A battery safety management method and system based on BMS

Abstract

The present invention relates to the field of charger technology, and in particular to a battery safety management method and system based on a BMS. The present invention collects battery operating data in real time, uses sliding window technology to segment continuous data streams, extracts a discretized state point set, extracts features from it and performs denoising processing to obtain an optimized feature vector set, uses a clustering algorithm to perform pattern division, matches the error pattern library to determine the potential error type, combines decision tree analysis to analyze the correlation between time and state dimensions, determines the correspondence between the cause of the abnormality and the operating stage, evaluates the severity of the error through conditional probability calculation, identifies the error stage, and finally extracts high-risk state points, compares historical data to verify accuracy, and outputs a safety management analysis report. The present invention achieves comprehensive monitoring of the battery operating status and abnormality warning, effectively improves the safety and reliability of battery use, and provides an important basis for the optimization of the battery management system.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery management and safety control, and specifically relates to a battery safety management method and system based on a BMS. Background Art

[0002] Traditional battery management systems rely on simple threshold judgments to identify anomalies when processing continuous data streams, and their ability to cope with complex error patterns is limited. Especially during battery operation, the dynamic and multi-causal nature of error states makes it easy for traditional methods to confuse anomalies of different causes, which may lead to misjudgments or missed judgments, affecting the effectiveness of refined management. In addition, existing technologies lack accuracy in feature extraction and error pattern matching, and anomalies with the same appearance may originate from different operating stages or conditions. If their causes cannot be effectively distinguished, potential risks may be amplified. Therefore, how to build a refined error analysis framework through time segmentation and state division has become an important direction for improving the safety management capabilities of BMS. Summary of the Invention

[0003] In order to solve the above technical problems, the present application is proposed. The embodiments of the present invention provide a battery safety management method and system based on BMS. The system first collects the voltage, current, and temperature parameters of the battery during operation in real time through sensors, forming a continuous data stream. A time series sampling method is used to obtain the original data set. Next, a sliding window technique is used to segment the continuous data stream, extracting subsequences based on preset time intervals in the time dimension to obtain a discretized state point set. Statistical features, including mean, variance, and rate of change, are extracted from the discretized state point set and combined with the physical model to generate an initial feature vector set. Subsequently, if noise interference is present in the initial feature vector set, the feature vectors are denoised using a wavelet transform method to obtain an optimized feature vector set. Next, a K-means clustering algorithm is used to perform pattern segmentation on the optimized feature vector set. The clustering results are matched against a pre-established error pattern library to determine the distribution of potential error types. A decision tree classification method is then used to analyze the correlation between the time dimension and the state dimension, determining the correspondence between the cause of the anomaly and the operating stage. Finally, a conditional probability calculation method is used to assess the error severity of each state point, identifying the error stage. The identification accuracy is verified by comparing historical data, and a final safety management analysis report is output. In this way, the error detection accuracy and safety management capabilities of BMS in complex dynamic scenarios can be effectively improved.

[0004] According to one aspect of the present invention, a battery safety management method based on a BMS is provided, which includes:

[0005] Acquire battery operation data collected by sensors in real time, including voltage, current, and temperature parameters, to form a continuous data stream, and use time series sampling methods to obtain the original data set;

[0006] The sliding window technology is used to segment the continuous data stream, extract subsequences according to the preset time interval in the time dimension, and obtain a discretized state point set;

[0007] Extract statistical features from the discretized state point set, including mean, variance, and rate of change, and generate an initial feature vector set based on the physical model;

[0008] If there is noise interference in the initial feature vector set, the feature vector is denoised by wavelet transform method to obtain the optimized feature vector set;

[0009] For the optimized feature vector set, the K-means clustering algorithm is used to perform pattern division, and the clustering results are matched with the pre-established error pattern library to determine the distribution of potential error types;

[0010] Analyze the correlation between the time dimension and the state dimension through the decision tree classification method to determine the corresponding relationship between the cause of the abnormality and the operation stage;

[0011] The conditional probability calculation method is used to evaluate the error severity of each state point and obtain the error stage identification result;

[0012] High-risk status points are extracted from the error stage identification results, and the identification accuracy is verified by comparing with historical data to output the final safety management analysis report.

[0013] In the above-mentioned BMS-based battery safety management method, the sliding window technology is used to segment the continuous data stream, and subsequences are extracted according to preset time intervals in the time dimension to obtain a discretized state point set, including:

[0014] Set the step size and window width of the sliding window to divide the continuous data stream into multiple overlapping or non-overlapping time periods;

[0015] Extract subsequences in each time period and map the subsequences into a multidimensional feature space to form a discretized state point set;

[0016] The step size and window width of the sliding window are dynamically adjusted according to the battery operating characteristics and data sampling frequency.

[0017] In the above-mentioned BMS-based battery safety management method, statistical features, including mean, variance, and rate of change, are extracted from the discretized state point set, and combined with the physical model to generate an initial feature vector set, including:

[0018] Calculate the mean and variance of each subsequence as the basic statistical features to characterize the battery operating status;

[0019] Calculate the rate of change of the subsequence to capture the dynamic change characteristics of the battery operating state;

[0020] Combined with the battery physical model, electrochemical parameters and thermodynamic parameters are introduced to generate an initial feature vector set containing multi-dimensional information.

[0021] In the above-mentioned BMS-based battery safety management method, if there is noise interference in the initial feature vector set, the feature vector is denoised by a wavelet transform method to obtain an optimized feature vector set, including:

[0022] Select appropriate wavelet basis functions and decomposition levels to perform multi-scale decomposition on the initial feature vector set;

[0023] At each scale after decomposition, a threshold denoising algorithm is applied to remove high-frequency noise;

[0024] The denoised eigenvectors are reconstructed through inverse wavelet transform to obtain the optimized eigenvector set.

[0025] In the above-mentioned BMS-based battery safety management method, the K-means clustering algorithm is used to perform pattern division on the optimized feature vector set. The clustering results are then matched with a pre-established error pattern library to determine the distribution of potential error types, including:

[0026] Initialize the number of cluster centers of the K-means clustering algorithm and set the maximum number of iterations;

[0027] Calculate the distance between each eigenvector in the optimized eigenvector set and the cluster center, and classify according to the minimum distance principle;

[0028] Update the cluster center position and repeat the above steps until convergence to obtain the final clustering result;

[0029] The clustering results are matched with the patterns in the error pattern library to determine the distribution of potential error types.

[0030] In the above-mentioned BMS-based battery safety management method, the correlation between the time dimension and the state dimension is analyzed through the decision tree classification method to determine the correspondence between the abnormality cause and the operation stage, including:

[0031] Build a decision tree model and select the key dimensions in the feature vector as split nodes;

[0032] Train the decision tree model based on the training data set to generate classification rules;

[0033] Use classification rules to predict the test data set and analyze the correlation between the time dimension and the state dimension;

[0034] Based on the prediction results, the correspondence between the abnormality cause and the battery operation stage is determined.

[0035] In the above-mentioned BMS-based battery safety management method, a conditional probability calculation method is used to evaluate the error severity of each state point to obtain the error stage identification results, including:

[0036] Define the conditional probability formula to calculate the probability of each state point occurring at different error stages;

[0037] According to the maximum value of the occurrence probability, determine the error stage to which the state point belongs;

[0038] The error stage identification results of all state points are summarized to generate an overall error stage distribution map.

[0039] According to another aspect of the present invention, a battery safety management system based on a BMS is provided, comprising:

[0040] The data acquisition module is used to obtain the battery operation data collected by the sensor in real time, including voltage, current and temperature parameters, forming a continuous data stream and using the time series sampling method to obtain the original data set;

[0041] The data segmentation module is used to segment the continuous data stream using the sliding window technology, extract subsequences according to the preset time interval in the time dimension, and obtain a discretized state point set;

[0042] The feature extraction module is used to extract statistical features from the discretized state point set, including mean, variance and rate of change, and generate an initial feature vector set in combination with the physical model;

[0043] A denoising module is used to denoise the feature vectors by wavelet transform if there is noise interference in the initial feature vector set to obtain an optimized feature vector set;

[0044] The pattern division module is used to perform pattern division based on the optimized feature vector set using the K-means clustering algorithm, and match the clustering results with the pre-established error pattern library to determine the distribution of potential error types;

[0045] The correlation analysis module is used to analyze the correlation between the time dimension and the state dimension through the decision tree classification method to determine the correspondence between the cause of the abnormality and the operation stage;

[0046] The error assessment module is used to evaluate the error severity of each state point using the conditional probability calculation method to obtain the error stage identification result;

[0047] The result verification module is used to extract high-risk state points from the error stage recognition results, verify the recognition accuracy through historical data comparison, and output the final safety management analysis report.

[0048] In the above-mentioned BMS-based battery safety management system, the data segmentation module is used to:

[0049] Set the step size and window width of the sliding window to divide the continuous data stream into multiple overlapping or non-overlapping time periods;

[0050] Extract subsequences in each time period and map the subsequences into a multidimensional feature space to form a discretized state point set;

[0051] The step size and window width of the sliding window are dynamically adjusted according to the battery operating characteristics and data sampling frequency.

[0052] In the above-mentioned BMS-based battery safety management system, the feature extraction module is used to:

[0053] Calculate the mean and variance of each subsequence as the basic statistical features to characterize the battery operating status;

[0054] Calculate the rate of change of the subsequence to capture the dynamic change characteristics of the battery operating state;

[0055] Combined with the battery physical model, electrochemical parameters and thermodynamic parameters are introduced to generate an initial feature vector set containing multi-dimensional information.

[0056] In the above-mentioned BMS-based battery safety management system, the mode division module is used to:

[0057] Initialize the number of cluster centers of the K-means clustering algorithm and set the maximum number of iterations;

[0058] Calculate the distance between each eigenvector in the optimized eigenvector set and the cluster center, and classify according to the minimum distance principle;

[0059] Update the cluster center position and repeat the above steps until convergence to obtain the final clustering result;

[0060] The clustering results are matched with the patterns in the error pattern library to determine the distribution of potential error types.

[0061] The beneficial effects of the present invention are as follows: This invention collects battery operation data in real time, uses sliding window technology to segment continuous data streams, extracts a set of discrete state points, extracts features from them and performs denoising processing to obtain an optimized feature vector set, uses a clustering algorithm to perform pattern division, matches the error pattern library to determine the potential error type, combines decision tree analysis to analyze the correlation between time and state dimensions, determines the correspondence between the cause of the anomaly and the operation stage, evaluates the severity of the error through conditional probability calculation, identifies the error stage, and finally extracts high-risk state points, compares historical data to verify accuracy, and outputs a safety management analysis report. The present invention achieves comprehensive monitoring of the battery operation status and abnormality warning, effectively improves the safety and reliability of battery use, and provides an important basis for the optimization of the battery management system. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not limit the invention in any way. A person skilled in the art can derive other drawings based on the following drawings without inventive effort.

[0063] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0064] The present invention is further described with reference to the following examples.

[0065] Depend on Figure 1 It can be seen that the embodiment of the present invention provides a battery safety management method and system based on BMS, which realizes comprehensive monitoring and safety management of battery operating status through the collaborative work of multiple modules. Figure 1 The module structure diagram is shown in the figure. The specific implementation of each module and the connection relationship, position relationship and coordination relationship between them will be described in detail below.

[0066] like Figure 1As shown in the figure, the system includes a data acquisition module, a data segmentation module, a feature extraction module, a denoising module, a pattern partitioning module, a correlation analysis module, an error assessment module, and a result verification module. These modules are connected in a logical sequence to form a complete data processing chain. The data acquisition module is responsible for acquiring battery operating data from sensors and passing it to the data segmentation module. The data segmentation module segments the continuous data stream and outputs a discretized set of state points to the feature extraction module. The feature extraction module generates an initial set of feature vectors and passes them to the denoising module. The denoising module denoises the feature vectors and outputs an optimized set of feature vectors to the pattern partitioning module. The pattern partitioning module performs cluster analysis and passes the results to the correlation analysis module. The correlation analysis module further analyzes the correlation between the time dimension and the state dimension and passes the results to the error assessment module. The error assessment module calculates the error severity of each state point and passes the results to the result verification module. Finally, the result verification module generates a safety management analysis report.

[0067] During implementation, the data acquisition module collects the voltage, current, and temperature parameters of the battery in real time during operation using voltage, current, and temperature sensors installed on the battery. These sensors are connected to the data acquisition module via signal lines, ensuring that the collected data can be directly transmitted to the module. A time series sampling unit is internally installed in the data acquisition module, which samples the collected continuous data stream at preset time intervals to generate the original data set. To ensure sampling accuracy, the sampling frequency of the time series sampling unit is dynamically adjusted based on the battery's operating characteristics and application scenarios. For example, in high-power charging and discharging scenarios, the sampling frequency is set to 100 times per second; in low-power or static scenarios, the sampling frequency can be reduced to 10 times per second. This dynamic adjustment mechanism ensures efficient and accurate data acquisition.

[0068] The data segmentation module receives the raw data set from the data acquisition module and segments the continuous data stream using a sliding window technique. The sliding window step size and window width are core parameters of this module, and their settings directly impact the accuracy and efficiency of subsequent data analysis. In practical applications, the sliding window step size is typically set to half the window width to ensure that the overlapping nature of data segmentation can capture subtle changes in the temporal dimension. For example, when the window width is set to 10 seconds, the step size is 5 seconds. The segmented subsequences are mapped into a multidimensional feature space, forming a set of discretized state points. This process is accomplished by an embedded processor, which includes a dedicated mathematical operation unit for performing subsequence extraction and mapping operations. Furthermore, the data segmentation module features a dynamic adjustment function that can modify the sliding window parameters in real time based on the battery's operating characteristics and data sampling frequency to adapt to different operating environments.

[0069] The feature extraction module receives the discretized state point set from the data segmentation module and extracts statistical features from it. Specifically, the module first calculates the mean and variance of each subsequence as the basic statistical features to characterize the battery operating status. The mean reflects the overall level of the subsequence, while the variance reflects the degree of fluctuation of the subsequence. On this basis, the feature extraction module further calculates the rate of change of the subsequence to capture the dynamic change characteristics of the battery operating status. The formula for calculating the rate of change is the difference between the end value and the starting value of the current subsequence divided by the time interval. In order to enhance the information content of the feature vector, the feature extraction module also introduces electrochemical parameters and thermodynamic parameters in combination with the battery physical model to generate an initial feature vector set containing multi-dimensional information. These parameters include the internal resistance, open circuit voltage and thermal conductivity of the battery, which are calculated and integrated through the built-in physical modeling unit.

[0070] The denoising module receives the initial feature vector set from the feature extraction module and performs denoising on it. Because the initial feature vector set may contain high-frequency noise, which can affect the accuracy of subsequent analysis, the denoising module uses a wavelet transform to denoise the feature vectors. The specific steps are as follows: First, an appropriate wavelet basis function and number of decomposition levels are selected to perform a multiscale decomposition of the initial feature vector set. Then, a threshold denoising algorithm is applied to each decomposed scale to remove high-frequency noise. Finally, an inverse wavelet transform is used to reconstruct the denoised feature vectors to obtain an optimized feature vector set. The denoising module includes a wavelet transform processor and a threshold calculation unit, which perform the multiscale decomposition and threshold denoising operations, respectively. The wavelet basis function is selected based on the spectral characteristics of the battery operating data, and the number of decomposition levels is dynamically adjusted based on the noise intensity. For example, in cases of strong noise, the number of decomposition levels is set to five; in cases of weak noise, the number of decomposition levels can be reduced to three.

[0071] The pattern division module receives the optimized feature vector set from the denoising module and uses the K-means clustering algorithm to perform pattern division. The module first initializes the number of cluster centers and sets the maximum number of iterations. The number of cluster centers is determined by the complexity of the battery's operating state and is usually in the range of 3 to 10. Subsequently, the pattern division module calculates the distance between each feature vector in the optimized feature vector set and the cluster center and classifies them according to the principle of minimum distance. After classification, the cluster center position is updated and the above steps are repeated until convergence to obtain the final clustering result. The clustering results are matched with a pre-established error pattern library to determine the distribution of potential error types. The error pattern library is stored in the internal memory of the pattern division module and contains a variety of typical error patterns and their corresponding feature vector distributions. The matching process is completed by the similarity calculation unit, which calculates the similarity based on the Euclidean distance between the feature vectors and selects the pattern with the highest similarity as the matching result.

[0072] The correlation analysis module receives the clustering results from the pattern partitioning module and analyzes the correlation between the time and state dimensions using a decision tree classification method. The module first constructs a decision tree model and selects key dimensions from the feature vector as splitting nodes. The selection of key dimensions is determined by the feature importance score, which is calculated by the built-in feature selection unit. The correlation analysis module then trains the decision tree model using a training dataset to generate classification rules. The training dataset is stored in the module's external database and contains a large amount of historical operating data and its corresponding annotation information. After training is complete, the classification rules are used to predict the test dataset and analyze the correlation between the time and state dimensions. The prediction results are output by the correlation analysis unit and used to determine the correspondence between the cause of the anomaly and the battery's operating stage.

[0073] The Error Assessment Module receives the prediction results from the Correlation Analysis Module and uses conditional probability calculation to assess the error severity of each state point. This module first defines a conditional probability formula to calculate the probability of occurrence of each state point under different error stages. This conditional probability formula is calculated based on Bayes' theorem and is performed using a built-in probability calculation unit. Subsequently, the error stage to which the state point belongs is determined based on the maximum probability of occurrence. The error stage identification results for all state points are aggregated to generate an overall error stage distribution map, which is displayed using a visualization unit. The Error Assessment Module also features a dynamic adjustment function that updates the conditional probability formula in real time based on new data input to improve assessment accuracy.

[0074] The result verification module receives the overall error stage distribution map from the error assessment module and extracts high-risk state points from it. The extraction of high-risk state points is determined by the error severity score, which is calculated by the built-in risk assessment unit. After the extraction is completed, the result verification module verifies the recognition accuracy by comparing historical data. The historical data is stored in the module's external database and contains a large number of known error cases and their corresponding annotation information. The verification process is completed by the similarity calculation unit, which calculates the similarity based on the difference in the feature vectors between the state points and selects the case with the highest similarity as the verification result. Finally, the result verification module generates a security management analysis report and transmits it to the user terminal through the output interface. The report content includes the error stage distribution map, a list of high-risk state points, and their corresponding error types and severity scores.

[0075] The above modules are interconnected through a data bus to ensure that data can be efficiently transmitted between modules. The data bus adopts a high-speed serial communication protocol and supports multi-channel concurrent transmission to meet the needs of large-scale data processing. In addition, a synchronization control unit is provided between each module to coordinate the operating rhythm between modules to avoid data loss or delay. The operating process of the entire system is uniformly scheduled by the main control unit, and the main control unit dynamically allocates computing resources according to the preset task priority to ensure that the system can operate stably in complex dynamic scenarios. In order to better enable relevant personnel in this technical field to fully understand and implement the present invention, the specific implementation principle of the present invention is further supplemented below in combination with a specific application scenario.

[0076] In practical applications, for example, the battery pack of a new energy vehicle operates in a complex and ever-changing environment, potentially facing high temperatures, high loads, and frequent charging and discharging. In these situations, a battery safety management system based on a BMS can achieve refined monitoring and safety management of the battery's operating status through the coordinated operation of multiple modules.

[0077] First, the data acquisition module collects operating parameters in real time using voltage, current, and temperature sensors installed on the battery pack. These sensors transmit the collected data via signal lines to the data acquisition module. The module's internal time series sampling unit dynamically adjusts the sampling frequency based on the current operating scenario. For example, when the vehicle is traveling at high speed, the battery is in a high-power discharge state, and the sampling frequency is set to 100 times per second. However, when the vehicle is traveling at low speed or parked, the sampling frequency is reduced to 10 times per second. This dynamic adjustment mechanism ensures efficient data acquisition while avoiding the generation of redundant data.

[0078] The data segmentation module then receives the raw data set from the data acquisition module and segments the continuous data stream using a sliding window technique. In this scenario, the sliding window step size is set to 5 seconds, and the window width is 10 seconds. This parameter setting ensures that the segmented subsequences maintain a certain degree of overlap in the temporal dimension, thereby capturing subtle changes in the battery's operating status. The segmented subsequences are mapped into a multidimensional feature space, forming a set of discretized state points. The mathematical operation unit in the embedded processor completes this process, ensuring the accuracy and real-time performance of the segmentation operation.

[0079] Next, the feature extraction module extracts statistical features from the discretized state point set. Specifically, for each subsequence, the module calculates its mean and variance, reflecting the overall level and degree of fluctuation of the subsequence, respectively. In addition, the module also calculates the rate of change of the subsequence to capture the dynamic changes in the battery's operating state. On this basis, electrochemical parameters and thermodynamic parameters are introduced in combination with the battery physical model to generate an initial feature vector set containing multi-dimensional information. For example, for operating data in a high-temperature environment, the module will focus on analyzing the changing trends of the battery's internal resistance and thermal conductivity to more accurately characterize the battery state.

[0080] After receiving the initial set of feature vectors, the denoising module uses a wavelet transform to denoise them. Because sensor data in high-temperature environments may be affected by high-frequency noise, the module selects appropriate wavelet basis functions and decomposition levels to perform a multi-scale decomposition of the feature vectors. In this scenario, the number of decomposition levels is set to five to effectively remove high-frequency noise. Subsequently, an inverse wavelet transform is used to reconstruct the optimized feature vector set to ensure the accuracy of subsequent analysis.

[0081] After receiving the optimized feature vector set, the pattern classification module uses the K-means clustering algorithm to classify it into patterns. In this scenario, the number of cluster centers is initialized to 5. The module calculates the distance between each feature vector and the cluster center and classifies them based on the principle of minimum distance. After classification, the cluster center position is updated and the above steps are repeated until convergence, resulting in the final clustering results. The clustering results are matched with the patterns in the error pattern library to determine the distribution of potential error types. For example, if a cluster result is highly similar to an overheating pattern, it is determined that there may be a risk of battery overheating during that time period.

[0082] After receiving the clustering results, the correlation analysis module analyzes the correlation between the time and state dimensions using a decision tree classification method. The module constructs a decision tree model, selects key dimensions from the feature vector as splitting nodes, and trains the model using a training dataset. In this scenario, the module's analysis found that abnormal conditions in high-temperature environments are primarily associated with increased battery internal resistance and decreased thermal conductivity, thereby determining the correspondence between the cause of the abnormality and the operating stage.

[0083] After receiving the prediction results, the error assessment module uses conditional probability calculation to assess the error severity of each state point. The module defines a conditional probability formula to calculate the probability of each state point occurring in different error stages. For example, if the probability of a state point occurring in the overheating stage is 0.85, significantly higher than in other stages, then the state point is considered to be in the overheating stage. The error stage identification results for all state points are aggregated to generate an overall error stage distribution map, which is displayed using the visualization unit.

[0084] Finally, the result verification module receives the overall error phase distribution map and extracts high-risk status points from it. The module verifies the recognition accuracy by comparing historical data and generates a safety management analysis report. The report includes the error phase distribution map, a list of high-risk status points, and their corresponding error types and severity scores. For example, the report indicates that in high-temperature environments, the battery pack's overheating risk is primarily concentrated during high-speed driving, and recommends taking cooling measures to mitigate the risk.

[0085] The above modules are interconnected via a data bus, and the main control unit dynamically allocates computing resources based on preset task priorities, ensuring the system can operate stably in complex and dynamic scenarios. Through these steps, this system can effectively improve the error detection accuracy and safety management capabilities of the BMS in complex and dynamic scenarios, providing reliable safety protection for new energy vehicle battery packs.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A battery safety management method based on BMS, characterized by: The following steps are involved: Acquire battery operation data collected by sensors in real time, including voltage, current, and temperature parameters, to form a continuous data stream, and use time series sampling methods to obtain the original data set; The sliding window technology is used to segment the continuous data stream, extract subsequences according to the preset time interval in the time dimension, and obtain a discretized state point set; Extract statistical features from the discretized state point set, including mean, variance, and rate of change, and generate an initial feature vector set based on the physical model; If there is noise interference in the initial feature vector set, the feature vector is denoised by wavelet transform method to obtain the optimized feature vector set; For the optimized feature vector set, the K-means clustering algorithm is used to perform pattern division, and the clustering results are matched with the pre-established error pattern library to determine the distribution of potential error types; The decision tree classification method is used to analyze the correlation between the time dimension and the state dimension composed of the initial feature vector set, and the corresponding relationship between the abnormal cause and the operation stage is determined; The conditional probability calculation method is used to evaluate the error severity of each state point and obtain the error stage identification result; High-risk status points are extracted from the error stage identification results, and the identification accuracy is verified by comparing with historical data to output the final safety management analysis report.

2. The battery safety management method based on BMS according to claim 1, characterized in that: The sliding window technology is used to segment the continuous data stream, extract subsequences according to the preset time interval in the time dimension, and obtain a discretized state point set, including: Set the step size and window width of the sliding window to divide the continuous data stream into multiple overlapping or non-overlapping time periods; Extract subsequences in each time period and map the subsequences into a multidimensional feature space to form a discretized state point set; The step size and window width of the sliding window are dynamically adjusted according to the battery operating characteristics and data sampling frequency.

3. The battery safety management method based on BMS according to claim 1, characterized in that: Statistical features, including mean, variance, and rate of change, are extracted from the discretized state point set and combined with the physical model to generate an initial feature vector set, including: Calculate the mean and variance of each subsequence as the basic statistical features to characterize the battery operating status; calculate the rate of change of the subsequence to capture the dynamic changes in the battery operating status; Combined with the battery physical model, electrochemical parameters and thermodynamic parameters are introduced to generate an initial feature vector set containing multi-dimensional information.

4. The battery safety management method based on BMS according to claim 1, characterized in that: If there is noise interference in the initial feature vector set, the feature vector is denoised by wavelet transform method to obtain the optimized feature vector set, including: Select appropriate wavelet basis functions and decomposition levels to perform multi-scale decomposition on the initial feature vector set; apply threshold denoising algorithm to remove high-frequency noise at each decomposed scale; The denoised eigenvectors are reconstructed through inverse wavelet transform to obtain the optimized eigenvector set.

5. The battery safety management method based on BMS according to claim 1, characterized in that: For the optimized feature vector set, the K-means clustering algorithm is used to perform pattern division. The clustering results are matched with the pre-established error pattern library to determine the distribution of potential error types, including: Initialize the number of cluster centers of the K-means clustering algorithm and set the maximum number of iterations; Calculate the distance between each eigenvector in the optimized eigenvector set and the cluster center, and classify according to the minimum distance principle; Update the cluster center position; Repeatedly calculate the distance between each eigenvector in the optimized eigenvector set and the cluster center, classify according to the minimum distance principle, and update the location of the cluster center until convergence to obtain the final clustering result; The clustering results are matched with the patterns in the error pattern library to determine the distribution of potential error types.

6. The battery safety management method based on BMS according to claim 1, characterized in that: The decision tree classification method is used to analyze the correlation between the time dimension and the state dimension, and determine the corresponding relationship between the abnormal cause and the operation stage, including: Build a decision tree model and select the key dimensions in the feature vector as split nodes; Train the decision tree model based on the training data set to generate classification rules; Use classification rules to predict the test data set and analyze the correlation between the time dimension and the state dimension; Based on the prediction results, the correspondence between the abnormality cause and the battery operation stage is determined.

7. A battery safety management system based on BMS, characterized by: Includes the following modules: The data acquisition module is used to obtain the battery operation data collected by sensors in real time, including voltage, current and temperature parameters, forming a continuous data stream and using a time series sampling method to obtain the original data set; The data segmentation module is used to segment the continuous data stream using the sliding window technology, extract subsequences according to the preset time interval in the time dimension, and obtain a discretized state point set; The feature extraction module is used to extract statistical features from the discretized state point set, including mean, variance and rate of change, and generate an initial feature vector set in combination with the physical model; A denoising module is used to denoise the feature vectors by wavelet transform if there is noise interference in the initial feature vector set to obtain an optimized feature vector set; The pattern division module is used to perform pattern division based on the optimized feature vector set using the K-means clustering algorithm, and match the clustering results with the pre-established error pattern library to determine the distribution of potential error types; The correlation analysis module is used to analyze the correlation between the time dimension and the state dimension composed of the initial feature vector set through the decision tree classification method, and determine the correspondence between the cause of the abnormality and the operation stage; The error assessment module is used to evaluate the error severity of each state point using the conditional probability calculation method to obtain the error stage identification result; The result verification module is used to extract high-risk state points from the error stage recognition results, verify the recognition accuracy through historical data comparison, and output the final safety management analysis report.

8. The BMS-based battery safety management system according to claim 7, characterized in that: The data segmentation module is used to: Set the step size and window width of the sliding window to divide the continuous data stream into multiple overlapping or non-overlapping time periods; Extract subsequences in each time period and map the subsequences into a multidimensional feature space to form a discretized state point set; The step size and window width of the sliding window are dynamically adjusted according to the battery operating characteristics and data sampling frequency.

9. The battery safety management system based on BMS according to claim 7, characterized in that: The feature extraction module is used to: Calculate the mean and variance of each subsequence as the basic statistical features to characterize the battery operating status; Calculate the rate of change of the subsequence to capture the dynamic change characteristics of the battery operating state; Combined with the battery physical model, electrochemical parameters and thermodynamic parameters are introduced to generate an initial feature vector set containing multi-dimensional information.

10. The battery safety management system based on BMS according to claim 7, characterized in that: The mode division module is used for: Initialize the number of cluster centers of the K-means clustering algorithm and set the maximum number of iterations; Calculate the distance between each eigenvector in the optimized eigenvector set and the cluster center, and classify according to the minimum distance principle; Update the cluster center position; Repeatedly calculate the distance between each eigenvector in the optimized eigenvector set and the cluster center, classify according to the minimum distance principle, and update the location of the cluster center until convergence to obtain the final clustering result; The clustering results are matched with the patterns in the error pattern library to determine the distribution of potential error types.

Images