Single power battery anomaly detection method and system based on space-time correlation analysis

Abstract

The invention discloses a time-space correlation analysis-based single power battery anomaly detection method and system, and belongs to the field of battery anomaly detection. The method comprises the following steps: collecting state data of a single battery; the collected data are preprocessed; correcting the preprocessed data through a preset dynamic environment compensation model; performing space-time correlation feature extraction on the corrected data; meanwhile, according to the corrected data, self-adaptive threshold values of various battery monomer parameters are calculated through a machine learning model; performing abnormal mode matching in a preset abnormal mode feature library according to the time-space correlation features; and performing anomaly detection and graded early warning in combination with the space-time correlation features and the adaptive threshold. The abnormal detection accuracy of the single power battery is improved, and the safety and reliability of the battery system are improved.

Description

Technical Field

[0001] This invention belongs to the field of battery anomaly detection. Specifically, this invention relates to a method and system for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis. Background Technology

[0002] Current power battery anomaly detection technology faces many shortcomings:

[0003] Insufficient environmental adaptability: Traditional methods do not fully consider the nonlinear effects of environmental parameters such as temperature and humidity on battery performance, resulting in significant fluctuations in false alarm rate under different climatic conditions;

[0004] Insufficient utilization of spatiotemporal features: Most schemes rely only on data analysis in a single time or space dimension, ignoring the spatiotemporal coupling characteristics of individual battery parameter changes, making it difficult to capture early weak abnormal signals;

[0005] Rigid threshold setting: Using a fixed threshold or a simple linear compensation model cannot adapt to the performance degradation pattern and changes in operating conditions throughout the battery's entire life cycle;

[0006] The early warning mechanism is too simple: the lack of a tiered early warning system makes it difficult to distinguish between anomalies of different severity, which affects the efficiency of fault handling.

[0007] For example, publication CN118457364A, dated August 9, 2024, discloses a cloud-based method for monitoring the health status of electric vehicle power batteries. This method involves uploading battery-related data from the vehicle's battery management system to a cloud server via an onboard terminal. Simultaneously, a battery capacity self-learning algorithm is created in the cloud to estimate the real-time health status of the power batteries in online vehicles, with a particular focus on monitoring vehicles with abnormal battery lifespan degradation and those nearing the end of their lifespan. This allows for early warnings, informing users to promptly visit a repair shop or replace the power battery. However, this published document does not overcome the aforementioned shortcomings.

[0008] Therefore, this invention proposes a method and system for detecting anomalies in power battery cells based on spatiotemporal correlation analysis. Summary of the Invention

[0009] This invention aims to overcome the shortcomings of existing technologies and proposes a method and system for detecting anomalies in power battery cells based on spatiotemporal correlation analysis, in order to achieve the following objectives: improve the accuracy of power battery cell anomaly detection and improve the safety and reliability of battery systems.

[0010] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis, the method comprising the following steps:

[0011] Step S1: Collect the status data of individual battery cells, including voltage, temperature, and ambient temperature and humidity;

[0012] Step S2: Preprocess the data collected in step S1;

[0013] Step S3: Correct the preprocessed data using a preset dynamic environment compensation model;

[0014] Step S4: Extract spatiotemporal correlation features from the corrected data, wherein the time dimension features include various battery cell parameters, and the spatial dimension features include the spatial deviation of various battery cell parameters; simultaneously, based on the corrected data, calculate the adaptive thresholds of various battery cell parameters using a machine learning model.

[0015] Step S5: Perform abnormal pattern matching in a preset abnormal pattern feature library based on the spatiotemporal correlation features;

[0016] Step S6: Combine the spatiotemporal correlation features and adaptive thresholds to perform anomaly detection and graded early warning.

[0017] Preferably, in step S1, a distributed acquisition network is constructed to acquire the status data of individual battery cells. The distributed acquisition network includes high-precision voltage sensors, temperature sensors, and ambient temperature and humidity sensors deployed in the battery system.

[0018] Preferably, in step S2, the preprocessing includes using a filtering algorithm to remove noise, aligning the data collected by different sensors in time, and removing obviously unreasonable outliers.

[0019] Preferably, in step S3, the dynamic environment compensation model is as follows:

[0020] ;

[0021] ;

[0022] in, Indicates the dynamic environmental compensation coefficient; , , , , The fitting coefficients are obtained through experimental fitting; T represents temperature; H represents humidity. This term is used to accurately model the nonlinear characteristics of battery parameters at low temperatures; Used to enhance the compensation effect under high humidity conditions; V represents the state data of a single battery cell before compensation; This indicates the state data of a single battery cell after compensation.

[0023] Preferably, in step S4, the time dimension feature extraction includes setting a sliding window of length L and step size of 1 second for the voltage sequence of the battery cells, and calculating various battery cell parameters in each window, including the mean voltage V_mean, standard deviation V_std, and the mean of the absolute values ​​of the first-order differences V_change_rate.

[0024] Preferably, in step S4, spatial dimension feature extraction includes constructing a distribution matrix of various battery cell parameters for all battery cells, and using the DBSCAN clustering algorithm to calculate the spatial deviation between various battery cell parameters and the cluster center in order to identify spatially anomalous battery cell parameters.

[0025] Preferably, step S5 includes: if there is no matching abnormal pattern, the spatiotemporal characteristics of the current battery cell, environmental parameter labels, and abnormal type labels set by the user after analysis are saved as new abnormal patterns in the feature library; at the same time, the feature library is periodically updated online using an incremental SVM algorithm.

[0026] Preferably, in step S6, a three-level early warning system is adopted, including:

[0027] Level 1 Warning:

[0028] a. Among the various parameters of a battery cell, at least one parameter exceeds the adaptive threshold of that parameter.

[0029] b. Among the various battery cell parameters, there is at least one spatially anomalous cell parameter.

[0030] When either a or b is triggered, the corresponding battery cell will receive a Level 1 warning.

[0031] Level 2 warning: If the number of battery cell parameters that trigger Level 1 warning is greater than a preset value, or if the duration of Level 1 warning for any battery cell parameter is greater than a preset first duration, then Level 2 warning will be issued.

[0032] Level 3 warning: A Level 3 warning is issued if any parameter of a battery cell changes abruptly, or if the duration of a Level 2 warning for any battery cell parameter exceeds the preset second duration.

[0033] Preferably, in the case of a Level 1 warning, the warning information is sent to the local BMS via the CAN bus, the instrument panel warning light is illuminated, and a detailed log is recorded; in the case of a Level 2 warning, the warning information is sent to the vehicle controller via the high-bandwidth bus, and the charging power is limited or the drive power is reduced, while a notification is pushed to the owner's APP via the T-Box; in the case of a Level 3 warning, the warning information is immediately uploaded to the cloud monitoring platform via the 4G / 5G network, the BMS executes the highest level of protection strategy, and the cloud platform automatically notifies the operator and emergency contacts to activate the emergency plan.

[0034] This embodiment also provides a power battery cell anomaly detection system based on spatiotemporal correlation analysis. Using the above-described power battery cell anomaly detection method based on spatiotemporal correlation analysis, the system includes:

[0035] The data acquisition module is used to collect status data of individual battery cells, including voltage, temperature, and ambient temperature and humidity.

[0036] The preprocessing module is used to preprocess the collected data;

[0037] The dynamic environment compensation module is used to correct the preprocessed data using a preset dynamic environment compensation model.

[0038] The spatiotemporal correlation feature and adaptive threshold calculation module is used to extract spatiotemporal correlation features from the corrected data; at the same time, based on the corrected data, adaptive thresholds for various battery cell parameters are calculated using a machine learning model.

[0039] An anomaly pattern matching module is used to perform anomaly pattern matching in a preset anomaly pattern feature library based on the spatiotemporal correlation features.

[0040] The graded early warning module is used to perform anomaly detection and graded early warning by combining the spatiotemporal correlation features and adaptive thresholds.

[0041] The technical effects of this invention are as follows:

[0042] This invention achieves multi-source heterogeneous spatiotemporal data fusion, constructs a high-precision acquisition network including multiple parameters such as voltage and temperature, and performs in-depth mining of spatiotemporal correlation features, providing reliable input for anomaly detection and improving detection accuracy. This invention constructs an environmental adaptive compensation model, which overcomes environmental influences and improves detection accuracy by compensating and correcting the acquired data. This invention constructs an intelligent hierarchical early warning mechanism, which improves anomaly handling efficiency and thus enhances the safety and reliability of the battery system by designing a multi-level early warning system and implementing differentiated early warning strategies based on the severity of anomalies. Attached Figure Description

[0043] Figure 1The flowchart illustrates a method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis, as provided in this embodiment of the invention. Detailed Implementation

[0044] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. This is to help those skilled in the art to have a more complete, accurate, and in-depth understanding of the inventive concept and technical solutions of the present invention, and to facilitate its implementation. It should be noted that the terms "first," "second," etc., used in this application are only for the convenience of describing the technical solutions and to distinguish components; the corresponding component configurations may be the same or different, and are not intended to limit the scope of this application. To make the technical solutions of the present invention clearer, the present invention will be explained and illustrated through the following embodiments.

[0045] This embodiment provides a method for detecting anomalies in power battery cells based on spatiotemporal correlation analysis. It aims to provide a high-precision and robust solution for detecting battery cell anomalies in applications such as electric vehicles and energy storage systems through multi-dimensional data acquisition, dynamic environmental compensation, spatiotemporal correlation feature extraction, adaptive threshold calculation, and intelligent early warning mechanisms, significantly improving the safety and reliability of battery systems. Figure 1 As shown, the method includes the following steps:

[0046] Step S1: Collect the status data of individual battery cells, including voltage, temperature, and ambient temperature and humidity;

[0047] Step S2: Preprocess the data collected in step S1;

[0048] Step S3: Correct the preprocessed data using a preset dynamic environment compensation model;

[0049] Step S4: Extract spatiotemporal correlation features from the corrected data, wherein the time dimension features include various battery cell parameters, and the spatial dimension features include the spatial deviation of various battery cell parameters; simultaneously, based on the corrected data, calculate the adaptive thresholds of various battery cell parameters using a machine learning model.

[0050] Step S5: Perform abnormal pattern matching in a preset abnormal pattern feature library based on the spatiotemporal correlation features;

[0051] Step S6: Perform hierarchical early warning by combining the spatiotemporal correlation features and adaptive thresholds.

[0052] Referring to step S1, this embodiment constructs a distributed acquisition network to collect state data of individual battery cells. The distributed acquisition network includes high-precision voltage sensors, temperature sensors, and ambient temperature and humidity sensors deployed within the battery system. During operation, the distributed acquisition network collects data at a sampling frequency of no less than 10Hz, while ensuring data synchronization accuracy of ±0.5ms. These sensors continuously collect data such as the voltage and temperature of individual battery cells, as well as the ambient temperature and humidity, providing a basis for subsequent analysis.

[0053] The collected data may contain issues such as noise, time asynchrony, and outliers. Therefore, before proceeding with the formal analysis, it is necessary to preprocess the data as described in step S2. Specifically, the preprocessing operations in this embodiment include using filtering algorithms to remove noise, aligning the data collected from different sensors in time to ensure consistency of the data in the time dimension, and removing obviously unreasonable outliers, thereby improving data quality and providing a reliable data foundation for subsequent analysis.

[0054] Most existing methods do not fully consider the coupled effects of multiple environmental parameters such as temperature, humidity, and air pressure, leading to significant fluctuations in detection performance under different climatic conditions. Therefore, referring to step S3, this embodiment introduces a dynamic environmental compensation model to correct the collected battery state data, thereby improving the accuracy of anomaly detection. Specifically, the dynamic environmental compensation model in this embodiment is as follows:

[0055] ;

[0056] ;

[0057] in, Indicates the dynamic environmental compensation coefficient; , , , , The fitting coefficients are obtained through experimental fitting; T represents temperature; H represents humidity. This term is used to accurately model the nonlinear characteristics of battery parameters at low temperatures; Used to enhance the compensation effect under high humidity conditions; V represents the state data of a single battery cell before compensation; This represents the state data of a single battery cell after compensation. The dynamic environmental compensation model obtains the compensation coefficients through fitting a large amount of experimental data. For example, , , , , .

[0058] Traditional solutions either rely solely on time-dimensional analysis (such as simple threshold comparison) or focus only on spatial distribution (such as single-cell voltage uniformity analysis), lacking in-depth mining of spatiotemporal coupling characteristics. Therefore, referring to step S4, this embodiment performs feature extraction through a dual spatiotemporal dimension. The time-dimensional feature extraction involves setting a sliding window of length L (typically 120s) and a step size of 1 second for the voltage sequence of individual battery cells. Within each window, various battery cell parameters are calculated, including the mean voltage V_mean, standard deviation V_std, and the mean of the absolute values ​​of the first-order differences V_change_rate. In specific implementations, other battery cell parameters can also be extracted according to actual needs.

[0059] Spatial dimension feature extraction includes constructing a distribution matrix of various battery cell parameters for all battery cells, and using the existing DBSCAN clustering algorithm to calculate the spatial deviation of various battery cell parameters from the cluster center in order to identify whether there are spatially anomalous battery cell parameters in the current battery cell.

[0060] Meanwhile, existing methods for detecting anomalies in individual battery cells have rigid threshold settings, often employing fixed thresholds or simple linear compensation models, which cannot adapt to the performance degradation patterns and changes in operating conditions throughout the battery's entire lifespan. Therefore, referring to step S4, this embodiment also introduces a machine learning model for adaptive threshold calculation. The machine learning model is pre-trained and optimized using a large amount of experimental data. After training and optimization, the machine learning model continuously adjusts the threshold based on real-time data, achieving adaptive threshold adjustment to adapt to the performance degradation patterns and changes in operating conditions throughout the battery's entire lifespan. For example, using the real-time charge / discharge current, SOC, and SOH of a single battery cell as input to the machine learning model, the corresponding machine learning model will output a dynamic and personalized adaptive threshold. Simultaneously, the model is incrementally updated periodically (e.g., daily) using newly confirmed normal data to ensure that the model can adapt to the performance degradation patterns and changes in operating conditions throughout the battery's entire lifespan.

[0061] Referring to step S5, this embodiment pre-constructs a feature library containing multiple anomaly patterns. When spatiotemporal features extracted from real-time data are input, the feature library identifies matching anomaly patterns accordingly. The identification methods include constructing an anomaly pattern classification model based on existing neural network models and analyzing the distribution characteristics of the data itself using existing unsupervised learning methods to discover anomalies. The specific identification methods adopt existing technologies, and their principles will not be elaborated here.

[0062] If no matching anomaly pattern is found, the spatiotemporal characteristics of the current battery cell, environmental parameter labels, and anomaly type labels set by the user after analysis are saved as new anomaly patterns in the feature library. At the same time, the feature library is periodically updated online using the existing incremental SVM algorithm to achieve online learning and optimization of the feature library and improve the accuracy of anomaly identification.

[0063] Referring to step S5, this embodiment designs a three-level early warning system to achieve differentiated alarms and improve anomaly handling efficiency. The three-level early warning system includes:

[0064] Level 1 Warning: a) Among the various battery cell parameters to which a battery cell belongs, at least one battery cell parameter exceeds the adaptive threshold of that parameter; b) Among the various battery cell parameters to which a battery cell belongs, at least one spatially abnormal cell parameter exists.

[0065] When either a or b is triggered, the corresponding battery cell will receive a Level 1 warning.

[0066] Level 2 warning: If the number of battery cell parameters that trigger Level 1 warning is greater than a preset value, or if the duration of Level 1 warning for any battery cell parameter is greater than a preset first duration, then Level 2 warning will be issued.

[0067] Level 3 warning: A Level 3 warning is issued when any parameter of a battery cell undergoes a sudden change (e.g., a significant increase or decrease in parameter change over a short period of time), or when the duration of a Level 2 warning for any battery cell parameter exceeds the preset second duration.

[0068] Correspondingly, in the case of a Level 1 warning, the warning information is sent to the local BMS via the CAN bus, the instrument panel warning lights are illuminated, and a detailed log is recorded. In the case of a Level 2 warning, the warning information is sent to the vehicle controller via a high-bandwidth bus, and charging power is limited or drive power is reduced. Simultaneously, a notification is pushed to the owner's app via the T-Box. In the case of a Level 3 warning, the warning information is immediately uploaded to the cloud monitoring platform via the 4G / 5G network. The BMS executes the highest level of protection strategy, and the cloud platform automatically notifies the operator and emergency contacts, activating the emergency plan. The hierarchical warning system constructed in this embodiment enables adaptive warnings based on the severity of the anomaly, optimizes system resources, and improves anomaly handling efficiency.

[0069] This embodiment also provides a power battery cell anomaly detection system based on spatiotemporal correlation analysis. Using the above-described power battery cell anomaly detection method based on spatiotemporal correlation analysis, the system includes:

[0070] The data acquisition module is used to collect status data of individual battery cells, including voltage, temperature, and ambient temperature and humidity.

[0071] The preprocessing module is used to preprocess the collected data;

[0072] The dynamic environment compensation module is used to correct the preprocessed data using a preset dynamic environment compensation model.

[0073] The spatiotemporal correlation feature and adaptive threshold calculation module is used to extract spatiotemporal correlation features from the corrected data; at the same time, based on the corrected data, adaptive thresholds for various battery cell parameters are calculated using a machine learning model.

[0074] An abnormal pattern matching module is used to perform abnormal pattern matching in a preset abnormal pattern feature library based on the spatiotemporal correlation features and adaptive threshold.

[0075] The graded early warning module is used to perform anomaly detection and graded early warning by combining the spatiotemporal correlation features and adaptive thresholds.

[0076] In summary, this embodiment achieves rapid identification and early warning of abnormal states of individual power battery cells by deeply integrating multi-source heterogeneous spatiotemporal data and combining an environmental adaptive compensation model and an intelligent hierarchical early warning mechanism, significantly improving the safety and reliability of the battery system.

[0077] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution; or the direct application of the inventive concept and technical solution to other situations without modification, are all within the protection scope of the present invention.

Claims

1. A method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis, characterized in that: The method includes the following steps: Step S1: Collect the status data of individual battery cells, including voltage, temperature, and ambient temperature and humidity; Step S2: Preprocess the data collected in step S1; Step S3: Correct the preprocessed data using a preset dynamic environment compensation model; Step S4: Extract spatiotemporal correlation features from the corrected data, wherein the time dimension features include various battery cell parameters, and the spatial dimension features include the spatial deviation of various battery cell parameters; simultaneously, based on the corrected data, calculate the adaptive thresholds of various battery cell parameters using a machine learning model. Step S5: Perform abnormal pattern matching in a preset abnormal pattern feature library based on the spatiotemporal correlation features; Step S6: Combine the spatiotemporal correlation features and adaptive thresholds to perform anomaly detection and graded early warning.

2. The method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis according to claim 1, characterized in that: In step S1, a distributed acquisition network is constructed to acquire the status data of individual battery cells. The distributed acquisition network includes high-precision voltage sensors, temperature sensors, and ambient temperature and humidity sensors deployed in the battery system.

3. The method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis according to claim 1, characterized in that: In step S2, the preprocessing includes using a filtering algorithm to remove noise, aligning the data collected by different sensors in time, and removing obviously unreasonable outliers.

4. The method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis according to claim 1, characterized in that: In step S3, the dynamic environment compensation model is as follows: ； ； in, Indicates the dynamic environmental compensation coefficient; , , , , The fitting coefficients are obtained through experimental fitting; T represents temperature; H represents humidity. This term is used to accurately model the nonlinear characteristics of battery parameters at low temperatures; Used to enhance the compensation effect under high humidity conditions; V represents the state data of a single battery cell before compensation; This indicates the state data of a single battery cell after compensation.

5. The method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis according to claim 1, characterized in that: In step S4, the time dimension feature extraction includes setting a sliding window of length L and step size of 1 second for the voltage sequence of the battery cells, and calculating various battery cell parameters in each window, including the mean voltage V_mean, standard deviation V_std, and mean of the absolute values ​​of the first-order differences V_change_rate.

6. The method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis according to claim 1, characterized in that: In step S4, spatial dimension feature extraction includes constructing a distribution matrix of various battery cell parameters for all battery cells, and using the DBSCAN clustering algorithm to calculate the spatial deviation between various battery cell parameters and the cluster center in order to identify spatially anomalous battery cell parameters.

7. The method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis according to claim 1, characterized in that: Step S5 includes: if no matching anomaly pattern is found, the spatiotemporal characteristics of the current battery cell, environmental parameter labels, and anomaly type labels set by the user after analysis are saved as new anomaly patterns in the feature library; at the same time, the feature library is periodically updated online using an incremental SVM algorithm.

8. The method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis according to claim 1, characterized in that: In step S6, a three-level early warning system is adopted, including: Level 1 Warning: a. Among the various parameters of a battery cell, at least one parameter exceeds the adaptive threshold of that parameter. b. Among the various battery cell parameters, there is at least one spatially anomalous cell parameter. When either a or b is triggered, the corresponding battery cell will receive a Level 1 warning. Level 2 warning: If the number of battery cell parameters that trigger Level 1 warning is greater than a preset value, or if the duration of Level 1 warning for any battery cell parameter is greater than a preset first duration, then Level 2 warning will be issued. Level 3 warning: A Level 3 warning is issued if any parameter of a battery cell changes abruptly, or if the duration of a Level 2 warning for any battery cell parameter exceeds the preset second duration.

9. The method for detecting anomalies in a single power battery cell based on spatiotemporal correlation analysis according to claim 8, characterized in that: In the event of a Level 1 warning, the warning information is sent to the local BMS via the CAN bus, the instrument panel warning light is illuminated, and a detailed log is recorded. In the event of a Level 2 warning, the warning information is sent to the vehicle controller via a high-bandwidth bus, and the charging power is limited or the drive power is reduced. At the same time, a notification is pushed to the owner's APP via the T-Box. In the event of a Level 3 warning, the warning information is immediately uploaded to the cloud monitoring platform via the 4G / 5G network. The BMS executes the highest level of protection strategy, and the cloud platform automatically notifies the operator and emergency contacts to activate the emergency plan.

10. A power battery cell anomaly detection system based on spatiotemporal correlation analysis, using the power battery cell anomaly detection method based on spatiotemporal correlation analysis according to any one of claims 1-9, characterized in that: The system includes: The data acquisition module is used to collect status data of individual battery cells, including voltage, temperature, and ambient temperature and humidity. The preprocessing module is used to preprocess the collected data; The dynamic environment compensation module is used to correct the preprocessed data using a preset dynamic environment compensation model. The spatiotemporal correlation feature and adaptive threshold calculation module is used to extract spatiotemporal correlation features from the corrected data; at the same time, based on the corrected data, adaptive thresholds for various battery cell parameters are calculated using a machine learning model. An anomaly pattern matching module is used to perform anomaly pattern matching in a preset anomaly pattern feature library based on the spatiotemporal correlation features. The graded early warning module is used to perform anomaly detection and graded early warning by combining the spatiotemporal correlation features and adaptive thresholds.

Images