Real-time detection method of multi-source anomaly information of real-time orbit determination of Beidou system

Abstract

The application discloses a multi-source abnormal information real-time detection method for a Beidou system real-time orbit determination, and relates to the field of satellite anomaly detection. The method takes new information obtained through real-time orbit determination filtering calculation as a detection factor, classifies and selects the detection factor through a phase jump detection threshold, cooperatively screens the abnormal detection factor through an isolated forest algorithm and a DBSCAN clustering analysis method, and determines the abnormal source through the proportion of the abnormal detection factor. The method can effectively detect abnormal information of different orders of magnitude, different characteristics and different sources, and greatly reduces the influence of abnormal information on the service capacity of the Beidou system.

Description

Technical Field

[0001] This invention relates to the field of satellite anomaly detection, specifically to a real-time detection method for multi-source anomaly information in the BeiDou system's real-time orbit determination. Background Technology

[0002] During their operation in orbit, BeiDou satellites face a variety of abnormal information streams with diverse characteristics that dynamically change over time, posing a serious threat to the high-precision service capabilities of the BeiDou system. Therefore, clarifying the specific sources and manifestations of multi-source abnormal information, constructing detection factors that are sensitive to multi-source anomalies, and researching detection factor modeling algorithms are of great significance for the stable operation of the BeiDou system and maintaining real-time high-precision services.

[0003] Hardware failures in the BeiDou satellite signal transponder can lead to abnormal transmitted signals, resulting in gross observations (observation anomalies). During long-term operation, GEO and IGSO satellites are subject to perturbations such as the Earth's non-spherical gravity, gradually deviating from their preset orbits and frequently maneuvering to maintain their positions (state anomalies). Furthermore, influenced by factors such as the atomic clock manufacturing principle, the master control station's active phase and frequency modulation, and changes in temperature and humidity, BeiDou satellite atomic clocks may experience phase and frequency jumps (state anomalies). These unknown, sudden, and highly outlier-like observation and state anomalies pose significant challenges to the reliability and integrity of the BeiDou system. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, the multi-source anomaly detection method for real-time orbit determination of the BeiDou system provided by this invention can quickly detect anomalies in BeiDou satellite observations and status.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A method for real-time detection of multi-source anomalies in the BeiDou system for real-time orbit determination is provided, which includes the following steps: S1. Calculate the predicted state value of the satellite in the current epoch by obtaining the state transition matrix of the satellite between the current epoch and the previous epoch. S2. Based on the satellite predicted state value and the satellite observation value, obtain the filtered innovation value; S3. Set the phase jump detection threshold, and determine whether there is a filtered innovation value greater than or equal to the phase jump detection threshold in the epoch to be detected. If so, use the filtered innovation value greater than or equal to the phase jump detection threshold as the detection factor; otherwise, use all filtered innovation values ​​in the epoch to be detected as the detection factor. S4. Calculate the outlier score corresponding to the detection factor using the isolated forest algorithm, and record the detection factor whose outlier score is greater than or equal to the score threshold as the outlier detection factor. S5. Cluster the detection factors using the DBSCAN clustering analysis method to obtain the anomaly detection factors; S6. By calculating the proportion of anomaly detection factors, the source of anomalies is determined, and real-time detection of multi-source anomaly information for real-time orbit determination of the BeiDou system is completed.

[0006] Furthermore, the specific method for obtaining the satellite's state transition matrix between the current epoch and the previous epoch in step S1 is as follows: Based on the perturbations from Earth, Sun and Moon, Earth's tides, relativistic effects, and solar radiation pressure experienced by the GEO, IGSO, and MEO satellites of the BeiDou system in orbit, perturbation motion equations are established, and the state transition matrix between the current epoch and the previous epoch is obtained using numerical methods. ; The expression for the satellite prediction state value within the current epoch in step S1 is:

[0007] in This represents the predicted satellite state value within the current epoch. These are the satellite state parameters for the previous epoch.

[0008] Furthermore, the expression for calculating the filtered innovation value is as follows:

[0009] in This is the filtered innovation value matrix for the current epoch. , The first in the current epoch m Each filtered information value; For designing the matrix; This is the matrix of satellite observations within the current epoch.

[0010] Furthermore, the specific method for calculating the outliers corresponding to the detection factors using the isolated forest algorithm in step S4 includes the following sub-steps: S4-1. Obtain several filter innovation values ​​using the same method as in step S2. Starting from the epoch when the filter innovation values ​​converge, select filter innovation values ​​within N epochs to form a training set. Train the isolated forest algorithm using the training set to obtain the trained isolated forest algorithm. S4-2. Using an epoch-by-epoch sequence, the probe factor traverses each isolated tree of the trained isolated forest algorithm and calculates the average height of the probe factor in the isolated forest. S4-3. Normalize the average height of all detection factors in the isolated forest and calculate the outlier score for each detection factor.

[0011] Furthermore, the specific method for training the Isolation Forest algorithm in step S4-1 includes the following sub-steps: S4-1-1, Randomly select from the training set Each filtered innovation value is used as a subset of samples; S4-1-2 For any subset of samples, perform a segmentation operation: Randomly select a feature, and then randomly select a split point within the value range of that feature to divide the current data into two parts; data with values ​​less than the split point in the randomly selected feature are placed in the left child node of the current node, and data with values ​​greater than or equal to the split point are placed in the right child node of the current node; the data in the first split is a sample subset; S4-1-3. Repeat the segmentation operation in step S4-1-2 until the data of the sample subset cannot be further divided, obtain the leaf nodes, and thus complete the construction of an isolated tree. S4-1-4. Repeat steps S4-1-1 to S4-1-3 until T isolated trees are generated, thus obtaining the trained isolated forest algorithm.

[0012] Furthermore, The value of is 256, and the value of T is 100.

[0013] Furthermore, the expression for calculating the outlier score corresponding to each detection factor in step S4-3 is as follows:

[0014]

[0015]

[0016] in It is the harmonic number; It is the natural logarithm; It is Euler's constant; The average height of the nodes in the isolated tree; For detection factor The corresponding outlier score; For detection factor The height after normalization; the score threshold in step S4 is 0.6.

[0017] Furthermore, in step S5, the detection factors are clustered using the DBSCAN clustering analysis method to obtain the anomaly detection factors. The specific method includes the following sub-steps: S4-1. Set the neighborhood radius R and the minimum number of points in the neighborhood minpoints; S4-2. Traverse all probe factors of the epoch to be probed. For probe factors with a number of points greater than or equal to minpoints within the radius, include them as core points in the core point list. Form temporary clusters corresponding to the core points by combining the probe factors within the radius of the core points. S4-3. For each temporary cluster, check whether the probe factor is a core point. If so, merge the temporary cluster corresponding to the probe factor that is the core point with the current temporary cluster until each probe factor in the temporary cluster is not in the core point list, or all probe factors within the neighborhood radius R of each probe factor in the temporary cluster are in the same temporary cluster as it, thus obtaining the cluster. S4-4. Detection factors that are not classified into clusters are considered as anomalous detection factors.

[0018] Furthermore, the specific method for determining the source of anomalies by calculating the proportion of anomaly detection factors in step S6 includes the following sub-steps: S5-1. Calculate the proportion of the anomaly detection factor, the expression of which is:

[0019] in This represents the proportion of anomaly detection factors; Indicates the total number of detection factors; Indicates the total number of anomaly detection factors; S5-2. When there is a filtered innovation value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of abnormal detection factors is 100%, it is determined that the corresponding satellite has a phase jump fault. When there is no filtered information value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of abnormal detection factors of each satellite in the BeiDou system is less than 90%, it is judged as an observation anomaly. If there is no filtered information value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of the satellite's abnormal detection factor is greater than or equal to 90% and less than 100%, then the satellite is determined to be in an abnormal state, and the abnormality is determined to be caused by the frequency jump of the satellite's atomic clock. If there is no filtered information value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of the satellite's abnormal detection factor is 100%, then the satellite is determined to be in an abnormal state, and the abnormality is determined to be caused by the satellite's orbital maneuver.

[0020] The beneficial effects of this invention are as follows: This method uses the information obtained from real-time orbit determination filtering as a detection factor, classifies and selects the detection factors by phase jump detection threshold, and uses the isolated forest algorithm and DBSCAN clustering analysis method to screen abnormal detection factors in a collaborative manner. The source of the anomaly is determined by the proportion of abnormal detection factors. This method can effectively detect anomalies of different magnitudes, characteristics and sources, and greatly reduce the impact of abnormal information on the service capabilities of the BeiDou system. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the method. Detailed Implementation

[0022] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0023] like Figure 1 As shown, the real-time detection method for multi-source anomaly information in the BeiDou system's real-time orbit determination includes the following steps: S1. Calculate the predicted state value of the satellite in the current epoch by obtaining the state transition matrix of the satellite between the current epoch and the previous epoch. S2. Based on the satellite predicted state value and the satellite observation value, obtain the filtered innovation value; S3. Set the phase jump detection threshold, and determine whether there is a filtered innovation value greater than or equal to the phase jump detection threshold in the epoch to be detected. If so, use the filtered innovation value greater than or equal to the phase jump detection threshold as the detection factor; otherwise, use all filtered innovation values ​​in the epoch to be detected as the detection factor; the current epoch is the epoch to be detected. S4. Calculate the outlier score corresponding to the detection factor using the isolated forest algorithm, and record the detection factor whose outlier score is greater than or equal to the score threshold as the outlier detection factor. S5. Cluster the detection factors using the DBSCAN clustering analysis method to obtain the anomaly detection factors; S6. By calculating the proportion of anomaly detection factors, the source of anomalies is determined, and real-time detection of multi-source anomaly information for real-time orbit determination of the BeiDou system is completed.

[0024] The specific method for obtaining the satellite's state transition matrix between the current epoch and the previous epoch in step S1 is as follows: Based on the perturbations from Earth, Sun and Moon, Earth's tides, relativistic effects, and solar radiation pressure experienced by the GEO, IGSO, and MEO satellites of the BeiDou system in orbit, perturbation motion equations are established, and the state transition matrix between the current epoch and the previous epoch is obtained using numerical methods. ; The expression for the satellite prediction state value within the current epoch in step S1 is:

[0025] in This represents the predicted satellite state value within the current epoch. These are the satellite state parameters for the previous epoch.

[0026] The expression for calculating the filtered innovation value is:

[0027] in This is the filtered innovation value matrix for the current epoch. , The first in the current epoch m Each filtered information value; For designing the matrix; This is the matrix of satellite observations within the current epoch.

[0028] The specific method for calculating the outliers corresponding to the probe factors using the Isolation Forest algorithm in step S4 includes the following sub-steps: S4-1. Obtain several filter innovation values ​​using the same method as in step S2. Starting from the epoch when the filter innovation values ​​converge, select filter innovation values ​​within N epochs to form a training set. Train the isolated forest algorithm using the training set to obtain the trained isolated forest algorithm. S4-2. Using an epoch-by-epoch sequence, the probe factor traverses each isolated tree of the trained isolated forest algorithm and calculates the average height of the probe factor in the isolated forest. S4-3. Normalize the average height of all detection factors in the isolated forest and calculate the outlier score for each detection factor.

[0029] The specific method for training the Isolation Forest algorithm in step S4-1 includes the following sub-steps: S4-1-1, Randomly select from the training set Each filtered innovation value is used as a subset of samples; S4-1-2 For any subset of samples, perform a segmentation operation: Randomly select a feature, and then randomly select a split point within the value range of that feature to divide the current data into two parts; data with values ​​less than the split point in the randomly selected feature are placed in the left child node of the current node, and data with values ​​greater than or equal to the split point are placed in the right child node of the current node; the data in the first split is a sample subset; S4-1-3. Repeat the segmentation operation in step S4-1-2 until the data of the sample subset cannot be further divided, obtain the leaf nodes, and thus complete the construction of an isolated tree. S4-1-4. Repeat steps S4-1-1 to S4-1-3 until T isolated trees are generated, thus obtaining the trained isolated forest algorithm.

[0030] The value of is 256, and the value of T is 100.

[0031] The expression for calculating the outlier score for each probe factor in step S4-3 is as follows:

[0032]

[0033]

[0034] in It is the harmonic number; It is the natural logarithm; It is Euler's constant; The average height of the nodes in the isolated tree; For detection factor The corresponding outlier score; For detection factor The height after normalization; the score threshold in step S4 is 0.6.

[0035] Step S5 involves clustering the detection factors using the DBSCAN clustering analysis method to obtain the anomaly detection factors. The specific method includes the following sub-steps: S4-1. Set the neighborhood radius R and the minimum number of points in the neighborhood minpoints; S4-2. Traverse all probe factors of the epoch to be probed. For probe factors with a number of points greater than or equal to minpoints within the radius, include them as core points in the core point list. Form temporary clusters corresponding to the core points by combining the probe factors within the radius of the core points. S4-3. For each temporary cluster, check whether the probe factor is a core point. If so, merge the temporary cluster corresponding to the probe factor that is the core point with the current temporary cluster until each probe factor in the temporary cluster is not in the core point list, or all probe factors within the neighborhood radius R of each probe factor in the temporary cluster are in the same temporary cluster as it, thus obtaining the cluster. S4-4. Detection factors that are not classified into clusters are considered as anomalous detection factors.

[0036] In this embodiment, both R and minpoints are set to 4.

[0037] The specific method for determining the source of anomalies by calculating the proportion of anomaly detection factors in step S6 includes the following sub-steps: S5-1. Calculate the proportion of the anomaly detection factor, the expression of which is:

[0038] in This represents the proportion of anomaly detection factors; Indicates the total number of detection factors; Indicates the total number of anomaly detection factors; S5-2. When there is a filtered innovation value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of abnormal detection factors is 100%, it is determined that the corresponding satellite has a phase jump fault. When there is no filtered information value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of abnormal detection factors of each satellite in the BeiDou system is less than 90%, it is judged as an observation anomaly. If there is no filtered information value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of the satellite's abnormal detection factor is greater than or equal to 90% and less than 100%, then the satellite is determined to be in an abnormal state, and the abnormality is determined to be caused by the frequency jump of the satellite's atomic clock. If there is no filtered information value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of the satellite's abnormal detection factor is 100%, then the satellite is determined to be in an abnormal state, and the abnormality is determined to be caused by the satellite's orbital maneuver.

[0039] In the specific implementation process, taking the anomaly detection results of the 102-epoch observation as an example, k is the satellite PRN number, i is the number of anomalous new information of the satellite in this epoch; m is the total number of new information of the satellite in this epoch, and P is the anomaly ratio. The detection results of the isolated forest algorithm are shown in Table 1 below.

[0040] Table 1

[0041] The detection results obtained by the DBSCAN clustering analysis method are shown in Table 2 below.

[0042] Table 2

[0043] During the anomaly detection process at 102 epochs, due to the influence of simulated satellite line-of-sight conditions or the failure to assign gross errors to some satellites when generating simulation gross errors, some satellites had no detection results.

[0044] In the specific implementation process, anomaly detection is performed at a certain epoch, where k is the satellite PRN number, i is the number of anomalous new information in that epoch of the satellite, m is the total number of new information in that epoch of the satellite, and P is the anomaly ratio. The detection results using the Isolation Forest algorithm are shown in Table 3 below.

[0045] Table 3

[0046] The detection results obtained by the DBSCAN clustering analysis method are shown in Table 4 below.

[0047]

[0048] As can be seen from Tables 3 and 4, PRN1 is an anomalous satellite that underwent maneuvering.

[0049] The frequency jump data of the simulated satellite atomic clock during the specific implementation process is shown in Table 5.

[0050] Table 5

[0051] The results of using the isolated forest algorithm to probe PRN1&23 are shown in Table 6.

[0052] Table 6

[0053] The results of using the isolated forest algorithm to probe PRN7&22 are shown in Table 7.

[0054] Table 7

[0055] The detection results of PRN1&23 using the DBSCAN clustering analysis method are shown in Table 8 below.

[0056] Table 8

[0057] The detection results of PRN7&22 using the DBSCAN clustering analysis method are shown in Table 9 below.

[0058] Table 9

[0059] The statistical results of PRN1 frequency jumps are shown in Table 10.

[0060] Table 10

[0061] The statistical results of PRN7 frequency jumps are shown in Table 11.

[0062] Table 11

[0063] As can be seen from the experimental data above, this method uses the information obtained from real-time orbit determination filtering as the detection factor, and classifies and selects the detection factor by phase jump detection threshold. Both DBSCAN clustering analysis method and isolated forest algorithm can effectively screen the detection factor for anomalies. The source of the anomaly is determined by the proportion of the results of the two co-screening methods. It can effectively detect anomalies of different magnitudes, characteristics and sources, and greatly reduce the impact of anomaly information on the service capability of Beidou system.

Claims

1. A method for real-time detection of multi-source anomaly information for real-time orbit determination in the BeiDou system, characterized in that, Includes the following steps: S1. Calculate the predicted state value of the satellite in the current epoch by obtaining the state transition matrix of the satellite between the current epoch and the previous epoch. S2. Based on the satellite predicted state value and the satellite observation value, obtain the filtered innovation value; S3. Set the phase jump detection threshold, and determine whether there is a filtered innovation value greater than or equal to the phase jump detection threshold within the epoch to be detected. If so, use the filtered innovation value greater than or equal to the phase jump detection threshold as the detection factor. Otherwise, all filtered innovation values ​​of the epoch to be detected will be used as detection factors; S4. Calculate the outlier score corresponding to the detection factor using the isolated forest algorithm, and record the detection factor whose outlier score is greater than or equal to the score threshold as the outlier detection factor. S5. Cluster the detection factors using the DBSCAN clustering analysis method to obtain the anomaly detection factors; S6. By calculating the proportion of anomaly detection factors, the source of anomalies is determined, and real-time detection of multi-source anomaly information for real-time orbit determination of the BeiDou system is completed. The specific method for calculating the outliers corresponding to the probe factors using the Isolation Forest algorithm in step S4 includes the following sub-steps: S4-1. Obtain several filter innovation values ​​using the same method as in step S2. Starting from the epoch when the filter innovation values ​​converge, select filter innovation values ​​within N epochs to form a training set. Train the isolated forest algorithm using the training set to obtain the trained isolated forest algorithm. S4-2. Using an epoch-by-epoch sequence, the probe factor traverses each isolated tree of the trained isolated forest algorithm and calculates the average height of the probe factor in the isolated forest. S4-3. Normalize the average height of all detection factors in the isolated forest and calculate the outlier score for each detection factor. The specific method for training the Isolation Forest algorithm in step S4-1 includes the following sub-steps: S4-1-1, Randomly select from the training set Each filtered innovation value is used as a subset of samples; S4-1-2 For any subset of samples, perform a segmentation operation: Randomly select a feature, and then randomly select a split point within the value range of that feature to divide the current data into two parts; data with values ​​less than the split point in the randomly selected feature are placed in the left child node of the current node, and data with values ​​greater than or equal to the split point are placed in the right child node of the current node; the data in the first split is a sample subset; S4-1-3. Repeat the segmentation operation in step S4-1-2 until the data of the sample subset cannot be further divided, obtain the leaf nodes, and thus complete the construction of an isolated tree. S4-1-4. Repeat steps S4-1-1 to S4-1-3 until T isolated trees are generated, thus obtaining the trained isolated forest algorithm.

2. The method for real-time detection of multi-source anomaly information in BeiDou system real-time orbit determination according to claim 1, characterized in that, The specific method for obtaining the satellite's state transition matrix between the current epoch and the previous epoch in step S1 is as follows: Based on the perturbations from Earth, Sun and Moon, Earth's tides, relativistic effects, and solar radiation pressure experienced by the GEO, IGSO, and MEO satellites of the BeiDou system in orbit, perturbation motion equations are established, and the state transition matrix between the current epoch and the previous epoch is obtained using numerical methods. ; The expression for the satellite prediction state value within the current epoch in step S1 is: in This represents the predicted satellite state value within the current epoch. These are the satellite state parameters for the previous epoch.

3. The method for real-time detection of multi-source anomaly information in BeiDou system real-time orbit determination according to claim 2, characterized in that, The expression for calculating the filtered innovation value is: in This is the filtered innovation value matrix for the current epoch. , The first in the current epoch m Each filtered information value; For designing the matrix; This is the matrix of satellite observations within the current epoch.

4. The method for real-time detection of multi-source anomaly information in BeiDou system real-time orbit determination according to claim 1, characterized in that, The value of is 256, and the value of T is 100.

5. The method for real-time detection of multi-source anomaly information for real-time orbit determination of the BeiDou system according to claim 1, characterized in that, The expression for calculating the outlier score for each probe factor in step S4-3 is as follows: in It is the harmonic number; It is the natural logarithm; It is Euler's constant; The average height of the nodes in the isolated tree; For detection factor The corresponding outlier score; For detection factor The height after normalization; the score threshold in step S4 is 0.

6.

6. The method for real-time detection of multi-source anomaly information in BeiDou system real-time orbit determination according to claim 1, characterized in that, Step S5 involves clustering the detection factors using the DBSCAN clustering analysis method to obtain the anomaly detection factors. The specific method includes the following sub-steps: S4-1. Set the neighborhood radius R and the minimum number of points in the neighborhood minpoints; S4-2. Traverse all probe factors of the epoch to be probed. For probe factors with a number of points greater than or equal to minpoints within the radius, include them as core points in the core point list. Form temporary clusters corresponding to the core points by combining the probe factors within the radius of the core points. S4-3. For each temporary cluster, check whether the probe factor is a core point. If so, merge the temporary cluster corresponding to the probe factor that is the core point with the current temporary cluster until each probe factor in the temporary cluster is not in the core point list, or all probe factors within the neighborhood radius R of each probe factor in the temporary cluster are in the same temporary cluster as it, thus obtaining the cluster. S4-4. Detection factors that are not classified into clusters are considered as anomalous detection factors.

7. The method for real-time detection of multi-source anomaly information for real-time orbit determination of the BeiDou system according to claim 1, characterized in that, The specific method for determining the source of anomalies by calculating the proportion of anomaly detection factors in step S6 includes the following sub-steps: S5-1. Calculate the proportion of the anomaly detection factor, the expression of which is: in This represents the proportion of anomaly detection factors; Indicates the total number of detection factors; Indicates the total number of anomaly detection factors; S5-2. When there is a filtered innovation value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of abnormal detection factors is 100%, it is determined that the corresponding satellite has a phase jump fault. When there is no filtered information value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of abnormal detection factors of each satellite in the BeiDou system is less than 90%, it is judged as an observation anomaly. If there is no filtered information value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of the satellite's abnormal detection factor is greater than or equal to 90% and less than 100%, then the satellite is determined to be in an abnormal state, and the abnormality is determined to be caused by the frequency jump of the satellite's atomic clock. If there is no filtered information value greater than or equal to the phase jump detection threshold within the epoch to be detected, and the proportion of the satellite's abnormal detection factor is 100%, then the satellite is determined to be in an abnormal state, and the abnormality is determined to be caused by the satellite's orbital maneuver.

Images