Battery safety performance detection method and system

Abstract

The invention relates to the technical field of new energy battery safety monitoring, and discloses a battery safety performance detection method and system, and the method comprises the steps: obtaining battery operation data, and carrying out the preprocessing, and obtaining preliminary structural data; performing feature compression and key feature extraction on the preliminary structured data to obtain a state correlation feature vector; according to the state correlation feature vectors, dynamic change rules of the features under different working conditions are analyzed, and abnormal trend signals are determined; comparing the state association feature vector corresponding to the abnormal trend signal with a preset abnormal threshold range to obtain an abnormal degree score; matching the abnormal degree score with a preset historical fault case library, determining an abnormal shutdown mode category, predicting a fault evolution path, and generating a risk level; and according to the risk level, generating corresponding early warning information, and determining a battery safety early warning response level. According to the method, early-stage accurate early warning of the battery fault can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of new energy battery safety monitoring, and in particular to a battery safety performance detection method and system. BACKGROUND

[0002] With the rapid development of new energy vehicles and energy storage industries, the large-scale application of lithium ion batteries makes the safety performance monitoring a key link to protect life and property safety and stable operation of energy systems. The real-time and accuracy of battery safety detection are directly related to the effectiveness of accident prevention. In particular, under complex working conditions, whether abnormal characteristics can be accurately identified and fault evolution path can be predicted for risk warning has become the core difficulty of industry technology breakthrough.

[0003] The prior art usually adopts an intelligent sensing system composed of intelligent sensing elements to collect battery operation data, acquires voltage, current, temperature and other signals through intelligent sensors, compares the collected data with a preset fixed threshold value, and when the data exceeds the threshold value, matches a shutdown strategy from a pre-stored historical fault case library and triggers a single parameter response control.

[0004] However, since the preset fixed threshold value does not consider the change of dynamic working conditions, the false positive rate of the intelligent sensing system is increased. At the same time, the existing method only analyzes the single-dimensional data collected by the intelligent sensor independently, ignores the multivariate coupling effect of voltage drop and temperature rise, and causes hidden fault missed detection. Therefore, the existing technology seriously restricts the real-time and accuracy of battery safety performance detection. SUMMARY

[0005] The present application provides a battery safety performance detection method and system to solve the technical problem of the real-time and accuracy of battery safety performance detection restricted by the prior art.

[0006] In the first aspect, to solve the above technical problem, the present application provides a battery safety performance detection method and system, comprising: acquiring battery operation data and preprocessing to obtain preliminary structured data; performing feature compression and key feature extraction on the preliminary structured data to obtain a state-related feature vector; analyzing the dynamic change rule of the feature under different working conditions according to the state-related feature vector to determine an abnormal trend signal; comparing the state-related feature vector corresponding to the abnormal trend signal with a preset abnormal threshold range to obtain an abnormality degree score; matching the abnormality degree score with a preset historical fault case library to determine an abnormal shutdown mode category; predicting a fault evolution path based on the abnormal shutdown mode category to generate a risk level; According to the risk level, corresponding early warning information is generated, and a battery safety early warning response level is determined.

[0007] In an optional implementation, the battery operation data is acquired and preprocessed to obtain preliminary structured data, including: The battery operation data is collected in layers from the voltage dimension and the current dimension to obtain original time series; The original time series is segmented to determine segmented data segments; The segmented data segments are smoothed to obtain a clean data set; The clean data set is grouped according to similar segments to generate the preliminary structured data.

[0008] In an optional implementation, the preliminary structured data is subjected to feature compression and key feature extraction to obtain a state-related feature vector, including: The high-dimensional operation data in the preliminary structured data is subjected to principal component analysis dimension reduction to obtain a compressed feature set; For the compressed feature set, a preliminary key feature vector is determined by filtering features highly related to the battery state through a preset coefficient calculation; If there is a redundant dimension in the preliminary key feature vector, the redundant dimension is removed using a preset mutual information metric to obtain the state-related feature vector.

[0009] In an optional implementation, the dynamic change law of the feature under different working conditions is analyzed according to the state-related feature vector to determine an abnormal trend signal, including: The feature values in the state-related feature vector are extracted using a sliding window method and grouped to obtain dynamic feature values; The dynamic feature values are subjected to weighted processing according to real-time acquired environmental influence factors to obtain weighted feature values; Signals with the weighted feature values exceeding a preset first abnormal threshold are marked as abnormal trend signals.

[0010] In an optional implementation, the state-related feature vector corresponding to the abnormal trend signal is compared with a preset abnormal threshold range to obtain an abnormality degree score, including: The state-related feature vector corresponding to the abnormal trend signal is obtained according to the abnormal trend signal; The state-related feature vector is compared with a preset second abnormal threshold to calculate a single abnormality degree of each state-related feature vector; All the single abnormality degrees are summed to obtain an abnormality degree score.

[0011] In an optional implementation, the matching the abnormality degree score with the preset historical fault case library to determine an abnormal shutdown mode category comprises: matching the abnormality degree score with the preset historical fault case library to obtain a preliminary abnormality trend matching result; when the preliminary abnormality trend matching result exceeds a preset third abnormality threshold, determining a preliminary shutdown mode category; refining the preliminary shutdown mode category classification result to output a final abnormality shutdown mode category.

[0012] In an optional implementation, the predicting a fault evolution path based on the abnormality shutdown mode category and generating a risk level comprises: calculating a conditional probability of an abnormality mode evolving into a fault according to the abnormality shutdown mode category by using Bayes' theorem; when the conditional probability exceeds a preset evolution threshold, marking a corresponding evolution path as a possible path; sequentially sorting the possible paths to determine a path with a conditional probability higher than a preset probability threshold as an evolution path sequence; performing a multiplication operation on the evolution path sequence and a preset future time parameter to obtain a cumulative risk value, and matching the cumulative risk value with a preset risk determination rule to determine the risk level.

[0013] In an optional implementation, the generating corresponding early warning information according to the risk level and determining a battery safety early warning response level comprises: according to the risk level, obtaining corresponding influence factors from a pre-established data set, analyzing multivariate coupling influence weights, and determining a key influence factor set; obtaining risk control parameters from the key influence factor set, and optimizing the risk control parameters by using a preset parameter adjustment logic to obtain an optimized control parameter combination; generating corresponding battery safety early warning response information according to the optimized control parameter combination; determining a final battery safety early warning response level according to the battery safety early warning response information and a preset early warning mapping table.

[0014] In a second aspect, the present application provides a battery safety performance detection system, comprising: a data acquisition module configured to acquire battery operation data and perform preprocessing to obtain preliminary structured data; a feature extraction module configured to perform feature compression and key feature extraction on the preliminary structured data to obtain a state-associated feature vector; an abnormality detection module configured to analyze dynamic change rules of the features under different working conditions according to the state-associated feature vector, and determine an abnormal trend signal; an abnormality scoring module configured to compare the state-associated feature vector corresponding to the abnormal trend signal with a preset abnormal threshold range, and obtain an abnormality degree score; a pattern matching module configured to match the abnormality degree score with a preset historical fault case library, and determine an abnormal shutdown mode category; a risk prediction module configured to predict a fault evolution path based on the abnormal shutdown mode category, and generate a risk level; a response control module configured to generate corresponding early warning information according to the risk level, and determine a battery safety early warning response level.

[0015] In a third aspect, the present application further provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the battery safety performance detection method and system according to any one of the above when executing the computer program.

[0016] In a fourth aspect, the present application further provides a computer readable storage medium comprising a stored computer program, wherein the computer readable storage medium controls a device where the computer readable storage medium is located to execute the battery safety performance detection method and system according to any one of the above when the computer program runs.

[0017] Compared with the prior art, the present application has the following beneficial effects: (1) The present application detects the battery safety performance, reduces the false positive rate and improves the implicit fault detection rate through dynamic threshold mechanism and multivariate coupling analysis. In view of the problem that the static threshold of the prior art is not adaptive to working condition changes, the present application uses a sliding window method to extract the feature dynamic change rule, combines with real-time weighted calculation of environmental factors, generates a working condition adaptive abnormal threshold range, reduces the false positive rate, and at the same time, extracts the multivariate coupling features of voltage sudden drop and temperature rise rate through principal component analysis and mutual information measurement, solves the limitations of single parameter analysis, and improves the implicit fault detection rate.

[0018] (2) The present application realizes early and accurate early warning of faults based on Bayesian path prediction. The abnormality degree score is matched with the historical fault case library, the conditional probability is calculated through the Bayesian theorem, and when the probability is greater than 0.85, it is marked as a possible path; the cumulative risk value is calculated by combining the future time parameter, the risk level is generated, the early warning can be triggered before the fault occurs, and the early warning accuracy is improved.

[0019] (3) The application optimizes the control parameter closed loop to improve the system safety. According to the risk level, the key influence factors are dynamically extracted, the risk control parameters are optimized, the hierarchical response strategy is generated, and the safety accident rate is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a flow chart of a battery safety performance detection method provided by the first embodiment of the application; Figure 2 is a schematic diagram of a battery safety performance detection system module provided by the second embodiment of the application. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0022] In the field of new energy battery safety monitoring technology, the real-time and accuracy of battery safety detection are directly related to the accident prevention efficiency. Especially under complex working conditions, whether abnormal characteristics can be accurately identified and fault evolution path can be predicted for risk warning has become the core difficulty of the industry technology breakthrough.

[0023] The prior art usually adopts an intelligent sensing system composed of intelligent sensing elements to collect battery operation data, acquires signals such as voltage, current and temperature through intelligent sensors, compares the collected data with a preset fixed threshold value, and when the data exceeds the threshold value, matches a shutdown strategy from a pre-stored historical fault case library, and triggers a single-parameter response control. However, since the preset fixed threshold value does not consider the change of dynamic working conditions, the false positive rate of the intelligent sensing system is increased. At the same time, the existing method only analyzes the single-dimensional data collected by the intelligent sensor independently, ignores the multivariate coupling effect of voltage drop and temperature rise, and causes poor hidden fault detection effect.

[0024] In order to solve the above problems, the battery safety performance detection method provided by the embodiments of the application will be described and explained in detail in the following specific embodiments.

[0025] Reference Figure 1 The first embodiment of the application provides a battery safety performance detection method and system, including the following steps: S101, acquiring battery operation data and preprocessing to obtain preliminary structured data; S102, performing feature compression and key feature extraction on the preliminary structured data to obtain a state correlation feature vector; S103, analyzing dynamic change rules of the features under different working conditions according to the state-associated feature vector, and determining an abnormal trend signal; S104, comparing the state-associated feature vector corresponding to the abnormal trend signal with a preset abnormal threshold range, and obtaining an abnormal degree score; S105, matching the abnormal degree score with a preset historical fault case library, and determining an abnormal shutdown mode category; S106, predicting a fault evolution path based on the abnormal shutdown mode category, and generating a risk level; S107, generating corresponding early warning information according to the risk level, and determining a battery safety early warning response level.

[0026] In step S101, battery operation data is obtained and preprocessed to obtain preliminary structured data, including: S1011, collecting the battery operation data from the voltage dimension and the current dimension in layers to obtain an original time sequence; S1012, performing sequence segmentation on the original time sequence to determine segmented data segments; S1013, performing smoothing processing on the segmented data segments to obtain a clean data set; S1014, grouping the clean data set according to similar segments to generate the preliminary structured data.

[0027] In step S1011, the battery operation data is collected from the voltage dimension and the current dimension in layers to obtain an original time sequence.

[0028] It should be noted that the battery operation data refers to a set of electrical parameters reflecting the working state of the battery obtained in real time by intelligent sensors, including voltage, current and other key indicators. These data have the characteristics of strong time sequence and high dimension correlation. Layered collection is a multi-granularity collection strategy for battery operation data. By setting different levels of collection frequency and data processing logic, the integrity of the original signal is ensured, and the efficient use of system resources is also considered.

[0029] In an embodiment, an embedded high-precision sensor is used at the bottom layer to collect the original voltage signal at a frequency of one per second, and the effective voltage range is set to 3.0V to 4.2V. When it is detected that the voltage instantaneous value exceeds 4.0V, middle-layer aggregation calculation is performed, and the average value is calculated for the original data at an hourly granularity to obtain the associated data set of the average voltage 3.7V and the corresponding current 2.5A. The original time sequence containing original fine-grained data and aggregated coarse-grained data is formed.

[0030] This hierarchical mechanism effectively solves the data redundancy problem caused by high-frequency acquisition, while retaining the key operating condition change characteristics through threshold-triggered middle-layer aggregation, providing a data foundation with both timeliness and computational efficiency for subsequent anomaly detection.

[0031] Specifically, when the voltage value fluctuates within the safe range, the system only stores aggregated data to save resources; when the voltage breaks through the threshold, the original sequence is retained for in-depth analysis, and this adaptive acquisition strategy improves the data value density.

[0032] This process is of core significance for building a high-reliability battery safety monitoring system, avoiding the risk of key signal loss caused by traditional uniform sampling, and reducing storage and computing costs through intelligent data compression, providing optimized input data for subsequent feature extraction and fault warning links.

[0033] In step S1012, the original time series is segmented to determine the segmented data segments.

[0034] It should be noted that sequence segmentation is a time series processing method for battery operation data, which divides continuous time series data into data segments with fixed time span, retaining the time sequence characteristics of the data and facilitating the extraction of local features and detection of abnormal fluctuations.

[0035] In one implementation, for time series segmentation of voltage, current and other multi-dimensional information, a sliding window is used for automatic segmentation processing, with 10 minutes set as the window size and 5 minutes as the step parameter. The voltage sequence [3.6, 3.7, 3.8,...] with one-hour sampling frequency of one second is segmented, the original sequence of 3600 data points is segmented into 6 overlapping subsequences, each containing 120 data points, and the statistical features of each window are calculated in real time, such as the mean of the first window being 3.65V and the variance being 0.02. Through the calculation of such local statistics, abnormal fluctuation patterns of voltage can be effectively captured. This segmentation method can automatically extract 144 data segments with fixed time span from 24 hours of continuous monitoring data, each segment containing multiple dimensional feature vectors such as voltage, current and temperature, forming the segmented data segments.

[0036] The processing procedure plays a key role in subsequent anomaly detection and state assessment. Local feature extraction through sliding windows avoids both transient anomalies that global statistics might mask and important events that window boundaries might truncate, providing an optimal data foundation for refined analysis of battery safety states. This temporal segmentation method is particularly suitable for battery operation data with strong temporal correlation, meeting the needs of real-time monitoring while providing sufficient data support for offline deep analysis, and is an important preprocessing step for building an efficient battery safety warning system.

[0037] In step S1013, the segmented data segments are smoothed to obtain a clean data set.

[0038] It should be noted that the smoothing process is a data purification method for noise interference and abnormal fluctuations in battery operation data, and its core purpose is to eliminate random errors and transient interference in the sensor acquisition process and retain effective signal features reflecting the true state of the battery.

[0039] In an implementation, median filtering is used to filter noise from the segmented current sequence, with 5A set as the abnormal value determination threshold. Mutant current data exceeding the threshold is automatically replaced with the median value of adjacent data, effectively suppressing sharp peak noise caused by electromagnetic interference or sensor transient failure. At the same time, the voltage data is normalized, linearly mapping the original voltage value range of 3.0V-4.2V to the [0,1] interval, and converting the typical working voltage of 3.7V to a standardized value of 0.5. This processing not only eliminates analysis bias caused by different dimensions, but also improves the convergence efficiency of subsequent machine learning algorithms. To further optimize the data structure, principal component analysis is used to reduce the dimensionality of the multi-dimensional features, retaining 90% of the original data variance and automatically eliminating possible redundant dimensions. Finally, a structured data table is generated in time window units, with each data entry containing time stamp, voltage mean, current variance, and other key features, such as {time stamp: 2023-10-0100:00, voltage-mean: 3.65, current-var: 0.1}. This data preprocessing procedure is implemented automatically through a cloud computing platform, ensuring efficient and reliable data collection from raw data to structured processing, and providing a high-quality data foundation for battery health state prediction and remaining life estimation.

[0040] The smoothing process plays a key role in improving the accuracy of subsequent analysis, effectively suppressing noise interference, and improving feature extraction efficiency through data standardization and dimensionality reduction, laying a solid data foundation for building a high-precision battery safety warning system. This refined data processing method is particularly suitable for battery management systems and other applications that require strict data quality, ensuring the reliability of state assessment and fault warning, and is a core technical link in realizing intelligent battery management.

[0041] In step S1014, the clean data set is grouped by similar segments to generate the preliminary structured data.

[0042] It should be noted that similar segment grouping is an intelligent clustering process based on the similarity of the time series characteristics of battery operation data, and its core is to identify data segments with the same change pattern under different working conditions, thereby establishing a data structure reflecting the typical working state of the battery.

[0043] In one implementation, a clustering method based on the dynamic time warping (DTW) algorithm is used to analyze the cleaned data segments. By calculating the shape similarity of voltage and current curves between segments, segments with a Euclidean distance less than 0.05 are automatically classified into the same group, for example, constant current charging segments and constant voltage charging segments are classified separately. This grouping process can effectively distinguish the characteristic performance of the battery under different working modes, such as normal discharge, rapid charging, high temperature operation, etc. Typical scenarios will be automatically identified and form independent data groups. Through this intelligent grouping, the system can build a structured data table containing multi-dimensional feature vectors, where each group of data is labeled with the corresponding working mode label and key statistical features, establishing a standardized data foundation for subsequent anomaly detection and state assessment.

[0044] This grouping process plays a key role in improving the accuracy of battery state recognition, not only avoiding confusion and interference from different working conditions data, but also quickly identifying abnormal working conditions through pattern matching, which is a core technical link in realizing intelligent battery management. This structured processing method based on data similarity can improve the training efficiency and prediction accuracy of subsequent machine learning algorithms, providing an optimized data organization form for building a high-reliability battery safety warning system.

[0045] In step S102, the preliminary structured data is subjected to feature compression and key feature extraction to obtain a state-related feature vector, including: S1021, performing principal component analysis dimensionality reduction on the high-dimensional operation data in the preliminary structured data to obtain a compressed feature set; S1022, for the compressed feature set, filtering features highly related to the battery state through a pre-set coefficient calculation to determine a preliminary key feature vector; S1023, if there is a redundant dimension in the preliminary key feature vector, a preset mutual information metric is used to remove the redundant dimension, and the state-related feature vector is obtained.

[0046] In step S1021, principal component analysis dimension reduction is performed on the high-dimensional operation data in the preliminary structured data to obtain a compressed feature set.

[0047] It should be noted that principal component analysis dimension reduction is a data processing method that projects high-dimensional data into a low-dimensional space through linear transformation. The core goal is to retain the most representative change patterns in the original data while eliminating redundant dimensions and noise interference.

[0048] In an implementation manner, based on the 15-dimensional original feature matrix in the time window feature table, the covariance matrix is calculated and its eigenvalues and eigenvectors are solved, the first four principal components are selected according to the cumulative variance contribution rate of 85%, and the original feature space is compressed from 15 dimensions to 4 dimensions, for example, the multi-dimensional data points containing voltage 3.8V, current 2.8A, temperature 48°C and other parameters are converted into low-dimensional vector [1.5, 0.9, -0.4, 0.2] representation.

[0049] This dimension reduction processing not only reduces the data dimension, but also automatically identifies the key change pattern affecting the battery state through matrix decomposition, providing a simple and effective data basis for subsequent feature selection.

[0050] In step S1022, for the compressed feature set, the features highly related to the battery state are calculated and screened through a preset coefficient, and a preliminary key feature vector is determined.

[0051] It should be noted that feature correlation screening is a process of quantitatively analyzing the correlation degree of each feature with the battery health state, thereby identifying the feature subset with the highest discriminant.

[0052] In an implementation manner, the statistical dependence between each compressed feature and the battery remaining capacity is quantified, and it is assumed that the mutual information value between the voltage fluctuation rate and the capacity attenuation is 0.75, which is significantly higher than the preset threshold 0.6, indicating that the feature has predictive ability for the battery state. Based on such correlation analysis, the features with mutual information values exceeding the threshold are selected to form the preliminary key feature vector, such as [voltage-fluc: 0.03, current-peak: 3.2, temperature-grad: 1.5]. These features can reflect the core working state and degradation trend of the battery.

[0053] In step S1023, if there is a redundant dimension in the preliminary key feature vector, a preset mutual information metric is used to remove the redundant dimension, and the state-related feature vector is obtained.

[0054] It should be noted that the feature redundancy elimination is a process of removing redundant dimensions of information overlap by analyzing the interdependence between features, and improving the discrimination efficiency of the feature set.

[0055] In an implementation manner, when it is detected that there is a feature pair with mutual information value greater than 0.8 in the preliminary key feature vector, such as a high correlation between the current peak value and the temperature gradient, one of the more representative features is retained and the other is removed, and finally a refined state-related feature vector is obtained. This processing ensures the independence of each dimension in the feature set, provides feature input for subsequent anomaly detection, and improves the accuracy and efficiency of state evaluation. The whole feature processing flow is automatically executed through the cloud computing platform, and is real-time linked with the battery energy management business system. When an abnormal feature is detected, the operating parameters can be dynamically adjusted to ensure that the battery system always works within the state range.

[0056] In step S103, the dynamic change rule of the features under different working conditions is analyzed according to the state-related feature vector, and an abnormal trend signal is determined, including: S1031, the feature values in the state-related feature vector are extracted by using a sliding window method, and are grouped to obtain dynamic feature values; S1032, the dynamic feature values are weighted processed according to the real-time acquired environmental influence factors to obtain weighted feature values; S1033, the signal of the weighted feature value exceeding a preset first abnormal threshold is marked as an abnormal trend signal.

[0057] In step S1031, the feature values in the state-related feature vector are extracted by using a sliding window method, and are grouped to obtain dynamic feature values.

[0058] It should be noted that the sliding window method is a technical means for analyzing time series data through a fixed time span, and the core is to divide the continuous state feature vector into data segments with overlapping characteristics, so as to capture the local dynamic change characteristics of the battery running state.

[0059] In an implementation manner, for the classified battery running state data, such as normal discharge, rapid charging, high temperature running condition, 5 minutes is set as the time span of the sliding window, and the state-related feature vector is segmented processed with 1 minute as the step. For example, when analyzing the voltage characteristics in the charging process, the voltage in the window jumps from 3.7V to 3.9V, and the voltage mutation rate of the window is calculated as 0.2V / minute, which is a dynamic feature value. At the same time, the current fluctuation amplitude, temperature change gradient and other characteristics in the same window are extracted and grouped synchronously to form a dynamic feature value set reflecting the behavior characteristics of the battery in this period.

[0060] This processing method can effectively identify the subtle changes of battery state under different working conditions, and provide high timeliness feature input for subsequent anomaly detection.

[0061] In step S1032, the dynamic characteristic value is weighted processed according to the real-time acquired environmental factors, and a weighted characteristic value is obtained.

[0062] It should be noted that the environmental factor weighting is a process of adjusting the importance of the dynamic characteristic value by quantifying the influence degree of external environmental conditions on the battery characteristics, and the purpose is to eliminate environmental interference and highlight the characteristic changes that truly reflect the battery health state.

[0063] In an implementation manner, the environmental data is collected in real time by a temperature sensor and a humidity sensor, and a weighting model is established, in which the influence weight of temperature on voltage fluctuation is 0.6, and the influence weight of humidity on current fluctuation is 0.4. For example, when the environmental temperature rises to 40°C, the detected voltage fluctuation characteristic value 0.12 is weighted processed, 0.12*0.6=0.072, and then is superimposed with the current characteristic value weighted by humidity, and finally the comprehensive weighted characteristic value 0.15 is obtained.

[0064] This dynamic weighting mechanism can automatically adjust the contribution degree of each characteristic according to the actual environmental conditions, improve the accuracy of state evaluation, and especially improve the reliability of anomaly detection under complex working conditions.

[0065] In step S1033, the signal whose weighted characteristic value exceeds a preset first anomaly threshold is marked as an anomaly trend signal.

[0066] It should be noted that the anomaly trend marking is a key step of identifying early signs of potential failure by comparing the weighted characteristic value with the preset first anomaly threshold.

[0067] In an implementation manner, the preset first anomaly threshold of voltage fluctuation and current mutation is set to 0.5, when the weighted characteristic value of a time window reaches 0.7, the system automatically marks it as an anomaly signal, and records the starting time, duration and intensity of the signal and other parameters. For example, under high-temperature operating conditions, the weighted characteristic values of 3 consecutive windows for 15 minutes all exceed the threshold, and an anomaly trend signal set with time continuity is generated. This marking mechanism not only can identify instantaneous anomalies, but also can capture slowly developing potential failures, and provide complete anomaly evolution data for subsequent trend prediction and risk assessment. The entire anomaly detection process is realized by a cloud analysis platform for real-time processing, and when an anomaly signal is detected, a warning mechanism is triggered immediately, and the anomaly characteristic vector is stored in a fault case library, and the accuracy of the detection model is continuously optimized.

[0068] The abnormality detection method based on dynamic feature analysis and environment adaptive weighting can improve the early warning capability and fault identification precision of the battery safety management system, and provides reliable technical support for preventive maintenance.

[0069] In step S104, the state-related feature vector corresponding to the abnormal trend signal is compared with the preset abnormal threshold range to obtain an abnormality degree score, including: S1041, obtaining the state-related feature vector corresponding to the abnormal trend signal according to the abnormal trend signal; S1042, comparing and calculating the state-related feature vector with the preset second abnormal threshold to obtain the single abnormality degree of each state-related feature vector; S1043, weighting and summing all the single abnormality degrees to obtain an abnormality degree score.

[0070] In step S1041, the state-related feature vector corresponding to the abnormal trend signal is obtained according to the abnormal trend signal.

[0071] It should be noted that this step is a process of locating the relevant feature vector from the running data by tracking the space-time characteristics of the abnormal signal.

[0072] In an implementation mode, when the system detects the voltage fluctuation rate abnormal signal in a certain time window, the complete feature vector set corresponding to the period is automatically extracted, the key parameters include voltage mean 3.65V, current peak 3.2A, temperature gradient 1.5°C / min, and these parameters jointly constitute the state-related feature vector reflecting the complete features of the abnormal event.

[0073] In step S1042, the state-related feature vector is compared with the preset second abnormal threshold to obtain the single abnormality degree of each state-related feature vector.

[0074] It should be noted that this is to realize the accurate calibration of the abnormality degree by establishing the quantitative comparison relationship between each feature parameter and the standard running interval.

[0075] In an implementation mode, for the voltage fluctuation rate feature, the preset second abnormal threshold is 0.05-0.15V / min, when the detection value is 0.2V / min, the linear interpolation method is used to calculate the single abnormality degree as (0.2-0.15) / (0.25-0.15)=0.5, wherein 0.25V / min is the preset upper limit of the dangerous threshold. Similarly, the current peak abnormality degree is 0.7, and the temperature gradient abnormality degree is 0.6, forming the abnormality quantization index of each feature.

[0076] In step S1043, all the single abnormality degrees are weighted and summed to obtain an abnormality degree score.

[0077] It should be noted that this is the process of generating a comprehensive risk assessment value by comprehensively considering the abnormal performance of each feature parameter and its influence weight on battery safety.

[0078] In an implementation manner, the voltage fluctuation rate weight is set as 0.4, the current peak value weight is set as 0.3, and the temperature gradient weight is set as 0.3, and the comprehensive abnormality degree score is calculated as 0.5*0.4+0.7*0.3+0.6*0.3=0.59. The score value will be used for subsequent risk classification, and when the score exceeds 0.5, a secondary warning is triggered, and when the score exceeds 0.7, a primary warning is triggered.

[0079] This evaluation method based on multi-feature weighted fusion can avoid false alarms caused by single feature misjudgment and improve the reliability of the early warning system.

[0080] In step S105, the abnormality degree score is matched with the preset historical failure case library to determine the abnormal shutdown mode category, including: S1051, match the abnormality degree score with the preset historical failure case library to obtain a preliminary abnormal trend matching result; S1052, when the preliminary abnormal trend matching result exceeds a preset third abnormal threshold, a preliminary shutdown mode category is determined; S1053, refining the preliminary shutdown mode category classification result, and outputting a final abnormal shutdown mode category.

[0081] In step S1051, the abnormality degree score is matched with the preset historical failure case library to obtain a preliminary abnormal trend matching result.

[0082] It should be noted that the preset historical failure case library is a structured database optimized by machine learning, which contains various typical failure cases and their feature vectors accumulated by the battery system in the past operation. Each case records key information such as abnormal score interval, feature parameter change trend, failure type, and disposal measures in detail. When the abnormality degree score is matched with the historical failure case library, the system performs clustering retrieval by calculating the cosine similarity, which can accurately evaluate the matching degree of the current abnormality and the historical cases.

[0083] In an implementation manner, when the system detects that the current abnormal score is 0.85, 32 most relevant cases with a similarity score greater than 0.8 are quickly searched from a historical failure case library containing 10,000 cases, including 15 "overload failure" cases, 12 "insulation aging" cases, and 5 "loose connection" cases. These cases are arranged in descending order of similarity to form a preliminary abnormal trend matching result set.

[0084] This intelligent matching method avoids the limitations of relying on a single threshold judgment in traditional fault diagnosis, and improves the accuracy of early fault identification through multi-case comparison and analysis, while providing a wealth of reference for subsequent pattern classification.

[0085] In step S1052, when the preliminary abnormal trend matching result exceeds the preset third abnormal threshold, the preliminary shutdown mode category is determined.

[0086] It should be noted that the preset third abnormal threshold is a dynamic judgment standard verified by a large amount of experimental data, which is self-adaptive to the battery type, service life and environmental conditions, etc. to ensure the accuracy of the judgment under different working conditions.

[0087] In one implementation, when the system detects that the preliminary matching result exceeds the third abnormal threshold 0.7, it will immediately analyze the key indicators of the "overload fault" category case with the highest matching degree, such as voltage fluctuation characteristics (matching degree 0.82), temperature change curve (matching degree 0.79) and current harmonic characteristics (matching degree 0.75), etc. The comprehensive matching degree is 0.78 by weighted fusion calculation. At this time, the system will generate a preliminary judgment result containing confidence evaluation, and classify the current abnormality as "overload risk" shutdown mode.

[0088] This stage of judgment process pays special attention to the time sequence correlation analysis of feature parameters, not only compares the static characteristic values, but also evaluates the similarity of the evolution trend of abnormal signals and historical cases, so as to avoid misjudgment caused by instantaneous interference. At the same time, the system will record all the intermediate results and judgment basis in the verification process, providing complete data chain support for subsequent refined classification and fault tracing.

[0089] In step S1053, the preliminary shutdown mode category is refined and classified, and the final abnormal shutdown mode category is output.

[0090] It should be noted that this stage can achieve accurate subdivision of fault modes by analyzing the time-frequency characteristics and dynamic evolution law of abnormal signals.

[0091] In one implementation, the system will analyze the duration, intensity change rate and related parameter coupling relationship of abnormal signals according to the preliminary judgment of "overload risk" category, and further refine it into more accurate subcategories, such as when the duration of abnormal signals exceeds 30 minutes and the temperature continues to rise at a rate of more than 0.5°C / min, it is determined as "continuous overload-high temperature type" shutdown mode, and automatically associates 23 historical disposal schemes in the knowledge base under this mode.

[0092] The three-level classification mechanism improves the overall accuracy of fault diagnosis through progressive analysis of case library matching, multi-feature verification, and intelligent refinement.

[0093] In step S106, the risk level is generated by predicting the fault evolution path based on the abnormal shutdown mode category, including: S1061, according to the abnormal shutdown mode category, the conditional probability of abnormal mode to fault evolution is calculated by Bayes theorem; S1062, when the conditional probability exceeds the preset evolution threshold, the corresponding evolution path is marked as a possible path; S1063, the possible path is processed by sequence sorting to determine the path with a conditional probability higher than the preset probability threshold as an evolution path sequence; S1064, the evolution path sequence is multiplied by the preset future time parameter to obtain a cumulative risk value, which is matched with the preset risk determination rule to determine the risk level.

[0094] In step S1061, the conditional probability of abnormal mode to fault evolution is calculated by Bayes theorem according to the abnormal shutdown mode category.

[0095] It should be noted that the Bayes theorem calculation is a core method for quantitatively evaluating the possibility of fault development by establishing the probability correlation between historical fault data and the current abnormal mode.

[0096] In one implementation, for the classified "overload-high temperature type" abnormal mode, the system extracts 100 abnormal records under this mode from the preset historical fault case library, of which 70 eventually lead to fault shutdown. The conditional probability P(fault|overload-high temperature) is calculated by Bayes formula to be 70%, and when this value exceeds the preset threshold of 50%, it is determined that the abnormal mode has risk.

[0097] This probability calculation method fully considers the statistical law of historical data and can objectively reflect the actual risk level of different abnormal modes.

[0098] In step S1062, when the conditional probability exceeds the preset evolution threshold, the corresponding evolution path is marked as a possible path.

[0099] It should be noted that the preset evolution threshold is a risk alert line set according to the criticality of the device and the operation and maintenance experience, which is used to screen potential fault paths that need to be focused on.

[0100] In an implementation, the evolution threshold is set to 55%, and when the condition probability of the "overload-high temperature type" anomaly is detected to be 70%, the system automatically marks it as a possible path and associates the typical evolution characteristics under the path, such as temperature rise rate > 0.5°C / min, voltage fluctuation continuously expanding.

[0101] This threshold-based path screening mechanism effectively focuses limited resources and ensures priority monitoring of high-risk paths.

[0102] In step S1063, the possible paths are sequentially sorted and processed, and the paths with condition probability higher than the preset probability threshold are determined as the evolution path sequence.

[0103] It should be noted that the sequential sorting process is a key step to build a complete fault development chain by analyzing the time sequence correlation of multiple anomaly patterns.

[0104] For example, the system identifies three possible paths as overload-high temperature type (70%), insulation aging (60%), and connection loosening (55%), and arranges them in descending order of probability value. Further analysis of the time evolution relationship shows that "overload-high temperature type" and "insulation aging" have a continuous occurrence history pattern, accounting for 45%, so they are merged into an evolution path sequence of "overload-insulation aging". This multi-path fusion analysis method improves the completeness of fault prediction.

[0105] In step S1064, the evolution path sequence is multiplied by the preset future time parameter to obtain the cumulative risk value, which is matched with the preset risk determination rule to determine the risk level.

[0106] It should be noted that the preset future time parameter refers to a risk amplification factor reflecting changes in device operation plan, which is used to dynamically adjust the static probability evaluation result.

[0107] In an implementation, when it is predicted that the device load rate will increase by 20% in the next 24 hours, the time parameter is set to 1.2, the original probability of the "overload-insulation aging" path is corrected to 70% x 60% = 42%, and the cumulative risk value is 50.4%. According to the preset risk determination rule, such as high risk > 45%, medium risk 30% ~ 45%, and low risk < 30%, the final risk level is determined as "high risk", triggering a red alert and recommending preventive measures such as load reduction operation.

[0108] This dynamic risk assessment mechanism realizes real-time response to changes in operating environment, greatly improving the timeliness and accuracy of the warning.

[0109] In step S107, according to the risk level, corresponding early warning information is generated, and the battery safety early warning response level is determined, including: S1071, according to the risk level, the corresponding influence factor is obtained from the pre-established data set, the multivariate coupling influence weight is analyzed, and the key influence factor set is determined; S1072, obtain the risk control parameter from the key influence factor set, and optimize the risk control parameter by using the preset parameter adjustment logic to obtain the optimized control parameter combination; S1073, according to the optimized control parameter combination, corresponding battery safety early warning response information is generated; S1074, according to the battery safety early warning response information, combined with the preset early warning mapping table, the final battery safety early warning response level is determined.

[0110] In step S1071, according to the risk level, the corresponding influence factor is obtained from the pre-established data set, the multivariate coupling influence weight is analyzed, and the key influence factor set is determined.

[0111] It should be noted that the multivariate coupling influence weight analysis is a core method for quantitatively evaluating the contribution of each factor to the safety risk by establishing the interaction relationship between the battery system parameters.

[0112] In one implementation, when the system determines that the current risk level is "high risk", the key parameters such as temperature, charge and discharge rate, and cycle number are extracted from the database containing 20 influence factors, the improved analytic hierarchy process is used to calculate the weight of each factor, for example, the temperature change weight is 0.45, the charge and discharge rate weight is 0.35, and the cycle number weight is 0.2. The temperature and charge and discharge rate with weight value exceeding threshold value 0.3 are selected as key influence factors.

[0113] This analysis method effectively identifies the main driving factors that cause risks, and provides precise targeting for subsequent parameter optimization.

[0114] In step S1072, the risk control parameter is obtained from the key influence factor set, and the risk control parameter is optimized by using the preset parameter adjustment logic to obtain the optimized control parameter combination.

[0115] It should be noted that the parameter adjustment logic is a dynamic adjustment mechanism based on the battery safety operation constraints and performance optimization targets, and its core is to seek a balance point between the safety threshold and the performance index through multi-objective optimization. The logic first establishes the relationship between the battery thermal runaway critical condition and the operating parameters, such as the temperature key factor and the cooling fan speed and the cooling liquid flow, and then calculates the adjustment amplitude of each parameter combined with the real-time working condition data.

[0116] In an implementation manner, for the identified temperature critical factor, the system obtains the current cooling fan speed (2500 rpm) and coolant flow rate (2 L / min) as control parameters, calculates the adjustment scheme through fuzzy PID control, for example, increases the fan speed to 3000 rpm (+20%), increases the coolant flow rate to 2.4 L / min (+20%), and limits the charging current to 80% of the rated value.

[0117] This optimized combination not only ensures the improvement of heat dissipation efficiency, but also avoids the excessive consumption of energy, achieving the balance between safety and economy.

[0118] In step S1073, the corresponding battery safety warning response information is generated according to the optimized control parameter combination.

[0119] It should be noted that the warning response information is a comprehensive guidance scheme integrating risk diagnosis results and disposal suggestions.

[0120] In an implementation manner, for example, the system automatically generates warning information containing the following contents according to the optimized control parameter combination: "high risk alarm: overload-high temperature risk detected (confidence 92%), it is recommended to immediately perform: 1) adjust the cooling fan to 3000 rpm; 2) increase the coolant flow rate to 2.4 L / min; 3) limit the charging current to ≤80A; 4) recheck the temperature change every 30 minutes. The historical similar case disposal success rate is 88%".

[0121] The information is pushed to the monitoring terminal through a multi-modal manner, and a one-key execution interface of parameter adjustment is attached, which improves the emergency response efficiency.

[0122] In step S1074, according to the battery safety warning response information and the preset warning mapping table, the final battery safety warning response level is determined.

[0123] It should be noted that the warning mapping table is a standardized grading system established according to risk severity and disposal urgency.

[0124] In an implementation manner, the system predicts the risk reduction to 0.35 according to the current risk value 0.72 and the optimized parameter execution effect, refers to the mapping table to determine the response level as "secondary response" (0.6-0.8 interval), and triggers the following measures: 1) automatically record the abnormal data snapshot; 2) notify the responsible engineer to confirm; 3) start the state inspection every 15 minutes; 4) lock the maximum output power to 90% of the rated value.

[0125] The scheme realizes accurate control of the battery safety risk through an intelligent hierarchical response mechanism, reduces the incidence of serious accidents, and avoids resource waste caused by excessive response.

[0126] To sum up, the application discloses a battery safety performance detection method, which comprises the following steps: obtaining battery operation data and preprocessing the data to obtain preliminary structured data; performing feature compression and key feature extraction on the preliminary structured data to obtain a state-related feature vector; analyzing the dynamic change rule of the features under different working conditions according to the state-related feature vector, and determining an abnormal trend signal; comparing the state-related feature vector corresponding to the abnormal trend signal with a preset abnormal threshold range to obtain an abnormality degree score; matching the abnormality degree score with a preset historical fault case library to determine an abnormal shutdown mode category, predict a fault evolution path, and generate a risk level; generating corresponding early warning information according to the risk level, and determining a battery safety early warning response level. The method can realize early and accurate early warning of battery faults.

[0127] Reference Figure 2 The second embodiment of the application provides a battery safety performance detection system, which comprises: A data acquisition module is configured to obtain battery operation data and preprocess the data to obtain preliminary structured data. A feature extraction module is configured to perform feature compression and key feature extraction on the preliminary structured data to obtain a state-related feature vector. An abnormality detection module is configured to analyze the dynamic change rule of the features under different working conditions according to the state-related feature vector, and determine an abnormal trend signal. An abnormality scoring module is configured to compare the state-related feature vector corresponding to the abnormal trend signal with a preset abnormal threshold range to obtain an abnormality degree score. A mode matching module is configured to match the abnormality degree score with a preset historical fault case library to determine an abnormal shutdown mode category. A risk prediction module is configured to predict a fault evolution path based on the abnormal shutdown mode category, and generate a risk level. A response control module is configured to generate corresponding early warning information according to the risk level, and determine a battery safety early warning response level.

[0128] It should be noted that the battery safety performance detection system provided by the embodiments of the application is used to execute all process steps of the battery safety performance detection method of the above-mentioned embodiments, and the working principles and beneficial effects of the two are one-to-one corresponding, so the working principles and beneficial effects of the two are not repeated.

[0129] The embodiment of the present application further provides an electronic device. The electronic device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, for example, a battery safety performance detection system. The processor implements the steps in each of the battery safety performance detection method embodiments when executing the computer program, for example Figure 1 The processor implements the functions of each module / unit in each of the device embodiments when executing the computer program, for example, a data acquisition module.

[0130] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0131] The electronic device can be a desktop computer, a notebook computer, a palm computer, a smart tablet and the like. The electronic device can comprise, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device, and the electronic device can comprise more or fewer components than the above, or combine certain components or different components, for example, the electronic device can further comprise an input / output device, a network access device, a bus and the like.

[0132] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor and the like. The processor is the control center of the electronic device, and connects all parts of the electronic device through various interfaces and lines.

[0133] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), and the like. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0134] The modules / units integrated in the electronic device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can realize the steps of the above-mentioned various method embodiments when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0135] It should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate units can or can not be physically separate, and the units displayed as units can or can not be physical units, i.e. can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. In addition, the connection relationship between the modules in the apparatus embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0136] The above specific embodiments further illustrate the purpose, technical scheme and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting safety performance of a battery, characterized by, The method comprises the following steps: acquiring battery operation data and preprocessing to obtain preliminary structured data; performing feature compression and key feature extraction on the preliminary structured data to obtain a state-related feature vector; analyzing the dynamic change rule of the features under different working conditions according to the state-related feature vector to determine an abnormal trend signal; comparing the state-related feature vector corresponding to the abnormal trend signal with a preset abnormal threshold range to obtain an abnormality degree score; matching the abnormality degree score with a preset historical fault case library to determine an abnormal shutdown mode category; predicting a fault evolution path based on the abnormal shutdown mode category to generate a risk level; generating corresponding warning information according to the risk level to determine a battery safety warning response level; wherein, according to the risk level, the corresponding influence factors are obtained from a pre-established data set, the multivariate coupling influence weight is analyzed to determine a key influence factor set; the risk control parameters are obtained from the key influence factor set, the preset parameter adjustment logic is used to optimize the risk control parameters to obtain an optimized control parameter combination; and the corresponding battery safety warning response information is generated according to the optimized control parameter combination; determining the final battery safety warning response level according to the battery safety warning response information combined with a preset warning mapping table.

2. The method of claim 1, wherein the battery safety performance is detected by measuring a voltage of the battery. The method of acquiring battery operation data and preprocessing to obtain preliminary structured data comprises the following steps: collecting the battery operation data from the voltage dimension and the current dimension to obtain an original time series; performing sequence segmentation on the original time series to determine segmented data segments; performing smoothing processing on the segmented data segments to obtain a clean data set; grouping the clean data set according to similar segments to generate the preliminary structured data.

3. The method of claim 1, wherein the battery safety performance is detected by measuring a voltage of the battery. The method of performing feature compression and key feature extraction on the preliminary structured data to obtain a state-related feature vector comprises the following steps: performing principal component analysis dimension reduction on the high-dimensional operation data in the preliminary structured data to obtain a compressed feature set; for the compressed feature set, filtering features highly related to the battery state by preset coefficient calculation to determine a preliminary key feature vector; if there is a redundant dimension in the preliminary key feature vector, removing the redundant dimension by using a preset mutual information metric to obtain the state-related feature vector.

4. The method of claim 1, wherein the battery safety performance is detected by measuring a voltage of the battery. The method of analyzing the dynamic change rule of the features under different working conditions according to the state-related feature vector to determine an abnormal trend signal comprises the following steps: extracting feature values in the state-related feature vector by using a sliding window method and grouping to obtain dynamic feature values; performing weighted processing on the dynamic feature values according to real-time acquired environmental influence factors to obtain weighted feature values; marking signals with weighted feature values exceeding a preset first abnormal threshold as abnormal trend signals.

5. The method of claim 1, wherein the battery safety performance is detected by measuring a voltage of the battery. The method of comparing the state-related feature vector corresponding to the abnormal trend signal with a preset abnormal threshold range to obtain an abnormality degree score comprises the following steps: obtaining the state-related feature vector corresponding to the abnormal trend signal; The state correlation feature vector is compared with a preset second abnormal threshold to obtain a single abnormality degree of each state correlation feature vector; All the single abnormality degrees are weighted and summed to obtain an abnormality degree score.

6. The method of claim 1, wherein the battery safety performance is detected by measuring a voltage of the battery. The abnormality degree score is matched with a preset historical fault case library to determine an abnormal shutdown mode category, including: The abnormality degree score is matched with a preset historical fault case library to obtain a preliminary abnormal trend matching result; When the preliminary abnormal trend matching result exceeds a preset third abnormal threshold, a preliminary shutdown mode category is determined; The preliminary shutdown mode category is refined and classified to output a final abnormal shutdown mode category.

7. The method of claim 1, wherein the battery safety performance is detected by measuring a voltage of the battery. Based on the abnormal shutdown mode category, a fault evolution path is predicted and a risk level is generated, including: According to the abnormal shutdown mode category, the conditional probability of the abnormal mode to the fault evolution is calculated by Bayes' theorem; When the conditional probability exceeds a preset evolution threshold, the corresponding evolution path is marked as a possible path; The possible paths are sequentially sorted to determine that the path with a conditional probability higher than a preset probability threshold is an evolution path sequence; The evolution path sequence is multiplied by a preset future time parameter to obtain a cumulative risk value, which is matched with a preset risk determination rule to determine the risk level.

8. A battery safety performance detection system, characterized by, A method for implementing the battery safety performance detection method of any one of claims 1 to 7, including: A data acquisition module for obtaining battery operation data and preprocessing to obtain preliminary structured data; A feature extraction module for feature compression and key feature extraction of the preliminary structured data to obtain a state correlation feature vector; An anomaly detection module for analyzing the dynamic change rule of the feature under different working conditions based on the state correlation feature vector to determine an abnormal trend signal; An abnormality scoring module for comparing the state correlation feature vector corresponding to the abnormal trend signal with a preset abnormal threshold range to obtain an abnormality degree score; A pattern matching module for matching the abnormality degree score with a preset historical fault case library to determine an abnormal shutdown mode category; A risk prediction module for predicting a fault evolution path based on the abnormal shutdown mode category and generating a risk level; A response control module for generating corresponding warning information according to the risk level and determining a battery safety warning response level.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the battery safety performance detection method of any one of claims 1 to 7 when the computer program is running. The computer-readable storage medium includes a stored computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the battery safety performance detection method of any one of claims 1 to 7 when the computer program is running.

Images