provided in the application embodiments, performing abnormal coupling pattern recognition based on the hierarchical coupling network further includes:

[0028] The feature decoupling layer is used to perform covariance decomposition of the multidimensional state vector to obtain the cross-correlation matrix between parameters. Strongly coupled parameter pairs are extracted from the cross-correlation matrix to establish a strongly coupled parameter subset, and the coupling weights of the parameter pairs within the strongly coupled parameter subset are calculated. The temporal coupling layer is used to perform temporal coherence spectrum calculation based on a sliding time window on the strongly coupled parameter subset to obtain a temporal coupling strength sequence. The temporal coupling strength sequence is weighted and superimposed using the parameter pair coupling weights to construct a global temporal coupling matrix. The global temporal coupling matrix is ​​used to capture the evolution trend of the coupling structure over time. The energy mapping layer is used to map the global temporal coupling matrix to the energy distribution space through Fourier transform to obtain the energy distribution matrix. The coupling pattern clustering is performed based on the energy distribution matrix to establish a coupling pattern set. Anomaly detection is performed on the coupling pattern set, and abnormal coupling patterns are output.

[0029] In this embodiment, a feature decoupling layer is first used to perform covariance decomposition on the input multidimensional state vector. By calculating the covariance and correlation coefficient between each monitored parameter, a sample covariance matrix and a Pearson correlation coefficient matrix are obtained. The sample covariance matrix reflects the joint fluctuation between parameters, while the Pearson correlation coefficient matrix eliminates the influence of dimensions through standardization, allowing direct comparison of the correlation between parameters. Subsequently, the absolute value of the correlation coefficient matrix is ​​taken and symmetricized and normalized to obtain a cross-correlation matrix, which is used to quantitatively describe the coupling strength between parameters. When the value in the cross-correlation matrix exceeds a preset threshold, it is determined that the parameter pair has a strong coupling relationship, thus filtering out strongly coupled parameter pairs and forming a strongly coupled parameter subset. Simultaneously, the coupling weight of the parameter pairs within the strongly coupled parameter subset is calculated. In this process, firstly, based on the absolute value of the covariance of each parameter pair in the covariance matrix, the joint fluctuation amplitude of the parameter pair is determined, and the maximum value is normalized to obtain the normalized covariance amplitude. Secondly, the absolute value of the Pearson correlation coefficient is taken as the dimensionless linear correlation strength index of the parameter pair. Then, to evaluate the stability of the parameter pairs over time, the standard deviation of the rolling correlation coefficient is calculated using a sliding time window, and a stability factor is constructed by taking its reciprocal. Finally, the covariance magnitude, correlation strength index, and stability factor are weighted and fused according to a preset ratio to obtain the comprehensive contribution of each parameter pair. All contributions are then normalized so that their sum is 1, thus forming the parameter pair coupling weights for each parameter pair.

[0030] Next, a temporal coupling layer is used to perform temporal coherence spectrum calculation on a subset of strongly coupled parameters based on a sliding time window. In this process, the subset of strongly coupled parameters is input, and a sliding time window method is used to analyze each pair of parameter signals. Within each time window, the signals are first subjected to mean and detrending processing to eliminate drift and noise, and then the coherence spectrum or cross-power spectral density is calculated to represent the coupling strength between the two parameter signals within that time period. As the time window slides along the time axis, a time-domain coupling strength sequence varying with time is obtained.

[0031] Subsequently, the multiple sets of temporal coupling strengths at each time point are weighted and superimposed according to the previously calculated coupling weights. Specifically, at any time point, the coupling strength values ​​of all strongly coupled parameter pairs are multiplied by their corresponding weights, and then summed to obtain the comprehensive coupling strength value. By arranging the comprehensive strength values ​​of consecutive time points in chronological order, a complete overall coupling strength time series can be obtained, and the weighted strength results of each parameter pair are recorded column-wise to form a global temporal coupling matrix with time as the row and parameter pairs as the column.

[0032] Then, an energy mapping layer is used to map the global temporal coupling matrix to the energy distribution space via Fourier transform. In this process, the time-series signal of the global temporal coupling matrix undergoes a Fourier transform, converting the time-domain information into a frequency-domain energy distribution. Before the Fourier transform, a time window, such as a Hanning window, is applied to the signal, and the transform result is normalized in amplitude and integrated in the frequency band to reduce spectral leakage and enhance the stability of the energy measurement. After the transform, the energy distribution matrix is ​​obtained.

[0033] Finally, coupling pattern clustering is performed based on the frequency domain characteristics of the energy distribution matrix. Energy feature vectors for each parameter combination in the energy matrix are extracted, and K-means or density clustering algorithms are used to classify the feature vectors, resulting in a set of coupling patterns. Next, anomaly detection is performed on the coupling pattern set. In this process, the coupling pattern set is compared with preset conditions; if the coupling pattern set meets any preset condition, it is determined to be an abnormal coupling pattern.

[0034] Furthermore, in the method provided in the application embodiments, the abnormal determination of the coupling mode set and the output of abnormal coupling modes further include:

[0035] If the set of coupling modes meets any preset condition, it is determined to be an abnormal coupling mode. The preset conditions are as follows: a: the cross-correlation between parameters exceeds twice the fluctuation range of the dynamic mean; b: the energy concentration region of the temporal coherence spectrum undergoes non-periodic drift; c: the energy density of the coupling mode cluster center changes abruptly, exceeding a preset threshold.

[0036] In this embodiment, an abnormal coupling mode is determined when the set of coupling modes meets any preset condition. Specifically, firstly, the dynamic change of the cross-correlation between parameters over time is calculated, and its mean and fluctuation range are statistically analyzed. When the cross-correlation value of any parameter pair exceeds twice the fluctuation range of the dynamic mean, it indicates that the correlation strength between the parameters has deviated significantly, and is determined to be abnormal. Secondly, the energy concentration region of the time-series coherence spectrum is monitored. When the main concentration region of the energy distribution shows a non-periodic drift in the time or frequency dimension, it indicates that the energy transfer characteristics have changed abnormally, and is also identified as abnormal. Finally, the energy density of the coupling mode cluster center is continuously tracked. When the energy density change exceeds a preset threshold in a short period of time, it indicates that the energy distribution state inside the cluster mode has changed abruptly, and is thus determined to be an abnormal coupling mode.

[0037] Step S300: After configuring abnormal attention according to the abnormal coupling mode, dynamically adjust the fault threshold curve of the battery pack and establish a real-time fault judgment standard.

[0038] In this embodiment, anomaly concern configuration is first executed based on the identified abnormal coupling patterns, and the fault threshold curve of the battery pack is dynamically adjusted accordingly. During this process, key parameters involved in the abnormal coupling patterns, such as voltage, current, temperature, electrolyte density, and plate deformation, are extracted. The time series of these parameters are denoised and smoothed, and the dynamic mean and dynamic standard deviation are calculated using a sliding time window method to obtain the baseline threshold. Subsequently, the average cross-correlation, temporal coupling strength, and energy distribution variation amplitude of each parameter with other parameters are statistically analyzed and used as adjustment reference values. When the average cross-correlation or energy distribution variation amplitude is high, it indicates a significant coupling change between the parameter and other parameters; in this case, the upper and lower limits of the baseline threshold are each reduced by 10%. When the variation amplitude is low, the original threshold range remains unchanged. The threshold range is recalculated and updated for each sampling period, forming a continuous fault threshold curve in the time dimension. Finally, the dynamically adjusted fault threshold curve is used as the real-time fault judgment standard. When the monitored value of any parameter exceeds the threshold range for three consecutive sampling points or the deviation exceeds twice the dynamic standard deviation, the parameter is determined to be abnormal.

[0039] Step S400: Read historical data on the health status of the battery pack and obtain a real-time monitoring dataset. Based on the real-time fault judgment criteria, the historical data, and the real-time monitoring dataset, identify fault anomalies and establish a first fault anomaly identification result.

[0040] In this embodiment, historical data on the health status of the battery pack is first read, and a real-time monitoring dataset is obtained. The two are then compared and analyzed to identify faults and anomalies. The historical data includes long-term changes in parameters such as voltage, current, temperature, electrolyte density, and plate deformation under different operating conditions, reflecting the normal fluctuation range and typical trends of the battery pack in a healthy state. The real-time monitoring dataset is generated by synchronously collecting the same type of parameters through sensors and then filtering and time-synchronized to form a standardized time series. Subsequently, the real-time monitoring data and historical data are compared point-by-point according to real-time fault judgment criteria. First, the deviation and rate of change of the real-time data relative to the historical mean are calculated to determine whether it exceeds the historical fluctuation range. When the deviation exceeds twice the historical standard deviation or the trend is opposite to the historical trend, it indicates an abnormal change in the parameter. Then, parameter deviation is continuously detected within a sliding time window. When multiple sampling results consistently exceed the range of the dynamic threshold curve, an abnormal trend in the parameter is confirmed. By utilizing the normal reference range provided by historical data and the real-time constraints of the dynamic threshold curve, the abnormal behavior in the real-time monitoring data is comprehensively judged, thereby establishing the first fault anomaly identification result.

[0041] Step S500: Synchronize the real-time fault judgment criteria to the joint identification model, use the joint identification model to perform joint simulation residual analysis, and establish a second fault anomaly identification result. The joint identification model is an electrochemical-electric joint identification model.

[0042] In this embodiment, the real-time fault determination criteria are synchronized to the joint identification model. The joint identification model is an electrochemical-electrical joint identification model. By reading the attribute dataset of the battery pack, an electrochemical sub-model and an electrical sub-model are constructed respectively, and configured as the joint identification model. Then, the joint discriminant layer of the joint identification model is initialized according to the real-time fault determination criteria. Afterwards, task execution simulations of the battery pack are performed using the electrochemical sub-model and the electrical sub-model respectively, obtaining a joint simulation dataset. Finally, the real-time monitoring dataset and the joint simulation dataset are synchronized to the initialized joint discriminant layer to perform residual analysis, outputting the second fault anomaly identification result.

[0043] Furthermore, in the method provided in the application embodiments, synchronizing the real-time fault determination criteria to the joint identification model and using the joint identification model to perform joint simulation residual analysis further includes:

[0044] Read the attribute dataset of the battery pack, construct an electrochemical sub-model and an electrical sub-model based on the attribute dataset, and configure the electrochemical sub-model and the electrical sub-model as a joint identification model; initialize the joint discrimination layer of the joint identification model according to the real-time fault judgment criteria; perform task execution simulation of the battery pack using the electrochemical sub-model and the electrical sub-model respectively, and establish a joint simulation dataset; synchronize the real-time monitoring dataset and the joint simulation dataset to the initialized joint discrimination layer, perform residual analysis, and output the second fault anomaly identification result.

[0045] In this embodiment, the pre-stored attribute dataset of the battery pack is first read, and parameters such as capacity, open-circuit voltage, internal resistance, electrolyte concentration, polarization characteristics, and temperature are uniformly organized. The data is then filtered, interpolated, and smoothed to eliminate noise and discontinuities. Next, an electrochemical sub-model and an electrical sub-model are constructed based on the attribute dataset. The electrical sub-model uses equivalent circuit analysis to establish the response relationship between voltage, current, and time, determining the variation patterns of resistance and capacitance to reflect the transient voltage response characteristics of the battery during charging and discharging. The electrochemical sub-model uses a dynamic relationship modeling method based on reaction rate and concentration changes. By fitting the open-circuit voltage curve and state-of-charge data, the correspondence between reaction rate, current, and temperature is established to characterize the internal electrochemical reaction characteristics and ion transport behavior of the battery. After completing the structural construction of the two sub-models, they are matched based on common input and output variables, and the electrochemical and electrical sub-models are coupled at the input and output ports to form a joint identification model. Input variables include current, temperature, and time; output variables include voltage.

[0046] Next, the joint discrimination layer of the joint identification model is initialized according to the real-time fault judgment criteria. The dynamic threshold curves of voltage, current, temperature, electrolyte density, and plate deformation parameters are matched one by one with the model output results. The average change trend and fluctuation range of each parameter are calculated using the sliding time window method to establish a dynamic threshold interval, enabling the joint discrimination layer to automatically update the judgment criteria under different operating conditions.

[0047] Subsequently, task execution simulations of the battery pack were performed using both electrochemical and electrochemical sub-models. Simulation inputs included charge / discharge current, temperature, and load conditions, while outputs included terminal voltage, current response, reaction rate, polarization voltage, and ion concentration changes. The two sets of simulation output data were integrated into a unified format using a time alignment method to establish a joint simulation dataset.

[0048] Finally, the real-time monitoring dataset and the co-simulation dataset are synchronized to the initialized co-discriminator layer, and residual analysis is performed. In this process, the real-time monitoring dataset and the co-simulation dataset are first time-aligned, and sliding registration is performed using the periodic characteristics of electrochemistry. Then, cross-domain feature mapping is performed on the co-simulation dataset to establish a cross-domain feature mapping matrix. This matrix is ​​then used to extract local residual information from the real-time monitoring dataset, forming a multi-dimensional residual set containing voltage-current residuals, temperature-response residuals, and deformation-impedance residuals. After setting adaptive weights based on the dynamic sensitivity of each residual dimension in the multi-dimensional residual set, weighted aggregation of the multi-dimensional residual set is performed to establish a residual aggregation matrix. Finally, cluster center drift analysis is performed using the residual aggregation matrix to identify anomaly focusing regions. Anomaly identification is then performed in conjunction with real-time fault determination criteria, and the final anomaly detection results are output.

[0049] Furthermore, in the method provided in the application embodiments, performing residual analysis further includes:

[0050] The real-time monitoring dataset and the co-simulation dataset are time-aligned, including performing sliding registration using the periodic characteristics of electrochemistry; cross-domain feature mapping is performed on the co-simulation dataset to establish a cross-domain feature mapping matrix; local residuals are extracted using the cross-domain feature mapping matrix and the real-time monitoring dataset to establish a multi-dimensional residual set, which includes voltage-current residuals, temperature-response residuals, and deformation-impedance residuals; after setting adaptive weights based on the dynamic sensitivity of each residual dimension in the multi-dimensional residual set, weighted aggregation of the multi-dimensional residual set is performed to establish a residual aggregation matrix; cluster center drift analysis is performed using the residual aggregation matrix to identify abnormal focusing intervals, and anomaly identification is performed based on the abnormal focusing intervals and the real-time fault judgment criteria.

[0051] In this embodiment, the real-time monitoring dataset and the co-simulation dataset are first time-aligned. During this process, the real-time monitoring dataset and the co-simulation dataset are resampled at a uniform sampling interval, and timestamp deduplication and linear interpolation are performed. A sliding registration window is set based on the periodic characteristics of electrochemistry, with a window length of 60 seconds and a step size of 1 second. Within each window, the terminal voltage plateau segment, charge / discharge duty cycle, and current pulse sequence are extracted as reference sequences. The Pearson correlation coefficient between the two sets of data is calculated, and the time offset corresponding to the maximum correlation is recorded. Phase correction is performed on the co-simulation dataset. If the maximum correlation is less than 0.8, the window length is increased by 10 seconds each time until a stable alignment offset is obtained. Finally, a one-to-one time index and the two synchronized sequences are output, completing the time alignment and sliding registration.

[0052] Subsequently, cross-domain feature mapping was performed on the co-simulation dataset. In this process, the units of voltage, current, temperature, impedance, and reaction quantities in the co-simulation dataset and the real-time monitoring dataset were unified and standardized with Z-scores. Then, least squares linear fitting was performed with the common physical quantities of the real-time monitoring dataset as independent variables and the target quantities of the co-simulation dataset as dependent variables to obtain mapping coefficients and offsets. These were then assembled in variable order to form a cross-domain feature mapping matrix. Based on this matrix, the co-simulation dataset was transformed to the same dimensions and scale as the real-time monitoring dataset, resulting in the cross-domain simulation feature sequence.

[0053] Next, local residuals are extracted using the cross-domain feature mapping matrix and the real-time monitoring dataset. In this process, a point-by-point interpolation method is used to extract local residuals and establish a multidimensional residual set. The simulated terminal voltage after subtracting the mapping from the measured terminal voltage and the simulated loop current after subtracting the mapping from the measured loop current are calculated and packaged in pairs as voltage-current residuals. The chemical reaction characterization quantity after subtracting the mapping from the measured temperature is calculated as the temperature-reaction residual. The equivalent impedance after subtracting the mapping from the measured deformation is calculated as the deformation-impedance residual. The three types of paired residuals are saved in chronological order to form a multidimensional residual set containing voltage-current residuals, temperature-reaction residuals, and deformation-impedance residuals.

[0054] Then, a residual aggregation matrix was established using a sliding time window statistical and weighted summation method. In this process, the window variance and over-threshold ratio of the three types of residuals were calculated separately within a fixed-length window (with the threshold of the real-time fault judgment standard as a reference). The variance ratio and over-threshold ratio were added together to obtain the dynamic sensitivity of each residual dimension, and the three-dimensional sensitivity was normalized to obtain adaptive weights. At each time point, the voltage-current residual, temperature-response residual, and deformation-impedance residual were weighted and summed with their corresponding adaptive weights to obtain the aggregated residual value at that time point, which was then arranged along time to form the residual aggregation matrix.

[0055] Finally, K-means clustering and threshold determination are used to perform cluster center drift analysis and anomaly identification. Specifically, vectors are extracted from the residual aggregation matrix using fixed-duration time slices. K-means clustering is performed on consecutive time slices, and the Euclidean distance and intra-cluster variance of cluster centers in adjacent time slices are calculated. When the distance between cluster centers is higher than the historical mean plus twice the standard deviation for multiple consecutive time slices, or when the intra-cluster variance increases by more than 50% in a single time slice compared to the previous time slice, it is marked as an anomaly focus interval. The aggregated residuals within the anomaly focus interval are compared point by point with the real-time fault determination criteria. If they continuously exceed the dynamic threshold band or the proportion of sample points exceeding the threshold exceeds half, it is determined as an anomaly identification, and the corresponding time period, trigger residual dimension, threshold, and deviation magnitude are output, completing the anomaly identification based on the anomaly focus interval and real-time fault determination criteria.

[0056] Step S600: Perform interactive verification of the first fault anomaly identification result and the second fault anomaly identification result, and report the fault anomaly.

[0057] Furthermore, in the method provided in the application embodiments, the interactive verification of the first fault anomaly identification result and the second fault anomaly identification result, and the reporting of the fault anomaly, further includes:

[0058] The warning level is matched using the interactive verification results to establish a warning level matching result; after adaptively configuring the warning signal using the warning level matching result and the fault type, the fault anomaly is reported.

[0059] In this embodiment, the first and second fault anomaly identification results are first interactively verified. During this process, the first and second fault anomaly identification results are synchronously compared on a unified time axis, and the abnormal parameter sets involved in both are extracted, including voltage anomalies, current anomalies, temperature anomalies, electrolyte density anomalies, and electrode deformation anomalies. Then, the duration and deviation of the same parameters in the two sets of results are compared, and the anomaly overlap ratio and average deviation value are calculated. When the anomaly overlap ratio is higher than a set threshold and the average deviation value is lower than the allowable range, the two sets of identification results are determined to be consistent, an interactive verification result is generated, and it is confirmed as a valid anomaly. If there is a significant deviation between the two sets of results, for example, one set of results shows an anomaly while the other set remains normal, or the difference in the anomaly magnitude exceeds a preset limit, the deviation information is recorded, and supplementary data collection is triggered for secondary verification to avoid misjudgment or omission, thereby obtaining the interactive verification result.

[0060] The interactive verification results are then used to perform early warning level matching. During this process, the interactive verification results are graded based on the magnitude of the anomaly, its duration, and the number of affected parameters. For example, a small anomaly with a short duration is classified as a general early warning; a moderate anomaly with a long duration and more than three affected parameters is classified as a severe early warning; and a large anomaly with simultaneous anomalies in multiple parameters such as temperature, voltage, and current is classified as a critical early warning. Through this grading process, early warning level matching results are established, ensuring that each type of anomaly has a clear risk level and triggering criteria.

[0061] Finally, the warning level matching results are combined with the fault type to adaptively configure the warning signal. Specifically, when the interactive verification results indicate that the fault type is an electrical performance abnormality, such as voltage abnormality or current overload, a warning output channel primarily based on voltage and current signals is configured; when the fault type is an electrochemical performance abnormality, such as temperature abnormality or reaction efficiency abnormality, a warning output channel primarily based on temperature and reaction signals is configured. The trigger threshold and response time of the warning signal are dynamically adjusted based on the warning level matching results. When the warning signal reaches the set trigger condition, the fault abnormality is immediately reported, thereby achieving real-time warning and accurate alarm for battery pack faults.

[0062] Furthermore, the method provided in the application embodiments, which involves interactive verification of the first fault anomaly identification result and the second fault anomaly identification result, further includes:

[0063] Calculate the consistency index between the first fault anomaly identification result and the second fault anomaly identification result to generate identification credibility; when the identification credibility is lower than the preset trust threshold, generate additional collection instructions; perform additional data collection and anomaly backtracking analysis according to the additional collection instructions to reconstruct interactive verification.

[0064] In this embodiment, the first and second fault anomaly identification results are first aligned on a unified time axis to ensure consistent anomaly states at each time point. Then, the abnormal parameters involved in the two sets of identification results, including voltage, current, temperature, electrolyte density, and electrode deformation, are compared one by one. Specifically, for each parameter, the proportion of overlap time between the two sets of results within the same time period is calculated to represent the total anomaly time for that parameter; this is recorded as the overlap ratio. Simultaneously, the average difference in the amplitude of the anomalies between the two sets of results within the same time period is calculated. After standardizing the amplitude difference, the standardized value is subtracted from 1 to obtain the amplitude consistency value. The amplitude consistency value is then added to the overlap ratio and divided by 2 to obtain the consistency index for that parameter dimension. The average of the consistency indices for all parameter dimensions is taken to form the consistency index, i.e., the identification reliability.

[0065] When the recognition confidence level is lower than the preset trust threshold, for example, when the threshold is set to 0.85, an additional acquisition instruction is generated to perform short-term high-frequency sampling of key parameters such as voltage, current, temperature, electrolyte density and plate deformation within a specific time period, for example, 10 times per second for a duration of 30 seconds, in order to obtain data with higher accuracy.

[0066] Finally, additional data acquisition and anomaly backtracking analysis are performed according to the additional acquisition instructions. In this process, the newly acquired data is first filtered, denoised, and missing value imputed. Then, it is time-aligned with the original real-time monitoring data to ensure that the monitoring data at the same time point corresponds to the newly acquired data. Subsequently, the difference between the new data and the original data is calculated parameter by parameter to form a local residual sequence. The magnitude, duration, and number of abnormal parameters of the residuals are analyzed to identify potential undetected anomalies or deviation trends, such as short-term current overload or sudden temperature rise. Finally, the anomaly backtracking analysis results are integrated with the original first fault anomaly identification results and second fault anomaly identification results to update the time period of the anomaly occurrence, the deviation magnitude, and the parameters involved, reconstructing the interactive verification results.

[0067] In summary, the embodiments of this application have at least the following technical effects:

[0068] This application simultaneously collects voltage, current, temperature, electrolyte density, and plate deformation data of the battery pack to establish a multi-parameter signal set. The multi-parameter signal set is input into a multi-parameter dynamic coupling analysis model to perform state change analysis of the battery pack and identify abnormal coupling modes. Based on the abnormal coupling modes, abnormal attention is configured, and the fault threshold curve of the battery pack is dynamically adjusted to establish a real-time fault judgment standard. Historical data on the health status of the battery pack is read, and a real-time monitoring dataset is obtained. Based on the real-time fault judgment standard, the historical data, and the real-time monitoring dataset, fault anomaly identification is performed to establish a first fault anomaly identification result. The real-time fault judgment standard is synchronized to a joint identification model, and joint simulation residual analysis is performed using the joint identification model to establish a second fault anomaly identification result. The joint identification model is an electrochemical-electrical joint identification model. The first and second fault anomaly identification results are interactively verified, and a fault anomaly is reported. This invention addresses the technical problems of low accuracy in battery pack fault detection and the inability to perform multi-parameter correlation analysis and early fault warning in existing technologies. By constructing a multi-parameter dynamic coupling analysis model and combining it with electrochemical-electrical joint simulation residual analysis, it achieves dynamic coupling identification of multi-dimensional parameters and fault correlation judgment, thereby improving the accuracy of fault identification and enabling early warning of battery pack faults.

[0069] Example 2: Based on the inventive concept of the intelligent fault detection method for a battery pack in the foregoing embodiments, this application also provides an electronic device, including: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of any of the methods described in Example 1 above.

[0070] Figure 2 This is a schematic diagram of the structure of an exemplary electronic device of this application. Figure 2 In this document, the bus architecture is represented by bus 300. Bus 300 may include any number of interconnected buses and bridges, and bus 300 connects various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 can be used to store data used by processor 302 during operation.

[0071] In Embodiment 3, based on the same inventive concept as the intelligent fault detection method for battery packs in the foregoing embodiments, this application also provides a computer-readable storage medium storing a computer program, which, when executed, implements the steps of any one of the methods in Embodiment 1 above.

[0072] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A smart fault detection method for battery packs, characterized in that, The method includes: Simultaneously collect data on the voltage, current, temperature, electrolyte density, and plate deformation of the battery pack to establish a multi-parameter signal set; The multi-parameter signal set is input into the multi-parameter dynamic coupling analysis model to perform state change analysis of the battery pack and identify abnormal coupling modes. After configuring abnormal attention according to the abnormal coupling mode, the fault threshold curve of the battery pack is dynamically adjusted to establish a real-time fault judgment standard. Read historical data on the health status of the battery pack and obtain a real-time monitoring dataset. Based on the real-time fault judgment criteria, the historical data, and the real-time monitoring dataset, identify fault anomalies and establish a first fault anomaly identification result. The real-time fault judgment criteria are synchronized to the joint identification model, and the joint identification model is used to perform joint simulation residual analysis to establish a second fault anomaly identification result. The joint identification model is an electrochemical-electric joint identification model. The first fault anomaly identification result and the second fault anomaly identification result are interactively verified, and the fault anomaly is reported.

2. The intelligent fault detection method for battery packs as described in claim 1, characterized in that, The real-time fault determination criteria are synchronized to the joint identification model, and the joint simulation residual analysis is performed using the joint identification model, including: Read the attribute dataset of the battery pack, construct an electrochemical sub-model and an electrical sub-model based on the attribute dataset, and configure the electrochemical sub-model and the electrical sub-model as a joint recognition model; The joint discrimination layer of the joint identification model is initialized according to the real-time fault determination criteria; The task execution simulation of the battery pack was carried out using the electrochemical sub-model and the electrical sub-model respectively, and a joint simulation dataset was established. The real-time monitoring dataset and the joint simulation dataset are synchronized to the initialized joint discriminant layer, residual analysis is performed, and the second fault anomaly identification result is output.

3. The intelligent fault detection method for a battery pack as described in claim 2, characterized in that, Performing residual analysis includes: The real-time monitoring dataset and the co-simulation dataset are time-aligned, and the time alignment includes performing sliding registration using the periodic characteristics of electrochemistry; Perform cross-domain feature mapping on the co-simulation dataset and establish a cross-domain feature mapping matrix; Local residuals are extracted using the cross-domain feature mapping matrix and the real-time monitoring dataset to establish a multidimensional residual set, which includes voltage-current residuals, temperature-response residuals, and deformation-impedance residuals. After setting adaptive weights based on the dynamic sensitivity of each residual dimension in the multidimensional residual set, weighted aggregation of the multidimensional residual set is performed to establish a residual aggregation matrix. Cluster center drift analysis is performed using the residual aggregation matrix to identify abnormal focusing intervals, and anomaly identification is performed based on the abnormal focusing intervals and the real-time fault judgment criteria.

4. The intelligent fault detection method for a battery pack as described in claim 1, characterized in that, The multi-parameter signal set is input into a multi-parameter dynamic coupling analysis model to perform state change analysis of the battery pack and identify abnormal coupling modes, including: The multi-parameter dynamic coupling analysis model is used to perform normalization and time synchronization preprocessing of the multi-parameter signal set through the preprocessing layer, and a multi-dimensional state vector is established. The multidimensional state vector is sent to the hierarchical coupling network of the multi-parameter dynamic coupling analysis model, the hierarchical coupling network including a feature decoupling layer, a temporal coupling layer, and an energy mapping layer; Abnormal coupling pattern identification is performed based on the hierarchical coupling network.

5. The intelligent fault detection method for a battery pack as described in claim 4, characterized in that, Performing abnormal coupling pattern recognition based on the hierarchical coupling network includes: The covariance decomposition of the multidimensional state vector is performed using the feature decoupling layer to obtain the cross-correlation matrix between parameters. Strongly coupled parameter pairs are extracted from the cross-correlation matrix to establish a strongly coupled parameter subset, and the coupling weight of the parameter pairs within the strongly coupled parameter subset is calculated. The temporal coherence spectrum of the strongly coupled parameter subset is calculated based on a sliding time window using a temporal coupling layer to obtain a temporal coupling strength sequence. The temporal coupling strength sequence is then weighted and superimposed using the parameters and coupling weights to construct a global temporal coupling matrix. This global temporal coupling matrix is ​​used to capture the evolution trend of the coupling structure over time. The energy mapping layer is used to map the global temporal coupling matrix to the energy distribution space via Fourier transform, thereby obtaining the energy distribution matrix; Based on the energy distribution matrix, coupling patterns are clustered to establish a coupling pattern set. Anomalies are determined in the coupling pattern set, and abnormal coupling patterns are output.

6. The intelligent fault detection method for a battery pack as described in claim 5, characterized in that, Anomaly detection is performed on the set of coupling patterns, and abnormal coupling patterns are output, including: If the set of coupling modes meets any preset condition, it is determined to be an abnormal coupling mode. The preset conditions are as follows: a: The cross-correlation between parameters exceeds twice the fluctuation range of the dynamic mean; b: Non-periodic drift occurs in the energy concentration region of the temporal coherence spectrum; c: The energy density of the cluster centers in the coupling mode changes abruptly beyond the preset threshold.

7. The intelligent fault detection method for a battery pack as described in claim 1, characterized in that, The first fault anomaly identification result and the second fault anomaly identification result are interactively verified, and a fault anomaly is reported, including: Use the interactive verification results to match early warning levels and establish early warning level matching results; After adaptively configuring the warning signal using the warning level matching result and the fault type, the fault anomaly is reported.

8. The intelligent fault detection method for a battery pack as described in claim 1, characterized in that, The interactive verification of the first fault anomaly identification result and the second fault anomaly identification result also includes: Calculate the consistency index between the first fault anomaly identification result and the second fault anomaly identification result to generate identification credibility; If the recognition credibility is lower than the preset trust threshold, an additional collection instruction is generated; Additional data collection and anomaly backtracking analysis are performed based on the additional collection instructions to reconstruct interactive verification.

9. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, implements the intelligent fault detection method for a battery pack according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements an intelligent fault detection method for a battery pack as described in any one of claims 1-8.

Images