A cluster-based satellite telemetry data sequence anomaly detection method

Abstract

The application discloses a satellite telemetry data sequence anomaly detection method based on clustering, and relates to the fields of satellite telemetry data analysis and artificial intelligence. In the application, first, historical telemetry data without anomaly is selected as a training sample, then, the telemetry data is processed in segments, and each segment of telemetry data is a sample. The segmented data sample is clustered by using a distance-based clustering method. On this basis, a one-dimensional Gaussian distribution is fitted by using the training error of the training data. Finally, the distance from the detected telemetry data to its nearest class is calculated, and whether the telemetry data is abnormal data is judged by using the fitted Gaussian distribution. In the application, the satellite operation mode commonly seen in normal telemetry data is automatically learned by performing distance-based clustering analysis on the satellite time sequence telemetry data. The sequence anomaly existing in the satellite telemetry data is found by detecting the abnormal operation mode. The method is used for real-time detection of the sequence anomaly of the satellite in orbit.

Description

Technical Field

[0001] This invention relates to the fields of satellite telemetry data analysis and artificial intelligence technology, and in particular to a clustering-based method for detecting anomalies in satellite telemetry data sequences. Background Technology

[0002] Satellite systems, or their subsystems, have multiple operating modes and may switch between them over time, such as between a solar season mode and a shadow season mode. However, when anomalies occur, the satellite system will operate in an unknown fault mode. Anomalies can be categorized into point anomalies and sequence anomalies based on data patterns. Sequence anomalies, compared to anomalies in a single data point, typically refer to anomalies occurring within a data sequence that cannot be identified by examining any single point. Typical sequence anomalies exist in satellite telemetry data. This invention utilizes Euclidean distance-based clustering to detect sequence anomalies in satellite time-series telemetry data; this method is an unsupervised learning approach. This method performs clustering learning on normal sequence data divided into time windows, and then uses distance calculations to detect sequence anomalies in new data. Summary of the Invention

[0003] The purpose of this invention is to propose a clustering-based method for detecting anomalies in satellite telemetry data sequences in order to solve the above-mentioned problems. This method can effectively detect sequence anomalies in satellite telemetry data.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A clustering-based method for detecting anomalies in satellite telemetry data sequences includes the following steps:

[0006] S1: Create a data table containing M telemetry variables from normal satellite telemetry data, with the data arranged in time series;

[0007] S2: Preprocess the data, including outlier removal and missing value completion;

[0008] S3: Generate N (N=MT) overlapping data sequences from the M-dimensional telemetry data, where T is the length of each data sequence, i.e., each data sequence contains telemetry data at T time points, forming N×T×M training data {x1, x2, ..., x...}. N};

[0009] S4: Use a clustering method based on Euclidean distance to learn clustering of the data and output K classes;

[0010] S5: Using the training error of the training data, fit a univariate Gaussian distribution N(μ′, σ′), and determine the arbitrary sequence of telemetry data x based on clustering prediction and Gaussian distribution. tIs it abnormal?

[0011] Preferably, the clustering method based on Euclidean distance includes:

[0012] Randomly initialize K class center vectors γ = {γ1, γ2, ..., γ3} k};

[0013] For each training data point, i = 1, ..., N, calculate its distance d from each cluster center based on Euclidean distance. i =||x i -γ k || 2 Choose the cluster center c(i) with the shortest distance as its class c(i) = min k ||x i -γ k || 2 ;

[0014] Recalculate the class center vector γ′ for each class, which is equal to the mean of all samples belonging to that class.

[0015] Repeat the above calculations until the class center vector stops updating.

[0016] Preferably, the univariate Gaussian distribution includes:

[0017] After clustering learning is complete, the calculation is performed for each training data x. i The distance to its cluster center, i.e., the training error d i =||x i -γ k || 2 The training error is used to estimate the univariate Gaussian distribution N(μ′, σ′) based on the maximum likelihood estimation method;

[0018] For any sequence of telemetry data x t The distance d from each cluster center is calculated based on Euclidean distance. t =||x t -γ k || 2 The cluster center c(t) with the shortest distance is selected as its class, and its anomaly score is... When a(x) t When ) > 2, determine the sequence telemetry data x t For abnormal data, when a(x) t When )≤2, determine the sequence telemetry data x t This is normal.

[0019] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0020] 1. This application utilizes a machine learning method to perform distance-based clustering analysis on satellite time-series telemetry data, automatically learning common satellite operation patterns found in normal telemetry data. By detecting abnormal operation patterns, sequence anomalies in the satellite telemetry data are identified. This method avoids the limitation of traditional satellite fault diagnosis methods that require threshold settings for all parameters to be detected.

[0021] 2. This application can effectively detect unknown fault modes, solving the problem that traditional threshold methods cannot detect non-exceeding anomalies, and solving the problem that traditional fault diagnosis methods can only detect known fault modes. Attached Figure Description

[0022] Figure 1 A flowchart of a clustering-based satellite telemetry data sequence anomaly detection method according to an embodiment of the present invention is shown. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 The present invention provides a technical solution:

[0025] A clustering-based method for detecting anomalies in satellite telemetry data sequences includes the following steps:

[0026] S1: Create a data table containing M telemetry variables from normal satellite telemetry data, with the data arranged in time series;

[0027] S2: Preprocess the data, including outlier removal and missing value completion;

[0028] S3: Generate N (N=MT) overlapping data sequences from the M-dimensional telemetry data, where T is the length of each data sequence, i.e., each data sequence contains telemetry data at T time points, forming N×T×M training data {x1, x2, ..., x...}. N};

[0029] S4: Use a clustering method based on Euclidean distance to learn clustering of the data and output K classes;

[0030] S5: Using the training error of the training data, fit a univariate Gaussian distribution N(μ′, σ′), and determine the arbitrary sequence of telemetry data x based on clustering prediction and Gaussian distribution.t Is it abnormal?

[0031] Specifically, such as Figure 1 As shown, clustering methods based on Euclidean distance include:

[0032] Randomly initialize K class center vectors γ = {γ1, γ2, ..., γ3} k};

[0033] For each training data point, i = 1, ..., N, calculate its distance d from each cluster center based on Euclidean distance. i =||x i -γ k || 2 Choose the cluster center c(i) with the shortest distance as its class c(i) = min k ||x i -γ k || 2 ;

[0034] Recalculate the class center vector γ′ for each class, which is equal to the mean of all samples belonging to that class.

[0035] Repeat the above calculations until the class center vector stops updating.

[0036] Specifically, such as Figure 1 As shown, the univariate Gaussian distribution includes:

[0037] After clustering learning is complete, the calculation is performed for each training data x. i The distance to its cluster center, i.e., the training error d i =||x i -γ k || 2 The training error is used to estimate the univariate Gaussian distribution N(μ′, σ′) based on the maximum likelihood estimation method;

[0038] For any sequence of telemetry data x t The distance d from each cluster center is calculated based on Euclidean distance. t =||x t -γ k || 2 The cluster center c(t) with the shortest distance is selected as its class, and its anomaly score is... When a(x) t When ) > 2, determine the sequence telemetry data x t For abnormal data, when a(x) t When )≤2, determine the sequence telemetry data x t This is normal.

[0039] This method utilizes a machine learning approach to perform distance-based clustering analysis on satellite time-series telemetry data, automatically learning common satellite operation patterns found in normal telemetry data. By detecting abnormal operation patterns, it identifies sequence anomalies in the satellite telemetry data. This method avoids the limitation of traditional satellite fault diagnosis methods that require threshold settings for all parameters to be detected. Furthermore, this method can effectively detect unknown fault modes, solving the problem that traditional threshold methods cannot detect anomalies within limits and that traditional fault diagnosis methods can only detect known fault modes.

[0040] The above description of the embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A clustering-based method for detecting anomalies in satellite telemetry data sequences, characterized in that, Includes the following steps: S1: Generate a data table containing M telemetry variables from normal satellite telemetry data, with the data arranged in a time series; S2: Preprocess the data, including outlier removal and missing value completion; S3: Generate N (N=MT) overlapping data sequences from the M-dimensional telemetry data, where T is the length of each data sequence, i.e., each data sequence contains telemetry data at T time points, forming N×T×M training data {x1,x2,…,v}. N }; S4: Use a clustering method based on Euclidean distance to learn clustering of the data and output K classes; S5: Using the training error of the training data, fit a univariate Gaussian distribution N(μ′,σ′), and determine the arbitrary sequence of telemetry data x based on clustering prediction and Gaussian distribution. t Is it abnormal? The Euclidean distance-based clustering method described in step S4 includes: Randomly initialize K class center vectors γ = {γ1, γ2, ..., γ k }; For each training data point, i = 1, ..., N, calculate its distance d from each cluster center based on Euclidean distance. i =||x i -γ k || 2 Choose the cluster center c(i) with the shortest distance as its class c(i) = min k ||x i -γ k || 2 ; Recalculate the class center vector γ′ for each class, which is equal to the mean of all samples belonging to that class. Repeat the above calculations until the class center vector stops updating.

2. The method for detecting anomalies in satellite telemetry data sequences based on clustering according to claim 1, characterized in that, The univariate Gaussian distribution includes: After clustering learning is complete, the calculation is performed for each training data x. i The distance to its cluster center, i.e., the training error d i =||x i -γ k || 2 The training error is used to estimate the univariate Gaussian distribution N(μ′,σ′) based on the maximum likelihood estimation method; For any sequence of telemetry data x t The distance d from each cluster center is calculated based on Euclidean distance. t =||x t -γ k || 2 The cluster center c(t) with the shortest distance is selected as its class, and its anomaly score is... When a(x) t When )>2, determine the sequence telemetry data x t For abnormal data, when a(x) t When )≤2, determine the sequence telemetry data x t This is normal.

Images