Interval dynamic fusion-based satellite telemetry data threshold fine mining method

Abstract

The invention discloses a satellite telemetry data threshold fine mining method based on interval dynamic fusion, and relates to a satellite telemetry data threshold fine mining method. The method aims at solving the problems that in the prior art, the threshold setting process often depends on domain knowledge, data-driven automation means are lacked, and adaptability is poor when a novel satellite or an unknown fault mode is faced. The method comprises the following steps: step 1, satellite telemetry data statistical feature extraction and outlier elimination; step 2, setting satellite telemetry data threshold mining without working conditions; and step 3, setting satellite telemetry data statistical feature extraction and threshold fine mining according to working conditions. The invention belongs to the technical field of satellite telemetry.

Description

Technical Field

[0001] The invention relates to a satellite telemetry data threshold refinement mining method, belonging to the technical field of satellite telemetry. Background Art

[0002] Satellites, as essential vehicles of modern science and technology, are widely used in fields such as communications, navigation, meteorological observation, Earth resource exploration, and scientific research. They play an irreplaceable role in the national economy, national defense, and the development of human society. With the rapid advancement of aerospace technology, satellite functions are becoming increasingly complex, and mission requirements are becoming more diverse. This places higher demands on satellite reliability and stability. During satellite operation, its status is directly related to mission success. Therefore, real-time monitoring of satellite health and anomaly warnings are particularly important. Telemetry data, as the core information reflecting satellite operational status, contains key parameters of each satellite subsystem. By analyzing this data, potential faults can be promptly identified and appropriate measures can be taken, thereby ensuring the safe operation of the satellite. However, with the continuous development of satellite systems, the scale and complexity of telemetry data are increasing. Traditional manual analysis methods are no longer able to meet the real-time and efficient requirements. Threshold-based satellite status monitoring methods have emerged. These methods determine whether telemetry data is abnormal by setting appropriate thresholds, providing an efficient and accurate technical means for satellite health management. They have become a research hotspot in the field of satellite status monitoring. In particular, numerous methods and techniques have been widely studied and applied in the statistical analysis and threshold extraction of telemetry data. Satellite telemetry threshold extraction methods mainly include statistical models, quantile regression, cluster analysis, and empirical threshold initialization. Statistical models extract thresholds based on methods such as Gaussian distribution and kernel density estimation, but are not robust enough to non-Gaussian distributions such as long-tailed data. Quantile regression uses quantiles to define threshold intervals, which can adapt to data distribution shifts but does not fully consider the temporal dependencies and spatial correlations between parameters. Cluster analysis uses unsupervised clustering methods such as DBSCAN to define data cluster boundaries as thresholds, but is prone to the curse of dimensionality when processing high-dimensional data. Empirical threshold initialization sets the initial threshold based on domain expert knowledge or statistical characteristics such as the mean, variance, and extreme values of historical data statistics, ensuring the physical meaning and engineering interpretability of the threshold. To further improve the adaptability of threshold extraction, statistical refinement combines historical data distribution characteristics (such as skewness, kurtosis, and long tail characteristics) with kernel density estimation or quantile regression to dynamically adjust the initial threshold. Multi-condition segmented optimization then performs segmented statistics on historical data based on satellite operating conditions (such as illuminated areas, shadowed areas, and device mode switching) to extract refined thresholds under different operating conditions, thereby reducing false positives caused by environmental interference. However, existing technologies still have many limitations, mainly reflected in the contradiction between static and dynamic thresholds, insufficient multi-parameter coupling processing, sensitivity to environmental interference, and scarcity of annotated data.Static thresholds are typically based on empirical historical data or ±3σ statistics. These cannot effectively distinguish between normal operating fluctuations and true anomalies, resulting in high false alarm rates or missed alerts. Dynamic thresholds, however, adjust thresholds based on sliding windows or time series prediction methods (such as ARIMA and LSTM). However, these methods are overly sensitive to sudden anomalies and have high computational complexity, making them difficult to meet real-time deployment requirements. Furthermore, the threshold setting process often relies on domain knowledge and lacks data-driven automation, making it less adaptable to new satellites or unknown failure modes. These issues limit the effectiveness of existing methods in practical applications. Summary of the Invention

[0003] In order to solve the problems in the existing technology that the threshold setting process often relies on domain knowledge, lacks data-driven automation means, and has poor adaptability when facing new satellites or unknown failure modes, the present invention proposes a satellite telemetry data threshold refinement mining method based on interval dynamic fusion.

[0004] The technical solution adopted by the present invention to solve the above problems is: the steps of the present invention include:

[0005] Step 1: Extract statistical features of satellite telemetry data and eliminate outliers;

[0006] Step 2: Setting satellite telemetry data threshold mining in the absence of working conditions;

[0007] Step 3: Set up satellite telemetry data statistical feature extraction and threshold refinement mining under working conditions.

[0008] Furthermore, step one specifically includes:

[0009] Step 1: Calculate a series of statistical features for a single telemetry data column;

[0010] Step 2: Eliminate outliers based on the 6σ interval of data distribution and data continuity.

[0011] Furthermore, the series of statistical features in step 1 are divided into the following categories:

[0012] Central Tendency Indicator

[0013] average value Reflects the concentration level of data, specifically:

[0014]

[0015] Medianx median : Evaluate the central position of the data;

[0016] Mode x mode : Indicates the value with the highest frequency;

[0017] Dispersion index

[0018] Minimum x min / Maximum value x max : Indicates the boundary of the data range;

[0019] Range: Indicates the difference between the maximum and minimum values, specifically:

[0020] Range=x max -x min ,

[0021] Standard deviation s: measures the degree of data dispersion, specifically:

[0022]

[0023] Variance 2 : The square of the standard deviation reflects the degree of deviation of the data point from the mean value, specifically:

[0024]

[0025] Coefficient of variation coeff: describes the relative degree of dispersion as the ratio of the standard deviation to the mean, specifically:

[0026]

[0027] Distribution morphology indicators

[0028] Skew: describes the symmetry of the data distribution. Positive skew means that the data is distributed longer on the right side, and negative skew means that the data is distributed longer on the left side. The skewness of a symmetrical distribution is close to 0. Specifically:

[0029]

[0030] Kurtosis: describes the sharpness of the data distribution. Kurtosis means that the data is concentrated near the mean, and low kurtosis means that the data distribution is relatively flat. Specifically:

[0031]

[0032] Furthermore, in step 2, the outlier elimination operation combining the 6σ interval of data distribution and data continuity is performed. First, the average value calculated in step 1 is used. With the standard deviation s, use the 6σ principle to calculate the upper bound upper_bound and lower bound lower_bound of the data, specifically:

[0033]

[0034] Traverse the input data column and mark the values that are less than the lower bound or greater than the upper bound as outliers; then find the outliers that last for 8 time points and mark them as normal, specifically:

[0035] Traverse the outlier labeling status of the data column. If the current i-th data is labeled as abnormal data, initialize the counter count to 1 to record the number of consecutive outliers. Then, start traversing the sub-data column from the current position. If subsequent data is still labeled as abnormal, increase the counter count. If the counter count is greater than 8, it means that abnormal values that have lasted for more than 8 time points have been found. Then, change the abnormal value marks of these consecutive data to normal and stop traversing the sub-data column. When the counter count is not greater than 8 and data marked as normal is encountered, the traversal of the sub-data column is also stopped. According to the outlier labeling of the data column obtained last, select the data marked as normal to form a new data column, and obtain a data column after eliminating outliers that combines the 6σ interval of data distribution and data continuity.

[0036] Furthermore, step 2 specifically includes:

[0037] Step 1: When the input working condition information is nan, that is, a null value, it is determined that the telemetry data does not have working condition configuration and is subsequently processed;

[0038] Step 2: Determine the number of valid data after removing outliers from the data column using the method in step 1. If the number of valid data is 0, there is no valid data available for analysis, and an empty result is returned.

[0039] Step 3: After using the method in step 1 to remove outliers from the data column, if valid data exists, use the maximum value of the expansion as the threshold range of the telemetry parameter without working condition setting.

[0040] Furthermore, step 3 specifically includes:

[0041]

[0042] where X filtered This indicates that the method in step ① is used to remove outliers from the original telemetry data column. Furthermore, since no relevant information such as the working condition name is provided, the default working condition name default is used as the mining working condition, and the above min and max values are used as the threshold range. Finally, the threshold mining results of satellite telemetry data without working condition settings are obtained in the form of a list, specifically:

[0043] ['default',min,max].

[0044] Furthermore, the algorithm in step three specifically includes:

[0045] Step 1: For telemetry parameters that provide working conditions and their corresponding threshold ranges, traverse the provided working condition list and obtain the working condition name r name and the minimum value r1 and the maximum value r2 of the corresponding threshold range;

[0046] Step 2: Obtain telemetry data within the working condition;

[0047] Step 3: Refine the mining threshold range.

[0048] Furthermore, step 2 specifically includes:

[0049] X r =r1 <X<r2,

[0050] Among them, X represents the telemetry raw data column, X r Indicates working condition r name Corresponding telemetry data column; if X r If the length of the data column is 0, it means that the telemetry data to be analyzed does not contain data under this working condition. In this case, the original setting of this working condition is added to the telemetry threshold result to be output, and the result is returned after all working condition information is processed.

[0051] Furthermore, step 3 specifically includes:

[0052]

[0053] For each working condition, the statistical characteristic parameters of the data within the working condition are calculated, and the threshold range for refined mining is obtained by combining the original threshold setting with the actual telemetry data value within the working condition; length The value range of the original threshold range r1 to r2 of this working condition is represented by the length of the interval, and interval is the length of each segment after the interval is divided into n equal parts. The value range of the i-th interval can be expressed as:

[0054] [lower_bound i ,upper_bound i ],

[0055] Among them, i∈Z and i∈[0,n-1], for each interval, count the number of data points in the interval interval , and the number of data points under this condition count r By comparison, the percentage of the data volume in this interval to the data volume of this working condition is obtained. For the i-th interval, it is specifically:

[0056] percentage i =count interval_i / count r ,

[0057] By setting a minimum data percentage threshold k, we count the intervals in n intervals that meet the requirement that the percentage of the intervals is greater than or equal to k, and then take the union of multiple intervals, and take the minimum value of the union as valid_boundaries min , take the maximum value of the union as valid_boundaries max , using these two values as the basis for refining the threshold results;

[0058] In order to make flexible adjustments based on the above threshold results, the boundary expansion parameter expand_range is introduced. Combining this parameter setting with the original threshold boundary values r1 and r2, the final interval refinement threshold minimum value r is calculated. min With the maximum value r max , specifically:

[0059]

[0060] The minimum value of valid_boundaries to be combined min Subtract the product of the boundary expansion multiple and the length of the single interval after n equal divisions, and take the larger value as the minimum value of the refined threshold compared with the original threshold minimum value, and set the maximum value of the union to valid_boundaries max The product of the boundary expansion factor and the length of the single interval after n equal divisions is added. Compared with the maximum value of the original threshold, the smaller value is taken as the maximum value of the refined threshold, ensuring that the refined threshold interval is a subset of the original threshold interval. At this point, the final result of refined threshold mining under this working condition is as follows:

[0061] [r name ,r min ,r max ],

[0062] The telemetry parameters contain the results of multiple working conditions. According to the distribution of the data of each working condition in the telemetry data, the refined threshold mining results of each working condition will be formed by steps 2 and 3 respectively, and summarized in the form of a list. Finally, the statistical feature extraction and threshold refined mining results of satellite telemetry data with working condition settings are obtained.

[0063] The beneficial effects of the present invention are:

[0064] 1. This invention combines historical data with statistical methods, integrating domain knowledge with data-driven methods. This not only preserves the interpretability of empirical thresholds, but also improves the refinement of thresholds through statistical methods.

[0065] 2. The present invention realizes multi-condition threshold optimization and extracts thresholds under different working conditions through segmented statistics, thus solving the problem of insufficient adaptability of traditional static thresholds in dynamic environments.

[0066] 3. The calculation subject of the present invention adopts statistical methods, which has low calculation complexity and is suitable for real-time processing requirements of satellites or ground stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a diagram of the BIE1 data curve of the telemetry parameters of the low-orbit satellite;

[0068] Figure 2 This is a schematic diagram of the results of satellite telemetry data threshold refinement mining based on intelligent statistics;

[0069] Figure 3 This is a diagram of the BIE1 data curve of the telemetry parameters of the low-orbit satellite;

[0070] Figure 4 Schematic diagram of satellite telemetry data threshold refinement mining results based on intelligent statistics;

[0071] Figure 5 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0072] Example 1

[0073] like Figure 1 As shown, a satellite telemetry data threshold refinement mining method based on interval dynamic fusion includes the following steps:

[0074] Step 1: Prepare satellite telemetry data;

[0075] Step 2: Parameter configuration of satellite telemetry data threshold refinement mining method based on intelligent statistics.

[0076] This embodiment uses the telemetry parameter BIE1 data of a low-orbit satellite. The telemetry data points of this telemetry parameter from 00:00:00 on July 1, 2022 to 23:59:00 on July 3, 2022 are 129,481. The telemetry data curve within the time range of 00:00:00 on July 1, 2022 to 00:05:42 on July 1, 2022 is as follows: Figure 1 The parameters required in the algorithm are as follows:

[0077] Telemetry data operating condition threshold information range, in list form: [["Not powered","0","0.2"],["Powered","0.2","0.8"]]

[0078] Number of equal parts of working condition interval n: 20

[0079] Minimum data percentage threshold k: 0.1

[0080] Boundary expansion parameter expand_range: 0

[0081] The results of threshold refinement mining of BIE1 data of telemetry parameters of a low-orbit satellite are as follows: Figure 2 In the method output, we can see that multiple statistical parameters of the BIE1 data as a whole and the corresponding data for the "unpowered" and "powered" operating conditions have been calculated and presented. The threshold value for the given telemetry data condition has been further narrowed, and the threshold range has been further refined.

[0082] Example 2

[0083] A satellite telemetry data threshold refinement mining method based on interval dynamic fusion includes the following steps:

[0084] Step 1: Prepare satellite telemetry data;

[0085] Step 2: Parameter configuration of satellite telemetry data threshold refinement mining method based on intelligent statistics.

[0086] This embodiment uses the telemetry parameter BIE1 data of a low-orbit satellite. The telemetry data points of this telemetry parameter from 00:00:00 on July 1, 2022 to 23:59:00 on July 3, 2022 are 129,481. The telemetry data curve within the time range of 00:00:00 on July 1, 2022 to 00:05:42 on July 1, 2022 is as follows: Figure 3 The parameters required in the algorithm are as follows:

[0087] Telemetry data working condition threshold information range: nan

[0088] Number of equal parts of working condition interval n: 20

[0089] Minimum data percentage threshold k: 0.1

[0090] Boundary expansion parameter expand_range: 0

[0091] The results of threshold refinement mining of BIE1 data of telemetry parameters of a low-orbit satellite are as follows: Figure 4 As shown in the method output, multiple statistical parameters of the BIE1 data as a whole are calculated and presented. Since the operating condition information of the input data is empty, the results of the threshold refined mining are provided in the form of the default value "default". The relationship between the threshold range and the statistical parameters of the BIE1 data as a whole is consistent with the method's technical solution.

[0092] It can be seen that the method proposed in this scheme can combine satellite telemetry data and knowledge in the fields of working condition settings, calculate a series of data statistical parameters, intelligently distinguish working condition information, and perform threshold-based fine-grained mining of data under different working conditions. It has good satellite telemetry data threshold-based fine-grained mining capabilities.

[0093] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with the present profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical content disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement of the above embodiments made according to the technical essence of the present invention, within the spirit and principles of the present invention, without departing from the content of the technical solution of the present invention, shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A satellite telemetry data threshold refinement mining method based on interval dynamic fusion, characterized in that: The specific steps include: Step 1: Extract statistical features of satellite telemetry data and eliminate outliers; Step 2: Setting satellite telemetry data threshold mining in the absence of working conditions; Step 3: Set up satellite telemetry data statistical feature extraction and threshold refinement mining under working conditions.

2. The satellite telemetry data threshold refinement mining method based on interval dynamic fusion according to claim 1 is characterized in that: Step 1 specifically includes: Step 1: Calculate a series of statistical features for a single telemetry data column; Step 2: Eliminate outliers based on the 6σ interval of data distribution and data continuity.

3. The satellite telemetry data threshold refinement mining method based on interval dynamic fusion according to claim 2 is characterized in that: The series of statistical features in step 1 are divided into feature categories: Central Tendency Indicator average value Reflects the concentration level of data, specifically: Medianx median : Evaluate the central position of the data; Mode x mode : Indicates the value with the highest frequency; Dispersion index Minimum x min / Maximum value x max : Indicates the boundary of the data range; Range: Indicates the difference between the maximum and minimum values, specifically: Range=x max -x min , Standard deviation s: measures the degree of data dispersion, specifically: Variance 2 : The square of the standard deviation reflects the degree of deviation of the data point from the mean value, specifically: Coefficient of variation coeff: describes the relative degree of dispersion as the ratio of the standard deviation to the mean, specifically: Distribution morphology indicators Skew: describes the symmetry of the data distribution. Positive skew means that the data is distributed longer on the right side, and negative skew means that the data is distributed longer on the left side. The skewness of a symmetrical distribution is close to 0. Specifically: Kurtosis: describes the sharpness of the data distribution. Kurtosis means that the data is concentrated near the mean, and low kurtosis means that the data distribution is relatively flat. Specifically:

4. The satellite telemetry data threshold refinement mining method based on interval dynamic fusion according to claim 2 is characterized in that: In step 2, the outlier elimination operation is combined with the 6σ interval of data distribution and data continuity. First, the average value calculated in step 1 is used. With the standard deviation s, use the 6σ principle to calculate the upper bound upper_bound and lower bound lower_bound of the data, specifically: Traverse the input data column and mark the values that are less than the lower bound or greater than the upper bound as outliers; then find the outliers that last for 8 time points and mark them as normal, specifically: Traverse the outlier labeling status of the data column. If the current i-th data is labeled as abnormal data, initialize the counter count to 1 to record the number of consecutive outliers. Then, start traversing the sub-data column from the current position. If subsequent data is still labeled as abnormal, increase the counter count. If the counter count is greater than 8, it means that abnormal values that have lasted for more than 8 time points have been found. Then, change the abnormal value marks of these consecutive data to normal and stop traversing the sub-data column. When the counter count is not greater than 8 and data marked as normal is encountered, the traversal of the sub-data column is also stopped. According to the outlier labeling of the data column obtained last, select the data marked as normal to form a new data column, and obtain a data column after eliminating outliers that combines the 6σ interval of data distribution and data continuity.

5. The satellite telemetry data threshold refinement mining method based on interval dynamic fusion according to claim 1 is characterized in that: Step 2 specifically includes: Step 1: When the input working condition information is nan, that is, a null value, it is determined that the telemetry data does not have working condition configuration and is subsequently processed; Step 2: Determine the number of valid data after removing outliers from the data column using the method in step 1. If the number of valid data is 0, there is no valid data available for analysis, and an empty result is returned. Step 3: After using the method in step 1 to remove outliers from the data column, if valid data exists, use the maximum value of the expansion as the threshold range of the telemetry parameter without working condition setting.

6. The satellite telemetry data threshold refinement mining method based on interval dynamic fusion according to claim 5 is characterized in that: Step 3 specifically includes: where X filtered This indicates that the method in step ① is used to remove outliers from the original telemetry data column. Furthermore, since no relevant information such as the working condition name is provided, the default working condition name default is used as the mining working condition, and the above min and max values are used as the threshold range. Finally, the threshold mining results of satellite telemetry data without working condition settings are obtained in the form of a list, specifically: ['default',min,max].

7. The satellite telemetry data threshold refinement mining method based on interval dynamic fusion according to claim 1 is characterized in that: The algorithm in step three specifically includes: Step 1: For telemetry parameters that provide working conditions and their corresponding threshold ranges, traverse the provided working condition list and obtain the working condition name r name and the minimum value r1 and the maximum value r2 of the corresponding threshold range; Step 2: Obtain telemetry data within the working condition; Step 3: Refine the mining threshold range.

8. The satellite telemetry data threshold refinement mining method based on interval dynamic fusion according to claim 7 is characterized in that: Step 2 specifically includes: X r =r1<X<r2, Among them, X represents the telemetry raw data column, X r Indicates working condition r name Corresponding telemetry data column; if X r If the length of the data column is 0, it means that the telemetry data to be analyzed does not contain data under this working condition. In this case, the original setting of this working condition is added to the telemetry threshold result to be output, and the result is returned after all working condition information is processed.

9. The satellite telemetry data threshold refinement mining method based on interval dynamic fusion according to claim 7 is characterized in that: Step 3 specifically includes: For each working condition, the statistical characteristic parameters of the data within the working condition are calculated, and the threshold range for refined mining is obtained by combining the original threshold setting with the actual telemetry data value within the working condition; length The value range of the original threshold range r1 to r2 for this working condition is represented by interval, and interval is the length of each segment after the interval is divided into n equal parts. The value range of the i-th interval can be expressed as: [lower_bound i ,upper_bound i ], Among them, i∈Z and i∈[0,n-1], for each interval, count the number of data points in the interval interval , and the number of data points under this condition count r By comparison, we can get the percentage of the data volume in this interval to the data volume of this working condition. For the i-th interval, it is specifically: percentage i =count interval_i / count r , By setting a minimum data percentage threshold k, we count the intervals in n intervals that meet the requirement that the percentage of the intervals is greater than or equal to k, and then take the union of multiple intervals, and take the minimum value of the union as valid_boundaries min , take the maximum value of the union as valid_boundaries max , using these two values as the basis for refining the threshold results; In order to make flexible adjustments based on the above threshold results, the boundary expansion parameter expand_range is introduced. Combining this parameter setting with the original threshold boundary values r1 and r2, the final interval refinement threshold minimum value r is calculated. min With the maximum value r max , specifically: The minimum value of valid_boundaries to be combined min Subtract the product of the boundary expansion multiple and the length of the single interval after n equal divisions, and take the larger value as the minimum value of the refined threshold compared with the original threshold minimum value, and set the maximum value of the union to valid_boundaries max The product of the boundary expansion factor and the length of the single interval after n equal divisions is added. Compared with the maximum value of the original threshold, the smaller value is taken as the maximum value of the refined threshold, ensuring that the refined threshold interval is a subset of the original threshold interval. At this point, the final result of refined threshold mining under this working condition is as follows: [r name ,r min ,r max ]， The telemetry parameters contain the results of multiple working conditions. According to the distribution of the data of each working condition in the telemetry data, the refined threshold mining results of each working condition will be formed by steps 2 and 3 respectively, and summarized in the form of a list. Finally, the statistical feature extraction and threshold refined mining results of satellite telemetry data with working condition settings are obtained.

Images