Method and system for detecting current anomaly caused by thermal runaway of lithium battery

Abstract

The embodiment of the invention provides a method and system for detecting current anomaly caused by thermal runaway of a lithium battery, and relates to the technical field of lithium batteries, and the method comprises the steps: obtaining the current data of the lithium battery; performing time domain processing and frequency domain processing on the current data to obtain a time domain feature set and a frequency domain feature set; and performing thermal runaway current anomaly detection on the lithium battery according to the time domain feature set and the frequency domain feature set to obtain an anomaly detection result. According to the method, time domain analysis and frequency domain analysis are carried out on the current data of the lithium battery, the time-frequency characteristics in the current data can be extracted more comprehensively and richly, and then the current anomaly detection of the lithium battery is realized by using the time-frequency characteristics, so that the accuracy of the current anomaly detection of the lithium battery can be effectively improved, and the detection efficiency of the lithium battery is improved. Preventive measures can be taken before thermal runaway of the lithium battery occurs, so that the overall safety and stability of a lithium battery system are protected.

Description

Technical Field

[0001] The present application relates to the field of lithium battery technology, and in particular to a method and system for detecting current anomaly caused by thermal runaway of a lithium battery. Background Art

[0002] Lithium batteries have been widely used in industry, grid energy storage and daily life due to their high energy density and long cycle life. In related technologies, the current of lithium batteries is usually monitored by sensors such as temperature sensors and gas analysis sensors to ensure the safety of lithium batteries. However, during the monitoring process, related technologies are easily interfered by noise such as switching frequency noise, resulting in distortion of lithium battery current signals, and the accuracy of current anomaly detection needs to be improved. Summary of the invention

[0003] The purpose of this application is to solve one of the technical problems existing in the prior art to at least a certain extent.

[0004] To this end, the purpose of the present application is to provide a method and system for detecting current anomaly caused by thermal runaway of a lithium battery, so as to realize current anomaly detection caused by thermal runaway of a lithium battery.

[0005] In order to achieve the above technical objectives, the technical solutions adopted by the embodiments of the present application include: On the one hand, an embodiment of the present application provides a method for detecting current anomaly caused by thermal runaway of a lithium battery, comprising the following steps: Get the current data of lithium battery; Performing time domain processing and frequency domain processing on the current data to obtain a time domain feature set and a frequency domain feature set; According to the time domain feature set and the frequency domain feature set, thermal runaway current anomaly detection is performed on the lithium battery to obtain an anomaly detection result.

[0006] Further, the current data is processed in the time domain and in the frequency domain to obtain a time domain feature set and a frequency domain feature set, including: Sampling and segmenting the current data of the lithium battery to obtain an initial current data set; the initial current data set includes a plurality of initial data subsets, each of which includes a plurality of data points; Extracting time domain features from each of the initial data subsets to obtain the time domain feature set; Frequency domain feature extraction is performed on each of the initial data subsets to obtain the frequency domain feature set.

[0007] Furthermore, the time domain feature set includes a first time domain feature set and a second time domain feature set; the step of extracting time domain features from each of the initial data subsets to obtain the time domain feature set includes: Obtaining a first time-domain data set according to each of the initial data subsets; Performing time domain feature extraction on the first time domain data set to obtain a first time domain integral feature set, a first time domain kurtosis feature set, and a first time domain entropy feature set as the first time domain feature set; Performing empirical filtering on each of the initial data subsets to obtain a second time domain data set; Time domain features are extracted from the second time domain data set to obtain a second time domain integral feature set, a second time domain kurtosis feature set, and a second time domain entropy feature set as the second time domain feature set.

[0008] Further, the frequency domain feature set includes a first frequency domain feature set and a second frequency domain feature set; the frequency domain feature extraction is performed on each of the initial data subsets to obtain the frequency domain feature set, including: Processing each of the initial data subsets to obtain frequency components corresponding to each of the initial data subsets; According to each of the frequency components, a first frequency domain integral feature set, a first frequency domain kurtosis feature set and a first frequency domain entropy feature set are obtained as the first frequency domain feature set; According to each of the frequency components and the preset frequency range, combined with the switching noise removal technology, a second frequency domain integral feature set, a second frequency domain kurtosis feature set and a second frequency domain entropy feature set are obtained as the second frequency domain feature set.

[0009] Further, the thermal runaway current anomaly detection is performed on the lithium battery according to the time domain feature set and the frequency domain feature set to obtain an anomaly detection result, including: Obtain a first feature set according to the first time domain feature set in the time domain feature set and the first frequency domain feature set in the frequency domain feature set; Obtain a second feature set according to the second time domain feature set in the time domain feature set and the second frequency domain feature set in the frequency domain feature set; A machine learning method is used to perform thermal runaway current anomaly detection on the lithium battery according to the first feature set and the second feature set to obtain the anomaly detection result.

[0010] Further, the first feature set includes a first integral feature set, a first kurtosis feature set and a first entropy feature set, and the first feature set is obtained according to the first time domain feature set in the time domain feature set and the first frequency domain feature set in the frequency domain feature set, including: Obtaining the first integral feature set according to the first time domain integral feature set in the first time domain feature set and the first frequency domain integral feature set in the first frequency domain feature set; Obtaining the first kurtosis feature set according to the first time domain kurtosis feature set in the first time domain feature set and the first frequency domain kurtosis feature set in the first frequency domain feature set; The first entropy feature set is obtained according to the first time domain entropy feature set in the first time domain feature set and the first frequency domain entropy feature set in the first frequency domain feature set.

[0011] Further, the second feature set includes a second integral feature set, a second kurtosis feature set and a second entropy feature set, and the second feature set is obtained according to the second time domain feature set in the time domain feature set and the second frequency domain feature set in the frequency domain feature set, including: Obtain the second integral feature set according to the second time domain integral feature set in the second time domain feature set and the second frequency domain integral feature set in the second frequency domain feature set; Obtain the second kurtosis feature set according to the second time domain kurtosis feature set in the second time domain feature set and the second frequency domain kurtosis feature set in the second frequency domain feature set; The second entropy feature set is obtained according to the second time domain entropy feature set in the second time domain feature set and the second frequency domain entropy feature set in the second frequency domain feature set.

[0012] Furthermore, the machine learning method includes a random forest algorithm.

[0013] Furthermore, the hyperparameters of the random forest algorithm are optimized by an artificial bee colony algorithm.

[0014] On the other hand, an embodiment of the present application provides a current anomaly detection system caused by thermal runaway of a lithium battery, comprising: A data acquisition module, used to acquire current data of the lithium battery; A feature extraction module, used for performing time domain processing and frequency domain processing on the current data to obtain a time domain feature set and a frequency domain feature set; The abnormality detection module is used to perform thermal runaway current abnormality detection on the lithium battery according to the time domain feature set and the frequency domain feature set to obtain an abnormality detection result.

[0015] The beneficial effects of the present application are as follows: the embodiments of the present application provide a method and system for detecting current anomalies caused by thermal runaway of a lithium battery. First, the current data of the lithium battery is obtained; then, the current data is processed in the time domain and the frequency domain to obtain a time domain feature set and a frequency domain feature set; finally, according to the time domain feature set and the frequency domain feature set, the lithium battery is subjected to thermal runaway current anomaly detection to obtain an anomaly detection result. The present application can more comprehensively and richly extract the time-frequency features in the current data by performing time domain analysis and frequency domain analysis on the current data of the lithium battery, and then use these time-frequency features to realize current anomaly detection of the lithium battery, which can effectively improve the accuracy of current anomaly detection of the lithium battery, and help to take preventive measures before thermal runaway of the lithium battery occurs, thereby protecting the overall safety and stability of the lithium battery system.

[0016] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a method for detecting abnormal current caused by thermal runaway of a lithium battery provided in the present application; Figure 2 It is a flow chart of determining a time domain feature set and a frequency domain feature set provided by the present application; Figure 3 is a schematic diagram of determining a first feature set provided by the present application; Figure 4 is a schematic diagram of determining a second feature set provided by the present application; Figure 5 is a flowchart of the random forest algorithm provided by this application; Figure 6 is a flow chart of the artificial bee colony algorithm provided by this application; Figure 7 It is a flowchart of optimizing the hyperparameters of the random forest algorithm by the artificial bee colony algorithm provided by the present application; Figure 8 The detection accuracy graph of the random forest algorithm optimized by the artificial bee colony algorithm when the current value provided in this application is 5A; Fig. 9 The detection accuracy graph of the random forest algorithm optimized by the artificial bee colony algorithm when the current value provided in this application is 8A; Fig.10 The detection accuracy graph of the support vector machine algorithm when the current value provided in this application is 5A; Fig.11The detection accuracy graph of the support vector machine algorithm when the current value provided in this application is 8A; Fig.12 The detection accuracy graph of the K nearest neighbor algorithm when the current value provided in this application is 5A; Fig.13 The detection accuracy graph of the K nearest neighbor algorithm when the current value provided in this application is 8A; Fig.14 The detection accuracy graph of the Naive Bayes algorithm when the current value provided in this application is 5A; Fig.15 The detection accuracy graph of the Naive Bayes algorithm when the current value provided in this application is 8A; Fig.16 The detection accuracy graph of the decision tree algorithm provided in this application when the current value is 5A; Fig.17 The detection accuracy diagram of the decision tree algorithm when the current value provided in this application is 8A. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] The present application is further described below in conjunction with the accompanying drawings and specific embodiments. The described embodiments should not be regarded as limiting the present application, and all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0020] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0022] In response to the problems and defects existing in the related technologies, the embodiments of the present application provide a method and system for detecting current anomalies caused by thermal runaway of a lithium battery. By performing time domain analysis and frequency domain analysis on the current data of the lithium battery, the time-frequency features in the current data can be extracted more comprehensively and richly, and then these time-frequency features are used to realize current anomaly detection of the lithium battery. This can effectively improve the accuracy of current anomaly detection of the lithium battery, and help to take preventive measures before thermal runaway of the lithium battery occurs, thereby protecting the overall safety and stability of the lithium battery system.

[0023] The following will describe in detail the implementation steps of a method for detecting current anomaly caused by thermal runaway of a lithium battery provided in an embodiment of the present application with reference to the accompanying drawings.

[0024] The current anomaly detection method caused by thermal runaway of lithium batteries proposed in this application can be applied to terminals, servers, or software running in terminals or servers. The terminal can be a tablet computer, a laptop computer, a desktop computer, etc., but is not limited to this. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks, and big data and artificial intelligence platforms.

[0025] Reference Figure 1 , Figure 1 This is a flow chart of a method for detecting current anomaly caused by thermal runaway of a lithium battery provided in the present application. The method for detecting current anomaly caused by thermal runaway of a lithium battery provided in an embodiment of the present application may include but is not limited to the following steps.

[0026] S101, obtaining current data of the lithium battery.

[0027] It should be noted that the current data of a lithium battery refers to the data signal formed by the change of current over time during the charging and discharging process of the battery. It can reflect the working status of the battery, such as the charging or discharging status and rate.

[0028] In this step, obtaining the current data of the lithium battery is helpful in determining whether the lithium battery is in normal working condition in subsequent analysis and preventing the lithium battery from thermal runaway.

[0029] S102, performing time domain processing and frequency domain processing on the current data to obtain a time domain feature set and a frequency domain feature set.

[0030] It should be noted that the time domain feature refers to the characteristics of the current data in the time dimension, which can reflect the trend of the current changing over time. In addition, the frequency domain feature refers to the characteristics of the frequency component of the current data, which can reflect the trend of the current frequency change.

[0031] In this step, time domain feature information and frequency domain feature information are extracted from the initially collected current data to form a time domain feature set and a frequency domain feature set, which helps to reveal potential periodic behaviors or other hidden patterns. These two feature sets together provide comprehensive data support for the next step of anomaly detection.

[0032] S103, performing thermal runaway current anomaly detection on the lithium battery according to the time domain feature set and the frequency domain feature set to obtain an anomaly detection result.

[0033] In this step, the current characteristics of the lithium battery are analyzed in depth using the time domain feature set and the frequency domain feature set to identify whether there is current anomaly caused by thermal runaway. If an anomaly is found, it indicates that the lithium battery may have safety hazards and further measures need to be taken to avoid accidents.

[0034] In some embodiments of the present application, reference Figure 2 , Figure 2 This is a flowchart for determining a time domain feature set and a frequency domain feature set provided by the present application. In the above step S102, the current data is processed in the time domain and the frequency domain to obtain the time domain feature set and the frequency domain feature set. The implementation process may include but is not limited to the following steps.

[0035] S201, sampling and segmenting the current data of the lithium battery to obtain an initial current data set; the initial current data set includes multiple initial data subsets, and each initial data subset includes multiple data points.

[0036] In this step, by sampling and segmenting the lithium battery current, the real-time current information of the battery during use can be obtained, and the real-time current information can be converted into a more structured initial current data set. The initial current data set is used for further feature extraction and efficient analysis, which can capture subtle changes in the battery current, ensure the integrity of the data, and facilitate subsequent feature extraction and anomaly detection.

[0037] Optionally, the sampling frequency may be set according to actual conditions, and the embodiments of the present application do not specifically limit this.

[0038] For example, the sampling frequency is 250 kHz, but is not limited thereto.

[0039] Optionally, the intervals between the divided data segments may be set according to actual conditions, and the embodiments of the present application do not impose any specific limitation on this.

[0040] For example, the interval between the divided data segments is 0.8 milliseconds, but the present invention is not limited thereto.

[0041] Optionally, the number of data points included in the data segment can be set according to actual conditions, and the embodiments of the present application do not specifically limit this.

[0042] For example, a data segment contains 200 data points, but is not limited thereto.

[0043] S202, extracting time domain features from each initial data subset to obtain a time domain feature set.

[0044] It should be noted that the time domain feature refers to the time characteristic of the current signal in each initial data subset, which can reflect the trend of the current signal changing over time.

[0045] In this step, the time domain characteristics of the lithium battery current are extracted to obtain a time domain feature set, in preparation for the subsequent current anomaly detection caused by thermal runaway of the lithium battery.

[0046] S203, extracting frequency domain features from each initial data subset to obtain a frequency domain feature set.

[0047] It should be noted that the frequency domain feature refers to the frequency characteristics of the current signal in each initial data subset, and the frequency domain feature can reveal the existence and relative strength of different frequency components in the signal.

[0048] In this step, the frequency domain characteristics of the lithium battery current are extracted to obtain a frequency domain feature set, in preparation for the subsequent current anomaly detection work caused by thermal runaway of the lithium battery.

[0049] In some embodiments of the present application, in the above step S202, the time domain feature extraction is performed on each initial data subset to obtain the time domain feature set, which may include but is not limited to the following steps.

[0050] S301, obtaining a first time domain data set according to each initial data subset.

[0051] In this step, by extracting the data of the current signal changing with time in each initial data subset as the first time domain data set, the original time series information of the unprocessed current signal is retained, ensuring that all subsequent processing and analysis are based on the most realistic battery behavior records, which is crucial for diagnosing potential problems. In addition, the first time domain data set provides unfiltered or processed benchmark data, which helps to compare and analyze the processed data results, thereby verifying the effectiveness of different preprocessing methods.

[0052] S302, performing time domain feature extraction on the first time domain data set to obtain a first time domain integral feature set, a first time domain kurtosis feature set, and a first time domain entropy feature set as a first time domain feature set.

[0053] It should be noted that the integral feature is often used to measure the overall strength or total amount of the current signal over a period of time, which helps to understand the average change of the battery over the entire sampling period. During the charging or discharging process of the lithium battery, if the integral feature of the current signal changes unusually, it may mean that the lithium battery has a current anomaly. The kurtosis feature is often used to describe the sharpness of the data distribution shape. For the current signal, a higher kurtosis value indicates that there are more extreme values ​​in the signal distribution, that is, there may be large instantaneous current fluctuations. Therefore, the kurtosis feature can help identify sudden changes in the current signal. The entropy feature can reflect the complexity and disorder of the current signal. A high entropy value usually means that the current signal is more complex and changeable, while a low entropy value means that the current signal is relatively simple and has strong regularity. In theory, the current signal of the lithium battery under normal working conditions should show a certain regularity. When the lithium battery has an abnormality such as thermal runaway, its current signal may become complex and unpredictable, and the entropy value of the current signal will increase. Therefore, the use of entropy features can help distinguish the difference between normal and abnormal states and improve the accuracy of lithium battery current anomaly detection.

[0054] In this step, the first time domain feature set is obtained by calculating the integral feature, kurtosis feature and entropy feature of the first time domain data set. The first time domain feature set helps to capture the characteristics of the current signal changing over time and provides valuable information for subsequent analysis.

[0055] Among them, the integral characteristic satisfies the following formula (1): (1); In formula (1), represents the integral characteristic, is the first Data elements at positions The number of is the number of sampling elements in the sampling period of each initial data subset; The kurtosis characteristic satisfies the following formula (2): (2); In formula (2), represents the kurtosis characteristic, is the mean of all data elements in each initial data subset; The entropy characteristic satisfies the following formula (3): (3); In formula (3), Represents entropy characteristics.

[0056] S303, performing empirical filtering processing on each initial data subset to obtain a second time domain data set.

[0057] It should be noted that the empirical filtering processing technology refers to the technology of filtering current data using the three sigma criterion, with the purpose of removing noise and outliers.

[0058] In this step, first, the mean value and standard deviation of each initial data subset are obtained based on the initial current data set. Then, based on the mean value and standard deviation of all data elements in each initial data subset, combined with the three sigma criterion, the effective range of each initial data subset is determined. Finally, data points within the effective range are extracted from each initial data subset to obtain the second time domain data set.

[0059] The upper limit of the effective range of the initial data subset satisfies the following formula (4): (4); In formula (4), Represents a constant based on The upper limit of the valid range of the initial data subset, is a constant that controls the effective range of the initial data subset, is the initial data subset, is the standard deviation of all data elements in the initial data subset, is the average value of all data elements in the initial data subset; The lower limit of the effective range of the initial data subset satisfies the following formula (5): (5); In formula (5), Based on constant The lower bound of the valid range of the initial data subset.

[0060] S304, performing time domain feature extraction on the second time domain data set to obtain a second time domain integral feature set, a second time domain kurtosis feature set, and a second time domain entropy feature set as a second time domain feature set.

[0061] It should be noted that the time domain feature set includes a first time domain feature set and a second time domain feature set.

[0062] In this step, the implementation process of extracting the second time domain feature set from the second time domain data set is the same as the implementation process of the aforementioned step S302, that is, the time domain feature extraction can be performed on the second time domain data set according to formula (1), formula (2) and formula (3), and the second time domain integral feature set, the second time domain kurtosis feature set and the second time domain entropy feature set are obtained as the second time domain feature set. In this way, the characteristics of the lithium battery current signal can be comprehensively characterized from multiple dimensions, and the normal current signal can be effectively distinguished from the abnormal current signal caused by thermal runaway, thereby improving the ability to monitor the safety status of the lithium battery.

[0063] In some embodiments of the present application, in the above step S203, the frequency domain feature extraction is performed on each initial data subset to obtain the frequency domain feature set, which may include but is not limited to the following steps.

[0064] S401, processing each initial data subset to obtain a frequency component corresponding to each initial data subset.

[0065] In this step, according to the preset frequency resolution and the number of frequency components, the current signal in each initial data subset is converted to the frequency domain to obtain components at different frequencies, namely, frequency components. These frequency components reveal the various frequencies contained in the current signal and their corresponding amplitudes, which are very useful for analyzing and diagnosing abnormal conditions in the current signal. When analyzing the lithium battery current, identifying and processing these frequency components helps to improve the accuracy of current anomaly detection caused by thermal runaway.

[0066] The preset frequency resolution satisfies the following formula (6): (6); In formula (6), represents the frequency resolution, Indicates the sampling rate; The number of preset frequency components satisfies the following formula (7): (7); In formula (7), Indicates the number of preset frequency components.

[0067] Optionally, the manner of processing each initial data subset to obtain frequency components may be set according to actual conditions, and the embodiment of the present application does not specifically limit this.

[0068] For example, the fast Fourier analysis method is used to process each initial data subset to obtain the frequency component, but it is not limited thereto.

[0069] Optionally, the preset frequency range can be set according to actual conditions, and the embodiments of the present application do not specifically limit this.

[0070] For example, the preset frequency range is 3 kHz to 30 kHz, but is not limited thereto.

[0071] S402: According to each frequency component, a first frequency domain integral feature set, a first frequency domain kurtosis feature set and a first frequency domain entropy feature set are obtained as a first frequency domain feature set.

[0072] In this step, the extraction process of the first frequency domain feature set is the same as the implementation process of the aforementioned step S302, that is, the frequency features of each frequency component can be extracted according to formula (1), formula (2) and formula (3), and the first frequency domain integral feature set, the first frequency domain kurtosis feature set and the first frequency domain entropy feature set are obtained as the first frequency domain feature set. The first frequency domain feature set retains the frequency information of the unprocessed current signal in each frequency component, which can reveal the hidden frequency characteristics in the signal and help identify the specific frequency components in the signal. This is because the frequency domain anomaly of the current signal is often manifested as a sudden change or enhancement of a specific frequency component.

[0073] S403, according to each frequency component and a preset frequency range, combined with a switching noise removal technology, a second frequency domain integral feature set, a second frequency domain kurtosis feature set and a second frequency domain entropy feature set are obtained as a second frequency domain feature set.

[0074] It should be noted that the frequency domain feature set includes a first frequency domain feature set and a second frequency domain feature set. In addition, the switching noise removal processing technology is a technology that reduces switching frequency noise, enhances basic feature extraction and precise filtering, and improves signal clarity by identifying and filtering out interfering switching noise, retaining information related to current anomalies.

[0075] In this step, by performing switching noise removal processing on each frequency component, it is ensured that each frequency component does not exceed a threshold of three times the average frequency, and the average frequency refers to the arithmetic mean of all frequency components contained in a preset frequency range. Then, the implementation process of obtaining the second frequency domain feature set based on the frequency components after the switching noise removal processing is the same as the implementation process of the aforementioned step S302, that is, the second frequency domain integral feature set, the second frequency domain kurtosis feature set and the second frequency domain entropy feature set can be obtained as the second frequency domain feature set through formulas (1) to (3).

[0076] In some embodiments of the present application, reference Figure 3 and Figure 4 , Figure 3 is a schematic diagram of determining the first feature set provided by the present application, Figure 4 This is a schematic diagram of determining the second feature set provided in the present application. In the above step S103, thermal runaway current anomaly detection is performed on the lithium battery according to the time domain feature set and the frequency domain feature set. The implementation process of obtaining the anomaly detection result may include but is not limited to the following steps.

[0077] S501, obtaining a first feature set according to a first time domain feature set in a time domain feature set and a first frequency domain feature set in a frequency domain feature set.

[0078] In this step, by combining the time domain features and frequency domain features that have not been specifically filtered as the first feature set, the advantages of time domain and frequency domain information are fully utilized to provide a more comprehensive feature representation, thereby improving the accuracy and robustness of current anomaly detection.

[0079] S502: Obtain a second feature set according to a second time domain feature set in the time domain feature set and a second frequency domain feature set in the frequency domain feature set.

[0080] In this step, by combining the time domain features filtered by the three-sigma criterion and the frequency domain features after the switching noise is removed as the second feature set, it can better cope with non-stationary signals and improve diagnostic accuracy for situations with significant dynamic changes, such as during the opening and closing of the electrode gap.

[0081] S503, performing thermal runaway current anomaly detection on the lithium battery using a machine learning method according to the first feature set and the second feature set to obtain an anomaly detection result.

[0082] In this step, by applying appropriate machine learning methods to the first feature set and the second feature set, and through an effective model training and evaluation process, effective detection of current anomalies caused by thermal runaway of lithium batteries can be achieved. This method not only improves the detection accuracy, but also enhances the robustness and reliability of the system.

[0083] In some embodiments of the present application, in the above step S501, the implementation process of obtaining the first feature set according to the first time domain feature set in the time domain feature set and the first frequency domain feature set in the frequency domain feature set may include but is not limited to the following steps.

[0084] S601, obtaining a first integral feature set according to a first time domain integral feature set in a first time domain feature set and a first frequency domain integral feature set in a first frequency domain feature set.

[0085] In this step, by integrating the integral characteristics of the first time domain feature set and the first frequency domain feature set, the overall strength or total amount characteristics of the current signal can be more comprehensively reflected, which helps to capture the comprehensive information of the current signal in the time dimension and frequency component, and provide richer and more accurate data support for current signal anomaly detection.

[0086] S602, obtaining a first kurtosis feature set according to a first time domain kurtosis feature set in the first time domain feature set and a first frequency domain kurtosis feature set in the first frequency domain feature set; In this step, by combining the kurtosis characteristics of the first time domain feature set and the first frequency domain feature set, it is possible to better identify possible instantaneous changes or sudden fluctuations in the current signal, which is particularly important for discovering current anomalies caused by thermal runaway.

[0087] S603: Obtain a first entropy feature set according to a first time domain entropy feature set in the first time domain feature set and a first frequency domain entropy feature set in the first frequency domain feature set.

[0088] It should be noted that the first feature set includes a first integral feature set, a first kurtosis feature set and a first entropy feature set.

[0089] In this step, combining the entropy features of the first time domain feature set and the first frequency domain feature set can help analyze the complexity and disorder of the current signal, thereby helping to distinguish the regular current signal of the lithium battery in a normal state from the irregular current signal caused by thermal runaway of the lithium battery.

[0090] In some embodiments of the present application, in the above step S502, the implementation process of obtaining the second feature set according to the second time domain feature set in the time domain feature set and the second frequency domain feature set in the frequency domain feature set may include but is not limited to the following steps.

[0091] S701, obtaining a second integral feature set according to a second time domain integral feature set in a second time domain feature set and a second frequency domain integral feature set in a second frequency domain feature set.

[0092] In this step, by integrating the integral features of the second time domain feature set and the second frequency domain feature set, which have both been subjected to noise removal processing, the overall strength or characteristics of the current signal can be more accurately reflected, which helps to capture the comprehensive information of the current signal in the time dimension and frequency component.

[0093] S702 , obtaining a second kurtosis feature set according to a second time-domain kurtosis feature set in the second time-domain feature set and a second frequency-domain kurtosis feature set in the second frequency-domain feature set.

[0094] In this step, by integrating the kurtosis characteristics of the second time domain feature set and the second frequency domain feature set, both of which have been processed by noise removal, it is possible to better identify possible instantaneous changes or sudden fluctuations in the current signal, especially those caused by thermal runaway, which helps to improve the accuracy of detecting current anomalies.

[0095] S703: Obtain a second entropy feature set according to the second time domain entropy feature set in the second time domain feature set and the second frequency domain entropy feature set in the second frequency domain feature set.

[0096] It should be noted that the second feature set includes a second integral feature set, a second kurtosis feature set and a second entropy feature set.

[0097] In this step, by integrating the entropy features of the second time domain feature set and the second frequency domain feature set, which have both been processed by noise removal, it can help analyze the complexity and disorder of the current signal, help distinguish the regular current pattern under normal operating conditions from the complex and unpredictable current behavior caused by thermal runaway, and thus improve the sensitivity to abnormal current conditions.

[0098] In some embodiments of the present application, in the above step S503, the machine learning method includes a random forest algorithm.

[0099] The random forest algorithm is an ensemble learning method based on decision trees, which determines the final classification by voting on the results of multiple decision trees. In lithium battery current anomaly detection, each decision tree in the random forest algorithm is trained on a different subset of the data, and a different subset of features is considered at each split. This mechanism effectively reduces the risk of overfitting and reduces the impact of noisy data points. In addition, random forests are able to capture the nonlinear relationship between input features and outputs, which is particularly important for identifying complex failure modes such as battery thermal runaway. Therefore, random forests excel in the field of lithium battery current anomaly detection due to their high accuracy and powerful feature processing capabilities. Combining time domain and frequency domain features, the random forest algorithm can more comprehensively analyze the working status of lithium batteries, effectively identify potential thermal runaway risks, and thus enhance the safety of lithium batteries.

[0100] Optionally, refer to Figure 5 , Figure 5 is a flow chart of the random forest algorithm provided in this application. The implementation process of the random forest algorithm may include but is not limited to the following steps.

[0101] First, the first feature set and the second feature set are divided to obtain the training set and test set , the training set , test set and the number of preset categories , the size of the tree , depth, hyperparameters , attribute screening rule and termination rule are used as inputs of the random forest algorithm.

[0102] Then, the bootstrap method is used to select the training set for each tree with replacement; for each tree, at each node, randomly select from all features The features are compared and the best feature is selected to divide the data set accordingly; each decision tree is generated recursively without pruning.

[0103] Among them, when each tree node is split, a subset of all features is randomly selected without replacement as the feature subset. The size of this feature subset is Usually much smaller than the number of all features ; Then, from the number of all features The size of the feature subset is randomly selected in the range ,calculate The information gain and other indicators under the split are used to select the best split feature and continue to divide it further as above. At this time, the candidate domain is .

[0104] Optionally, the method of selecting the best classification node from the feature subset can be set according to actual conditions, and the embodiment of the present application does not make any specific limitation on this.

[0105] For example, there are three methods to select the best classification node from the feature subset, namely based on information gain, information gain rate or Gini coefficient. The corresponding decision tree types are ID3, C4.5 and CART. The feature with the best corresponding value is selected as the best split feature. If the feature value is discrete, the next split can be carried out. If the feature value is a continuous variable, the best split point needs to be selected again.

[0106] Specifically, first of all, according to the definition of entropy in information theory, the greater the information entropy, the higher the purity of the data. The information entropy satisfies the following formula (8): (8); In formula (8), Representation sample The information entropy of Indicates category The proportion of samples in the total number of samples, Represents the number of preset categories; Then, in the feature After the action, the sample Divided into Subset ,sample The information entropy satisfies the following formula (9): (9); In formula (9), Representation characteristics Sample after action The information entropy of For subset The number of samples, For sample The number of samples; feature Function sample The information gain after θ satisfies the following formula (10): (10); In formula (10), Features Sample after action Information gain of According to the characteristics Sample after action The information gain of Sample after action The information gain rate, feature Sample after action The information gain rate satisfies the following formula (11): (11); In formula (11), Features Sample after action The split entropy, characteristic Sample after action The splitting entropy satisfies the following formula (12): (12); Furthermore, the sample The Gini coefficient satisfies formula (13): (13); In formula (13), For sample The Gini coefficient of feature Sample after action The Gini coefficient satisfies the following formula (14): (14); In formula (14), Features Sample after action The Gini coefficient, for The Gini coefficient of Furthermore, the calculation is done on the test set In the unknown sample Classification The probability of an unknown sample Classification The probability satisfies the following formula (15): (15); In formula (15), For unknown samples Classification The probability of For the A decision tree considers unknown samples Classification probability.

[0107] Afterwards, the majority voting method is used to determine the unknown samples Category, unknown sample The category satisfies the following formula (16): (16); In formula (16), For unknown samples Category.

[0108] Finally, according to the classification error Tune the random forest algorithm and return the anomaly detection results.

[0109] In some embodiments of the present application, the hyperparameters of the random forest algorithm are optimized by an artificial bee colony algorithm.

[0110] The artificial bee colony algorithm helps find the best hyperparameter combination through global search capabilities, thereby significantly improving the prediction accuracy of the random forest model. For lithium battery current anomaly detection, this means that potential safety hazards can be identified more accurately and timely measures can be taken to avoid accidents. The optimized random forest model not only performs well on training data, but also has stronger generalization ability on unseen data, reducing the risk of overfitting. This is very important for dealing with the complex and changeable operating environment of lithium batteries, ensuring that the model can adapt to abnormal situations under various working conditions. The artificial bee colony algorithm is used to realize automatic hyperparameter tuning and form an end-to-end solution, which greatly reduces the workload of manual parameter adjustment and speeds up the modeling process. This allows researchers and engineers to devote more energy to data analysis and other key links, improving overall work efficiency. Therefore, the application of the artificial bee colony algorithm to optimize the hyperparameters of the random forest algorithm not only improves the performance of the lithium battery current anomaly detection system, but also promotes the intelligent and efficient development of the entire process, which plays an important role in ensuring the safety of lithium battery use and promoting the healthy development of the new energy industry.

[0111] Optionally, refer to Figure 6 , Figure 6 : is a flow chart of the artificial bee colony algorithm provided in this application. The implementation process of the artificial bee colony algorithm may include but is not limited to the following steps.

[0112] First, in the initialization phase, the parameters of the artificial bee colony algorithm are initialized. Specifically, the number of honey sources of the artificial bee colony algorithm is initialized. 、Maximum number of stagnation times of nectar source and the maximum number of iterations . Secondly, initialize the honey source, that is, randomly generate SN initial honey sources within the solution space search range. Furthermore, set the current number of iterations And initialize the tag vector of each honey source , randomly generate initial solutions in the search space. The number of initial solutions is the number of nectar sources. The initial solutions represent the locations of nectar sources and employed bees.

[0113] The initial solution satisfies the following formula (17): (17); In formula (17), Indicates honey source In the The value of the dimension; depending on , Represents the number of nectar sources; depending on , Indicates the number of dimensions of each nectar source in the search space; is A random number in the range, and They represent the upper and lower limits of the nectar source search range respectively.

[0114] Then, in the hired bee stage, each hired bee will search for a new nectar source near the current nectar source, and then calculate the fitness value of the new nectar source. If the fitness value of the new nectar source is higher than that of the original nectar source, it means that the quality of the new nectar source is better than the current nectar source. Therefore, the new nectar source can be directly used to replace the current nectar source, and Otherwise, update the flag vector The new nectar source satisfies the following formula (18): (18); In formula (18), Represents hired bees In the The new position of the dimension, Represents hired bees In the The position of the dimension, Represents hired bees In the The position of the dimension, depending on ,and , is A random number in the range.

[0115] Afterwards, in the observation bee stage, the observation bees select hired bees through the roulette strategy, that is, the higher the quality of the nectar source where the hired bees are, the greater the possibility that they will be selected and followed by the observation bees. Subsequently, the observation bees look for new nectar sources around the nectar sources where the hired bees live, and generate new nectar sources for the observation bees according to formula (18). After calculating the fitness value of the new nectar source, the new nectar source is compared with the current nectar source based on the greedy criterion. If the fitness value of the new nectar source is higher than that of the current nectar source, the new nectar source replaces the current nectar source, and the new nectar source is replaced. Otherwise, update the flag vector The probability of an employed bee being selected in this process satisfies the following formula (19): (19); In formula (19), Represents hired bees The probability of being selected, depending on , Represents hired bees The fitness value of the nectar source.

[0116] Finally, during the scout bee stage, at the nectar source Logo vector Reached the maximum number of stalls , this nectar source will be abandoned, and the employed bee at the nectar source position will become a scout bee, randomly looking for a new nectar source from the search space, and finally directly replace the abandoned nectar source with the new nectar source. In this process, the new nectar source of the scout bee is generated according to formula (17). The best nectar source found by all bees is recorded, that is, the global optimal solution. The current number of iterations is ,if , the algorithm ends; otherwise, it returns to the hired bee stage to continue execution.

[0117] Optionally, refer to Figure 7 , Figure 7 This is a flowchart of optimizing the hyperparameters of a random forest algorithm through an artificial bee colony algorithm provided in the present application. The implementation process of optimizing the hyperparameters of a random forest algorithm through an artificial bee colony algorithm may include the following steps.

[0118] S801, establishing a mapping relationship between the values ​​of feasible hyperparameters in landslide risk assessment using the optimized random forest algorithm and the parameters of the random forest algorithm.

[0119] S802, randomly generating initial values ​​of pointers in landslide risk assessment using an optimized random forest algorithm, and calculating an initial Kappa coefficient for evaluating landslide risk assessment results.

[0120] S803, using an artificial bee colony algorithm to search for optimal hyperparameters of the random forest algorithm, and using the Kappa coefficient to calculate the fitness value of the feasible hyperparameters.

[0121] S804, using an improved adaptive step size method to update the values ​​of feasible hyperparameters and optimize the ability of the landslide risk assessment solution based on the random forest algorithm.

[0122] S805, iteratively complete the optimal solution of the hyperparameters of the random forest algorithm, and use the random forest algorithm under the optimal hyperparameters to carry out landslide risk assessment.

[0123] In addition, an embodiment of the present application provides a current anomaly detection system caused by thermal runaway of a lithium battery. The current anomaly detection system caused by thermal runaway of a lithium battery provided by the embodiment of the present application may include but is not limited to the following modules.

[0124] The data acquisition module 901 is used to acquire the current data of the lithium battery.

[0125] The feature extraction module 902 is used to perform time domain processing and frequency domain processing on the current data to obtain a time domain feature set and a frequency domain feature set.

[0126] The anomaly detection module 903 is used to perform thermal runaway current anomaly detection on the lithium battery according to the time domain feature set and the frequency domain feature set to obtain an anomaly detection result.

[0127] The following examples will be used to verify the effectiveness and advancement of the current anomaly detection method and system caused by thermal runaway of a lithium battery provided in the present application.

[0128] First, four classic machine learning algorithms are selected as comparison algorithms for the random forest algorithm (ABC-RF) optimized based on the artificial bee colony algorithm provided in this application. The comparison algorithms include the following four.

[0129] Support Vector Machine (SVM) is a supervised learning algorithm for binary and multi-class classification. It attempts to find an optimal hyperplane to separate sample points of different categories so that the interval between the two categories is maximized. The support vector machine algorithm has a good effect in solving small sample, nonlinear and high-dimensional pattern recognition problems.

[0130] K-Nearest Neighbors (KNN) algorithm calculates the distance between each sample and finds the K samples closest to each sample in the feature space. If most of the k nearest samples of a sample in the feature space belong to a certain category, then the sample also belongs to this category.

[0131] Naive Bayes (NB) is a classification algorithm based on Bayes' theorem. It assumes that features are independent of each other and uses Bayes' theorem to solve the problem of maximizing the posterior probability. Although this assumption is not always true in the real world, Naive Bayes still performs well in many practical problems.

[0132] Decision Tree (DT) is a tree structure in which each internal node represents an attribute test, each branch represents a test output, and each leaf node represents a category label. Decision trees are constructed by recursively partitioning a data set in order to make the partitioned subsets as pure as possible.

[0133] In order to show the performance differences of the above five algorithms at different current values ​​and switching frequencies, this application selects 5A and 8A as current values, and 5kHz, 10kHz, 15kHz and 20kHz as switching frequencies respectively.

[0134] Therefore, based on the above selected current value and switching frequency, this application selects the following six feature combinations as the data set for the above algorithm, referring to Figure 8-Figure 17 , where “RAW TIME&RAW FREQUENCY INTEGRAL” represents the combination of the first time domain integral feature set and the first frequency domain integral feature set, i.e., the first integral feature set; “RAW TIME&RAW FREQUENCYKURTOSIS” represents the combination of the first time domain kurtosis feature set and the first frequency domain kurtosis feature set, i.e., the first kurtosis feature set; “RAW TIME&RAW FREQUENCY SHANNON ENTROPY” represents the combination of the first time domain entropy feature set and the first frequency domain entropy feature set, i.e., the first entropy feature set; “EMPIRICAL TIME&SNR FREQUENCY INTEGRAL” represents the combination of the second time domain integral feature set and the second frequency domain integral feature set, i.e., the second integral feature set; “EMPIRICAL TIME&SNRFREQUENCY KURTOSIS” represents the combination of the second time domain kurtosis feature set and the second frequency domain kurtosis feature set, i.e., the second kurtosis feature set; “EMPIRICAL TIME&SNR FREQUENCY SHANNON ENTROPY” represents the combination of the first time domain entropy feature set and the first frequency domain entropy feature set, i.e., the first entropy feature set; "ENTROPY" represents the combination of the second time domain entropy feature set and the second frequency domain entropy feature set, that is, the second entropy feature set.

[0135] The random forest algorithm optimized based on the artificial bee colony algorithm provided in the embodiment of the present application and the above four comparison algorithms are respectively trained and verified on the above selected feature sets, and the detection accuracy of the random forest algorithm optimized based on the artificial bee colony algorithm provided in the embodiment of the present application is compared with the above four comparison algorithms to verify the effectiveness and advancement of the random forest algorithm optimized based on the artificial bee colony algorithm provided in the embodiment of the present application.

[0136] It can be seen that the random forest algorithm optimized based on the artificial bee colony algorithm provided in the present application has achieved the most stable and excellent detection accuracy under the conditions of different feature sets, current values ​​and switching frequencies.

[0137] First, by analyzing the feature sensitivity of the above algorithms, it can be found that the ABC-RF algorithm proposed in this application performs best and most stably when processing diversified and complex feature sets. This is because random forests can reduce noise and overfitting problems through the voting of multiple decision trees. The SVM algorithm, KNN algorithm, and NB algorithm are very sensitive to features. Once they are trained and tested on feature sets that do not provide obvious distinguishing information, their performance will drop significantly. The DT algorithm can perform nonlinear division based on features and can handle complex features, but because the single-tree algorithm is prone to overfitting, it performs worse than the ABC-RF algorithm.

[0138] Furthermore, based on different current values ​​and switching frequencies, the performance of the above five algorithms on the second filtered feature set is better than that on the unprocessed first feature set, which proves that the second time domain feature set after empirical filtering and the second frequency domain feature set after switching noise removal can better reflect the abnormal current situation and help improve the detection accuracy and precision.

[0139] Then, based on different current values ​​and switching frequencies, the above five algorithms performed best on the first entropy feature set and the second entropy feature set, and performed best and worst on the first kurtosis feature set and the second kurtosis feature set, proving that extracting entropy features helps improve the discrimination of current abnormal information, can reflect the uncertainty and complexity of current signals, and thus help the algorithm make more accurate classifications. However, kurtosis features may not provide enough information to distinguish between normal and abnormal states of current signals.

[0140] In addition, most of the above five algorithms perform better with the increase of current value, because the increase of current value makes the signal characteristics more obvious and the classification boundary clearer. Most of the above five algorithms also perform better with the increase of switching frequency, which shows that high-frequency signals contain more useful information and can improve the algorithm's ability to distinguish.

[0141] In summary, the embodiments of the present application can provide the following technical effects: On the one hand, by performing time domain analysis and frequency domain analysis on the current data of lithium batteries, the present application can extract the time-frequency characteristics in the current data more comprehensively and richly, and then use these time-frequency characteristics to realize current anomaly detection of lithium batteries. This can effectively improve the accuracy of current anomaly detection of lithium batteries, and help take preventive measures before thermal runaway of lithium batteries occurs, thereby protecting the overall safety and stability of the lithium battery system.

[0142] On the other hand, by optimizing the random forest algorithm through the artificial bee colony algorithm and using the optimized random forest algorithm to analyze the time-frequency characteristics of the current signal of the lithium battery, it is possible to maintain a high detection rate in a complex and changing environment, improve the reliability and robustness of the system, and ensure that current anomalies can be effectively detected under various operating conditions. By timely detecting current anomalies, this application helps to take preventive measures before thermal runaway of the lithium battery occurs, thereby protecting the overall safety and stability of the lithium battery system.

[0143] Although the embodiments of the present invention have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0144] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A method for detecting abnormal current caused by thermal runaway of a lithium battery, characterized in that: The steps include: Get the current data of lithium battery; Performing time domain processing and frequency domain processing on the current data to obtain a time domain feature set and a frequency domain feature set; According to the time domain feature set and the frequency domain feature set, thermal runaway current anomaly detection is performed on the lithium battery to obtain an anomaly detection result.

2. The method for detecting abnormal current caused by thermal runaway of a lithium battery according to claim 1, characterized in that: The performing time domain processing and frequency domain processing on the current data to obtain a time domain feature set and a frequency domain feature set includes: Sampling and segmenting the current data of the lithium battery to obtain an initial current data set; the initial current data set includes a plurality of initial data subsets, each of which includes a plurality of data points; Extracting time domain features from each of the initial data subsets to obtain the time domain feature set; Frequency domain feature extraction is performed on each of the initial data subsets to obtain the frequency domain feature set.

3. The method for detecting abnormal current caused by thermal runaway of a lithium battery according to claim 2, characterized in that: The time domain feature set includes a first time domain feature set and a second time domain feature set; the time domain feature extraction is performed on each of the initial data subsets to obtain the time domain feature set, including: Obtaining a first time-domain data set according to each of the initial data subsets; Performing time domain feature extraction on the first time domain data set to obtain a first time domain integral feature set, a first time domain kurtosis feature set, and a first time domain entropy feature set as the first time domain feature set; Performing empirical filtering on each of the initial data subsets to obtain a second time domain data set; Time domain features are extracted from the second time domain data set to obtain a second time domain integral feature set, a second time domain kurtosis feature set, and a second time domain entropy feature set as the second time domain feature set.

4. The method for detecting abnormal current caused by thermal runaway of a lithium battery according to claim 2, characterized in that: The frequency domain feature set includes a first frequency domain feature set and a second frequency domain feature set; extracting frequency domain features from each of the initial data subsets to obtain the frequency domain feature set includes: Processing each of the initial data subsets to obtain frequency components corresponding to each of the initial data subsets; According to each of the frequency components, a first frequency domain integral feature set, a first frequency domain kurtosis feature set and a first frequency domain entropy feature set are obtained as the first frequency domain feature set; According to each of the frequency components and the preset frequency range, combined with the switching noise removal technology, a second frequency domain integral feature set, a second frequency domain kurtosis feature set and a second frequency domain entropy feature set are obtained as the second frequency domain feature set.

5. The method for detecting abnormal current caused by thermal runaway of a lithium battery according to claim 1, characterized in that: The step of performing thermal runaway current anomaly detection on the lithium battery according to the time domain feature set and the frequency domain feature set to obtain an anomaly detection result includes: Obtain a first feature set according to the first time domain feature set in the time domain feature set and the first frequency domain feature set in the frequency domain feature set; Obtain a second feature set according to the second time domain feature set in the time domain feature set and the second frequency domain feature set in the frequency domain feature set; A machine learning method is used to perform thermal runaway current anomaly detection on the lithium battery according to the first feature set and the second feature set to obtain the anomaly detection result.

6. The method for detecting abnormal current caused by thermal runaway of a lithium battery according to claim 5, characterized in that: The first feature set includes a first integral feature set, a first kurtosis feature set, and a first entropy feature set, and the first feature set is obtained according to the first time domain feature set in the time domain feature set and the first frequency domain feature set in the frequency domain feature set, including: Obtain the first integral feature set according to the first time domain integral feature set in the first time domain feature set and the first frequency domain integral feature set in the first frequency domain feature set; Obtaining the first kurtosis feature set according to the first time domain kurtosis feature set in the first time domain feature set and the first frequency domain kurtosis feature set in the first frequency domain feature set; The first entropy feature set is obtained according to the first time domain entropy feature set in the first time domain feature set and the first frequency domain entropy feature set in the first frequency domain feature set.

7. The method for detecting abnormal current caused by thermal runaway of a lithium battery according to claim 5, characterized in that: The second feature set includes a second integral feature set, a second kurtosis feature set and a second entropy feature set, and the second feature set is obtained according to the second time domain feature set in the time domain feature set and the second frequency domain feature set in the frequency domain feature set, including: Obtain the second integral feature set according to the second time domain integral feature set in the second time domain feature set and the second frequency domain integral feature set in the second frequency domain feature set; Obtain the second kurtosis feature set according to the second time domain kurtosis feature set in the second time domain feature set and the second frequency domain kurtosis feature set in the second frequency domain feature set; The second entropy feature set is obtained according to the second time domain entropy feature set in the second time domain feature set and the second frequency domain entropy feature set in the second frequency domain feature set.

8. The method for detecting abnormal current caused by thermal runaway of a lithium battery according to claim 5, characterized in that: The machine learning method includes a random forest algorithm.

9. The method for detecting abnormal current caused by thermal runaway of a lithium battery according to claim 8, characterized in that: The hyperparameters of the random forest algorithm are optimized using an artificial bee colony algorithm.

10. A current anomaly detection system caused by thermal runaway of a lithium battery, characterized in that: include: A data acquisition module, used to acquire current data of the lithium battery; A feature extraction module, used for performing time domain processing and frequency domain processing on the current data to obtain a time domain feature set and a frequency domain feature set; The abnormality detection module is used to perform thermal runaway current abnormality detection on the lithium battery according to the time domain feature set and the frequency domain feature set to obtain an abnormality detection result.

Images