Battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition

Abstract

The present invention provides a battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition, comprising the following steps: S1. collecting battery bottom echo data during the use of lithium-ion batteries through ultrasonic equipment; S2. pre-processing the collected battery bottom echo data; S3. performing feature engineering on the pre-processed battery bottom echo data to extract features valuable for battery discharge evaluation; S4. constructing and training a sliding window isolation forest detection model based on time series; S5. selecting the optimal model; S6. using the isolation forest detection model to evaluate the discharge degree and over-discharge performance of lithium-ion batteries; the above method can be used to reflect the internal changes of the battery, and the trained model can be used in other online batteries, and lithium batteries with different positive electrode materials and structures all have this bottom wave characteristic property, so it has high universality and can provide technical support for avoiding battery damage.

Description

Technical Field

[0001] The invention relates to the technical field of battery detection, and in particular to a battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition. Background Art

[0002] Lithium-ion batteries have become core components of electronic consumer products, electric vehicles, and energy storage systems due to their superior energy density, long cycle life, and no memory effect. However, with the widespread use of lithium-ion batteries, ensuring their safety and extending their service life has become an urgent issue. Among them, overcharging and over-discharging are important considerations, which may lead to permanent loss of capacity, irreversible damage to the internal structure, and in severe cases may even cause battery failure or safety accidents.

[0003] Many scholars have conducted a lot of research on different battery types and over-discharge depths. The electrochemical behavior of batteries with positive electrode materials such as nickel cobalt manganese oxide, lithium iron phosphate, lithium cobalt oxide and lithium manganese oxide during over-discharge was observed by neutron diffraction, scanning electron microscopy (SEM) and X-ray diffraction (XRD). Although the over-discharge behavior of lithium-ion batteries at different depths was compared in the study, and the changes in battery ions and materials after deep over-discharge were analyzed, in actual applications, slight over-discharge is more common and cannot be ignored. In actual operation, lithium-ion batteries are often in a state of slight abuse due to factors such as consistency differences between battery modules, potential failures of the battery management system (BMS), and fluctuations in external environmental conditions. Although the battery can still work normally in this state, the long-term accumulated damage will cause significant damage to the overall health of the battery. In addition, with the increasingly obvious trend of global warming, the temperature changes of lithium-ion batteries during charging and discharging are particularly drastic under extreme temperature conditions. The coupling effect between slight over-discharge and ambient temperature will accelerate the decomposition rate of chemical substances inside the battery, and may cause stress changes in the internal structure, further deteriorating the health of the battery. The combined effect of these factors has brought unprecedented challenges to battery operation safety, and the importance of lithium-ion battery over-discharge detection is self-evident.

[0004] At present, over-discharge detection of lithium-ion batteries mainly relies on external parameters, including voltage, current, temperature and integrated monitoring of the battery management system. However, this method has a certain lag and is difficult to reflect subtle changes inside the battery in a timely manner. In addition, the electrochemical impedance spectroscopy, optical fiber sensing technology, X-ray tomography technology, chemical sensors, scanning electron microscopes and other equipment used in traditional internal detection methods are cumbersome to operate, and even require disassembly of the battery, resulting in damage to the battery structure and performance, or customized assembly during battery manufacturing, which is difficult to promote in practical applications. Therefore, the above methods are difficult to cope with the common detection needs of batteries, especially in dealing with complex and changeable battery applications.

[0005] Ultrasonic testing technology has been gradually applied to the battery field in recent years. It can not only be used to evaluate battery power, health and other indicators, but its sensitive testing characteristics can also evaluate battery abuse. Studies have shown that the physical properties of ultrasound can effectively characterize the chemical and structural changes of batteries during the charging and discharging process. Based on this principle, ultrasonic testing technology has important application potential in the optimization of lithium battery management.

[0006] Based on domestic and foreign detection methods, in order to improve battery safety, this application proposes a battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition, which can combine the internal detection advantages of ultrasonic signals and use abnormal detection technology to identify battery warning status. This method will be able to establish a high-precision, high-stability engineering detection capability for lithium-ion batteries, thereby overcoming the limitations of existing technologies and meeting the needs of future battery safety testing. Summary of the invention

[0007] The purpose of the present invention is to provide a battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition, which solves the discharge detection and estimation problems of lithium-ion batteries under complex conditions by combining multi-dimensional ultrasonic features, principal component analysis and anomaly detection algorithms. Through systematic data collection and feature extraction, an innovative, efficient and reliable solution is provided for the status assessment of lithium-ion batteries.

[0008] To achieve the above object, the present invention provides the following solutions: The present invention provides a battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition, comprising the following steps: S1. Collecting echo data of the bottom surface of the lithium-ion battery during use by ultrasonic equipment; S2. Preprocessing the collected battery bottom echo data; S3. Perform feature engineering on the preprocessed battery bottom echo data to extract features valuable for battery discharge evaluation; S4. Build and train a sliding window isolation forest detection model based on time series; S5. Screening the optimal model; S6. Use the isolation forest detection model to evaluate the discharge degree and over-discharge performance of lithium-ion batteries.

[0009] Preferably, step S1 is specifically: collecting the battery bottom surface echo data during the use of the lithium-ion battery through the ultrasonic acquisition experimental platform to obtain a data set ,in Representative A sampled waveform, is the current total number of samples, and the collected data is bandpass filtered.

[0010] Preferably, the ultrasonic acquisition experimental platform comprises a battery charging and discharging module, an ultrasonic detection module and a temperature control module, and the battery charging and discharging module, the ultrasonic detection module and the temperature control module all establish communication with a host computer.

[0011] Preferably, step S2 specifically includes: processing the collected battery bottom surface echo data by discrete wavelet transform and aligning the data scale. The discrete wavelet transform formula is:

[0012] in, is the wavelet coefficient, is the original signal, is the wavelet coefficient, and are the scaling factor and the translation factor respectively.

[0013] Preferably, step S3 comprises: S31. Obtain the corresponding time domain envelope and frequency domain signal by Hilbert transform and Fourier transform of the data set; S32. Extract the time-frequency domain features related to the changes in the internal structure of the battery based on the sampling points to obtain a feature set ,in is the number of features; S33. Apply principal component analysis technology to further extract the most effective features from the feature set and integrate these feature data.

[0014] Preferably, in step S32, the time-frequency domain features include the time-domain mean, the time-domain waveform index, the time-domain kurtosis index, the frequency-domain mean, the frequency-domain skewness, and the frequency-domain spectrum fluctuation degree.

[0015] Preferably, step S33 includes: S331. Through kernel function Calculate the inner product between multi-dimensional feature sample sets, process the feature data with the same trend together, and obtain the kernel matrix , where each element Features and Inner product after mapping high-dimensional space; S332. Through the formula Center the kernel matrix formed after the inner product, where For one A column vector of all ones, For one Column vector, representing the mean of each column, For one Row vector, representing the mean of each row; S333. The centralized core matrix Perform eigenvalue decomposition: , where the eigenvector The principal component of the corresponding feature set in the feature space; S334. According to the size of the eigenvalue, the original feature data is projected to the front Principal components: , draw the characteristic value curve, find the characteristic value position with obvious drop, and make corrections according to the characteristic trend of the acoustic bottom wave. Assign values ​​to obtain the feature set after dimensionality reduction .

[0016] Preferably, step S4 comprises: S41. The dimension-reduced ultrasound feature set is used to construct feature vectors for each column of data based on the sliding window to obtain ,in, is the starting position of the window, is the window length; S42. Calculate the abnormality score of the constructed feature vector: ,in, For data points The average path length in the isolation forest, is the normalization constant, ; S43. Set the forest pollution rate for the isolated forest detection model and increase the model training weight of the feature change rate.

[0017] Preferably, in step S41, a dynamic window method is adopted, and the minimum and maximum values ​​of the window are 3 and 60 respectively.

[0018] Preferably, in step S43, during training, the forest determines the abnormal score threshold according to the contamination rate. , the threshold is set to the quantile of the anomaly score, if , then determine As an outlier, the pollution rate of the forest is set to 0.2.

[0019] Compared with the prior art, the present invention has achieved the following beneficial technical effects: The present invention provides a battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition, comprising the following steps: S1. collecting battery bottom echo data during the use of lithium-ion batteries through ultrasonic equipment; S2. pre-processing the collected battery bottom echo data; S3. performing feature engineering on the pre-processed battery bottom echo data to extract features valuable for battery discharge evaluation; S4. constructing and training a sliding window isolation forest detection model based on time series; S5. selecting the optimal model; S6. using the isolation forest detection model to evaluate the discharge degree and over-discharge performance of lithium-ion batteries; the above method can be used to reflect the internal changes of the battery, and the trained model can be used in other online batteries, and lithium batteries with different positive electrode materials and structures all have this bottom wave characteristic property, so it has high universality and can provide technical support for avoiding battery damage. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 A flowchart of a battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition provided by the present invention; Figure 2 This is a schematic diagram of the test results of the method of the present application under standard working conditions with a nuclear order of d=2; Figure 3 This is a schematic diagram of the test results of the method of the present application under standard working conditions with a nuclear order of d=3; Figure 4 This is a schematic diagram of the test results of the method of the present application under standard working conditions with a nuclear order of d=4; Figure 5 This is a schematic diagram of the test results of the nuclear order d=5 under standard working conditions using the method of the present application. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] The purpose of the present invention is to provide a battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition to solve the problems existing in the prior art.

[0024] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Embodiment 1: This embodiment provides a battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition, such as Figure 1 As shown, the following steps are included: S1. Collect the echo data of the bottom surface of the lithium-ion battery during use by ultrasonic equipment; specifically: collect the echo data of the bottom surface of the lithium-ion battery during use by ultrasonic acquisition experimental platform to obtain a data set ,in Representative A sampled waveform, is the total number of current samples, and performs bandpass filtering on the collected data; The ultrasonic acquisition experimental platform includes a battery charging and discharging module, an ultrasonic detection module and a temperature control module, and the battery charging and discharging module, the ultrasonic detection module and the temperature control module all establish communication with the host computer; In this embodiment, the battery charging and discharging module adopts a high-precision Xinwei battery tester (CT-4008-5V30A-NA), and the battery charging and discharging parameters and test steps are set by the host computer; the ultrasonic detection module uses an Olympus piezoelectric probe (Olympus), and the probe is placed in the center of the battery, and cooperates with the M2M ultrasonic phased array transmitter and receiver and the Multi2000 ultrasonic acquisition software to complete the excitation and reception of ultrasonic reflected waves; in order to avoid the influence of the external ambient temperature of the battery, the acquisition is carried out in a constant temperature box; the American magnetic flux silicon-free coupling agent (UltragelⅡ) is used to ensure the transmission efficiency of the ultrasonic wave and ensure the continuity of the measurement work; in addition, a 500g weight is placed above the probe to apply a constant pressure to ensure that the ultrasonic wave can fully penetrate the lithium battery and reduce experimental errors; due to the sensitivity and real-time nature of ultrasonic signal acquisition, the data points can be saved in real time in the host computer; The bandpass filtering process is realized based on the second-order Butterworth bandpass filter designed by Multi2000 software in the host computer. The probe frequency is 1Mhz. The corresponding bandpass filter can be set based on the probe frequency, and the frequency range is 0.84 MHz and 1.65 MHz. Ultrasonic time-frequency domain data is based on filtered data, and then the data processing of missing values ​​is completed, which is realized by DSP signal processing software developed based on Matlab-GUI; S2. Preprocess the collected battery bottom echo data; specifically: process the collected battery bottom echo data through discrete wavelet transform and align the data scale. The discrete wavelet transform formula is:

[0026] in, is the wavelet coefficient, is the original signal, is the wavelet coefficient, and are the scaling factor and the translation factor respectively; S3. Perform feature engineering on the pre-processed battery bottom echo data to extract features valuable for battery discharge assessment; including: S31. Obtain the corresponding time domain envelope and frequency domain signal through Hilbert transform and Fourier transform of the data set, which can be realized based on the script developed by Matlab; The time-frequency domain analysis is based on the superposition of ultrasonic waveforms at each sampling moment during the battery discharge process, combined with the feature extraction interval obtained by Pearson correlation coefficient analysis. The formula is:

[0027] in, and There are two sets of data. and Especially their means, is the data size; S32. Extract the time-frequency domain features related to the changes in the internal structure of the battery based on the sampling points to obtain a feature set , where N is the number of features; the features show two different trends during the charge and discharge process: one is first decreasing and then increasing, and the other is the opposite. During the extraction process, multidimensional features can be extracted according to the data processing method, including time domain mean, time domain waveform index, time domain kurtosis index, frequency domain mean, frequency domain skewness, and frequency domain spectrum fluctuation degree; see Table 1 for details: Table 1 Bottom wave characteristic parameters:

[0028] in, represents the bottom wave signal in the time domain, is the spectrum of the signal after Fourier transform; For time, is the frequency; is the mean value of the signal strength in the time domain, is the mean value of the signal strength in the frequency domain; is the length of the time domain signal, is the length of the spectrum, is the standard deviation in the frequency domain; S33. Apply principal component analysis technology to further extract the most effective features from the feature set and integrate these feature data; The bottom wave feature can reflect the internal change trend of the battery discharge state, but after batch feature extraction, it is found that it has different overall trends and there are worthless interference features. In order to better extract features and retain abnormal change trends, the kernel principal component analysis method is selected for dimensionality reduction; Kernel principal component analysis (KPCA) is a nonlinear extension of principal component analysis (PCA). PCA is a linear dimensionality reduction technology that reduces the dimension of data by finding the direction of the maximum variance of the data, while trying to retain most of the original data information. However, the ultrasonic feature in the battery discharge process is a complex nonlinear sequence data. In order to better retain feature information and reduce the impact of noise, KPCA maps the data to a high-dimensional feature space by introducing a kernel function. PCA analysis in the high-dimensional space can better extract the time series changes under internal influences and retain the abnormal trend of battery discharge to the greatest extent; it specifically includes: S331. Through kernel function Calculate the inner product between multi-dimensional feature sample sets, process the feature data with the same trend together, and obtain the kernel matrix Each element Features and Inner product after mapping high-dimensional space; S332. Through the formula Center the kernel matrix formed after the inner product, where For one A column vector of all ones, For one Column vector, representing the mean of each column, For one Row vector, representing the mean of each row; S333. The centralized core matrix Perform eigenvalue decomposition: , where the eigenvector The principal component of the corresponding feature set in the feature space; S334. According to the size of the eigenvalue, the original feature data is projected to the front Principal components: , draw the characteristic value curve, find the characteristic value position with obvious drop, and make corrections according to the characteristic trend of the acoustic bottom wave. Assign values ​​to obtain the feature set after dimensionality reduction ; S4. Construct and train a sliding window isolation forest detection model based on time series; Ultrasonic time series features have the ability to distinguish different stages of battery discharge, but in order to capture and detect abnormal changes in features, traditional anomaly detection algorithms such as statistical Z-score, density-based LOF and distance-based KNN need to assume that the data obeys a certain characteristic distribution, which cannot capture the nonlinear relationship of the data well, and also have obvious disadvantages and time-consuming problems in processing real-time and large-scale data; Isolation Forest (IF) uses decision tree characteristics to identify anomalies without prior knowledge of data distribution, and for real-time data streams such as ultrasonic data, with the help of low-cost computing overhead and sliding window design, it can well match data characteristics, not only can be used to detect outliers in classification applications, but also can be used to determine local abnormal trends in time series processes; Specifically include: S41. The dimension-reduced ultrasound feature set is used to construct feature vectors for each column of data based on the sliding window to obtain ,in, is the starting position of the window, Window length; This embodiment adopts a dynamic window method, and the minimum and maximum values ​​of the window are 3 and 60 respectively; S42. Calculate the abnormality score of the constructed feature vector: ,in, Data Points The average path length in the isolation forest, is the normalization constant, ; S43. Set the forest pollution rate for the isolated forest detection model and increase the model training weight of the feature change rate. During training, the forest will determine the anomaly score threshold based on the pollution rate. , the threshold will be set to the quantile of the anomaly score to ensure that the "contaminated" proportion of samples are classified as anomalies. The threshold will be automatically adjusted during training to ensure that the model meets this proportion. , then determine is an outlier, and the pollution rate of the forest in this embodiment is set to 0.2; In processing time series data, the time complexity of Isolation Forest is close to linear. For large-scale time series data sets, the calculation speed is fast and it is suitable for real-time or near-real-time application scenarios. In addition, Isolation Forest requires fewer parameters, which reduces the workload of parameter adjustment and makes the model easier to use. It has no assumptions about data distribution and can handle non-Gaussian distribution data well. This is especially important for time series data because such data has complex dynamic characteristics and nonlinear relationships. S5. Screening the optimal model: In order to improve the effectiveness of the model, a multi-model screening strategy was adopted to select the model with the prediction effect closest to the change node, and adjust the model parameters according to the results to obtain the optimal model; S6. Use the isolation forest detection model to evaluate the discharge degree and over-discharge performance of lithium-ion batteries.

[0029] Please refer to Figure 2-Figure 5 , taking 0.5C discharge and 25℃ ambient temperature as an example, Figure 2-Figure 5 The dimensionality reduction feature detection effect with a polynomial kernel order of 2 to 5 is demonstrated. Compared with electrical parameters such as voltage, the method of the present application has a more stable and sensitive detection effect. Even if the trends and features obtained by dimensionality reduction are different, better results can be achieved through isolated forest anomaly segmentation. The results show that this method can effectively identify over-discharge warning and abuse warning nodes with a discharge depth of 80% during battery discharge, and its effectiveness and stability are verified under multiple factors. The method of the present application can also be extended to other online batteries, has high universality, and can provide a non-invasive solution for lithium battery safety monitoring and battery performance assurance.

[0030] Compared with conventional methods, firstly, the present application does not use commonly used electrical parameters as characteristic factors for discharge detection and estimation, but adopts non-destructive, non-destructive ultrasonic reflection wave signals, and the monitoring performance is significantly more forward-looking than that of electrical parameters; secondly, the characteristic indicators adopted in the present application are also different from the acoustic indicators (TOF, SA, etc.) commonly used in previous literature, but are more stable and more diverse acoustic indicators in the field of ultrasonic detection obtained by further selection through kernel principal component analysis; finally, the present application adopts the strategy of combining kernel principal component analysis with isolation forest, takes advantage of the advantage of isolation forest in detecting local anomalies, and innovatively uses isolation forest for anomaly detection and analysis of time series, achieving unexpected technical effects.

[0031] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition, characterized in that: The following steps are involved: S1. Collecting echo data of the bottom surface of the lithium-ion battery during use by ultrasonic equipment; S2. Preprocessing the collected battery bottom echo data; S3. Perform feature engineering on the preprocessed battery bottom echo data to extract features valuable for battery discharge evaluation; S4. Build and train a sliding window isolation forest detection model based on time series; S5. Screening the optimal model; S6. Use the isolation forest detection model to evaluate the discharge degree and over-discharge performance of lithium-ion batteries.

2. The battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition according to claim 1 is characterized in that: Step S1 is specifically: using the ultrasonic acquisition experimental platform to collect the battery bottom echo data during the use of the lithium-ion battery to obtain a data set ,in Representative sampling waveforms, m is the total number of current samples, and the collected data is bandpass filtered.

3. The battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition according to claim 2 is characterized in that: The ultrasonic acquisition experimental platform includes a battery charging and discharging module, an ultrasonic detection module and a temperature control module. The battery charging and discharging module, the ultrasonic detection module and the temperature control module all establish communication with a host computer.

4. The battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition according to claim 2 is characterized in that: Step S2 is specifically: processing the collected battery bottom echo data by discrete wavelet transform and aligning the data scale. The discrete wavelet transform formula is: in, is the wavelet coefficient, is the original signal, is the wavelet coefficient, and The scaling factor and translation factor are respectively, is the total number of sampled waveforms. Since the count starts from 0, .

5. The battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition according to claim 3 is characterized in that: Step S3 includes: S31. Obtain the corresponding time domain envelope and frequency domain signal by Hilbert transform and Fourier transform of the data set; S32. Extract the time-frequency domain features related to the changes in the internal structure of the battery based on the sampling points to obtain a feature set , where N is the number of features; S33. Apply principal component analysis technology to further extract the most effective features from the feature set and integrate these feature data.

6. The battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition according to claim 5 is characterized in that: In step S32, the time-frequency domain features include the time-domain mean, the time-domain waveform index, the time-domain kurtosis index, the frequency-domain mean, the frequency-domain skewness, and the frequency-domain spectrum fluctuation degree.

7. The battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition according to claim 5 is characterized in that: Step S33 includes: S331. Through kernel function Calculate the inner product between multidimensional feature sample sets, where and is a feature sample, and It is a nonlinear mapping function that maps data to a high-dimensional feature space. Represents the inner product in the high-dimensional feature space, and processes the feature data with the same trend together to obtain the kernel matrix , where each element Features and Inner product after mapping high-dimensional space; S332. Through the formula Center the kernel matrix formed after the inner product, where For one A column vector of all ones, For one Column vector, representing the mean of each column, For one Row vector, representing the mean of each row; S333. The centralized core matrix Perform eigenvalue decomposition: ,in, is the eigenvalue, eigenvector The principal component of the corresponding feature set in the feature space; S334. According to the size of the eigenvalue, the original feature data is projected to the front Principal components: ,in , is the The eigenvector components, draw the characteristic value curve, find the position of the characteristic value with obvious decrease, and Assign values ​​to obtain the feature set after dimensionality reduction .

8. The battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition according to claim 7 is characterized in that: Step S4 includes: S41. The dimension-reduced ultrasound feature set is used to construct feature vectors for each column of data based on the sliding window to obtain ,in, is the starting position of the window, is the window length; S42. Calculate the abnormality score of the constructed feature vector: ,in, Data Points The average path length in the isolation forest, is the normalization constant; S43. Set the forest pollution rate for the isolated forest detection model and increase the model training weight of the feature change rate.

9. The battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition according to claim 8 is characterized in that: In step S41, a dynamic window method is adopted, and the minimum and maximum values ​​of the window are 3 and 60 respectively.

10. The battery over-discharge monitoring and early warning method based on ultrasonic bottom wave feature recognition according to claim 8 is characterized in that: In step S43, during training, the forest determines the anomaly score threshold based on the contamination rate , the threshold is set to the quantile of the anomaly score, if , then determine As an outlier, the pollution rate of the forest is set to 0.2.

Images