Real-time anomaly detection method for lithium ion battery module in energy storage system

Abstract

The invention discloses a real-time anomaly detection method for a lithium ion battery module in an energy storage system, and aims to solve the problems of complex battery operation state, insufficient anomaly detection real-time performance and strong dependence on an abnormal sample, the method combines the space-time characteristics of battery operation data, and takes a local anomaly factor algorithm as a core to construct a detection framework. And a detection threshold calculation method based on data features is designed. Through analyzing local density difference and distribution characteristics of battery operation data, a threshold value is calculated according to actual characteristics of the data, and accurate identification and positioning of abnormal points are realized. The abnormal lithium ion battery in the energy storage system can be rapidly and accurately detected in combination with the local abnormal factor algorithm and the optimized detection threshold value, the method adapts to complex abnormal modes possibly occurring in the battery operation process, the false alarm rate and the missing report rate are remarkably reduced, and important technical guarantee is provided for safety and stability of the energy storage system.

Description

Technical Field

[0001] The present invention is applicable to the field of energy storage technologies. More specifically, it is a real-time anomaly detection method for lithium-ion battery modules in an energy storage system based on local outlier factors and spatio-temporal data characteristics. Background Art

[0002] Lithium-ion batteries have been widely used and rapidly developed in the global energy storage field due to their high energy density, long cycle life, high efficiency, and relatively low cost. However, during operation, lithium-ion batteries may cause safety problems due to factors such as overcharging, over-discharging, and current overload, such as short circuits, thermal runaway, and even fires or explosions. Therefore, in order to effectively ensure the safety of the energy storage system, it is necessary to monitor the operating state of the battery in real time to ensure efficient and rapid detection at the early stage of anomalies, thereby preventing the expansion of potential safety hazards. In recent years, in response to safety problems such as overcharging, over-discharging, short circuits, and thermal runaway that may occur during the operation of lithium-ion batteries, researchers have proposed a variety of advanced anomaly detection methods, including techniques based on statistical analysis, machine learning, and physical models.

[0003] Statistical analysis methods identify abnormal states by analyzing the statistical characteristics of battery operation data. For example, using K-means clustering, Z-score, and 3σ screening strategies, abnormal battery cells can be located; a short-circuit fault diagnosis method based on cosine similarity can achieve second-level fault detection by detecting voltage changes and cosine similarity; principal component analysis and cumulative sum methods are also used for temperature anomaly monitoring. These methods have the advantages of simple implementation and intuitive operation, but there are computational efficiency problems when dealing with complex large-scale data sets, and it is difficult to accurately capture the distribution characteristics of the data.

[0004] With the development of big data technology, machine learning has been widely used in the field of battery anomaly detection. For example, an improved variational autoencoder can efficiently extract battery anomaly features for high-precision detection; by combining a BP neural network with a multi-layer screening strategy, abnormal changes in the voltage of battery cells can be effectively detected; a dynamic deep learning model analyzes the charging state and battery response through an autoencoder, significantly improving the sensitivity of anomaly detection. However, machine learning methods usually rely on a large amount of high-quality training data, and the interpretability of the model is relatively low. In addition, the computational overhead of complex models is relatively high, making it difficult to meet the requirements of the energy storage system for real-time performance and efficiency.

[0005] The physical model-based method judges whether there is an anomaly by modeling the operating mechanism of lithium-ion batteries and comparing the residuals between the model estimation values and the actual observed values. For example, equivalent circuit models, electrochemical models, and thermal models are commonly used modeling methods. The advantage of this type of method lies in its high theoretical accuracy and the ability to describe in detail the physical and chemical processes inside the battery. However, the construction of physical models usually requires a large amount of experimental data and prior knowledge, with high model complexity and large computational overhead, making it difficult to be applied in real time to large-scale energy storage systems.

[0006] Based on the above background, there is an urgent need for a new real-time detection method that combines the spatio-temporal characteristics of lithium-ion battery operating data and local outlier factors. Summary of the Invention

[0007] The present invention proposes a real-time detection method for abnormal lithium batteries based on local outlier factors and data spatio-temporal characteristics. This method should be able to efficiently extract abnormal features, achieve accurate and rapid monitoring of the battery operating state, and have high scalability to meet the operating requirements of energy storage systems under complex working conditions. By improving the real-time, accuracy, and efficiency of detection, this method will provide strong guarantees for the safety and reliability of energy storage systems, and at the same time promote the application and development of energy storage technologies in a wider range of fields.

[0008] In the anomaly measurement section, the present invention adopts the Local Outlier Factor (LOF) technology to perform local density analysis on the characteristics of battery cluster operating data. By calculating the local density difference between each data point and its neighborhood data, the anomaly degree of the data point is evaluated and potential anomaly points are identified.

[0009] In the threshold calculation section, the present invention designs a threshold setting mechanism based on the data distribution characteristics. This mechanism uses statistical methods to fit the distribution of the data, and through comprehensive analysis of the skewness and kurtosis characteristics of the data distribution, selects the distribution model that best conforms to the data characteristics. On the basis of determining the fitted distribution, the LOF detection threshold that conforms to the actual characteristics of the data is calculated. When the distribution of the data shifts or fluctuates, the threshold can be recalculated to ensure the stability and adaptability of the model's detection effect in different data scenarios, thereby improving the accuracy of anomaly detection.

[0010] Through the above innovative design, the present invention realizes efficient, accurate, and real-time anomaly detection, overcomes the problem of insufficient adaptability of traditional methods in complex scenarios, and provides an important technical guarantee for the safety and stability of energy storage systems. By making full use of the time-series pattern of the operation data of lithium-ion batteries and the spatial consistency characteristics among individual batteries, the real-time and efficient identification and accurate positioning of abnormal batteries are achieved. The analysis in the time dimension can effectively identify anomalies caused by long-term battery aging or short-term severe fluctuations; while the analysis in the spatial dimension, through the consistency characteristics among individual batteries and combined with the structure of the energy storage system, can accurately locate the source of anomalies. In addition, the present invention makes full use of the advantages of unsupervised learning, overcomes the limitation of scarce abnormal samples, and gets rid of the dependence on a large amount of labeled data, enabling it to still achieve accurate anomaly detection in the case of insufficient abnormal samples. Through the above innovative technologies, the present invention provides an important technical support for the efficient and accurate anomaly detection of lithium-ion batteries, significantly improving the safety and stability of energy storage systems.

[0011] The technical solution adopted by the present invention is specifically as follows:

[0012] A real-time anomaly detection method for lithium-ion battery modules in an energy storage system, the method comprising:

[0013] Obtain the voltage signal segment data of each battery in the lithium-ion battery module in the energy storage system and calculate the mean value as the statistical feature of the corresponding battery;

[0014] For each battery, calculate the Euclidean distance from the battery to other batteries based on the statistical feature, and determine the k-distance and k-distance neighborhood of each battery; where k is equal to the first detection threshold;

[0015] For each battery, calculate the reachable distance from the battery to each neighborhood point within the corresponding k-distance neighborhood;

[0016] For each battery, define the local reachability density of the battery to describe the relative density of the battery in its k-distance neighborhood;

[0017] For each battery, calculate the local outlier factor of the battery to evaluate the degree of anomaly of the battery. When the local outlier factor of the battery is greater than 1, it is determined that the battery may be an abnormal battery;

[0018] Further determine whether it is abnormal based on the voltage mean value of the battery that may be abnormal: If the voltage mean value of the battery that may be abnormal exceeds the corrected left and right quantile mean values, determine that the battery is an abnormal battery; or calculate the deviation degree of the voltage mean value of the battery that may be abnormal. If the deviation degree of the voltage mean value exceeds the second detection threshold, determine that the battery is an abnormal battery; the corrected left and right quantiles and the second detection threshold are obtained by using the Cornish-Fisher expansion formula, comprehensively considering the skewness and kurtosis characteristics of the data, and correcting the quantiles of the standard normal distribution fitted to the statistical characteristics of each battery in the lithium-ion battery module in the energy storage system under normal operating conditions.

[0019] Further, the specific method for obtaining the voltage signal segment data of each battery in the lithium-ion battery module of the energy storage system is as follows:

[0020] Real-time collect the voltage signal sequences of each battery in the lithium-ion battery module of the energy storage system;

[0021] Use a sliding window to slide on the voltage signal sequence at a fixed step size, and extract the voltage signal segment data for analysis each time it moves.

[0022] By introducing the sliding window mechanism, the continuously arriving battery operation data is processed in segments. The sliding window effectively reduces the overall computational complexity by restricting the data processing range, and at the same time realizes a fast response to abnormal points, ensuring that the algorithm has good real-time detection ability and meets the real-time requirements of the energy storage system.

[0023] The present invention restricts the calculation range within the sliding window, and at the same time utilizes the local density characteristics in the multi-dimensional space, reduces the dependence on the global data analysis, greatly reduces the computational cost, and provides technical support for the online application of large-scale energy storage systems.

[0024] Further, the corrected left and right quantile mean values and the second detection threshold are calculated as follows:

[0025] Obtain the statistical characteristics of each battery in the lithium-ion battery module of the energy storage system under normal operating conditions as a data set;

[0026] Fit the standard normal distribution based on the data set, and calculate the skewness S and kurtosis K of the samples in the data set based on the fitting result of the normal distribution;

[0027] Use the Cornish-Fisher expansion formula to correct the standard normal quantiles by comprehensively considering the skewness and kurtosis characteristics. The calculation formula is:

[0028] ;

[0029] Where Z is the quantile of the standard normal distribution, and the quantiles are divided into left quantiles and right quantiles. is the corrected quantile;

[0030] Calculate the mean values of the corrected left and right quantiles based on the corrected left and right quantiles.

[0031] Combine the mean and standard deviation of the normal distribution with the corrected left and right quantiles to calculate the values on the standard normal distribution and take the mean as the second detection threshold.

[0032] Furthermore, it is characterized in that the calculation formula of skewness S is as follows:

[0033] ;

[0034] Where X represents the statistical characteristics of the sample; E represents the expectation, and μ and σ are the mean and standard deviation of the normal distribution respectively.

[0035] Furthermore, the calculation formula of kurtosis K is as follows:

[0036] ;

[0037] Where X represents the statistical characteristics of the sample; E represents the expectation, and μ and σ are the mean and standard deviation of the normal distribution respectively.

[0038] Furthermore, the quantiles of the standard normal distribution are determined according to the actual failure rate of the lithium-ion battery modules in the energy storage system.

[0039] Furthermore, regularly or irregularly re-acquire the statistical characteristic data of each battery in the lithium-ion battery modules in the energy storage system under the current normal operating conditions, and update the corrected left and right quantiles and the second detection threshold.

[0040] Furthermore, the first detection threshold is determined according to the empirical method (that is, the k value accounts for 1% to 10% of the dataset size).

[0041] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the real-time anomaly detection method for lithium-ion battery modules in an energy storage system as described above.

[0042] A storage medium containing computer-executable instructions, and when the computer-executable instructions are executed by a computer processor, they implement the real-time anomaly detection method for lithium-ion battery modules in an energy storage system as described above.

[0043] A computer program product includes a computer program / instructions. When the computer program / instructions are executed by a processor, the steps of the real-time anomaly detection method for a lithium-ion battery module in an energy storage system are implemented.

[0044] The present invention has the following beneficial effects:

[0045] (1) By combining the time series characteristics of battery operation data and the spatial consistency among individual batteries, the present invention uses the LOF method to perform refined analysis on the local density characteristics of the data, thereby achieving efficient identification and accurate positioning of abnormal batteries.

[0046] (2) Calculating the detection threshold based on the actual characteristics of the public data enables the present invention to detect abnormal battery monomers more quickly and accurately, significantly reducing the false alarm rate and the missed alarm rate.

[0047] (3) The present invention does not rely on a large number of abnormal samples or preset empirical parameters, has strong robustness and adaptability, and can be widely applied to the anomaly detection of lithium-ion battery clusters in different scenarios, providing a reliable guarantee for the safe and efficient operation of the energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments.

[0049] Figure 1 is a flowchart of a real-time anomaly detection method for a lithium-ion battery module in an energy storage system provided by the present invention.

[0050] Figure 2 is a probability-probability plot (P-P plot) of different distribution fittings of data in an embodiment of the present invention; Figure 2 In (a) is the probability-probability plot of the normal distribution fitting of data in an embodiment of the present invention, Figure 2 In (b) is the probability-probability plot of the Laplace distribution fitting of data in an embodiment of the present invention.

[0051] Figure 3 is a statistical chart of the battery replacement rate trend of 20,000 vehicles in a certain community from 2011 to 2023.

[0052] Figure 4 is the cumulative distribution function curve, probability distribution function curve, and detection threshold of data in an embodiment of the present invention.

[0053] Figure 5 is a detection result and abnormal point index result diagram of different methods in an embodiment of the present invention; Figure 5 In (a) is the detection result and abnormal point index result diagram using the K-means clustering method; Figure 5In (b), it is the detection result and the outlier index result graph using the random forest method, Figure 5 In (c), it is the detection result and the outlier index result graph using the Shannon entropy method, Figure 5 In (d), it is the detection result and the outlier index result graph using the cosine similarity method, Figure 5 In (e), it is the detection result and the outlier index result graph using the autoencoder method, Figure 5 In (f), it is the detection result and the outlier index result graph using the separate LOF method, Figure 5 In (g), it is the detection result and the outlier index result graph of the method of the present invention. Detailed implementation manners

[0054] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all the implementation manners consistent with the present application. On the contrary, they are merely examples of the devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0055] The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application.

[0056] The singular forms "a", "the" and "said" used in the present application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0057] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to determining".

[0058] The present invention provides a real-time anomaly detection method for lithium-ion battery modules in an energy storage system, as Figure 1 shown, the method includes the following steps:

[0059] (1) Obtain the voltage signal segment data of each battery in the lithium-ion battery module in the energy storage system and calculate the mean value as the statistical feature of the corresponding battery;

[0060] In a specific embodiment, voltage signal sequences of each battery in the lithium-ion battery module of the energy storage system are collected in real time, the original data is sliced or segmented, and a suitable time window is selected to slide on the data sequence with a fixed step size (for example, 1 minute or an adaptive time length determined according to the data characteristics) using a sliding window. Each time it moves, voltage signal segment data is extracted for analysis. Then, these statistical features are used for distribution fitting, and a suitable theoretical distribution model is selected to fit the statistical characteristics of the data. In this embodiment, the mean value is calculated as the statistical feature corresponding to the battery.

[0061] (2) For each battery, based on the statistical features, calculate the distance from the battery to other batteries, and determine the k-distance and k-distance neighborhood of each battery; specifically, the statistical features of each battery are regarded as a data point, and the k-distance of point p is defined as the distance to its k-th nearest neighbor point to determine the neighborhood of point p. The present invention selects the Euclidean distance as the way to measure the distance between data points: first calculate the Euclidean distance from point p to all other points in the data set, then sort them in ascending order, and take the k-th smallest distance as the k-distance of p, where k is determined by the first detection threshold. And the set of all points whose distance to point p is the k-distance of p is called the k-distance neighborhood of p, denoted as .

[0062] (3) For each battery, calculate the reachable distance from the battery to each neighborhood point within the corresponding k-distance neighborhood to avoid the problem of local density calculation deviation caused by too large or too small distances between neighboring points. Define the reachable distance from point p to point q, which is defined as:

[0063] (1)

[0064] Where, is the reachable distance from point p to point q, is the k-distance of point q, is the Euclidean distance between point p and point q.

[0065] (4) For each battery, define the local reachable density of the battery to describe the relative density of the battery in its k-distance neighborhood; it is expressed as:

[0066] (2)

[0067] Where, is the local reachable density of point p, is the size of the k-nearest neighbor set of point p.

[0068] (5) For each battery, calculate the local outlier factor of the battery. Define the local outlier factor of point p as:

[0069] (3)

[0070] Among them, is the local outlier factor of point p, is the local reachability density of point q. If , it indicates that the density of point p is similar to the density of its neighborhood, and it belongs to a normal point; if and is significantly higher than 1, it indicates that the density of point p is lower than the average density of the neighborhood, and it may be an outlier. The larger the value, the higher the degree of outlier. To reduce the missed judgment, in the present invention, when the local outlier factor of the battery is greater than 1, it is determined that the battery may be an abnormal battery; and then further judgment is made in combination with a threshold.

[0071] (6) Further judge whether it is abnormal based on the voltage mean value of the battery that may be abnormal: If the voltage mean value of the battery that may be abnormal exceeds the mean value of the corrected left and right quantiles, it is determined that the battery is an abnormal battery; or calculate the deviation degree of the voltage mean value of the battery that may be abnormal, if the deviation degree of the voltage mean value exceeds the second detection threshold, it is determined that the battery is an abnormal battery;

[0072] Traditional threshold calculation methods usually assume that the data strictly follows a normal distribution, and calculate the threshold based on the quantiles (Z-score) of the standard normal distribution. However, the actual data often has a certain deviation on the basis of following the normal distribution, showing non-normal characteristics such as skewness or kurtosis, resulting in the thresholds calculated by traditional methods lacking accuracy and adaptability. To solve the above problems, the present invention proposes an improved method. By introducing the Cornish-Fisher expansion, the skewness and kurtosis characteristics of the data are comprehensively considered, and the standard normal quantiles are corrected, so as to obtain a more accurate threshold estimate.

[0073] First, obtain the statistical characteristics of each battery in the lithium-ion battery module in the energy storage system under normal operating conditions as the data set; fit the standard normal distribution based on the data set;

[0074] In practical applications, common distribution types include normal distribution, Laplace distribution, lognormal distribution, exponential distribution, gamma distribution, etc. Different distribution types have their own applicability and advantages in different application scenarios. Regarding the voltage data characteristics of lithium-ion batteries, the present invention analyzes their distribution characteristics. Battery voltage data usually exhibits static characteristics of small fluctuations and symmetric distribution, and may also show dynamic spike characteristics under charge and discharge conditions. Therefore, the present invention focuses on considering two models, normal distribution and Laplace distribution, in the selection of the distribution model. To evaluate the fitting effect, the Kolmogorov-Smirnov (KS) test and P-P plot are used to quantify and visualize the goodness of fit. The KS test selects the most suitable model by quantifying the maximum deviation between the empirical distribution and the fitting model; the P-P plot intuitively reflects the goodness of the fitting effect by comparing the cumulative distributions of the theoretical distribution and the sample data. In this embodiment, the applicability of the two distributions is compared and verified through strict distribution fitting and statistical tests, and it is determined that the normal distribution can better fit the voltage mean data distribution. The results of a specific implementation case of fitting and statistical tests (KS test and P-P plot) are shown in Table 1 and Figure 2 as follows.

[0075] Table 1 KS statistics and P-values of the fitted distributions

[0076]

[0077] Based on the fitting results of the normal distribution, its mean μ and standard deviation σ are calculated.

[0078] Then, the skewness and kurtosis characteristics of the data sample are calculated. Skewness measures the degree of asymmetry of the distribution, and its calculation formula is:

[0079] (4)

[0080] where X represents the statistical characteristics of the data sample, μ is the sample mean, and σ is the sample standard deviation. E represents the expectation; positive skewness indicates that the right tail of the distribution is longer, and negative skewness indicates that the left tail is longer. Kurtosis measures the sharpness and tail thickness of the distribution. In the present invention, the excess kurtosis of the data is calculated, and its formula is:

[0081] (5)

[0082] Positive kurtosis indicates that the distribution is sharper or the tail is thicker than the normal distribution, and negative kurtosis indicates that the distribution is flatter or the tail is thinner. Based on the obtained skewness and kurtosis values, the present invention uses the Cornish-Fisher expansion to correct the standard normal quantiles, and the corrected quantile calculation formula is:

[0083] (6)

[0084] where Z is the quantile of the standard normal distribution, including the left quantile and the right quantile. This formula includes a skewness correction term, a kurtosis correction term, and a high-order interaction correction term, which can comprehensively consider the characteristics of data distribution deviating from the normal distribution. Specifically, the first term Z in the formula represents the original quantile of the standard normal distribution and is the basic reference value; the second term is the skewness correction term. When the data distribution shows an asymmetric characteristic, this term preliminarily adjusts the Z value through the skewness coefficient S, so that the quantile extends appropriately in the skewed direction of the distribution; the third term is the kurtosis correction term, which mainly targets the peakedness and tail thickness characteristics of the distribution. When the distribution shows kurtosis characteristics different from the normal distribution, this term will correspondingly adjust the size of the extreme quantiles; the fourth term is the high-order interaction correction term, which mainly considers the second-order effect of skewness, prevents overcorrection of the far-tail quantiles in the case of high skewness, and ensures the accuracy and stability of the overall correction effect.

[0085] The quantiles of the standard normal distribution can be determined according to the actual failure rate of the lithium-ion battery modules in the energy storage system. Specifically, in recent years, the proportion of lithium-ion batteries used in electric vehicles replaced due to failures has decreased significantly. From Figure 3 it can be seen that the failure rate has decreased significantly after 2015, especially stabilizing between 0.1% and 0.5% after 2016. After calculation, the average failure rate from 2016 to 2023 is 0.25%. Referring to the vehicle battery failure data, and usually having extremely high reliability requirements for the operating batteries (such as electric vehicles, grid energy storage systems), so the proportion of the normal lithium battery operation data segments in all segments of the energy storage system is set to 99.75%. Then the left quantile is 0.125 and the right quantile is 99.875.

[0086] Finally, based on the corrected value, the second detection threshold is calculated, and its formula is:

[0087] (7)

[0088] The technical solution of the present invention will be further described below in conjunction with a specific embodiment and the accompanying drawings, but the present invention is not limited to the following embodiments.

[0089] (1) Obtain the voltage signal segment data of each battery in the lithium-ion battery module of the energy storage system;

[0090] The data set used in this embodiment is from a certain energy storage power station. Each battery cluster is composed of 7 modules connected in series, and each module is further composed of 52 lithium-ion batteries connected in series. The standard voltage of each battery is 3.2V. The collected data includes the real-time voltage signal of each battery , , , and the charge and discharge current of the entire battery cluster , , where n represents the total number of batteries, n = 7 × 52 = 364. As shown in Table 2. The data set contains the operation data of 4 normal battery clusters and the data of 1 battery cluster containing abnormal data points. In this embodiment, the real-time voltage signal of the battery obtaining the above data is used to implement anomaly detection.

[0091] Table 2 Battery Cluster Operation Data Samples

[0092]

[0093] Note: 60 data are collected per minute during data acquisition, and the data in the table are arranged in sequence.

[0094] (2) Introduce a sliding window processing mechanism to the voltage data stream of the operation data of the 4 normal battery clusters collected. Divide the continuous voltage data stream into multiple subsequences according to a fixed length w (window size). In this embodiment, w = 60. The sliding window slides on the data sequence with a fixed step size, and each time it moves, a new subset is extracted for analysis.

[0095] Slice the voltage subsequences generated by the sliding window. Randomly extract multiple segments from the normal operation data of the battery. Each segment contains w = 60 data points, and calculate the mean of each segment as the representative statistical feature of voltage fluctuation. Fit the distribution of the segment mean data. The results of fitting and statistical tests (KS test and P-P plot) are shown in Table 1 and Figure 2 .

[0096] Based on the fitting result of the normal distribution, calculate its mean μ and standard deviation σ. Further, calculate the second detection threshold T according to formulas (4)-(7). Among them, for the standard normal distribution, the left quantile is 0.125 and the right quantile is 99.875.

[0097] As Figure 4 shown, calculate the second detection threshold . The present invention realizes an accurate mapping relationship between distribution characteristics and quantile calculation through this mathematical formula. When facing the actual data distribution, this formula can adaptively adjust the standard normal quantiles according to the specific skewness and kurtosis characteristics of the data. For example, for a positively skewed distribution ( ), the formula will appropriately increase the positive Z value and decrease the negative Z value to accurately reflect the characteristic of the extended right tail of the distribution; while for a high peak and thick tail distribution ( ), the formula will enhance the extremity of extreme quantiles, thus accurately reflecting the characteristic that the probability of tail events in the distribution is higher than that of the normal distribution. This adjustment mechanism is based on the principle of moment expansion in mathematical statistics, expressing the quantiles of non-normal distributions as polynomial functions of the quantiles of normal distributions, which not only maintains the simplicity of calculation but also significantly improves the accuracy of the results.

[0098] (3) For the data subsequence after processing the data of a battery cluster containing abnormal data points collected by a sliding window, calculate the mean value as the statistical feature of the corresponding battery.

[0099] (4) For each battery, calculate the distance from the battery to other batteries based on the statistical features, and determine the k-distance and k-distance neighborhood of each battery; where k is equal to the first detection threshold, and the first detection threshold is determined by the empirical method; in this embodiment, k = 25 is determined through multiple experiments, which is 6% of the dataset size;

[0100] (5) For each battery, calculate the reachable distance from the battery to each neighborhood point within the corresponding k-distance neighborhood.

[0101] (6) For each battery, define the local reachability density of the battery to describe the relative density of the battery in its k-distance neighborhood.

[0102] (7) For each battery, calculate the local outlier factor of the battery to evaluate the degree of abnormality of the battery. When the local outlier factor of the battery is greater than 1, it is determined that the battery may be an abnormal battery.

[0103] (8) Further determine whether it is abnormal based on the voltage mean value of the battery that may be abnormal: In this embodiment, calculate the deviation degree of the voltage mean value of the battery that may be abnormal , where is the mean value of each voltage segment. If the deviation degree of the voltage mean value exceeds the second detection threshold, it is determined that the battery is an abnormal battery and used as the basis for triggering an alarm.

[0104] To verify the detection ability of the method of the present invention, comparative experiments were conducted with five other detection methods, including K-means clustering, random forest, Shannon entropy, cosine similarity, and autoencoder, and compared with the conventional LOF algorithm based on IQR (this method calculates the threshold based on the quartiles of the data distribution: calculates the difference between the third quartile (Q3) and the first quartile (Q1), representing the distribution range of the middle 50% of the data, and marks the data points exceeding "Q3 + 3.0 × IQR" as upper-bound anomalies, and identifies the points below "Q1 – 3.0 × IQR" as lower-bound anomalies). The embodiments of the present invention performed anomaly detection analysis on 60 data points in the range of 3920 to 3960. To ensure the fairness of the comparison, the neighborhood density calculated by LOF was converted into an anomaly score in the experiment, and the detection results of other methods were standardized. The detection effects and running times of each method were recorded in Table 3 and Figure 5 in which the running time is the average time of ten runs of each method. The abnormal battery indices are 155 and 364, and the rest are normal points.

[0105] Table 3 Detection effects and running times of each method

[0106]

[0107] The method of the present invention can accurately detect two abnormal points. During the entire detection process, no false detection occurred, showing extremely high detection accuracy. In addition, the method of the present invention also shows superiority in terms of running efficiency, and it only takes an average of 0.0106 seconds to complete the data detection of a sliding window, meeting the requirements of real-time monitoring of energy storage batteries.

[0108] Compared with other methods, the method of the present invention has significant advantages in both detection performance and efficiency. Specifically, the K-means clustering method successfully detected two types of abnormal points, but the detection process took a long time and the efficiency was lower than that of the method proposed in the present invention. Although the random forest method can identify anomalies, the false detection rate is relatively high, and 6 single cells were misdetected in the experiment, significantly reducing the reliability of the detection results. The Shannon entropy method is mainly applicable to the identification of long-term trend anomalies, but it fails to effectively capture the anomalies with short-term fluctuation characteristics, and its detection ability has limitations. The cosine similarity method performs relatively balanced in terms of detection accuracy and efficiency, can detect two abnormal points and has a short running time, but its stability in complex scenarios is slightly inferior to that of the method of the present invention. The autoencoder method has low detection efficiency due to the need for complex pre-training, is difficult to meet the requirements of real-time detection, and has limited applicability.

[0109] Compared with the conventional LOF algorithm based on IQR, the detection threshold calculation method proposed by the present invention not only significantly improves the detection accuracy, but also effectively enhances the running speed. From the detection results, the conventional LOF algorithm detected 29 abnormal points, while the method of the present invention accurately identified two abnormal points with battery indices 155 and 364, indicating that the method of the present invention can more precisely screen out the true abnormal points, reduce the false alarm rate, and improve the reliability of detection. The conventional LOF algorithm relies on the calculation of quartiles of data distribution to determine the threshold when judging abnormal points. This method has a high calculation cost when the data volume is large or the data distribution is complex, resulting in low detection efficiency and difficulty in adapting to the complex and dynamically changing battery voltage data in energy storage power stations. By adopting an optimized detection threshold calculation method, the present invention realizes the rapid and accurate identification of abnormal points, ensures the detection is completed in an extremely short time, and avoids redundant calculations, thus improving the overall operation efficiency. In contrast, the method of the present invention can not only capture battery anomalies in the energy storage system faster, but also effectively reduce false alarms, enabling maintenance personnel to timely discover potential faults and quickly take measures to ensure the safe and stable operation of the energy storage power station, providing more advantageous technical support for the efficient management and data analysis of the energy storage system.

[0110] Corresponding to the foregoing embodiment of a real-time anomaly detection method for lithium-ion battery modules in an energy storage system, the present invention also provides an embodiment of an electronic device implementing a real-time anomaly detection method for lithium-ion battery modules in an energy storage system.

[0111] An electronic device provided by an embodiment of the present invention includes one or more processors for implementing a real-time anomaly detection method for lithium-ion battery modules in an energy storage system in the foregoing embodiment.

[0112] The embodiment of the electronic device of the present invention can be applied to any device with data processing capabilities, and the any device with data processing capabilities can be a device or apparatus such as a computer.

[0113] The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful device, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory and running. In terms of hardware, in addition to the processor, memory, network interface, and non-volatile memory, any device with data processing capabilities where the device in the embodiment is located usually also includes other hardware according to the actual functions of the any device with data processing capabilities, which will not be elaborated here.

[0114] The implementation processes of the functions and roles of each unit in the above device are specifically described in detail in the implementation processes of the corresponding steps in the above method, and will not be elaborated here.

[0115] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial descriptions of the method embodiments. The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. Those of ordinary skill in the art can understand and implement it without creative work.

[0116] The embodiments of the present invention also provide a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, a real-time anomaly detection method for a lithium-ion battery module in an energy storage system in the above embodiments is implemented.

[0117] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be any device with data processing capabilities, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capabilities and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store the data that has been output or will be output.

[0118] As described above, only some embodiments of the present invention are given, and the present invention is not limited in other forms. Those skilled in the art can make various modifications or supplements to the specific embodiments described or use similar ways to replace them. However, any simple modifications and equivalent changes made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A real-time anomaly detection method for lithium-ion battery modules in an energy storage system, characterized in that, The method includes: Obtaining the voltage signal segment data of each battery in the lithium-ion battery module of the energy storage system and calculating the mean value as the statistical feature of the corresponding battery; For each battery, calculating the Euclidean distance from the battery to other batteries based on the statistical feature, and determining the k-distance and k-distance neighborhood of each battery; where k is equal to the first detection threshold; For each battery, calculating the reachable distance from the battery to each neighborhood point within the corresponding k-distance neighborhood; For each battery, defining the local reachability density of the battery to describe the relative density of the battery in its k-distance neighborhood; For each battery, calculating the local outlier factor of the battery to evaluate the degree of abnormality of the battery. When the local outlier factor of the battery is greater than 1, it is determined that the battery may be an abnormal battery; Further determining whether it is abnormal based on the voltage mean value of the battery that may be abnormal: If the voltage mean value of the battery that may be abnormal exceeds the corrected left and right quantile mean values, it is determined that the battery is an abnormal battery; or calculating the deviation degree of the voltage mean value of the battery that may be abnormal. If the voltage mean deviation degree exceeds the second detection threshold, it is determined that the battery is an abnormal battery; The corrected left and right quantiles and the second detection threshold are obtained by using the Cornish-Fisher expansion formula, comprehensively considering the skewness and kurtosis characteristics of the data, and correcting the quantiles of the standard normal distribution fitted based on the statistical features of each battery in the lithium-ion battery module of the energy storage system under normal operating conditions.

2. The method according to claim 1, characterized in that, The obtaining of the voltage signal segment data of each battery in the lithium-ion battery module of the energy storage system is specifically: Real-time collecting the voltage signal sequences of each battery in the lithium-ion battery module of the energy storage system; Using a sliding window to slide on the voltage signal sequence at a fixed step length, and extracting the voltage signal segment data for analysis each time it moves.

3. The method according to claim 1, characterized in that, The corrected left and right quantile mean values and the second detection threshold are calculated and obtained through the following method: Obtaining the statistical features of each battery in the lithium-ion battery module of the energy storage system under normal operating conditions as a data set; Fitting a standard normal distribution based on the data set, and calculating the skewness S and kurtosis K of the samples in the data set based on the fitting result of the normal distribution; Using the Cornish-Fisher expansion formula to correct the standard normal quantiles by comprehensively considering the skewness and kurtosis characteristics, and its calculation formula is: ； Where Z is the quantile of the standard normal distribution, and the quantiles are divided into left quantiles and right quantiles, is the corrected quantile; the corrected left and right quantile means are calculated based on the corrected left and right quantiles; Combining the mean value, standard deviation of the normal distribution and the corrected left and right quantiles to calculate the values on the standard normal distribution and taking the mean value as the second detection threshold.

4. The method according to claim 3, wherein The calculation formula of skewness S is as follows: ； Where X represents the statistical feature of the sample; E represents the expectation, and μ and σ are the mean value and standard deviation of the normal distribution respectively.

5. The method according to claim 3, wherein The calculation formula of kurtosis K is as follows: ； Where X represents the statistical feature of the sample; E represents the expectation, and μ and σ are the mean value and standard deviation of the normal distribution respectively.

6. The method according to claim 3, characterized in that The quantiles of the standard normal distribution are determined according to the failure rate of the actual operation of the lithium-ion battery module in the energy storage system.

7. The method according to claim 1, wherein The first detection threshold is based on the empirical method.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a real-time anomaly detection method for a lithium-ion battery module in an energy storage system as described in any one of claims 1-7.

9. A storage medium containing computer-executable instructions, which, when executed by a computer processor, implement a real-time anomaly detection method for a lithium-ion battery module in an energy storage system as described in any one of claims 1-7.

10. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by a processor, the steps of a real-time anomaly detection method for a lithium-ion battery module in an energy storage system as described in any one of claims 1-7 are implemented.

Images