A Clustering-Based Fault Diagnosis Method for Satellite Telemetry Data

Abstract

The present invention relates to a satellite telemetry data fault diagnosis method based on clustering. First, the model selection of Gaussian mixture model is performed by using the YES information criterion to determine the number of mixture models. Then, the Gaussian mixture model is trained by using normal multivariate satellite telemetry data. On this basis, a univariate Gaussian distribution is learned by using the likelihood score on the training set. When the likelihood score of the data meets a given condition, the data is determined to be faulty data. The present invention is used for real-time detection of on-orbit satellite faults.

Description

Technical Field

[0001] The invention relates to a satellite telemetry data fault diagnosis method based on clustering, belonging to the technical field of satellite telemetry data analysis. Background Art

[0002] The satellite system or each of its subsystems has multiple different working modes and will change from one mode to another over time, such as the switching between the illumination season mode and the earth shadow season mode. However, when an abnormality occurs, the satellite system will work in an unknown fault mode. Therefore, data modeling of common working operations and detecting unknown working modes can achieve the purpose of fault diagnosis.

[0003] Current fault diagnosis techniques only detect anomalies on a small number of parameters and are unable to respond to unexpected anomalies. Summary of the invention

[0004] The technical problem solved by the present invention is: to overcome the deficiencies of the prior art and to propose a satellite telemetry data fault diagnosis method based on clustering to effectively detect faults in satellite telemetry data.

[0005] The solution of the present invention is:

[0006] A clustering-based satellite telemetry data fault diagnosis method, the steps are as follows:

[0007] (1) Normal satellite telemetry data is converted into a data table containing M telemetry variables and N samples as training data for the model;

[0008] (2) Preprocess the data, including outlier removal and missing value filling;

[0009] (3) Use the Bayesian information criterion to infer the number of mixture components K in the Gaussian mixture model from the training data, and select the best model by testing different K values;

[0010] (4) Using the training data, train a Gaussian mixture model with a hyperparameter K;

[0011] (5) Using the expectation maximization algorithm to estimate the parameters of the Gaussian mixture model, starting from an initial state of the parameter set π to be estimated by the Gaussian mixture model, iteratively updating π until the model converges;

[0012] (6) After the model parameters are estimated, the likelihood score on the training set is used to estimate the univariate Gaussian distribution N(μ',σ') based on the maximum likelihood estimation method;

[0013] (7) According to the z-score algorithm, for any telemetry data x t , whose anomaly score is p is the probability, when a(x t )>2σ', determine the telemetry data x t For fault data.

[0014] Furthermore, the Gaussian mixture model described in step (4) is Where x is the input vector set, π is the parameter of the Gaussian mixture model, K is the number of sub-Gaussian models in the mixture model, α i ,i=1,....,k is the weight of the sub-Gaussian model, g(x|μ i ,∑ i ) is the density function of the sub-Gaussian model, μ i is the mean vector, ∑ i is the covariance matrix.

[0015] Furthermore, π={α i ,μ i ∑ i}.

[0016] Further, satisfy

[0017] Furthermore,

[0018] Furthermore, each iteration consists of an E step and an M step.

[0019] Step E uses the formula Calculate the posterior probability of each data sample in each Gaussian component, g(x t |μ i ,∑ i ) is the probability value generated by the i-th sub-Gaussian model, g(x t |μ j ,∑ j ) is the sum of the probabilities generated by all sub-Gaussian models,

[0020] Among them, α i ,i=1,....,k is the weight of the sub-Gaussian model,

[0021] M step uses the formula and Update the parameters π of the Gaussian mixture model when the log-likelihood: Convergence is achieved when there is almost no significant change from one iteration to the next.

[0022] N is the number of samples, μ i is the mean vector, ∑ i is the covariance matrix.

[0023] The beneficial effects of the present invention compared with the prior art are:

[0024] (1) The present invention uses a Gaussian mixture model to perform cluster analysis on satellite telemetry data, automatically learning common satellite operation modes in normal telemetry data; by detecting abnormal operation modes, faults in satellite telemetry data are discovered; this method avoids the limitation of traditional satellite fault diagnosis methods that require threshold setting for all parameters to be detected, greatly reducing the workload;

[0025] (2) The present invention can effectively detect any unknown fault mode, thus overcoming the shortcoming that traditional fault diagnosis methods can only detect known fault modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0027] The present invention will be further described below in conjunction with the embodiments.

[0028] The satellite telemetry data fault diagnosis method based on clustering described in this embodiment is as follows: Figure 1 As shown, specifically including:

[0029] (1) Normal satellite telemetry data is converted into a data table containing M telemetry variables and N samples as training data for the model;

[0030] (2) Preprocess the data, including outlier removal and missing value filling;

[0031] (3) Use the Bayesian Information Criterion to infer the number of mixture components K in the Gaussian mixture model from the training data. By testing different K values, the best model is selected.

[0032] (4) Using the training data, train the Gaussian mixture model with hyperparameter K Where x is the input vector set, π={α i ,μ i ∑ i} is the parameter of the Gaussian mixture model, K is the number of sub-Gaussian models in the mixture model, α i , i=1,...., k is the weight of the sub-Gaussian model, and satisfies is the density function of the sub-Gaussian model, μ i is the mean vector, ∑ i is the covariance matrix;

[0033] (5) Parameter estimation of the Gaussian mixture model is performed using the expectation maximization algorithm. The expectation maximization algorithm is an iterative algorithm that starts from a certain initial state of π and then iteratively updates π until the model converges. Each iteration consists of an E step and an M step;

[0034] (6) Step E uses the formula Calculate the posterior probability of each data sample in each Gaussian component;

[0035] (7) M step uses the formula and Update the parameters π of the Gaussian mixture model;

[0036] (8) When the log-likelihood Convergence is achieved when there is almost no significant change from one iteration to the next.

[0037] (9) After the model parameters are estimated, the likelihood score of the newly observed telemetry data is generated using the formula in step (8). The univariate Gaussian distribution N(μ',σ') is estimated based on the maximum likelihood estimation method using the negative log likelihood score on the training set.

[0038] (10) For any telemetry data x t , whose anomaly score is When a(x t )>2σ', determine the telemetry data x t For fault data.

[0039] The present invention performs cluster analysis on satellite telemetry data through a Gaussian mixture model, automatically learns common satellite operation modes in normal telemetry data, and detects faults in satellite telemetry data by detecting abnormal operation modes. This method avoids the limitation of traditional satellite fault diagnosis methods that require threshold setting for all parameters to be detected, greatly reducing the workload.

[0040] The present invention can effectively detect any unknown fault mode, thus overcoming the shortcoming that the traditional fault diagnosis method can only detect known fault modes.

[0041] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

Claims

1. A satellite telemetry data fault diagnosis method based on clustering, characterized in that: The steps are as follows: (1) Convert normal satellite telemetry data into M telemetry variables and N The data table of samples is used as the training data of the model; (2) Preprocess the data, including removing outliers and filling in missing values; (3) Use the Bayesian Information Criterion to infer the mixture components in the Gaussian mixture model from the training data K The number of K value, select the best model; (4) Using the training data, the training hyperparameters are K Gaussian mixture model of (5) Use the expectation maximization algorithm to estimate the parameters of the Gaussian mixture model. Starting from an initial state, iterative update , until the model converges; Each iteration consists of an E step and an M step. Step E uses the formula Calculate the posterior probability of each data sample in each Gaussian component, is the probability value generated by the i-th sub-Gaussian model, is the sum of the probabilities generated by all sub-Gaussian models, in, is the weight of the sub-Gaussian model, M step uses the formula , and , update the parameters of the Gaussian mixture model , when the log-likelihood: Convergence is achieved when there is almost no significant change from one iteration to the next. N is the number of samples, is the mean vector, is the covariance matrix; (6) After the model parameters are estimated, the likelihood score on the training set is used to estimate the univariate Gaussian distribution based on the maximum likelihood estimation method. ; (7) According to the z-score algorithm, for any telemetry data , whose anomaly score is , p is the probability, when When the telemetry data is determined is the fault data; The Gaussian mixture model described in step (4) is ,in is the input vector set, are the parameters of the Gaussian mixing mode, K is the number of sub-Gaussian models in the mixture model, is the weight of the sub-Gaussian model, is the density function of the sub-Gaussian model, is the mean vector, is the covariance matrix.

2. A clustering-based satellite telemetry data fault diagnosis method according to claim 1, characterized in that: 。 3. The satellite telemetry data fault diagnosis method based on clustering according to claim 1 is characterized in that: satisfy .

4. The satellite telemetry data fault diagnosis method based on clustering according to claim 1, characterized in that: 。

Images