A boron-11 electron gas analysis method for trace impurity identification

Abstract

The present application relates to the technical field of data processing, and more particularly to a boron-11 electron gas analysis method for trace impurity identification, which obtains historical time sequence data sequences of the concentration of each gas phase element in a container storing boron-11 electron gas at a current monitoring time; each historical time sequence data sequence is periodically divided to obtain suspected abnormal data in each data segment; according to the suspected abnormal data in the recent time sequence data sequence of the concentration of each gas phase element, the recent monitoring data distribution change abnormality degree and the multiple category correlation change abnormality degree of the concentration of each gas phase element are obtained, and the adaptive concentration threshold value at the current monitoring time is correspondingly obtained; and the boron-11 electron gas at the current monitoring time is monitored for trace impurity abnormality according to the adaptive concentration threshold value corresponding to the concentration of each gas phase element, so that the accuracy of trace impurity abnormality monitoring is improved through dynamic adjustment of the fixed threshold value.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a boron-11 electron gas analysis method for the identification of trace impurities. Background Technology

[0002] Boron-11 electronic gas typically refers to the highly electronegative boron difluoride gas. Most molecules in naturally occurring boron trifluoride gas contain boron-11 (BF3). Boron trifluoride is an important specialty gas in the electronics industry, widely used in critical processes such as semiconductor manufacturing, ion implantation, and chemical vapor deposition. Because it comes into direct contact with high-precision devices during application, the purity and impurity content of the gas directly affect chip performance and yield. Furthermore, boron trifluoride is highly corrosive and toxic. The presence of moisture, oxygen, or other impurities in the gas can cause equipment corrosion or process abnormalities, and even threaten operator safety. Therefore, the identification, detection, and analysis of trace impurities in boron-11 electronic gas are crucial.

[0003] Current technologies for analyzing and monitoring boron-11 electron gas typically employ a fixed threshold method, which involves monitoring the concentration of trace impurity elements to see if it exceeds a fixed concentration threshold. However, this method has several drawbacks. First, it lacks timeliness, triggering only after the impurity concentration has reached the set threshold, while many sensitive processes (such as plasma etching) may already be affected below the threshold. Second, it is susceptible to noise interference, causing the abnormal data detected by the fixed threshold to be noise data, thus reducing monitoring accuracy. In other words, using a high fixed threshold can lead to the failure to detect abnormal concentrations in sensitive processes in a timely manner, while using a low fixed threshold can result in false detections due to noise.

[0004] Therefore, improving the timeliness and accuracy of analyzing and monitoring boron-11 electron gas using the fixed threshold method has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a boron-11 electron gas analysis method for the identification of trace impurities, in order to solve the problem of improving the timeliness and accuracy of boron-11 electron gas analysis and monitoring using a fixed threshold method.

[0006] This invention provides a boron-11 electron gas analysis method for identifying trace impurities, the method comprising the following steps:

[0007] Monitor the concentrations of at least two gaseous elements in a container holding boron-11 electron gas, and obtain a historical time series data sequence of the concentration of each gaseous element at the current monitoring time;

[0008] Each of the historical time-series data sequences is periodically divided to obtain multiple data segments, and suspected abnormal data in each data segment is obtained based on the data changes in each data segment.

[0009] For any gaseous element concentration, a preset number of recent data segments are obtained from the data segment corresponding to the gaseous element concentration to form a recent time series data sequence. Based on the distribution of suspected abnormal data, the distribution of normal data, and the number of suspected abnormal data in each recent data segment in the recent time series data sequence, the degree of abnormality in the distribution of recent monitoring data is obtained.

[0010] Based on the time difference between each suspected abnormal data in the recent time series data sequence and its corresponding nearest neighbor data in the historical time series data sequences of other gaseous element concentrations, the degree of abnormality of multiple types of associated changes is obtained; by combining the degree of abnormality of recent monitoring data distribution changes and the degree of abnormality of multiple types of associated changes, the adaptive concentration threshold of any gaseous element concentration at the current monitoring time is obtained.

[0011] Based on the adaptive concentration threshold corresponding to the concentration of each gaseous element, the boron-11 electron gas at the current monitoring time is monitored for trace impurities.

[0012] Preferably, the step of periodically dividing each of the historical time-series data sequences to obtain multiple data segments includes:

[0013] The periodic components of each historical time series data sequence are extracted using Fourier transform to obtain the principal period length of each historical time series data sequence. Based on the principal period length of each historical time series data sequence, each historical time series data sequence is adaptively divided into multiple data segments.

[0014] Preferably, obtaining suspected abnormal data in each data segment based on data changes in each data segment includes:

[0015] For any data in any data segment, based on the position of the data in the data segment, data at the same position in other data segments belonging to the same historical time series data sequence as the data segment are obtained and recorded as reference data. The absolute value of the difference between the data and each reference data is calculated, and the mean of the absolute value of the difference is recorded as the degree of difference of the data.

[0016] Obtain the difference degree of each data in the historical time series data sequence to which any data segment belongs, and obtain the maximum difference degree and the minimum difference degree. Use the maximum difference degree and the minimum difference degree to normalize the difference degree of any data to obtain the suspected anomaly degree of any data.

[0017] If the suspected abnormality level of any data exceeds a preset suspected abnormality level threshold, then the data is determined to be suspected abnormal data.

[0018] Preferably, the step of determining the degree of abnormality in the distribution of recent monitoring data based on the distribution of suspected abnormal data, the distribution of normal data, and the number of suspected abnormal data in each recent data segment includes:

[0019] The absolute value of the difference in the number of suspected anomalous data between the first and last recent data segments in the recent time series data sequence is calculated and denoted as the anomalous change feature value. The absolute values ​​of the difference in the number of suspected anomalous data between every two adjacent recent data segments in the recent time series data sequence are accumulated to obtain the overall anomalous change feature value. The product of the anomalous change feature value and the overall anomalous change feature value is normalized to obtain the degree of periodic anomalous fluctuation.

[0020] For any recent data segment in the recent time series data sequence, perform one-dimensional clustering on the recent data segment to obtain a suspected abnormal cluster composed of suspected abnormal data and a normal cluster composed of normal data. Based on the data differences in each suspected abnormal cluster, obtain the abnormal data distribution dispersion and the mean cluster interval distance of the any recent data segment.

[0021] By combining the dispersion of abnormal data distribution and the mean clustering interval distance of each recent data segment in the recent time series data sequence, the stability of abnormal distribution is obtained; based on the sum of the periodic abnormal fluctuation degree and the stability of abnormal distribution, the degree of abnormality of recent monitoring data distribution changes is obtained.

[0022] Preferably, the step of obtaining the anomalous data distribution dispersion and the mean cluster interval distance of any recent data segment based on the data differences in each suspected anomalous cluster includes:

[0023] For any data point in any suspected anomaly cluster, calculate the absolute value of the data difference between the data point and other data points in the suspected anomaly cluster to obtain the minimum absolute value of the data difference. Then, sum the minimum absolute values ​​of the data differences for each data point in the suspected anomaly cluster to obtain a first accumulated value. Finally, sum the first accumulated values ​​for each suspected anomaly cluster in any recent data segment to obtain the anomalous data distribution dispersion of the recent data segment.

[0024] Obtain the absolute value of the difference between the cluster center of each suspected abnormal cluster and the cluster center of each normal cluster, and record it as the interval distance. Use the average of all interval distances as the average cluster interval distance of any recent data segment.

[0025] Preferably, the step of obtaining the stability of the abnormal distribution by combining the dispersion of the abnormal data distribution and the mean clustering interval distance of each recent data segment in the recent time series data sequence includes:

[0026] The number of suspected anomalous clusters in each recent data segment of the recent time series data sequence is used as a weight to perform a weighted summation of the anomalous data distribution dispersion of all recent data segments, and the standard deviation of the mean cluster interval distance of all recent data segments is calculated. The negative of the product of the weighted summation and the standard deviation is used as the independent variable of an exponential function with the natural constant as the base to obtain the stability of the anomalous distribution.

[0027] Preferably, the step of obtaining the degree of various types of correlated changes based on the time difference between each suspected abnormal data in the recent time series data sequence and its corresponding nearest neighbor data in the historical time series data sequences of other gaseous element concentrations includes:

[0028] For any suspected abnormal data in the recent time series data sequence, the time of the suspected abnormal data is obtained, the time interval between the time of each suspected abnormal data in the historical time series data sequence of any other gaseous element concentration and the time of the suspected abnormal data is calculated, the minimum time interval corresponding to the concentration of any other gaseous element is obtained, and the minimum time interval corresponding to the concentration of each other gaseous element is accumulated to obtain the accumulated value of the time interval of the suspected abnormal data.

[0029] The time intervals of each suspected abnormal data in the recent time series are accumulated to obtain the overall time difference. The negative of the overall time difference is used as the independent variable of an exponential function with the natural constant as the base to obtain the degree of abnormality of various types of correlation changes.

[0030] Preferably, the step of obtaining the adaptive concentration threshold of any gaseous element at the current monitoring time by comprehensively considering the degree of abnormality in the distribution of recent monitoring data and the degree of abnormality in various types of correlations includes:

[0031] Based on the degree of abnormality in the distribution of recent monitoring data and the degree of abnormality in the correlation of multiple types, the degree of abnormality in recent monitoring data is obtained. Using the degree of abnormality in recent monitoring data, the fixed concentration threshold of any gaseous element is adjusted to obtain the adaptive concentration threshold of any gaseous element at the current monitoring time.

[0032] Preferably, the step of obtaining the degree of anomaly in recent monitoring data based on the degree of anomaly in the distribution of recent monitoring data and the degree of anomaly in the correlation of multiple categories includes:

[0033] The degree of anomaly in recent monitoring data is obtained by averaging the degree of anomaly in the distribution of half of the recent monitoring data with the degree of anomaly in the correlation of multiple categories.

[0034] Preferably, the step of adjusting the fixed concentration threshold of any gaseous element concentration based on the degree of anomaly in recent monitoring data to obtain an adaptive concentration threshold of any gaseous element concentration at the current monitoring time includes:

[0035] If the degree of abnormality of recent monitoring data is less than or equal to the preset threshold for the degree of abnormality of recent monitoring data, then the difference between constant 1 and the degree of abnormality of recent monitoring data is obtained, the product of the preset adjustment coefficient and the difference is calculated, the sum of constant 1 and the product is used as the adjustment coefficient, and the product of the fixed concentration threshold of any gaseous element concentration and the adjustment coefficient is used as the adaptive concentration threshold of any gaseous element concentration at the current monitoring time.

[0036] If the degree of abnormality in recent monitoring data is greater than the preset threshold for the degree of abnormality in recent monitoring data, then the product of the preset adjustment coefficient and the degree of abnormality in recent monitoring data is obtained, and the difference between the constant 1 and the product is used as the adjustment coefficient. The product of the fixed concentration threshold of any gaseous element concentration and the adjustment coefficient is used as the adaptive concentration threshold of any gaseous element concentration at the current monitoring time.

[0037] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0038] This invention monitors the concentrations of at least two gaseous elements in a container storing boron-11 electron gas, obtaining historical time-series data sequences of each gaseous element concentration at the current monitoring time. Each historical time-series data sequence is periodically divided into multiple data segments. Based on data changes within each data segment, suspected abnormal data is obtained for each data segment. For any given gaseous element concentration, a preset number of recent data segments are obtained from the data segment corresponding to that gaseous element concentration to form a recent time-series data sequence. Based on the distribution of suspected abnormal data, the distribution of normal data, and the number of suspected abnormal data segments in each recent data sequence, the degree of abnormality in the distribution of recent monitoring data is obtained. Based on the time difference between each suspected abnormal data in the recent time-series data sequence and its nearest neighbor data in the historical time-series data sequences of other gaseous element concentrations, the degree of abnormality of various types of correlated changes is obtained. Combining the degree of abnormality in the distribution of recent monitoring data and the degree of abnormality of various types of correlated changes, an adaptive concentration threshold for the given gaseous element concentration at the current monitoring time is obtained. Based on the adaptive concentration threshold corresponding to each gaseous element concentration, trace impurity anomaly monitoring is performed on the boron-11 electron gas at the current monitoring time. By analyzing the distribution characteristics of recent historical data, the degree of anomaly in recent monitoring data that characterizes the abnormal concentration of trace impurities in boron-11 electron gas is obtained. This allows for dynamic adjustment of fixed concentration thresholds. When noise may exist at the current monitoring time, the concentration threshold is increased; when a real anomaly may exist at the current monitoring time, the concentration threshold is decreased. This enables anomalies below the threshold to be detected in advance, while preventing more noise from being misidentified as anomalies. This improves the timeliness and accuracy of trace impurity analysis and monitoring of boron-11 electron gas. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a boron-11 electron gas analysis method for identifying trace impurities, provided in Embodiment 1 of the present invention. Detailed Implementation

[0041] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0042] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0043] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0044] See Figure 1 This is a flowchart of a boron-11 electron gas analysis method for identifying trace impurities, provided in Embodiment 1 of the present invention. Figure 1 As shown, the method may include:

[0045] Step S101 involves monitoring the concentrations of at least two gaseous elements within the container holding boron-11 electron gas, obtaining a historical time-series data sequence of the concentration of each gaseous element at the current monitoring moment. When monitoring and analyzing trace impurities in boron-11 electron gas, a fixed threshold is primarily used to monitor whether the impurities exceed the limit. However, a fixed threshold may lead to poor monitoring timeliness and the monitoring accuracy may be affected by noise interference. Therefore, it is necessary to improve the timeliness and accuracy of fixed threshold monitoring. Thus, in this embodiment of the invention, an RGA device is used to monitor the concentrations of each gaseous element within the container holding boron-11 electron gas, including... Hydrocarbons , Inert gases such as , And so on, and the dimensions of the concentration of each gaseous element are all . The system monitors once per second, storing the concentrations of various gaseous elements in a data terminal. Each gaseous element concentration is stored as a time-series sequence. Each time new data is acquired, it is inserted into the corresponding time-series sequence of the gaseous element concentration. This allows for the generation of historical time-series data sequences for each gaseous element concentration at the current monitoring moment. These sequences are used to dynamically adjust fixed thresholds, thereby improving the accuracy of monitoring the concentrations of various gaseous elements at the current monitoring moment. It should be noted that each historical time-series data sequence does not include monitoring data from the current monitoring moment.

[0046] Step S102: Periodically divide each historical time series data sequence to obtain multiple data segments. Based on the data changes in each data segment, obtain the suspected abnormal data in each data segment.

[0047] When boron trifluoride-11 is supplied stably, the pressure regulator / flow controller undergoes an adsorption-desorption process, causing the concentration monitoring values ​​of all gaseous elements to fluctuate regularly, forming a certain periodicity. Therefore, it is not possible to directly analyze abnormal data based on data fluctuations. Instead, each historical time series data sequence is first divided into periods to eliminate the influence of periodic interference. Then, based on the data changes in each data segment, potential data with suspected anomalies are analyzed.

[0048] The periodic components of each historical time-series data sequence are extracted using Fourier transform to obtain the principal period length of each sequence. Based on the principal period length, each historical time-series data sequence is adaptively divided into multiple data segments, with each segment corresponding to one period. It should be noted that using Fourier transform for periodic component extraction is existing technology and will not be elaborated upon here.

[0049] Because boron-11 electron gas requires extremely high purity, its similarity across multiple cycles is generally high. However, when a seal leak leads to impurities during the gas supply process, the data at the gas supply time point will differ from the historical data, resulting in low similarity. Slight seal leaks cause small changes in impurity concentration, which are not easily detected by fixed thresholds, leading to a risk of increased seal leaks in the future. Therefore, this invention analyzes suspected abnormal data in each data segment of the historical time-series data sequence for each gaseous element concentration based on node differences within historical cycles, in order to predict potential anomalies in advance.

[0050] Therefore, taking a historical time-series data sequence of gaseous element concentrations as an example, for any data segment in this historical time-series data sequence, by analyzing the data differences between the data within that data segment and the data at the same time position in other data segments, the degree of suspected anomaly for each data point can be analyzed. This is because, under normal circumstances, the similarity of data changes within a period is high, and the similarity of data at each corresponding time position is also high. However, when data anomalies occur, the similarity of data at the time position corresponding to the abnormal data will be low, and normal data accounts for a large proportion of historical data. Therefore, for any data point in any data segment, based on the position of that data point in that data segment, data at the same position in other data segments belonging to the same historical time-series data segment is obtained and recorded as reference data. For example, if the second data point in any data segment is the second data point in another data segment, then the reference data is the second data point in another data segment. The absolute value of the difference between that data point and each of the reference data points is calculated, and the mean of the absolute values ​​of the differences is recorded as the degree of difference for that data point. The formula for calculating the degree of difference for any data point is: in, This represents the difference of the q-th data point in any data segment. This indicates the number of data segments into which a historical time-series data sequence is divided. Indicates the number of data segments other than any other data segment. This indicates the data in the i-th other data segment that is in the same position as the q-th data, which is the reference data. | represents the absolute value symbol.

[0051] It should be noted that, Used to characterize the data difference between two data points at the same location within two periods. This is used to characterize the cumulative difference between the q-th data point and the data at the same position in each period. The larger the cumulative difference, the greater the difference in periodic changes between the q-th data point and most normal data, and the more likely the q-th data point is to have an abnormal trend risk, and the greater its degree of difference.

[0052] Similarly, the difference degree of each data point in the historical time-series data sequence to which any data segment belongs is obtained, yielding the maximum and minimum difference degrees. The difference degree of any data point is then normalized using the maximum and minimum difference degrees to obtain the suspected anomaly level of that data point. The formula for calculating the suspected anomaly level of any data point is as follows: in, This indicates the degree of suspected anomaly of the q-th data point in any data segment. This represents the difference of the q-th data point in any data segment. Indicates the minimum degree of difference. This represents the maximum degree of difference. Since noise, or potentially abnormal data, is inevitable during data monitoring and collection, after normalizing all data to its maximum and minimum values, data with potential anomalies (including actual abnormal data and noise) and normal data without potential anomalies are divided into two ends of the value range. Therefore, the middle value of the value range, 0.5, is taken as the threshold for the degree of potential anomaly. If the degree of potential anomaly of any data exceeds the preset threshold, then that data is determined to be potentially abnormal, indicating that it possesses potential anomaly characteristics.

[0053] Similarly, suspected anomalous data in each data segment of each historical time series data sequence is obtained, and all suspected anomalous data in each historical time series data sequence is also identified.

[0054] Step S103: For any gaseous element concentration, obtain a preset number of recent data segments from the data segment corresponding to any gaseous element concentration to form a recent time series data sequence. Based on the distribution of suspected abnormal data, the distribution of normal data, and the number of suspected abnormal data in each recent data segment in the recent time series data sequence, obtain the degree of abnormality in the distribution of recent monitoring data.

[0055] Boron-11 electronic gas is highly corrosive. Even when using corrosion-resistant RGA equipment for monitoring, the lifespan of electronic components can be affected by the strong corrosiveness of boron trifluoride gas, resulting in noise in the monitoring data. This noise can also deviate from normal periodicity. Therefore, it is necessary to further distinguish between noise and true anomalies when monitoring trace impurities in boron-11 electronic gas.

[0056] When genuine anomalies occur (such as leaks), the leak will gradually enlarge due to corrosion from boron trifluoride gas. This leakage will intensify the monitored anomalies, leading to a higher concentration of abnormal trace impurities in the transported gas. However, if the monitoring equipment is simply experiencing lifespan damage and has a certain probability of failure, the monitored anomalies will not exhibit this pattern. Furthermore, the changes in trace impurity concentration caused by genuine anomalies are gradual, resulting in a relatively dense clustering of suspected anomalies across multiple data segments. In contrast, monitoring noise caused by equipment lifespan damage is more random, leading to a more dispersed distribution of suspected anomalies across multiple data segments.

[0057] Therefore, for any gaseous element concentration, the abnormality index of recent monitoring data can be obtained by analyzing the changes and distribution of the proportion of suspected abnormal data in multiple recent historical data segments. Thus, a predetermined number of recent data segments are obtained from the data segments corresponding to any gaseous element concentration and arranged chronologically to form a recent time-series data sequence. Preferably, in this embodiment of the invention, the 10 data segments closest to the current monitoring time are set as recent data segments. Then, based on the distribution of suspected abnormal data, the distribution of normal data, and the number of suspected abnormal data in each recent data segment in the recent time-series data sequence, the degree of abnormality in the distribution of recent monitoring data for any gaseous element concentration is obtained. The method for obtaining the degree of abnormality in the distribution of recent monitoring data for any gaseous element concentration is as follows:

[0058] The absolute value of the difference in the number of suspected anomalous data between the first and last recent data segments in the recent time series data sequence is calculated and denoted as the anomalous change feature value. The absolute values ​​of the difference in the number of suspected anomalous data between any two adjacent recent data segments in the recent time series data sequence are accumulated to obtain the overall anomalous change feature value. The product of the anomalous change feature value and the overall anomalous change feature value is normalized to obtain the degree of periodic anomalous fluctuation.

[0059] The formula for calculating the degree of abnormal periodic fluctuations is as follows: in, Indicates the degree of abnormal fluctuations in the cycle. Represents the normalization function. This indicates the number of suspected outliers in the last recent data segment of a recent time series data sequence. This indicates the number of suspected outliers in the first recent data segment of a recent time series data sequence. Represents the absolute value symbol. This represents the number of suspected outliers in the i-th recent data segment of a recent time series data sequence. This represents the number of suspected outliers in the (i+1)th recent data segment of a recent time series data sequence. This indicates the number of recent data segments contained in a recent time series data sequence.

[0060] It should be noted that when the difference in the number of suspected anomalous data between adjacent recent data segments is greater, and the difference in the number of suspected anomalous data between the two recent data segments corresponding to the starting point is greater, it indicates that the suspected anomalous data has the characteristic of gradually increasing true anomalousness. The greater the fluctuation difference of suspected anomalous data between corresponding periods, the greater the degree of periodic anomalous fluctuation.

[0061] To analyze the clustering characteristics of suspected abnormal data within recent data segments, this embodiment of the invention employs a K-means clustering algorithm to perform one-dimensional clustering on any recent data segment within the recent time-series data sequence. This results in suspected abnormal clusters composed of suspected abnormal data and normal clusters composed of normal data. Furthermore, based on the data differences within each suspected abnormal cluster, the distribution dispersion of abnormal data and the mean cluster interval distance for any recent data segment are obtained. For any data point within any suspected abnormal cluster, the distance between that data point and the mean cluster interval distance is calculated. The absolute value of the data difference between other data in a suspected anomalous cluster is used to obtain the minimum absolute value of the data difference. The minimum absolute value of the data difference corresponding to each data point in any suspected anomalous cluster is accumulated to obtain a first accumulated value. The first accumulated value of each suspected anomalous cluster in any recent data segment is accumulated to obtain the anomalous data distribution dispersion of that recent data segment. The absolute value of the difference between the cluster center of each suspected anomalous cluster and the cluster center of each normal cluster is obtained and denoted as the interval distance. The average of all interval distances is used as the average cluster interval distance of any recent data segment. .

[0062] The formula for calculating the dispersion of outlier data distribution in any recent data segment is as follows: in, This represents the dispersion of outlier data distribution in the i-th recent data segment. This represents the number of suspected anomalous clusters in the i-th recent data segment. This represents the number of data points within the j-th suspected anomalous cluster in the i-th recent data segment. This represents the absolute value of the minimum data difference corresponding to the z-th data within the j-th suspected anomaly cluster in the i-th recent data segment, which is also the closest distance between the z-th data and all other data in the j-th suspected anomaly cluster. It should be noted that... The smaller the value, the closer the nearest distance of each data point within the j-th suspected anomaly cluster, and the denser the distribution of suspected anomaly data within the j-th suspected anomaly cluster. If the distribution of suspected anomaly data within all suspected anomaly clusters in the i-th recent data segment is denser, the smaller the dispersion of anomaly data distribution in the i-th recent data segment, and the more consistent it is with the characteristics of a true anomaly.

[0063] Similarly, the dispersion of outlier data distribution and the mean cluster interval distance of each recent data segment in the recent time series data sequence are obtained. Then, the stability of the outlier distribution is obtained by combining the dispersion of outlier data distribution and the mean cluster interval distance of each recent data segment in the recent time series data sequence: the number of suspected outlier clusters in each recent data segment in the recent time series data sequence is used as a weight to perform a weighted summation of the dispersion of outlier data distribution of all recent data segments to obtain a weighted summation result; the standard deviation of the mean cluster interval distance of all recent data segments is calculated, and the negative of the product of the weighted summation result and the standard deviation is used as the independent variable of an exponential function with the natural constant as the base to obtain the stability of the outlier distribution.

[0064] The formula for calculating the stability of the abnormal distribution is as follows: in, Indicates the stability of the anomalous distribution. This represents an exponential function with the natural constant as the base, used for inverse proportional normalization. This represents the number of suspected anomalous clusters in the i-th recent data segment. This represents the dispersion of outlier data distribution in the i-th recent data segment. This represents the standard deviation of the mean cluster interval distance corresponding to each recent data segment in a recent time series data sequence. This indicates the number of recent data segments contained in a recent time series data sequence.

[0065] It should be noted that the smaller the dispersion of the abnormal data distribution in each recent data segment of the recent time series data sequence, the higher the proximity of the suspected abnormal data, the more concentrated the distribution of the suspected abnormal data, and the more consistent it is with the distribution characteristics of the real abnormal data. When the interval distance between the suspected abnormal clusters and the normal clusters in each recent data segment is more similar, the smaller the standard deviation, the more stable the distribution change of the suspected abnormal clusters, and there will be no fluctuating differences between the abnormal clusters and the normal clusters caused by the random distribution of monitoring noise. The corresponding abnormal distribution is more stable.

[0066] Furthermore, the degree of anomalous change in the recent monitoring data distribution of any gaseous element concentration is obtained by summing the periodic abnormal fluctuation degree and the abnormal distribution stability degree. The degree of anomalous change in the recent monitoring data distribution of any gaseous element concentration is then determined. The calculation formula is: The greater the recent abnormal changes in the distribution of monitoring data, the more concentrated the distribution of suspected abnormal data in the recent time series data sequence is, which does not conform to the random distribution characteristics of monitoring noise caused by equipment lifespan damage.

[0067] Step S104: Based on the time difference between each suspected abnormal data in the recent time series data sequence and its corresponding nearest neighbor data in the historical time series data sequence of other gaseous element concentrations, obtain the degree of abnormality of multiple types of correlation changes; combine the degree of abnormality of recent monitoring data distribution changes and the degree of abnormality of multiple types of correlation changes to obtain the adaptive concentration threshold of any gaseous element concentration at the current monitoring time.

[0068] To obtain more accurate anomaly prediction results, this embodiment of the invention combines multidimensional correlation features and uses the time difference between each suspected anomaly data in the recent time series data sequence corresponding to any gaseous element concentration and its corresponding nearest neighbor data in the historical time series data sequence of other gaseous element concentrations to analyze the degree of anomaly in the multi-type correlation changes of any gaseous element concentration. Since a genuine leak anomaly can cause simultaneous changes in the concentrations of multiple trace impurities, while noise anomalies are manifested in individual trace impurities, in this embodiment of the invention, for any suspected anomaly data in a recent time series data sequence of any gaseous element concentration, the time of the suspected anomaly data is obtained. The interval time between the time of each suspected anomaly data in a historical time series data sequence of any other gaseous element concentration and the time of the suspected anomaly data is calculated to obtain the minimum interval time corresponding to the other gaseous element concentration. Other gaseous element concentrations refer to gaseous element concentrations other than any one gaseous element concentration. The minimum interval time corresponding to each other gaseous element concentration is accumulated to obtain the accumulated interval time value of the suspected anomaly data. The accumulated interval time value of each suspected anomaly data in the recent time series data sequence is accumulated to obtain the overall time difference. The negative of the overall time difference is used as the independent variable of an exponential function with the natural constant as the base to obtain the degree of anomalies in multiple types of correlation changes.

[0069] The formula for calculating the degree of abnormality in multiple class association changes is as follows: Indicates the degree of abnormality in the association between multiple classes. This represents an exponential function with the natural constant as its base. This represents the number of suspected outliers in a recent time-series data sequence of any gaseous element concentration. This indicates the number of different types of other gaseous elements in terms of concentration. The first element in a recent time series data sequence representing the concentration of any gaseous element. At a moment when suspected abnormal data was collected. Indicates the first Historical time series data of other gaseous element concentrations are related to the first The most recent time of suspected anomalous data. Indicates the minimum interval time. Represents the absolute value symbol.

[0070] It should be noted that the closer the time of occurrence of suspected anomalies in the recent time series data of any gaseous element concentration is to the time of occurrence of suspected anomalies in other types of gaseous element concentrations, the higher the confidence level of the suspected anomaly in that gaseous element concentration, and the more likely it is to be a genuine anomaly. Therefore, when The smaller the value, the greater the degree of abnormality in the multi-category correlation changes, and the greater the possibility that the suspected abnormal data appearing at any gaseous element concentration is actually abnormal data.

[0071] Furthermore, by comprehensively considering the degree of abnormality in the distribution of recent monitoring data of any gaseous element concentration and the degree of abnormality in various related changes, an adaptive concentration threshold for any gaseous element concentration at the current monitoring time is obtained. This is used to improve the accuracy of monitoring trace impurity anomalies at the current monitoring time. The specific method is as follows:

[0072] (1) Based on the degree of abnormality in the distribution of recent monitoring data and the degree of abnormality in the correlation of multiple types, obtain the degree of abnormality in recent monitoring data and utilize the degree of abnormality in recent monitoring data.

[0073] Specifically, the degree of anomaly in the recent monitoring data is obtained by averaging the anomalies in the distribution of half of the recent monitoring data and the anomalies in the correlations among various categories. The formula for calculating the degree of anomaly in the recent monitoring data is as follows: Where M represents the degree of anomaly in recent monitoring data, W represents the degree of anomaly in the distribution changes of recent monitoring data, and P represents the degree of anomaly in the changes of multiple class associations.

[0074] It should be noted that the greater the degree of abnormality in the distribution of recent monitoring data, and the greater the degree of abnormality in the correlation of multiple types, the more likely the abnormal data in the recent time series of any gaseous element concentration is to be real abnormal data rather than noise. Therefore, the greater the degree of abnormality in recent monitoring data. In order to normalize the value range to the interval [0, 1], (2) the fixed concentration threshold of any gaseous element concentration is adjusted to obtain the adaptive concentration threshold of any gaseous element concentration at the current monitoring time.

[0075] Specifically, after obtaining the recent monitoring data anomaly level of any gaseous element concentration, the real-time fixed threshold can be adaptively adjusted based on the recent monitoring data anomaly level. Since the recent monitoring data anomaly level can distinguish abnormal data from noisy data at both ends of the value range, the midpoint of the value range, 0.5, is used as the recent monitoring data anomaly level threshold. If the recent monitoring data anomaly level is greater than the preset recent monitoring data anomaly level threshold, it is considered that the recent data of any gaseous element concentration has a real anomaly risk, and the corresponding monitoring fixed threshold needs to be reduced, that is, the fixed concentration threshold of any gaseous element concentration, in order to detect anomalies in advance. Therefore, the product of the preset adjustment coefficient and the recent monitoring data anomaly level is obtained, and the difference between the constant 1 and the product is used as the adjustment coefficient. The product of the fixed concentration threshold of any gaseous element concentration and the adjustment coefficient is used as the adaptive concentration threshold of any gaseous element concentration at the current monitoring time.

[0076] If the degree of anomaly in recent monitoring data is less than or equal to the preset threshold for the degree of anomaly in recent monitoring data, it is considered that there is noise anomaly in the recent data of any gaseous element concentration. It is necessary to increase the fixed monitoring threshold to avoid mismonitoring noise as a real anomaly. Therefore, the difference between constant 1 and the degree of anomaly in recent monitoring data is obtained, the product of the preset adjustment coefficient and the difference is calculated, the sum of constant 1 and the product is used as the adjustment coefficient, and the product of the fixed concentration threshold of any gaseous element concentration and the adjustment coefficient is used as the adaptive concentration threshold of any gaseous element concentration at the current monitoring time.

[0077] In one embodiment, the formula for calculating the adaptive concentration threshold of any gaseous element at the current monitoring time is: in, This represents the adaptive concentration threshold for any gaseous element at the current monitoring time. This represents the fixed concentration threshold before adjustment. 1 indicates a constant, and 0.5 indicates a preset adjustment coefficient to prevent over-adjustment. This means that, based on the original data, the smaller the degree of anomaly in recent monitoring data, the higher the concentration threshold will be adjusted, to prevent noise from being mistaken for real anomalies. This means that, based on the original data, the greater the degree of anomaly in recent monitoring data, the lower the concentration threshold will be adjusted, so that anomalies below the concentration threshold can be detected in advance.

[0078] Similarly, based on the method for obtaining the adaptive concentration threshold of any gaseous element concentration at the current monitoring time, the adaptive concentration threshold of each gaseous element concentration at the current monitoring time can be obtained. Thus, the adaptive concentration threshold of each gaseous element concentration at the current monitoring time can be obtained, which can be used to monitor the abnormality of trace impurities at the current monitoring time.

[0079] Step S105: Based on the adaptive concentration threshold corresponding to the concentration of each gaseous element, perform trace impurity anomaly monitoring on the boron-11 electron gas at the current monitoring time.

[0080] After obtaining the adaptive concentration threshold for each gaseous element at the current monitoring time, the concentration data of each gaseous element in the container storing boron-11 electron gas at the current monitoring time are obtained. Then, the adaptive concentration threshold is used to compare the concentration data at the current monitoring time. If the concentration data of any gaseous element at the current monitoring time is greater than the corresponding adaptive concentration threshold, it is determined that the trace impurities in the boron-11 electron gas exceed the standard, and an abnormality warning is immediately issued and relevant personnel are notified to carry out maintenance.

[0081] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A boron-11 electron gas analysis method for identifying trace impurities, characterized in that, The method includes: Monitor the concentrations of at least two gaseous elements in a container storing boron-11 electron gas, and obtain a historical time series data sequence of the concentration of each gaseous element at the current monitoring time. The gaseous elements include hydrocarbons and inert gases. Each of the historical time-series data sequences is periodically divided to obtain multiple data segments, and suspected abnormal data in each data segment is obtained based on the data changes in each data segment. For any gaseous element concentration, a preset number of recent data segments are obtained from the data segment corresponding to the gaseous element concentration to form a recent time series data sequence. Based on the distribution of suspected abnormal data, the distribution of normal data, and the number of suspected abnormal data in each recent data segment in the recent time series data sequence, the degree of abnormality in the distribution of recent monitoring data is obtained. Based on the time difference between each suspected abnormal data in the recent time series data sequence and its corresponding nearest neighbor data in the historical time series data sequences of other gaseous element concentrations, the degree of abnormality of multiple types of associated changes is obtained; by combining the degree of abnormality of recent monitoring data distribution changes and the degree of abnormality of multiple types of associated changes, the adaptive concentration threshold of any gaseous element concentration at the current monitoring time is obtained. Based on the adaptive concentration threshold corresponding to the concentration of each gaseous element, the boron-11 electron gas at the current monitoring time is monitored for trace impurities. Based on the degree of anomalous changes in the distribution of recent monitoring data and the degree of anomalous changes in various related types, an adaptive concentration threshold for the concentration of any gaseous element at the current monitoring time is obtained, including: The degree of anomaly in the recent monitoring data is obtained by averaging the degree of anomaly in the distribution of half of the recent monitoring data and the degree of anomaly in the correlation of multiple categories. If the degree of abnormality of recent monitoring data is less than or equal to the preset threshold for the degree of abnormality of recent monitoring data, then the difference between constant 1 and the degree of abnormality of recent monitoring data is obtained, the product of the preset adjustment coefficient and the difference is calculated, the sum of constant 1 and the product is used as the adjustment coefficient, and the product of the fixed concentration threshold of any gaseous element concentration and the adjustment coefficient is used as the adaptive concentration threshold of any gaseous element concentration at the current monitoring time. If the degree of abnormality in recent monitoring data is greater than the preset threshold for the degree of abnormality in recent monitoring data, then the product of the preset adjustment coefficient and the degree of abnormality in recent monitoring data is obtained, and the difference between the constant 1 and the product is used as the adjustment coefficient. The product of the fixed concentration threshold of any gaseous element concentration and the adjustment coefficient is used as the adaptive concentration threshold of any gaseous element concentration at the current monitoring time.

2. The boron-11 electron gas analysis method for identifying trace impurities according to claim 1, characterized in that, The process of periodically dividing each of the historical time-series data sequences to obtain multiple data segments includes: The periodic components of each historical time series data sequence are extracted using Fourier transform to obtain the principal period length of each historical time series data sequence. Based on the principal period length of each historical time series data sequence, each historical time series data sequence is adaptively divided into multiple data segments.

3. The boron-11 electron gas analysis method for identifying trace impurities according to claim 1, characterized in that, The step of obtaining suspected abnormal data in each data segment based on data changes in each data segment includes: For any data in any data segment, based on the position of the data in the data segment, data at the same position in other data segments belonging to the same historical time series data sequence as the data segment are obtained and recorded as reference data. The absolute value of the difference between the data and each reference data is calculated, and the mean of the absolute value of the difference is recorded as the degree of difference of the data. Obtain the difference degree of each data in the historical time series data sequence to which any data segment belongs, and obtain the maximum difference degree and the minimum difference degree. Use the maximum difference degree and the minimum difference degree to normalize the difference degree of any data to obtain the suspected anomaly degree of any data. If the suspected abnormality level of any data exceeds a preset suspected abnormality level threshold, then the data is determined to be suspected abnormal data.

4. The boron-11 electron gas analysis method for identifying trace impurities according to claim 1, characterized in that, The step of determining the degree of abnormality in the distribution of recent monitoring data based on the distribution of suspected abnormal data, the distribution of normal data, and the number of suspected abnormal data in each recent data segment includes: The absolute value of the difference in the number of suspected anomalous data between the first and last recent data segments in the recent time series data sequence is calculated and denoted as the anomalous change feature value. The absolute values ​​of the difference in the number of suspected anomalous data between every two adjacent recent data segments in the recent time series data sequence are accumulated to obtain the overall anomalous change feature value. The product of the anomalous change feature value and the overall anomalous change feature value is normalized to obtain the degree of periodic anomalous fluctuation. For any recent data segment in the recent time series data sequence, perform one-dimensional clustering on the recent data segment to obtain a suspected abnormal cluster composed of suspected abnormal data and a normal cluster composed of normal data. Based on the data differences in each suspected abnormal cluster, obtain the abnormal data distribution dispersion and the mean cluster interval distance of the any recent data segment. By combining the dispersion of abnormal data distribution and the mean clustering interval distance of each recent data segment in the recent time series data sequence, the stability of abnormal distribution is obtained; based on the sum of the periodic abnormal fluctuation degree and the stability of abnormal distribution, the degree of abnormality of recent monitoring data distribution changes is obtained.

5. The boron-11 electron gas analysis method for identifying trace impurities according to claim 4, characterized in that, The step of obtaining the anomalous data distribution dispersion and the mean cluster interval distance of any recent data segment based on the data differences in each suspected anomalous cluster includes: For any data point in any suspected anomaly cluster, calculate the absolute value of the data difference between the data point and other data points in the suspected anomaly cluster to obtain the minimum absolute value of the data difference. Then, sum the minimum absolute values ​​of the data differences for each data point in the suspected anomaly cluster to obtain a first accumulated value. Finally, sum the first accumulated values ​​for each suspected anomaly cluster in any recent data segment to obtain the anomalous data distribution dispersion of the recent data segment. Obtain the absolute value of the difference between the cluster center of each suspected abnormal cluster and the cluster center of each normal cluster, and record it as the interval distance. Use the average of all interval distances as the average cluster interval distance of any recent data segment.

6. The boron-11 electron gas analysis method for identifying trace impurities according to claim 4, characterized in that, The method of obtaining the stability of the abnormal distribution by combining the dispersion of the abnormal data distribution and the mean clustering interval distance of each recent data segment in the recent time series data sequence includes: The number of suspected anomalous clusters in each recent data segment of the recent time series data sequence is used as a weight to perform a weighted summation of the anomalous data distribution dispersion of all recent data segments, and the standard deviation of the mean cluster interval distance of all recent data segments is calculated. The negative of the product of the weighted summation and the standard deviation is used as the independent variable of an exponential function with the natural constant as the base to obtain the stability of the anomalous distribution.

7. The boron-11 electron gas analysis method for identifying trace impurities according to claim 1, characterized in that, The method involves obtaining the degree of various types of correlated changes based on the time difference between each suspected abnormal data point in the recent time-series data sequence and its corresponding nearest neighbor data in the historical time-series data sequences of other gaseous element concentrations, including: For any suspected abnormal data in the recent time series data sequence, the time of the suspected abnormal data is obtained, the time interval between the time of each suspected abnormal data in the historical time series data sequence of any other gaseous element concentration and the time of the suspected abnormal data is calculated, the minimum time interval corresponding to the concentration of any other gaseous element is obtained, and the minimum time interval corresponding to the concentration of each other gaseous element is accumulated to obtain the accumulated value of the time interval of the suspected abnormal data. The time intervals of each suspected abnormal data in the recent time series are accumulated to obtain the overall time difference. The negative of the overall time difference is used as the independent variable of an exponential function with the natural constant as the base to obtain the degree of abnormality of various types of correlation changes.

Images