A lithium battery anomaly detection method and system based on deep learning

Abstract

The present invention provides a lithium battery anomaly detection method and system based on deep learning, which relates to the technical field of data processing. The method comprises: obtaining original data sets of lithium batteries under various working conditions; performing multi-layer preprocessing on the original data sets to obtain basic data sets; performing step-by-step feature extraction on the basic data sets to obtain feature data sets; dividing the feature data sets into state data sets of charging, discharging, and static working according to the working state of the battery; obtaining a decomposed data set according to the state data sets; constructing a multidimensional state model of the lithium battery, inputting the decomposed data set into the multidimensional state model, and obtaining a matching result set; performing comprehensive analysis according to the matching result set to generate abnormal discrimination data; analyzing detection information characterizing the abnormal position and severity according to the abnormal discrimination data to generate abnormal identification information; the present invention has a good abnormality detection effect.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a lithium battery anomaly detection method and system based on deep learning. Background Art

[0002] With the rapid development of electric vehicles and energy storage systems, lithium batteries have been widely used as energy storage devices with high energy density. However, in actual use, lithium batteries may experience abnormal conditions such as performance degradation, overheating or internal short circuits due to various reasons, which may even lead to safety accidents in severe cases. In the prior art, abnormality detection for lithium batteries mainly relies on real-time monitoring of single physical parameters such as voltage, current and temperature, and the health status of the battery is judged by analyzing the changing trends of these parameters. However, this method often lacks accuracy when dealing with anomalies caused by multiple factors, especially when the external environment is complex and the load changes frequently, misjudgment or missed judgment may occur.

[0003] To improve the accuracy of detection, some technologies use multi-parameter fusion algorithms based on machine learning to conduct comprehensive analysis of various data such as voltage, current, temperature, etc. to identify abnormal conditions of lithium batteries. However, these methods are often limited to the extraction of shallow features and do not deeply explore the complex relationships between data. Especially in the process of dealing with battery cycle aging, existing methods may find it difficult to accurately predict sudden changes in battery performance caused by aging. This defect is particularly evident in long-term energy storage power station applications, because in the later stages of battery performance degradation, traditional shallow features are difficult to capture small but critical feature changes, which may lead to missed detection of battery abnormalities. Summary of the invention

[0004] The purpose of the present invention is to provide a lithium battery abnormality detection method and system based on deep learning, aiming to solve the problems mentioned in the background technology.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] In a first aspect, a lithium battery abnormality detection method based on deep learning is provided, the method comprising:

[0007] Acquire raw data sets of lithium batteries under various working conditions, wherein the raw data sets include voltage, temperature, and current information;

[0008] Perform multi-layer preprocessing on the original data set to obtain a basic data set, wherein the multi-layer preprocessing includes filtering of noise data, removing of abnormal points, and time synchronization of different physical parameters;

[0009] Performing step-by-step feature extraction on the basic data set to obtain a feature data set, wherein the step-by-step feature extraction includes basic feature extraction and associated feature extraction of physical parameters;

[0010] According to the working state of the battery, the feature data set is divided into charging, discharging, and static working state data sets;

[0011] According to the state data set, by calculating the synergistic relationship of different parameters, the key feature subsets under each state are extracted to obtain the decomposed data set;

[0012] Construct a multidimensional state model of lithium batteries, input the decomposed data set into the multidimensional state model, match the sub-features of the decomposed data set level by level, and obtain a matching result set;

[0013] Perform comprehensive analysis based on the matching result set to generate abnormality discrimination data, wherein the abnormality discrimination data identifies the specific time period in which the abnormality occurs and the physical parameter information related to the abnormality by analyzing the feature comparison data;

[0014] Based on the abnormality discrimination data, the detection information characterizing the abnormality location and severity is analyzed to generate abnormality identification information.

[0015] Preferably, the original data set is subjected to multi-layer preprocessing to obtain a basic data set, including:

[0016] The collected data of each physical parameter in the original data set are filtered one by one, and the sampling values ​​that do not meet the dynamic threshold are eliminated to generate a preliminary data set; the dynamic threshold is calculated by the following formula:

[0017] ,in, is the dynamic threshold, For in time The standard deviation of For in time The mean of and is the adjustment factor;

[0018] Sampling adjustment is performed on each type of data in the preliminary data set, and the data of different physical parameters are time-aligned by uniform interpolation within the acquisition period to obtain a synchronized data set; time alignment is calculated using the following formula:

[0019] ,in, is the synchronization data after interpolation, For the sampling data points, is the total number of sampled data points;

[0020] Perform threshold detection on each parameter of the synchronized data set, remove abnormal data points that are significantly deviated from the expected value according to the preset change rate standard, and generate a basic data set; the change rate is calculated by the following formula:

[0021] ,in, is the rate of change, and is the data value of the current and previous sampling points.

[0022] Preferably, the basic data set is subjected to step-by-step feature extraction to obtain a feature data set, including:

[0023] According to the basic data set, the mean, variance and cycle peak value change in each sampling period are extracted for the voltage, temperature and current information to generate basic feature data; the mean, variance and cycle peak value change are calculated by the following formula:

[0024] ,in, is the mean, For the data points;

[0025] ,in, is the variance, For the data points, is the mean;

[0026] ,in, is the peak value, is the maximum value in the data point set;

[0027] According to the basic characteristic data, the correlation information of each physical parameter in different states is extracted; the correlation information between voltage, temperature and current is expressed by the following formula:

[0028] ,in, is the correlation information between voltage, temperature and current, is the voltage, is the temperature, is the current, , , , , , is the weight coefficient, is a constant;

[0029] The basic features and associated information are combined into a set to generate a feature data set, which is used to describe the internal dynamic characteristics of the battery under different states.

[0030] Preferably, according to the state data set, by calculating the synergistic relationship of different parameters, a key feature subset in each state is extracted to obtain a decomposed data set, including:

[0031] In the charging state, the changing relationship between voltage and temperature is analyzed, the synergistic cumulative value of voltage and temperature is extracted, and the charging overheating feature subset is screened out; the synergistic cumulative value of voltage and temperature is calculated by the following formula:

[0032] ,in, is the synergistic cumulative value of voltage and temperature, , For time When the voltage and temperature, , is the mean value of voltage and temperature, , The start time and end time are respectively;

[0033] In the discharge state, the synchronization of the change of current and voltage over time is evaluated, the synchronization ratio of current and voltage in the discharge process is extracted, and the abnormal discharge feature subset is generated; the synchronization ratio of current and voltage is calculated by the following formula:

[0034] ,in, is the synchronous ratio of current to voltage, is the current change, is the voltage change;

[0035] In the static state, the fluctuation amplitude of temperature and voltage is recorded, the features that can reflect the self-discharge characteristics are identified, and the self-discharge feature subset is generated; the fluctuation amplitude of temperature and voltage is calculated by the following formula:

[0036] ,in, , is the fluctuation range of temperature and voltage, , are the maximum and minimum values ​​of temperature, , is the maximum and minimum value of voltage;

[0037] The charging overheating feature subset, the abnormal discharge feature subset and the self-discharge feature subset are combined to obtain a decomposed data set.

[0038] Preferably, a multidimensional state model of the lithium battery is constructed, the decomposed data set is input into the multidimensional state model, and the sub-features of the decomposed data set are matched level by level to obtain a matching result set, including:

[0039] Based on the historical original data set, the Gaussian distribution is used to establish the normal characteristic distribution range of each state, and a multidimensional state model is generated with vector Represents the characteristic distribution of lithium batteries in different states, where Indicates voltage, Indicates temperature, represents the current; the multidimensional state model is represented by the probability density function of the multidimensional Gaussian distribution, where the formula of the probability density function is:

[0040] ,in, is the probability density of the multidimensional Gaussian distribution, is the feature vector, is the mean vector of multi-dimensional features, is the covariance matrix of the eigenvectors;

[0041] Input the feature vector of each state in the decomposed data set into the multidimensional state model, and calculate the deviation value and standard deviation between the real-time feature value and the benchmark feature;

[0042] According to the first preset threshold, the deviation value and the standard deviation, the abnormal features are screened and marked to obtain the abnormal features;

[0043] Generate matching result sets in different states based on abnormal characteristics.

[0044] Preferably, a comprehensive analysis is performed based on the matching result set to generate abnormality discrimination data, including:

[0045] Perform time series analysis on the matching result set, and calculate the cumulative deviation of abnormal features based on the change of feature trends. The persistence and severity of the abnormal features are judged based on the cumulative deviation to obtain the cumulative deviation data. The cumulative deviation is calculated using the following formula:

[0046] ,in, is the cumulative deviation, is the time decay weight, For the data points, is the mean, is the attenuation coefficient, , For the current time and The time of the data point, is the total number of time points;

[0047] When the accumulated deviation data exceeds the second preset threshold, the specific time period of the anomaly is recorded, and the physical parameters related to the anomaly are subjected to multi-layer feature decomposition to determine the abnormal change pattern;

[0048] According to the abnormal change pattern and the accumulated deviation data, abnormality discrimination data is generated, wherein the abnormality discrimination data includes the specific time period when the abnormality occurs, the fluctuation range of the abnormal characteristics, and the physical parameter information associated with the abnormal characteristics.

[0049] Preferably, based on the abnormality discrimination data, the detection information characterizing the abnormality location and severity is analyzed to generate abnormality identification information, including:

[0050] The deviation amplitude of the time period and physical parameters in the abnormal discrimination data is quantified, and the deviation amplitude, occurrence frequency and duration of each abnormal feature are graded to generate graded abnormal detection information; the abnormal detection information is graded according to the following formula:

[0051] ,in, is the risk level, is the deviation amplitude, is the frequency of deviation, is the duration, , , , is the weight coefficient;

[0052] Determine the specific location of the abnormality in the battery module or battery pack according to the abnormal feature location in the abnormality detection information, and generate location identification information;

[0053] The severity of the anomaly detection information is evaluated, and anomalies with more drastic feature changes are marked to obtain anomaly identification information.

[0054] In a second aspect, a lithium battery abnormality detection system based on deep learning is provided, the system comprising:

[0055] A data acquisition module, used to acquire original data sets of lithium batteries under various working conditions, wherein the original data sets include voltage, temperature and current information;

[0056] A preprocessing module is used to perform multi-layer preprocessing on the original data set to obtain a basic data set, wherein the multi-layer preprocessing includes filtering of noise data, removing of abnormal points, and time synchronization of different physical parameters;

[0057] A feature extraction module is used to extract features from the basic data set step by step to obtain a feature data set, wherein the step by step feature extraction includes basic feature extraction and associated feature extraction of physical parameters;

[0058] A data partitioning module is used to partition the feature data set into multiple state data sets according to the battery working state, including charging, discharging, and static working state data sets;

[0059] The feature decomposition module is used to extract the key feature subsets in each state based on the state data set by calculating the synergistic relationship of different parameters to obtain the decomposed data set;

[0060] The feature matching module is used to construct a multidimensional state model of the lithium battery, input the decomposed data set into the multidimensional state model, match the sub-features of the decomposed data set level by level, and obtain a matching result set;

[0061] A feature analysis module is used to perform a comprehensive analysis based on the matching result set to generate abnormality discrimination data. The abnormality discrimination data identifies the specific time period when the abnormality occurs and the physical parameter information related to the abnormality by analyzing the feature comparison data;

[0062] The anomaly identification module is used to analyze the detection information representing the location and severity of the anomaly based on the anomaly discrimination data and generate anomaly identification information.

[0063] The above solution of the present invention includes at least the following beneficial effects:

[0064] First, the present invention obtains raw data sets under various working conditions, including voltage, temperature and current information, so that the system can monitor the real-time status of the battery from multiple dimensions. Compared with the prior art, single parameter monitoring can usually only capture one aspect of the battery status, while the present invention is based on real-time acquisition and synchronous processing of multiple parameters, so that the detection method has a more comprehensive data foundation. Through the joint analysis of multiple physical parameters, the present invention can more accurately reflect the working status of lithium batteries and provide more reliable data support for abnormality detection.

[0065] Secondly, in the data preprocessing stage, the present invention significantly improves the accuracy and consistency of the data through multi-layer preprocessing methods such as noise filtering, outlier removal and time synchronization. Data preprocessing in the prior art is mostly limited to simple noise processing, which is difficult to deal with the fluctuation of data and the interference of outliers on the detection results in complex working conditions. The present invention can effectively remove invalid data and ensure the time alignment of different physical parameters through dynamic threshold and change rate removal mechanism, thereby improving the accuracy and availability of data, laying a solid foundation for subsequent feature extraction.

[0066] In terms of feature extraction, the present invention adopts a step-by-step feature extraction method, which not only extracts the basic features of voltage, temperature and current, such as mean, variance and peak, but also mines the correlation information between different physical parameters to form a multidimensional feature data set. Most of the existing technologies are limited to the simple extraction of basic features, while ignoring the synergistic relationship between parameters. The present invention further quantifies the synergistic relationship between voltage, temperature and current through a nonlinear relationship formula, so that the dynamic changes inside the battery can be captured more comprehensively, especially in the early stage of abnormality, the mutual influence between different parameters can be quickly identified. This refined feature extraction method significantly enhances the sensitivity and accuracy of anomaly detection.

[0067] The present invention also constructs a multidimensional state model of the lithium battery, uses a multidimensional Gaussian distribution to model the normal characteristic distribution under charging, discharging and static states, and uses it as a reference state model for real-time data matching. Compared with the simple single-dimensional model in the prior art, the multidimensional state model of the present invention can simultaneously consider the multidimensional characteristic distribution of voltage, temperature and current, providing a more accurate abnormality identification benchmark. By inputting real-time data into the multidimensional state model step by step and calculating the degree of deviation, the present invention can capture subtle abnormal changes in a timely manner and achieve high-precision abnormality detection. This multidimensional modeling method can adaptively perform feature comparison under different battery states, improving the robustness and detection effect of the system.

[0068] In terms of abnormality identification, the present invention significantly enhances the ability to judge the persistence and severity of abnormal characteristics through time series analysis and cumulative deviation calculation methods. In the prior art, abnormalities are usually judged by feature deviations at a single time point, while the present invention enables the system to identify abnormal features that have gradually accumulated over a long period of time through cumulative deviation analysis on a time series, which has obvious advantages in particular in the case of battery aging or long-term self-discharge. At the same time, the abnormality identification data of the present invention includes the abnormal time period, feature fluctuation range and related physical parameter information, providing detailed data support for subsequent fault warning and maintenance. In contrast, the prior art is relatively insufficient in terms of the persistence and accuracy of abnormality identification, making it difficult to identify potential risks in a timely manner.

[0069] In summary, the present invention has formed a systematic, highly accurate and robust lithium battery anomaly detection method through comprehensive data acquisition, sophisticated feature extraction, multi-dimensional state modeling, step-by-step matching analysis and time series anomaly discrimination, which can significantly improve the accuracy and reliability of detection, avoid the potential safety hazards caused by the failure to discover battery anomalies in time, and provide effective protection for the long-term safe operation of lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1It is a flowchart of a lithium battery abnormality detection method based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0072] like Figure 1 As shown, an embodiment of the present invention proposes a lithium battery abnormality detection method based on deep learning, the method comprising:

[0073] Acquire raw data sets of lithium batteries under various working conditions, wherein the raw data sets include voltage, temperature, and current information;

[0074] Perform multi-layer preprocessing on the original data set to obtain a basic data set, wherein the multi-layer preprocessing includes filtering of noise data, removing of abnormal points, and time synchronization of different physical parameters;

[0075] Performing step-by-step feature extraction on the basic data set to obtain a feature data set, wherein the step-by-step feature extraction includes basic feature extraction and associated feature extraction of physical parameters;

[0076] According to the working state of the battery, the feature data set is divided into charging, discharging, and static working state data sets;

[0077] According to the state data set, by calculating the synergistic relationship of different parameters, the key feature subsets under each state are extracted to obtain the decomposed data set;

[0078] Construct a multidimensional state model of lithium batteries, input the decomposed data set into the multidimensional state model, match the sub-features of the decomposed data set level by level, and obtain a matching result set;

[0079] Perform comprehensive analysis based on the matching result set to generate abnormality discrimination data, wherein the abnormality discrimination data identifies the specific time period in which the abnormality occurs and the physical parameter information related to the abnormality by analyzing the feature comparison data;

[0080] Based on the abnormality discrimination data, the detection information characterizing the abnormality location and severity is analyzed to generate abnormality identification information.

[0081] In the embodiment of the present invention, the refined monitoring and abnormal identification of the operating status of the lithium battery are realized through multi-level data processing and feature extraction. First of all, the present invention can comprehensively capture the operating status of the battery and provide a reliable data basis for subsequent detection by acquiring the original data set of the lithium battery under various working conditions, covering key physical parameters such as voltage, temperature and current. Compared with the monitoring method of a single parameter or simple feature in the prior art, the present invention is based on the acquisition of multi-parameter data sets, so that the detection results are more comprehensive.

[0082] In the data preprocessing stage, the present invention adopts a multi-layer preprocessing method, including noise filtering, outlier removal and time synchronization. This processing method can effectively remove random noise and outliers in the original data, and ensure the time alignment between different physical parameters, making the data more accurate and suitable for multi-parameter analysis. The data preprocessing in the prior art is usually limited to simple noise filtering, while the present invention improves the accuracy and consistency of the data through multi-level preprocessing, thereby enhancing the effect of subsequent feature extraction.

[0083] In terms of feature extraction, the present invention fully explores the intrinsic correlation and deep-level characteristics between various parameters through step-by-step feature extraction, including basic features and associated features. In particular, through the construction of a multidimensional feature data set, the different state characteristics of the battery can be more comprehensively captured in the subsequent state division and feature analysis. In contrast, the prior art rarely involves correlation analysis between different parameters, and it is difficult to provide accurate early warning signals when the battery state is abnormal. The present invention not only provides basic characteristic information of the battery, but also captures the interactive influence of multi-dimensional data through associated features, greatly improving the sensitivity and accuracy of detection.

[0084] By dividing the data sets of different states, the present invention further refines the analysis of abnormal characteristics of the battery in the states of charging, discharging and static. In the existing methods, the monitoring of battery state changes is mostly an overall analysis, lacking in-depth analysis of abnormal characteristics under different working states. However, by dividing the states, the present invention can accurately identify abnormal characteristics under different working states, greatly improving the accuracy and response speed of abnormal detection.

[0085] The multidimensional state model constructed by the present invention further improves the accuracy of anomaly detection. Through the multidimensional state model, the present invention can judge whether the battery state is normal or not based on the synergistic relationship of each feature subset. The multidimensional state matching method can not only accurately identify the specific time period when the anomaly occurs, but also can combine the multidimensional distribution characteristics to quantitatively analyze the anomaly, thereby achieving more accurate anomaly location and identification. Overall, the detection method of the present invention has higher accuracy and sensitivity than the prior art, and can effectively improve the safety of lithium batteries under complex working conditions.

[0086] Specifically, the original data set of the lithium battery under various working conditions is obtained, and the original data set includes voltage, temperature and current information. In this step, the system monitors the physical parameters such as voltage, temperature and current of the lithium battery in real time through multiple sensors. These parameters can reflect the performance of the lithium battery in different working conditions, such as charging, discharging and static conditions. By continuously collecting these data, a raw data set can be formed as the basis for subsequent abnormality detection.

[0087] The voltage sensor can monitor the change of battery voltage, the temperature sensor can monitor the surface or internal temperature of the battery, and the current sensor can monitor the charge and discharge current. Connect the sensor to the data acquisition module, set the sampling period, and record the data in real time at the set interval.

[0088] Specifically, according to the working state of the battery, the feature data set is divided into charging, discharging, and static working state data sets. The purpose of this step is to classify the multidimensional feature data sets in the basic data set according to the working state of the battery, so that subsequent analysis can perform more accurate processing and anomaly detection for feature data sets in different states. The data in the charging, discharging, and static states have different characteristics, so feature analysis by state is more helpful in discovering abnormal characteristics.

[0089] The system determines the working state of the battery by monitoring the direction and magnitude of the current in real time. When the current is positive, it is in the charging state; when the current is negative, it is in the discharging state; and when the current is close to zero, it is in the static state. According to the state recognition results, the feature data sets are classified into the charging, discharging and static state data sets respectively, so that the data of each state can be analyzed independently in the future. A state label is added to each feature data record to ensure that the data can be called and analyzed according to the state at any time. This state separation method provides a targeted feature set for subsequent abnormality recognition.

[0090] In a preferred embodiment of the present invention, the original data set is subjected to multi-layer preprocessing to obtain a basic data set, including:

[0091] The collected data of each physical parameter in the original data set are filtered one by one, and the sampling values ​​that do not meet the dynamic threshold are eliminated to generate a preliminary data set; the dynamic threshold is calculated by the following formula:

[0092] ,in, is the dynamic threshold, For in time The standard deviation of For in time The mean of and is the adjustment factor;

[0093] Sampling adjustment is performed on each type of data in the preliminary data set, and the data of different physical parameters are time-aligned by uniform interpolation within the acquisition period to obtain a synchronized data set; time alignment is calculated using the following formula:

[0094] ,in, is the synchronization data after interpolation, For the sampling data points, is the total number of sampled data points;

[0095] Perform threshold detection on each parameter of the synchronized data set, remove abnormal data points that are significantly deviated from the expected value according to the preset change rate standard, and generate a basic data set; the change rate is calculated by the following formula:

[0096] ,in, is the rate of change, and is the data value of the current and previous sampling points.

[0097] In the embodiment of the present invention, an innovative design is made in data preprocessing, and the accuracy and consistency of the data are effectively improved through multi-layer preprocessing of the original data set. First, in response to the noise problem during battery operation, the present invention uses a dynamic threshold noise filtering method to eliminate noise data that exceeds the upper and lower limits through an adaptive dynamic threshold, so that the data set can more truly reflect the actual state of the battery. Compared with the simple fixed threshold in the prior art, the dynamic threshold can better adapt to different working conditions and improve the flexibility and accuracy of noise filtering.

[0098] In addition, the present invention adopts interpolation processing in the time alignment of data. The data of different physical parameters are aligned through the interpolation method to ensure consistent voltage, temperature and current information at the same time point. This processing method not only improves the time synchronization of data, but also facilitates the correlation analysis of subsequent feature extraction. In contrast, most of the prior art is limited to the time acquisition of a single parameter, lacking time alignment and synchronization processing of multiple parameters. The present invention uses a time alignment method to make the data have higher integrity and accuracy during the analysis process.

[0099] In terms of outlier removal, the present invention detects anomalies based on the rate of change, and removes sampling points that exceed the normal range by setting a rate of change threshold, thereby avoiding the interference of outliers on the overall feature extraction. The prior art usually only uses simple noise filtering, without fully considering the impact of outliers on the analysis, while the present invention fully considers the characteristics of battery data in each step of preprocessing, making the preprocessed data more representative, laying a solid foundation for subsequent feature extraction.

[0100] In a preferred embodiment of the present invention, the basic data set is subjected to step-by-step feature extraction to obtain a feature data set, including:

[0101] According to the basic data set, the mean, variance and cycle peak value change in each sampling period are extracted for the voltage, temperature and current information to generate basic feature data; the mean, variance and cycle peak value change are calculated by the following formula:

[0102] ,in, is the mean, For the data points;

[0103] ,in, is the variance, For the data points, is the mean;

[0104] ,in, is the peak value, is the maximum value in the data point set;

[0105] According to the basic characteristic data, the correlation information of each physical parameter in different states is extracted; the correlation information between voltage, temperature and current is expressed by the following formula:

[0106] ,in, is the correlation information between voltage, temperature and current, is the voltage, is the temperature, is the current, , , , , , is the weight coefficient, is a constant;

[0107] The basic features and associated information are combined into a set to generate a feature data set, which is used to describe the internal dynamic characteristics of the battery under different states.

[0108] In an embodiment of the present invention, the basic features and associated features of the battery are fully explored during the feature extraction process. Through a step-by-step feature extraction method, multi-dimensional feature extraction is performed on voltage, temperature and current, which can more comprehensively describe the dynamic characteristics inside the battery. By extracting the mean, variance and peak value of voltage, temperature and current, the present invention can obtain the basic working state characteristics of the battery. These basic features can not only directly reflect the operating conditions of the battery, but also provide an important basis for subsequent association analysis. Compared with the single feature extraction in the prior art, the present invention performs multi-dimensional and multi-angle analysis on the basic features to make the detection results more reliable.

[0109] In terms of correlation feature extraction, the present invention further explores the synergistic relationship between voltage, temperature and current, quantifies the dynamic relationship between voltage, temperature and current through nonlinear relationship formulas, and realizes quantitative analysis of the synergistic relationship between the three. Compared with the existing technology that mostly only considers the changes of a single feature, the present invention makes anomaly detection more accurate through multi-parameter collaborative analysis, especially when multiple parameter changes affect each other, the correlation features of the present invention can effectively capture the subtle changes of the battery in different states, providing a more effective basis for anomaly detection.

[0110] In a preferred embodiment of the present invention, according to the state data set, by calculating the synergistic relationship of different parameters, a key feature subset under each state is extracted to obtain a decomposed data set, including:

[0111] In the charging state, the changing relationship between voltage and temperature is analyzed, the synergistic cumulative value of voltage and temperature is extracted, and the charging overheating feature subset is screened out; the synergistic cumulative value of voltage and temperature is calculated by the following formula:

[0112] ,in, is the synergistic cumulative value of voltage and temperature, , For time When the voltage and temperature, , is the mean value of voltage and temperature, , The start time and end time are respectively;

[0113] In the discharge state, the synchronization of the change of current and voltage over time is evaluated, the synchronization ratio of current and voltage in the discharge process is extracted, and the abnormal discharge feature subset is generated; the synchronization ratio of current and voltage is calculated by the following formula:

[0114] ,in, is the synchronous ratio of current to voltage, is the current change, is the voltage change;

[0115] In the static state, the fluctuation amplitude of temperature and voltage is recorded, the features that can reflect the self-discharge characteristics are identified, and the self-discharge feature subset is generated; the fluctuation amplitude of temperature and voltage is calculated by the following formula:

[0116] ,in, , is the fluctuation range of temperature and voltage, , are the maximum and minimum values ​​of temperature, , is the maximum and minimum value of voltage;

[0117] The charging overheating feature subset, the abnormal discharge feature subset and the self-discharge feature subset are combined to obtain a decomposed data set.

[0118] In the embodiment of the present invention, the key features in each state are deeply analyzed by constructing charging, discharging and static state data sets for the abnormal features of the battery in different working states. In the charging state, the present invention screens out the charging overheat feature subset through the coordinated cumulative value analysis of voltage and temperature, which can quickly identify temperature anomalies in the charging process and avoid overcharging risks. In the discharging state, the synchronous changes of current and voltage are evaluated, and the abnormal discharge feature subset is extracted to effectively capture the abnormal discharge features in the discharge process. Compared with the single feature monitoring in the prior art, the present invention makes abnormal detection more targeted through feature subset analysis in different states.

[0119] In a static state, the present invention identifies the self-discharge characteristics through the fluctuation amplitude of temperature and voltage, and can accurately detect the self-discharge behavior of the battery in a static state. The existing technology mostly focuses on abnormal detection in a working state, while the present invention conducts a detailed analysis of the self-discharge in a static state, so that the abnormal detection of the battery is not limited to the charging and discharging process. Through the combined analysis of different state characteristics, the present invention effectively improves the comprehensiveness and accuracy of abnormal detection.

[0120] In a preferred embodiment of the present invention, a multidimensional state model of a lithium battery is constructed, a decomposed data set is input into the multidimensional state model, and sub-features of the decomposed data set are matched step by step to obtain a matching result set, including:

[0121] Based on the historical original data set, the Gaussian distribution is used to establish the normal characteristic distribution range of each state, and a multidimensional state model is generated with vector Represents the characteristic distribution of lithium batteries in different states, where Indicates voltage, Indicates temperature, represents the current; the multidimensional state model is represented by the probability density function of the multidimensional Gaussian distribution, where the formula of the probability density function is:

[0122] ,in, is the probability density of the multidimensional Gaussian distribution, is the feature vector, is the mean vector of multi-dimensional features, is the covariance matrix of the eigenvectors;

[0123] Input the feature vector of each state in the decomposed data set into the multidimensional state model, and calculate the deviation value and standard deviation between the real-time feature value and the benchmark feature;

[0124] According to the first preset threshold, the deviation value and the standard deviation, the abnormal features are screened and marked to obtain the abnormal features;

[0125] Generate matching result sets in different states based on abnormal characteristics.

[0126] Specifically, in the charging state, the voltage rising trend and temperature change are matched layer by layer to identify the abnormal situation of overcharging;

[0127] In the discharge state, the current drop rate and the voltage drop rate are compared step by step to determine the abnormal discharge rate;

[0128] In the static state, the fluctuation amplitudes of temperature and voltage are matched step by step to identify the abnormal conditions of self-discharge and generate matching result sets under different states.

[0129] In an embodiment of the present invention, the accuracy of abnormality detection is further improved by constructing a multidimensional state model of the lithium battery. The multidimensional state model is based on the combination of feature subsets in the decomposed data set to form a multidimensional feature space that reflects the characteristics of different states, and uses a multidimensional Gaussian distribution to model the normal feature distribution of each state, so that the model can effectively characterize the various working states of the lithium battery. Most of the existing technologies use linear models or single distribution models, while the present invention uses a multidimensional Gaussian distribution to make the model more flexible and accurate, and can better adapt to the complex distribution of features under different working states.

[0130] When the characteristics of the actual working state exceed the normal distribution range of the multi-dimensional Gaussian distribution, the multi-dimensional state model of the present invention can identify abnormal conditions of the battery in different states through a step-by-step matching process. In the charging state, the present invention can accurately identify overcharging by comparing the changing trends of voltage and temperature layer by layer; in the discharging state, the change rates of current and voltage are compared step by step to determine the discharge abnormality; in the static state, the comparison of temperature and voltage fluctuation amplitudes can effectively identify the self-discharge phenomenon. Overall, the multi-dimensional state model of the present invention provides accurate abnormality detection capabilities, making the detection system more robust.

[0131] Specifically, the covariance matrix is ​​an important mathematical tool used to describe the relationship and change trend between multiple variables. Covariance describes the linear relationship between two variables, indicating how the two variables change together.

[0132] In a preferred embodiment of the present invention, a comprehensive analysis is performed based on the matching result set to generate abnormality discrimination data, including:

[0133] Perform time series analysis on the matching result set, and calculate the cumulative deviation of abnormal features based on the change of feature trends. The persistence and severity of the abnormal features are judged based on the cumulative deviation to obtain the cumulative deviation data. The cumulative deviation is calculated using the following formula:

[0134] ,in, is the cumulative deviation, is the time decay weight, For the data points, is the mean, is the attenuation coefficient, , For the current time and The time of the data point, is the total number of time points;

[0135] When the accumulated deviation data exceeds the second preset threshold, the specific time period of the anomaly is recorded, and the physical parameters related to the anomaly are subjected to multi-layer feature decomposition to determine the abnormal change pattern;

[0136] According to the abnormal change pattern and the accumulated deviation data, abnormality discrimination data is generated, wherein the abnormality discrimination data includes the specific time period when the abnormality occurs, the fluctuation range of the abnormal characteristics, and the physical parameter information associated with the abnormal characteristics.

[0137] In the embodiment of the present invention, in the generation of abnormal discrimination data, the recognition accuracy of abnormal features is effectively improved through time series analysis and cumulative deviation calculation method. Time series analysis enables the present invention to track the development trend of abnormal features over time, and through the cumulative deviation calculation method, the persistence and severity of the abnormality are judged according to the cumulative changes of the deviation. Compared with the single moment judgment in the prior art, the present invention makes the detection more accurate and reliable through the cumulative deviation analysis of the time series, and can identify long-term abnormalities with potential risks.

[0138] When the cumulative deviation exceeds the preset threshold, the present invention records the specific time period of the anomaly, and identifies the abnormal physical parameters through multi-layer feature decomposition, and further determines the pattern of abnormal changes. This multi-layer analysis method can not only detect abnormal features, but also identify the development trend of the anomaly, so that the detection system can detect potential risks earlier.

[0139] Specifically, Used to calculate the cumulative deviation By giving higher weights to the most recent time points, the decay coefficient allows the system to pay more attention to abnormal features that occur in a short period of time, thereby more sensitively detecting current or recent abnormalities. This dynamic weight mechanism is suitable for identifying sudden changes in battery status, such as sudden abnormalities such as overheating and overcurrent.

[0140] when When it is small, the time weight decays slowly, and abnormal data from earlier times still has a greater impact on the cumulative deviation. This setting is suitable for scenarios where you need to detect and focus on persistent anomalies, such as long-term deviations of temperature or current from normal values ​​during battery aging. Slow decay allows the system to pay more attention to abnormal data over a period of time, which helps detect long-term trend anomalies.

[0141] when When it is larger, the time weight decays faster, recent abnormal data has a greater impact, and the weight of older abnormal data decreases rapidly. This setting is suitable for identifying sudden anomalies, such as sudden temperature rise or current exceeding expected values ​​during charging or discharging. Rapid decay makes the system more sensitive to abnormal changes in a short period of time, facilitating real-time monitoring of emergencies.

[0142] The system can dynamically adjust according to factors such as the working status of the lithium battery, ambient temperature and historical abnormal data For example, under high temperature or high-speed discharge conditions, set a larger , making the system more sensitive to current anomalies. In low temperature or static conditions, set a smaller , making the system pay attention to long-term trend anomalies.

[0143] Assume that during a discharge process, the battery current gradually deviates from the normal value, and the deviation gradually accumulates. , the system can more accurately determine whether the deviation is a short-term sudden abnormality or a persistent long-term abnormality. When the system pays more attention to the accumulation of deviations in a longer time frame, When the system is adjusted, it focuses more on the changes in the last few minutes. This adjustment method makes the system's cumulative deviation calculation more flexible and adaptable, suitable for different types of anomaly detection needs.

[0144] Specifically, when the accumulated deviation data exceeds the second preset threshold, the specific time period of the anomaly is recorded, and the physical parameters related to the anomaly are decomposed into multiple layers to determine the abnormal change pattern. The goal of this step is to identify possible abnormal situations by the accumulated deviation data exceeding the threshold, record the time of occurrence of the anomaly, and further analyze the abnormal parameters to identify the abnormal change pattern. This can provide in-depth analysis of anomaly detection.

[0145] A second preset threshold is set based on historical data to determine the risk level of the cumulative deviation. When the cumulative deviation exceeds the threshold, the system automatically records the start and end time of the anomaly to form a complete record of the anomaly time period.

[0146] Conduct in-depth analysis of the physical parameters of the abnormality, decompose the characteristic changes during the abnormal time period, and determine whether it belongs to a specific mode such as temperature abnormality, overcharge abnormality or discharge abnormality.

[0147] Specifically, based on the abnormal change pattern and the accumulated deviation data, abnormal discrimination data is generated, which includes the specific time period when the abnormality occurs, the fluctuation range of the abnormal characteristics, and the physical parameter information associated with the abnormal characteristics. This step generates abnormal discrimination data, including the abnormal time period, characteristic fluctuation range, and associated parameter information, for subsequent abnormal classification and risk assessment. The output data of this step can be used as the basis for further diagnosis of the cause of the abnormality.

[0148] During the abnormal time period, the fluctuation amplitude of the deviation data is analyzed to quantify the variation range of the abnormal characteristics. Physical parameters such as voltage, temperature and current related to the abnormal characteristics are recorded to form abnormal discrimination data. Discrimination data containing abnormal characteristics, time period, fluctuation range and related parameters are generated to provide data support for abnormal type classification.

[0149] In a preferred embodiment of the present invention, based on the abnormality discrimination data, the detection information characterizing the abnormality location and severity is analyzed to generate abnormality identification information, including:

[0150] The deviation amplitude of the time period and physical parameters in the abnormal discrimination data is quantified, and the deviation amplitude, occurrence frequency and duration of each abnormal feature are graded to generate graded abnormal detection information; the abnormal detection information is graded according to the following formula:

[0151] ,in, is the risk level, is the deviation amplitude, is the frequency of deviation, is the duration, , , , is the weight coefficient;

[0152] Determine the specific location of the abnormality in the battery module or battery pack according to the abnormal feature location in the abnormality detection information, and generate location identification information;

[0153] The severity of the anomaly detection information is evaluated, and anomalies with more drastic feature changes are marked to obtain anomaly identification information.

[0154] In an embodiment of the present invention, when generating abnormal identification information, multiple factors such as the abnormal time period, deviation amplitude, occurrence frequency and duration are integrated to classify the abnormal detection information. By grading the risk level of each abnormal feature, the present invention can respond to high-risk abnormal situations in priority according to the size of the risk level, thereby improving the response speed of the detection system. In practical applications, for example, when the abnormal detection detects a high-frequency and long-term deviation, the present invention can identify it as a high-risk abnormality and issue an alarm to ensure the safe operation of the battery.

[0155] In addition, the present invention determines the specific location of the anomaly through location identification information, which facilitates fault location and maintenance. The anomaly identification information not only records the characteristic parameters and location of the anomaly, but also includes the risk level, providing a basis for subsequent maintenance and risk prediction. Compared with the single anomaly recording method of the prior art, the grading and positioning method of the present invention significantly improves the practicality and accuracy of anomaly detection.

[0156] Specifically, according to the abnormal feature position in the abnormality detection information, the specific position of the abnormality in the battery module or battery pack is determined, and the position identification information is generated. The purpose of this step is to map the abnormal situation to a specific battery module or battery pack by locating the physical position of the abnormal feature. This positioning information can provide guidance for subsequent fault repair and battery management.

[0157] Associate the abnormal feature information with the physical location of the battery module. Label the abnormal feature with a specific battery module or battery pack number and generate location identification information so that the abnormal location can be quickly located when a problem occurs. Store the location identification information in the abnormality detection system for subsequent query and record analysis.

[0158] Specifically, the severity of the anomaly detection information is comprehensively evaluated, and the anomaly with more drastic feature changes is marked to obtain anomaly identification information. This step comprehensively evaluates the severity of the anomaly detection information, determines the risk level of the anomaly, and marks the situation where the feature changes drastically. This process ensures that the system can prioritize high-risk anomaly information and improve security capabilities.

[0159] Quantify the deviation amplitude, frequency and duration in the anomaly detection information, and generate risk levels based on the previously set thresholds. High-risk anomalies are marked according to the risk level to ensure priority processing. Output anomaly identification information, including anomaly location, risk level and anomaly type, for anomaly alarm and risk notification.

[0160] An embodiment of the present invention further provides a lithium battery abnormality detection system based on deep learning, the system comprising:

[0161] A data acquisition module, used to acquire original data sets of lithium batteries under various working conditions, wherein the original data sets include voltage, temperature and current information;

[0162] A preprocessing module is used to perform multi-layer preprocessing on the original data set to obtain a basic data set, wherein the multi-layer preprocessing includes filtering of noise data, removing of abnormal points, and time synchronization of different physical parameters;

[0163] A feature extraction module is used to extract features from the basic data set step by step to obtain a feature data set, wherein the step by step feature extraction includes basic feature extraction and associated feature extraction of physical parameters;

[0164] A data partitioning module is used to partition the feature data set into multiple state data sets according to the battery working state, including charging, discharging, and static working state data sets;

[0165] The feature decomposition module is used to extract the key feature subsets in each state based on the state data set by calculating the synergistic relationship of different parameters to obtain the decomposed data set;

[0166] The feature matching module is used to construct a multidimensional state model of the lithium battery, input the decomposed data set into the multidimensional state model, match the sub-features of the decomposed data set level by level, and obtain a matching result set;

[0167] A feature analysis module is used to perform a comprehensive analysis based on the matching result set to generate abnormality discrimination data. The abnormality discrimination data identifies the specific time period when the abnormality occurs and the physical parameter information related to the abnormality by analyzing the feature comparison data;

[0168] The anomaly identification module is used to analyze the detection information representing the location and severity of the anomaly based on the anomaly discrimination data and generate anomaly identification information.

[0169] It should be noted that the system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0170] The embodiment of the present invention further provides a computing device, including: a processor, a memory storing a computer program, and when the computer program is executed by the processor, the method described above is executed. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0171] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the above method. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0172] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A lithium battery abnormality detection method based on deep learning, characterized in that: The method comprises: Acquire raw data sets of lithium batteries under various working conditions, wherein the raw data sets include voltage, temperature, and current information; Perform multi-layer preprocessing on the original data set to obtain a basic data set, wherein the multi-layer preprocessing includes filtering of noise data, removing of abnormal points, and time synchronization of different physical parameters; Performing step-by-step feature extraction on the basic data set to obtain a feature data set, wherein the step-by-step feature extraction includes basic feature extraction and associated feature extraction of physical parameters; According to the working state of the battery, the feature data set is divided into charging, discharging, and static working state data sets; According to the state data set, by calculating the synergistic relationship of different parameters, the key feature subsets under each state are extracted to obtain the decomposed data set; Construct a multidimensional state model of lithium batteries, input the decomposed data set into the multidimensional state model, match the sub-features of the decomposed data set level by level, and obtain a matching result set; Perform comprehensive analysis based on the matching result set to generate abnormality discrimination data, wherein the abnormality discrimination data identifies the specific time period in which the abnormality occurs and the physical parameter information related to the abnormality by analyzing the feature comparison data; Analyze the detection information representing the abnormal location and severity based on the abnormality discrimination data to generate abnormality identification information; According to the state data set, by calculating the synergistic relationship of different parameters, the key feature subsets in each state are extracted to obtain the decomposed data set, including: In the charging state, the changing relationship between voltage and temperature is analyzed, the synergistic cumulative value of voltage and temperature is extracted, and the charging overheating feature subset is screened out; the synergistic cumulative value of voltage and temperature is calculated by the following formula: ,in, is the synergistic cumulative value of voltage and temperature, , For time When the voltage and temperature, , is the mean value of voltage and temperature, , The start time and end time are respectively; In the discharge state, the synchronization of the change of current and voltage over time is evaluated, the synchronization ratio of current and voltage in the discharge process is extracted, and the abnormal discharge feature subset is generated; the synchronization ratio of current and voltage is calculated by the following formula: ,in, is the synchronous ratio of current to voltage, is the current change, is the voltage change; In the static state, the fluctuation amplitude of temperature and voltage is recorded, the features that can reflect the self-discharge characteristics are identified, and the self-discharge feature subset is generated; the fluctuation amplitude of temperature and voltage is calculated by the following formula: ,in, , is the fluctuation range of temperature and voltage, , are the maximum and minimum values ​​of temperature, , is the maximum and minimum value of voltage; The charging overheating feature subset, the abnormal discharge feature subset and the self-discharge feature subset are combined to obtain a decomposed data set; Construct a multidimensional state model of lithium batteries, input the decomposed data set into the multidimensional state model, match the sub-features of the decomposed data set step by step, and obtain a matching result set, including: Based on the historical original data set, the Gaussian distribution is used to establish the normal characteristic distribution range of each state, and a multidimensional state model is generated with vector Represents the characteristic distribution of lithium batteries in different states, where Indicates voltage, Indicates temperature, represents the current; the multidimensional state model is represented by the probability density function of the multidimensional Gaussian distribution, where the formula of the probability density function is: ,in, is the probability density of the multidimensional Gaussian distribution, is the feature vector, is the mean vector of multi-dimensional features, is the covariance matrix of the eigenvectors; Input the feature vector of each state in the decomposed data set into the multidimensional state model, and calculate the deviation value and standard deviation between the real-time feature value and the benchmark feature; According to the first preset threshold, the deviation value and the standard deviation, the abnormal features are screened and marked to obtain the abnormal features; Generate matching result sets in different states based on abnormal features; Comprehensive analysis is performed based on the matching result set to generate abnormal identification data, including: Perform time series analysis on the matching result set, and calculate the cumulative deviation of abnormal features based on the change of feature trends. The persistence and severity of the abnormal features are judged based on the cumulative deviation to obtain the cumulative deviation data. The cumulative deviation is calculated using the following formula: ,in, is the cumulative deviation, is the time decay weight, For the data points, is the mean, is the attenuation coefficient, , For the current time and The time of the data point, is the total number of time points; When the accumulated deviation data exceeds the second preset threshold, the specific time period of the anomaly is recorded, and the physical parameters related to the anomaly are subjected to multi-layer feature decomposition to determine the abnormal change pattern; Generate abnormality discrimination data according to the abnormal change pattern and the accumulated deviation data, wherein the abnormality discrimination data includes the specific time period when the abnormality occurs, the fluctuation range of the abnormal characteristics, and the physical parameter information associated with the abnormal characteristics; Based on the abnormality discrimination data, the detection information characterizing the abnormality location and severity is analyzed to generate abnormality identification information, including: The deviation amplitude of the time period and physical parameters in the abnormal discrimination data is quantified, and the deviation amplitude, occurrence frequency and duration of each abnormal feature are graded to generate graded abnormal detection information; the abnormal detection information is graded according to the following formula: ,in, is the risk level, is the deviation amplitude, is the frequency of deviation, is the duration, , , , is the weight coefficient; Determine the specific location of the abnormality in the battery module or battery pack according to the abnormal feature location in the abnormality detection information, and generate location identification information; The severity of the anomaly detection information is evaluated, and anomalies with more drastic feature changes are marked to obtain anomaly identification information.

2. The method for detecting abnormality of a lithium battery based on deep learning according to claim 1, characterized in that: The original data set is preprocessed in multiple layers to obtain the basic data set, including: The collected data of each physical parameter in the original data set are filtered one by one, and the sampling values ​​that do not meet the dynamic threshold are eliminated to generate a preliminary data set; the dynamic threshold is calculated by the following formula: ,in, is the dynamic threshold, For in time The standard deviation of For in time The mean of and is the adjustment factor; Each type of data in the preliminary data set is sampled and adjusted separately, and the data of different physical parameters are time-aligned by uniform interpolation within the acquisition period to obtain a synchronized data set; A threshold test is performed on each parameter of the synchronized data set, and abnormal data points that deviate significantly from the expected value are eliminated according to the preset change rate standard to generate a basic data set.

3. The method for detecting abnormality of a lithium battery based on deep learning according to claim 2, characterized in that: Perform feature extraction on the basic data set step by step to obtain a feature data set, including: Based on the basic data set, the mean, variance and cycle peak value change of the voltage, temperature and current information in each sampling period are extracted to generate basic feature data; According to the basic characteristic data, the correlation information of each physical parameter in different states is extracted; the correlation information between voltage, temperature and current is expressed by the following formula: ,in, is the correlation information between voltage, temperature and current, is the voltage, is the temperature, is the current, , , , , , is the weight coefficient, is a constant; The basic features and associated information are combined into a set to generate a feature data set, which is used to describe the internal dynamic characteristics of the battery under different states.

4. A lithium battery anomaly detection system based on deep learning, characterized in that: Applied to the method according to any one of claims 1 to 3, the system comprises: A data acquisition module, used to acquire original data sets of lithium batteries under various working conditions, wherein the original data sets include voltage, temperature and current information; A preprocessing module is used to perform multi-layer preprocessing on the original data set to obtain a basic data set, wherein the multi-layer preprocessing includes filtering of noise data, removing of abnormal points, and time synchronization of different physical parameters; A feature extraction module is used to extract features from the basic data set step by step to obtain a feature data set, wherein the step by step feature extraction includes basic feature extraction and associated feature extraction of physical parameters; A data partitioning module is used to partition the feature data set into multiple state data sets according to the battery working state, including charging, discharging, and static working state data sets; The feature decomposition module is used to extract the key feature subsets in each state based on the state data set by calculating the synergistic relationship of different parameters to obtain the decomposed data set; The feature matching module is used to construct a multidimensional state model of the lithium battery, input the decomposed data set into the multidimensional state model, match the sub-features of the decomposed data set level by level, and obtain a matching result set; A feature analysis module is used to perform a comprehensive analysis based on the matching result set to generate abnormality discrimination data. The abnormality discrimination data identifies the specific time period when the abnormality occurs and the physical parameter information related to the abnormality by analyzing the feature comparison data; The anomaly identification module is used to analyze the detection information representing the location and severity of the anomaly based on the anomaly discrimination data and generate anomaly identification information.

5. A computing device, characterized in that: include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, which implements the method according to any one of claims 1 to 3 when executed by a processor.

Images