CN111611294B - Star sensor data anomaly detection method - Google Patents
Star sensor data anomaly detection method Download PDFInfo
- Publication number
- CN111611294B CN111611294B CN202010361728.2A CN202010361728A CN111611294B CN 111611294 B CN111611294 B CN 111611294B CN 202010361728 A CN202010361728 A CN 202010361728A CN 111611294 B CN111611294 B CN 111611294B
- Authority
- CN
- China
- Prior art keywords
- data
- telemetry data
- detection
- quaternion
- period
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2462—Approximate or statistical queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2474—Sequence data queries, e.g. querying versioned data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Fuzzy Systems (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention provides a star sensor data anomaly detection method, relates to an anomaly data detection method, and can solve the problems that satellite attitude and operation orbit are influenced by star sensor working anomaly, so that satellite attitude deviates from a control range and even turns over, and serious loss is caused, more false alarm and false alarm missing problems are frequently caused. The specific technical scheme is as follows: selecting telemetry sample data for preprocessing to obtain target telemetry data; selecting the detection quantity and obtaining a confidence interval of the detection quantity of the target telemetry data through confidence analysis; and detecting that the target telemetry data exceeds the confidence interval on line to be abnormal data. The method is used for detecting and managing the data anomalies of the high-orbit satellite sensor.
Description
Technical Field
The disclosure relates to the field of spacecraft fault diagnosis, in particular to a star sensor data anomaly detection method.
Background
The fault diagnosis of the spacecraft is an important means for timely finding and processing the anomaly of the spacecraft, and plays an important role in long-term management of the spacecraft. The star sensor can provide accurate space orientation and reference for satellites, has autonomous navigation capability, and is an important measurement component of a satellite attitude control subsystem. The abnormal operation of the star sensor can influence the satellite attitude and the running orbit, so that the satellite attitude deviates from the control range and even turns over, and great loss is caused. In daily measurement and control, the monitoring of relevant parameters of the star sensor is very important, and the quaternion is a core attitude parameter of the star sensor. Machine learning is a multi-domain interdisciplinary, and research computers simulate human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the machine learning. The machine learning is characterized by being based on data, feature extraction and autonomous updating, can be used for designing a time sequence model of a machine learning algorithm for determining the change of a telemetry parameter according to time, analyzing the deviation between the model and measured data, and can be approximately used as a real change rule of the parameter in a certain deviation range for subsequent detection, prediction and evaluation.
The fault diagnosis logic and reasoning method mainly comprises a rule-based reasoning method, a model-based reasoning method and a case-based reasoning method, is applied to actual technologies such as threshold judgment, expert system, logic judgment and the like, plays an important role, and still has the following limitations: the threshold for anomaly detection is often rough depending on experience and design principles of spacecraft development units;
the expert system has the problems of unsophisticated weight distribution, inexact logic, insufficient basis and the like, and logic judgment is often unable to meet the field value or burst state conversion, so that more false alarms and false alarms are often generated in the application of the spacecraft fault diagnosis system.
Disclosure of Invention
The invention establishes a star sensor data anomaly detection method, which is used for solving the problem of telemetry data detection which cannot be dealt with based on a threshold judgment method in the existing diagnostic system. The problem that more false alarms and missed alarms frequently occur in the application of the spacecraft fault diagnosis system is solved.
The invention is based on the basic principle of machine learning, combines the basic ideas of big data and data mining, takes massive telemetry data as objects, and randomly divides the telemetry data into a training set, a testing set and a detecting set according to the independence principle through a telemetry data preprocessing algorithm. Two types of detection quantities are extracted according to the quaternion characteristics of the star sensor and serve as identification features of remote measurement parameter anomalies, a quaternion time sequence model of the star sensor is fitted through machine learning, verification and confidence analysis are carried out on the fitted model, and a method for moving a time window is designed for anomaly detection.
According to a first aspect of an embodiment of the present disclosure, there is provided a star sensor data anomaly detection method, including:
selecting telemetry sample data for preprocessing to obtain target telemetry data; selecting the detection quantity and obtaining a confidence interval of the detection quantity of the target telemetry data through confidence analysis; detecting that the target telemetry data exceeds the confidence interval on line to be abnormal data;
the confidence interval is the confidence interval of the single parameter model error detection amount; or, a confidence interval of the correlation coefficient detection amount;
the detection quantity is selected as single-parameter detection quantity; or, selecting the multi-dimensional telemetry association detection quantity.
In one embodiment, telemetry data preprocessing is to cull telemetry sample data outliers, obtain target telemetry data segments, and periodically segment and diversity the target telemetry data segments.
The abnormal value rejection is to reject abnormal values near zero values or reject abnormal values exceeding the upper limit and the lower limit of the telemetry data;
in one embodiment, the removing of the abnormal value near the zero value means scanning from the time starting point of the target telemetry data at least in one moving time window period, sorting the data in the moving time window according to the size, obtaining the median of the sorted sequence, adding three times of standard deviation of the data in the moving time window by taking the median as a reference to obtain an upper limit interval and a lower limit interval, and removing the data outside the upper limit interval and the lower limit interval;
the method comprises the steps of eliminating abnormal values exceeding the upper limit and the lower limit of telemetry data, namely carrying out statistical analysis on quaternion telemetry data of a satellite star sensor through a computer program, determining the upper limit and the lower limit of a quaternion telemetry data sequence, and eliminating abnormal values outside the upper limit and the lower limit of the quaternion telemetry data.
In one embodiment, the period segmentation of the target telemetry data segment means that the time sequence change period of the target telemetry data can be obtained through a machine learning mode, scanning is performed from the starting point to the end point of the target telemetry data segment in a moving time window, and the selected period of the telemetry data segment and the starting point and the end point of the period are identified through judging the maximum value or the minimum value in at least one corresponding period of monotonicity of data in the time window, so that telemetry data is divided according to the period;
the diversity of the target telemetry data segment means that the target telemetry data is divided into 3n groups according to the period, wherein n is more than or equal to 10, and the 3n groups of data are randomly divided into 3 groups which are respectively a training set, a testing set and a detecting set; the training set is used for machine learning of a time sequence model of the star-sensitive quaternion, the testing set is used for rechecking of the quaternion model, and the model is obtained by learning of the training set; the detection set telemetry data is at least more than or equal to one period, and can pre-pick out the whole period array with the fault time period or is real-time telemetry data of a satellite; the target telemetry data is a test set telemetry data.
In one embodiment, the test set telemetry data is passed through a fitting model to obtain a target telemetry data timing model;
the fitting model is a time sequence model which is obtained by extracting periodic characteristics and time sequence function characteristics of target telemetry data by a machine learning method and adopting a computer trigonometric function fitting program.
In one embodiment, the fitted model value of the target telemetry time sequence model is used as the detection value by selecting the difference value between the fitted model value and the target telemetry data in at least one complete period of different time periods. In one embodiment, the detected quantity is subjected to confidence analysis to obtain a confidence interval of the detected quantity of the target telemetry data; the confidence interval is obtained by carrying out statistical analysis on the detection quantity sequences of the telemetry data of N samples, wherein N is more than or equal to 1, the mean value and standard deviation of the detection quantity sequences are obtained, the significance test level is preset, and the confidence interval of the error detection quantity is obtained through calculation, and is the confidence interval of the target telemetry data detection quantity.
Preferably, online detection is carried out at least in a moving time window, the difference value between the simulated model value and the target telemetry data is monitored, and when the difference value exceeds the confidence interval range of the detection quantity of the target telemetry data, the abnormality of the target telemetry data is judged; and judging that the abnormal value is greater than 10 times, and judging that the target telemetry data is abnormal data in a monotonous increasing way for 30 times.
According to a second aspect of the embodiments of the present disclosure, in a confidence interval of a target telemetry data detection amount, correlation coefficients of the quaternion telemetry data related to each other in at least one period in at least one time period are selected as detection amounts, and correlation coefficients of M samples are calculated, wherein M is greater than or equal to 1, a significance test level is preset according to a statistical method, and a confidence interval of the quaternion correlation coefficients under the level is obtained.
Preferably, confidence intervals of quaternion correlation coefficients under the level are used for online detection
Judging abnormal data when the target telemetry data exceeds the confidence interval range of the quaternion correlation coefficient under the level;
the abnormal data is used for selecting quaternion sample telemetry data of at least one period in the same period, and obtaining the correlation coefficient of the quaternion, wherein the correlation coefficient of the quaternion is target telemetry data.
According to the embodiment of the disclosure, the difficulty faced by the fault diagnosis field can be effectively changed through the abnormality detection technology based on machine learning, the telemetry data is an important basis for truly reflecting satellite states, and the machine learning starts from massive historical telemetry data, fully excavates potential information of the historical telemetry data, and can sensitively and accurately discover the abnormality.
According to the invention, the quaternion telemetry data characteristics of the satellite sensor are extracted through a machine learning method, the quaternion time sequence curve of the satellite in a normal working state is modeled, the model value is compared with telemetry data in a satellite sensor fault period, the information of the data can be fully utilized for carrying out anomaly detection, an obvious effect is obtained, and when a given detection level is 5%, the detection accuracy is higher than 95%.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure.
FIG. 1 is a block diagram of a star sensor data anomaly detection method provided by an embodiment of the present disclosure;
FIG. 2 is a block diagram of a star sensor single parameter detection data anomaly detection method provided by an embodiment of the present disclosure;
FIG. 3 is a diagram illustrating a preprocessing of star sensor telemetry sample data according to an embodiment of the present disclosure;
FIG. 4 is a logic structure diagram of a confidence interval for error detection of a single parameter model of a star sensor according to an embodiment of the present disclosure;
FIG. 5 is a logic structure diagram of determining abnormal data in a confidence interval of a single parameter model error detection quantity of a star sensor according to an embodiment of the present disclosure;
FIG. 6 is a logic structure diagram of a star sensor multidimensional telemetry association detection data confidence interval provided in an embodiment of the present disclosure;
FIG. 7 is a diagram of a single multi-dimensional telemetry correlation detection data confidence interval decision exception logic structure provided by an embodiment of the present disclosure;
FIG. 8 is a logical block diagram of a multi-dimensional telemetry Guan Lianjian measurement selected confidence interval decision exception data provided in an embodiment of the present disclosure;
FIG. 9 is a schematic diagram of telemetry anomalies around zero after telemetry data format conversion in an embodiment of the present disclosure.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims.
The embodiment of the disclosure provides a star sensor data anomaly detection method by applying quaternion anomaly detection of a high orbit satellite star sensor, which is shown in figure 1,
embodiment 1,
101, selecting telemetry sample data for preprocessing to obtain target telemetry data;
in one embodiment, telemetry data preprocessing is to reject outliers of telemetry sample data, obtain target telemetry data segments, and perform periodic segmentation and diversity on the target telemetry data segments;
the abnormal value elimination is carried out in two ways, one is to eliminate abnormal values near zero values, and the other is to eliminate abnormal values exceeding the upper limit and the lower limit of the telemetry data;
in one embodiment, the removing of the abnormal value near the zero value means scanning from the time starting point of the target telemetry data at least in one moving time window period, sorting the data in the moving time window according to the size, obtaining the median of the sorted sequence, adding three times of standard deviation of the data in the moving time window by taking the median as a reference to obtain an upper limit section and a lower limit section, and removing the data outside the upper limit section and the lower limit section; such as: rejecting near zero points appearing as abrupt on the sinusoidal function curve with telemetry data less than 10 -5 The values are outliers, and the outlier case around the telemetry data zero value in the disclosed embodiment is shown in fig. 9, which illustrates the case that the star sensor quaternion telemetry value is around zero, which is the target that we need to reject, and the normal telemetry value should appear as a curve trend in the graph following the time sequence change, and the value/values around zero in the graph are the reject targets.
In one embodiment, the other method is to reject abnormal values exceeding the upper limit and the lower limit of the telemetry data, namely, performing statistical analysis on the quaternion telemetry data of the satellite sensor through a computer program, determining the upper limit and the lower limit of the quaternion telemetry data sequence, and rejecting abnormal values outside the upper limit and the lower limit of the quaternion telemetry data.
In one embodiment, the period division of the target telemetry data segment means that the time sequence change period of the target telemetry data is obtained through a machine learning mode, the target telemetry data segment is scanned from the starting point to the end point of the target telemetry data segment in a moving time window, and the position of the monotonicity change is a maximum value or a minimum value in one period by judging the maximum value or the minimum value in at least one corresponding period of the monotonicity of the data in the moving time window, and the length between two adjacent maximum values or minimum values is one period. Identifying the period of the selected telemetry data segment and the starting point and the end point of the period, and dividing telemetry data according to the period;
in one embodiment, the diversity of the target telemetry data segments means that the target telemetry data are divided into 3n groups according to the period, n is more than or equal to 10, and the 3n groups of data are randomly divided into 3 groups which are respectively a training set, a test set and a detection set; the training set is used for machine learning of a time sequence model of the star-sensitive quaternion, the testing set is used for rechecking of the quaternion model, and the model is obtained by learning of the training set; the detection set telemetry data is at least more than or equal to one period, and can pre-pick out an entire period array with a fault time period or is real-time telemetry data of a satellite; the detection set telemetry data is target telemetry data.
102, selecting a confidence interval of the detection quantity of the target telemetry data obtained by confidence analysis; the detection quantity is selected in two modes;
in one embodiment, the first detection amount is selected as a single-parameter detection amount, and the detection set telemetry data is subjected to confidence analysis to obtain a target telemetry data time sequence model through a fitting model; the fitting model is a time sequence model which is obtained by extracting periodic characteristics and time sequence function characteristics of target telemetry data by a machine learning method and adopting a computer trigonometric function fitting program.
And selecting the difference value between the simulated model value and the target telemetry data as a detection quantity in at least one complete period of different time periods. Obtaining a confidence interval which is the confidence interval of the single parameter model error detection quantity;
the confidence interval is obtained by carrying out statistical analysis on the detection volume sequences of the telemetry data of the N samples to obtain the mean value and standard deviation of the detection volume sequences, presetting the significance test level, wherein the significance test level is manually given and is not calculated. According to the statistical theory, the corresponding relation between the detection level and the detection interval is obtained through a formula and a table lookup, and the detection interval is wider generally given a larger detection level; and calculating to obtain a confidence interval of the error detection quantity, wherein the confidence interval of the error detection quantity is a confidence interval of the target telemetry data detection quantity, and N is more than or equal to 1.
The second detection amount is selected from multi-dimensional telemetry Guan Lianjian measurement, and confidence intervals of correlation coefficient detection amounts are obtained through confidence analysis;
selecting the detection quantity and obtaining a confidence interval of the detection quantity of the target telemetry data through confidence analysis; and selecting the correlation coefficients of the quaternion telemetry data in at least one period in the same time period as detection quantity, calculating the correlation coefficients of M samples, and presetting a significance test level according to a statistical method to obtain a confidence interval of the quaternion correlation coefficients under the level, wherein M is more than or equal to 1.
103 on-line detection of target telemetry data exceeding confidence interval as abnormal data
In one embodiment, there are two ways to detect that the target telemetry data exceeds the confidence interval as anomalous data;
in the first mode, on-line detection is performed based on the confidence interval of the detected amount of the target telemetry data as the confidence interval of the detected amount of the error detection, at least in one moving time window, the difference value between the simulated model value and the target telemetry data is monitored, and when the difference value exceeds the confidence interval range of the detected amount of the target telemetry data, abnormal data of the target telemetry data is judged; and judging that the abnormal value is greater than 10 times, and judging that the target telemetry data is abnormal data when the abnormal value is monotonically increased to 30 times.
The second mode is based on the confidence interval of the target telemetry data detection amount being the confidence interval of the quaternion correlation coefficient under the level, and the confidence interval of the quaternion correlation coefficient under the level is used for judging abnormal data when the online detection judgment target telemetry data exceeds the confidence interval range of the quaternion correlation coefficient under the level; the abnormal data is quaternion sample telemetry data of at least one period in the same period, and a correlation coefficient of the quaternion is calculated and is used as target telemetry data.
According to the satellite sensor data anomaly detection method, the quaternion time sequence curve of the satellite in a normal working state is modeled, the simulated model value is compared with remote target telemetry data in a satellite sensor fault period, the anomaly detection can be carried out by fully utilizing the information of the data, an obvious effect is obtained, the detection accuracy is higher than 95%, and the problems of more false alarms and false alarms frequently occurring in the application of a spacecraft fault diagnosis system are well solved.
Example two
A star sensor data anomaly detection method, in one embodiment, is shown in FIG. 2.
201. The telemetry sample data is selected for preprocessing and consists of data format conversion, data cleaning, data period division and data diversity. As shown in the figure 3 of the drawings,
2011. telemetry sample data format conversion
From the data perspective, satellite telemetry data is uneven and discontinuous, has certain interference to machine learning, and needs to be evenly characterized according to time granularity. From a modeling perspective, satellite telemetry data time series are datetime-type, which is not conducive to mathematical expression. In order to adapt to the requirement of model fitting, the starting time of a telemetry sample is selected as a starting point, datetime type data are converted into numerical type data, and telemetry time data are uniformly sampled.
Completion of satellite A star sensor quaternion q using computer program 0 、q 1 、q 2 、q 3 Telemetry data format conversion: selecting a certain period starting point Ts i As a time reference, the time is subtracted from the subsequent time, and the numerical conversion of the time series is completed in seconds.
2012. Telemetry sample data cleaning
The satellite works in a severe space environment, the telemetry data transmission path is far, the signal is weak, and certain measurement errors exist between the satellite and ground equipment, so that errors, error values and wild values exist in the satellite telemetry data measurement values. If the satellite telemetry data is not subjected to cleaning treatment, the machine learning algorithm brings the error information into the model, the error knowledge is learned, and the fitted model generates errors, so that the satellite telemetry data must be cleaned to obtain the data which can truly reflect the change rule of telemetry parameters. Outlier rejection includes two classes: such as outlier rejection exceeding + -1 and outlier rejection with a value difference from neighboring values of greater than 0.01 and less than 0.01.
2013. Telemetry target data period segmentation and diversity
The remote measurement target data of the quaternion of the star sensor has time sequence periodicity, and the time sequence change period of the remote measurement data of the star sensor can be obtained through a machine learning mode. Scanning from the starting point to the end point of the target telemetry data segment by adopting a moving time window, and identifying the period of the selected telemetry data segment and the starting point and the end point of the period by judging monotonicity and maximum value or minimum value of data in the time window, so as to divide telemetry data according to the period. Sample telemetry data are divided into 3n groups (n is larger than or equal to 10) according to the period, and in order to ensure the independence of diversity data, the 3n groups of data are randomly divided into 3 groups: the system comprises a training set, a testing set and a detecting set, wherein the training set is used for machine learning of a time sequence model of a star-sensitive quaternion, the testing set is used for rechecking of a quaternion model, and the quaternion model is obtained by learning of the training set; the detection set is at least more than or equal to one period, and can be a whole period array of fault time periods in advance selected, or real-time telemetry data of satellites.
In one embodiment, the starting point of a cycle of the sequence is determined from monotonicity and extrema of the telemetry data, and for one cycle, a maximum or minimum is determined, the fitting model is a sinusoidal function of half cycles, where the monotonicity changes to a maximum or minimum, and a cycle length is between two adjacent maxima or minima. But for a number of cycles are referred to as maxima or minima, i.e. collectively referred to as extrema; dividing the period, for example, selecting 100 groups of whole period data (at least one period and must satisfy the whole period) of the satellite A in a normal working state to form a data set R, randomly selecting and dividing the data set R into three groups: training set T, test set C and test set D, labeled as follows:
The three data sets are independent, i.e
T∪C∪D＝R
202. The detection set telemetry data is used for obtaining a target telemetry data time sequence model through a fitting model, and the target telemetry data time sequence model is model fitting based on machine learning. From the satellite orbit characteristics, the quaternion telemetry data of the high orbit satellite sensor is changed into a trigonometric function of a half period, and the general form of a time sequence function is as follows:
q(t)＝a(sinωt+b)+c+N 0 (t)
wherein q (t) represents a quaternion, ω is a quaternion period, N 0 (t) is the high-order noise quantity with negligible modeling, and a, b and c are undetermined coefficients, and can be obtained by learning and statistically averaging historical telemetry data of quaternions of the satellite star sensor. Because datetime-type time data (in the form of "2018-09-05 19:51:12.953") is inconvenient to mathematically express, a certain time starting point T0 of telemetry data is selected as a zero point, and datetime-type data is changed into a form easy to mathematically express according to a concept similar to "product seconds" in aerospace. The period of the quaternion time sequence model is calculated by a program, the starting point and the end point of the period of the telemetry data are identified by the program, and the function is fitted in the period.
In one embodiment, the training set T is fed into a learning machine, and the star sensor quaternion time series function model is fitted by using a python trigonometric function fitting program to obtain the period of the function and the values of the coefficients a, b and c. With quaternion q of star sensor 0 For example, some of the elements of the sample are as follows:
...
42108,0.7060563325
42109,0.7060572448
42110,0.7060581562
42111,0.7060590667
42112,0.7060599762
fitting to obtain a periodic time sequence function model of the sample as follows:
the fitting coefficients are:
| coefficient to be estimated | Fitting value of coefficient to be estimated |
| a | 0.706512158432 |
| b | 0.0000487812804619 |
| c | 0.0000288515321792 |
After calculating the fitting function period and coefficients of each sample, respectively carrying out statistical average on coefficient results obtained by fitting each sample, and taking the statistical average of each coefficient as a final fitting result.
203. Selecting single-parameter detection quantity and obtaining a confidence interval through confidence analysis;
according to the characteristics of the remote measurement data of the star sensor, two methods different from the existing manual setting of the detection threshold are designed: the two methods of single parameter detection respectively correspond to two detection amounts, namely error detection amounts, and the two parts comprise detection amount selection and confidence analysis; as shown in fig. 4.
2031 single parameter telemetry data detection amount selection,
the single parameter detection amount is selected, the accuracy of the fitting model can be measured by errors of the real model and the fitting model, and according to the definition of the error amount, a time function with errors can be obtained:
δ(t)＝q 0 (t)-q 0 * (t)
in the above, q 0 (t) true time function representing the quaternion of the star sensor, q 0 * (t) represents a fitting time function of the star sensor quaternion, and the error amount is a detection amount of the single parameter detection method.
In one embodiment, a detection amount sequence is obtained, the time sequences corresponding to the telemetry data sequences of the test set C are respectively substituted into the fitting model to obtain a quaternion model value sequence of the star sensor, and the value sequence is respectively combined with the respective time sequences to obtain a fitting model set G. Subtracting the quaternion telemetry data sequence of the star sensor in the test set C and the simulation value sequence in the fitting matrix G according to corresponding moments to obtain an error sequence E, wherein the sequence is reflection of the difference between the fitting model and the real telemetry data, and the smaller the value of the error sequence is, the higher the fitting degree of the fitting model and the telemetry data is, and the more accurate the fitting model is. The fitting model set G is:
the error sequence is expressed as follows:
wherein the method comprises the steps of
2032 single parameter telemetry data confidence analysis
Because of the single parameter model error detection amount, and because the test set and training set data are of different time periods, they can be considered independent. And taking the test set data as real data of the quaternion of the star sensor in the test period, taking the data obtained by calculating the fitting model according to the test period as model data, subtracting the two data according to the corresponding time, and obtaining an error sequence of the model in the test period, wherein the error sequence reflects the coincidence degree of the fitting function and the real function. According to statistical knowledge, presetting a significance test level alpha, taking an error sequence obtained by N samples, wherein N is more than or equal to 1, calculating to obtain a confidence interval of error detection quantity, and fitting the confidence coefficient of the model to be 1-alpha on the confidence interval.
2033. Obtaining single parameter telemetry confidence intervals
In one embodiment, the confidence analysis, the distribution function of the overall delta is set as
Obeying normal distribution, i.e. delta.N (mu) 1 ，σ 1 ) Wherein
According to the relevant conclusion of the statistical t distribution, the calculation is carried out to obtain
In the above-mentioned method, the step of,
is->
S is the sample mean value of (1) 1 Is->
From statistical knowledge, mu is obtained 1 A confidence interval with a confidence level of 1-alpha:
taking alpha=0.05 and n=10 to calculate the mean value and standard deviation of each detection quantity sequence to obtain the following table,
a confidence interval (-0.00007,0.00019) with a confidence level of 0.95 is calculated for the detected quantity e, and the interval can be used as a detection quantity judgment threshold to detect the abnormality of the quaternion of the star sensor.
204. The on-line detection is as shown in figure 5,
in one embodiment of the present invention, in one embodiment,
an anomaly is determined at least when the difference between the simulated model value and the target telemetry data for the one moving time window exceeds a confidence interval range for the detected amount of target telemetry data.
2041. The single parameter detection based on fitting model errors adopts a method of moving a time window, the size of the time window can be set according to detection requirements, if abnormality is found within 5 minutes, the window is selected to be 300 seconds, after the detection window is set, the cycle starting point of current real-time telemetry data is selected as an origin, the time window moves along with natural time, the difference value between the model value and the telemetry data in the time window is compared, the difference value exceeds a confidence interval of single parameter model error detection quantity and is marked as an abnormal value, and when the difference value exceeds a threshold for a specified number, the telemetry data is judged to be abnormal. Because the satellite works in a very severe environment, the quaternion telemetry data of the star sensor can have the condition of data missing or individual outliers, and the moving time window detection method designed by the invention has certain fault tolerance capability for the condition that the telemetry data has a small part of outliers, and the detection is real-time.
In one embodiment, the detection data set D is formed by taking the detection data set D as online data and supplementing a part of fault period data * The detection uses a moving time window, which is assumed to be 5min longThe end point is the current time and the start point of the window is 300 seconds ago.
2042. Taking the starting point of the current period of the detection set as an origin, moving a time window along with natural time, monitoring the difference value between the simulated model value and the real-time telemetry data, and judging abnormality when the difference value exceeds the range of the confidence interval of the detection quantity; the difference exceeds the upper and lower values of the confidence interval by more than 10 times;
2043. if the difference value is that the monotone increment abnormal value exceeds 30, the telemetry data is considered to be abnormal, and an alarm popup window is opened.
The invention provides a satellite sensor data anomaly detection method, which is a high orbit satellite sensor quaternion anomaly detection method based on machine learning, can fully utilize historical telemetry data resources, solves the problems of low accuracy, more false alarms and missed alarms of anomaly detection technology based on an experience threshold, improves satellite fault detection capability, and provides a new solution for anomaly detection of other parameters and components of a satellite.
Example III
Measuring confidence interval range judgment abnormal data of quaternion related data under the condition that confidence interval online detection target telemetry data exceeds the indicated level by the multidimensional telemetry Guan Lianjian; as shown in figure 6 of the drawings,
301. selecting telemetry sample data for preprocessing
The processing manner and steps for preprocessing the 201 telemetry sample data to obtain target telemetry data in fig. 2 are not repeated here.
302. The multi-dimensional telemetry association detection amount is selected and the confidence coefficient analysis is carried out to obtain a confidence interval,
after preprocessing the telemetry sample data, selecting a correlation coefficient detection quantity, and selecting a multidimensional telemetry correlation detection quantity, for example, when a satellite is in a normal working state, working parameters of components with correlation relations have determined correlation, the working parameters are mathematically represented as the magnitude of the correlation coefficient, when the correlation coefficient approaches 1 or-1, the correlation of the two parameters is very strong, and when the correlation coefficient approaches 0, the correlation of the two parameters is very weak or irrelevant. The correlation coefficient between the quaternions of the star sensor can be used for detecting whether the correlations between the quaternions are normal, and a normalized 4-order covariance matrix can be established according to the correlation coefficient and used for detecting satellite parameter anomalies.
The correlation coefficient detection amount is calculated by using the following formula of the correlation coefficient of two columns of data i and j according to statistical knowledge:
in the above, S ij For sample covariance, S i ，S j Respectively q i ，q j Is a sample standard deviation of (2). And selecting the quaternion of the whole-period star sensor without faults in different time periods as a training set, respectively calculating sample correlation coefficients of the star sensor fusion quaternion to obtain a corresponding quaternion correlation coefficient sequence, presetting a significance test level beta, calculating a confidence interval of correlation coefficient detection quantity, and ensuring that the confidence coefficient of the star sensor quaternion correlation coefficient is 1-beta in the confidence interval.
3021. Multidimensional telemetry correlation detection volume selection as shown in FIG. 7
And detecting the condition of the correlation coefficient of the whole period of the target telemetry data according to the calculated confidence interval according to the significance test level beta of the quaternion correlation coefficient, and judging that the satellite is abnormal when the correlation coefficient calculated by the telemetry data in the detection set, namely the target telemetry data exceeds the confidence interval.
It should be noted that if the quaternion sample data of the star sensor is less than 1 period, the calculated correlation coefficient cannot be used for detection because the expression of the correlation information is incomplete, so that during real-time detection, the data of at least one period must be taken backward from the detection time to perform correlation calculation, and the detection is non-real-time.
In one embodiment, training set D includes satellite A star sensor quaternion q 0 、q 1 、q 2 、q 3 By suitable combination of
T * ：{(time,q 0 、q 1 、q 2 、q 3 ) i },i∈(1,2，...，m)
The formula of the correlation coefficient of each quaternion in the same period is set as
In the above formula, i, j=0, 1,2,3 are each quaternion telemetry value sequences at the same time, S ij For sample covariance, S i ，S j Is the standard deviation of the samples. Calculating sample correlation coefficients r of quaternions of star sensors respectively 01 ，r 02 ，r 03 ，r 12 ，r 13 ，r 23 The value interval of the correlation coefficient is between-1 and 1,1 represents that two samples are completely correlated, -1 represents that two samples are completely negatively correlated, and 0 represents that two samples are uncorrelated.
In one embodiment, the correlation coefficients of the quaternions in the same period are the correlation coefficients of the quaternion q0 and the quaternion q1, the correlation coefficients of the quaternion q0 and the quaternion q2, the correlation coefficients of the quaternion q0 and the quaternion q3, the correlation coefficients of the quaternion q1 and the quaternion q2, and the correlation coefficients of the quaternion q2 and the quaternion q 3.
The closer to zero, the weaker the correlation of the two samples, the closer to 1 or-1, the stronger the correlation of the two variables. The quaternion correlation coefficient of the satellite A star sensor is used as the detection quantity of multi-dimensional telemetry correlation detection, and when the detection quantity is in the condition of positive correlation and negative correlation (boundary +/-1), the confidence interval is single-sided.
3022. Confidence analysis
In one embodiment, the integration region is compressed from (- + -infinity, + -infinity) to (-1, 1) assuming that the correlation coefficient between the sensor quaternions for satellite A star is normally distributed over (-1, 1) during the no-fault period. The correlation coefficient between fused quaternions has the problem of both double confidence intervals (near 0) and the problem of single confidence intervals (r ij →±1). Below, the quaternion q is inversely related 0 And q 1 For example, the detected quantity r is analyzed 12 Is a detection interval confidence problem.
3023. Multidimensional telemetry association confidence interval
Let the distribution function of the population P be F (r 12 The method comprises the steps of carrying out a first treatment on the surface of the Delta) is subjected to normal distribution, namely P-N (mu) 2 ，σ 2 )，P(r 12 The method comprises the steps of carrying out a first treatment on the surface of the δ) contains an unknown parameter δ, statistics determined by the training set T for a preset confidence level β
Satisfy the following requirements
Mu according to the statistically relevant conclusion 2 The confidence interval of (2) is:
taking m=10, the result of calculating the phase relation number of each sample is as follows:
calculated to obtain
S 2 Taking β=0.05, looking up the table, t when m=10 β (9) = 1.8331, a confidence interval (-1, -0.994847) with a confidence level of 0.95, that is, the quaternion q, is obtained 0 And q 1 The correlation coefficient of (2) is within a section (-1, -0.994847) with a 95% probability, and the section can be used for thresholding to detect the fused quaternion anomaly.
303. On-line detection
Judging abnormal data when the target telemetry data exceeds the confidence interval range of the quaternion correlation coefficient under the inspection level; as shown in FIG. 8
3031. Acquiring the correlation coefficient of the quaternion by quaternion sample telemetry data of at least one period in the same period;
3032. judging abnormal data when the confidence interval range of the quaternion correlation coefficient under the detection level is exceeded, obtaining the confidence interval of the corresponding detection amount through learning according to the significance detection level beta of the quaternion correlation coefficient, comparing the confidence interval with the whole period correlation coefficient of the target telemetry data, and judging that the satellite is abnormal when the corresponding quaternion correlation coefficient exceeds the confidence interval.
It should be noted that if the quaternion sample data of the star sensor is less than 1 period, the calculated correlation coefficient cannot be used for detection because the expression of the correlation information is incomplete, so that during real-time detection, the data of at least one period must be taken backward from the detection time to perform correlation calculation, and the detection is non-real-time.
In one embodiment, the online data of the detection set D is used to calculate the coefficient of each quaternion correlation of the detection set D, and the coefficients are compared with the confidence intervals corresponding to the quaternion correlation coefficients respectively, and the judgment that the coefficient falls outside the confidence intervals is abnormal. Loading data of abnormal period, and calculating to obtain r 01 = -0.990832, when the correlation coefficient falls outside the confidence interval, the detection program determines as abnormal.
The invention provides a satellite sensor data anomaly detection method, which is a high orbit satellite sensor quaternion anomaly detection method based on machine learning, can fully utilize historical telemetry data resources, solves the problems of low accuracy, more false alarms and missed alarms of anomaly detection technology based on an experience threshold, improves satellite fault detection capability, and provides a new solution for anomaly detection of other parameters and components of a satellite.
Fourth embodiment,
In one embodiment, model fusion is performed separately on the basis of the second or third embodiment.
The quaternion time sequence model (target telemetry data time sequence model) of the star sensor obtained by the data is not invariable, quaternion telemetry data is generated in real time, the new data also express the change rule of quaternion, the learning machine should refine the characteristics expressed by the new data, the characteristics are added into the model through data fusion, the real-time performance of machine learning is reflected, and the steps consist of information judgment and model fusion.
1. The new information judgment is carried out, and the normal telemetry data are fused to all normal data sequences;
the newly generated telemetry data is detected, and after confirmation of non-abnormal data of the telemetry data, the telemetry data can be added into a learning machine as innovation. If the satellite A newly downloads the telemetry data of the star sensor, after confirming that no abnormality exists in two detection modes, the telemetry data with the abnormality is added into the white list in a whole period mode, the telemetry data with the abnormality is added into the suspected abnormal yellow list in the whole period mode, and the blacklist adding for checking is confirmed to be the fault data.
2. Model fusion
After the new information judgment, the normal telemetry data are fused to all normal data sequences, and the data are used as a training set, so that the learning machine is trained to extract data characteristics and knowledge and fused to the model to obtain a new model for subsequent detection. Such as: and inputting the newly added data of the white list into a learning machine, and fusing the newly added data with the previous model to obtain a new model.
The invention provides a star sensor data anomaly detection method, which is characterized in that a quaternion telemetry data characteristic is extracted by performing machine learning fitting on massive historical telemetry data to a time sequence model of a quaternion of a star sensor, a quaternion correlation model is established, and compared with the prior modeling, the historical telemetry data is fully utilized; two detection quantities are designed, a confidence interval and a confidence coefficient of the detection quantity are calculated by adopting sufficient sample data, the problem of quaternion anomaly detection is supported by a strict mathematical method, the defect of manually setting a parameter threshold is overcome, and the method is more convincing than an expert system which relies on experience to assign judgment weights. The multi-dimensional telemetry association detection technology breaks the limit thinking of a single parameter fixed threshold, regards the quaternion detection of the star sensor as a whole, does not need to detect each quaternion, and can detect the fault phenomenon generated by abnormal association relation among quaternions; the two quaternion detection technologies are organically combined, can mutually provide a certificate, and have the characteristics of strong fault tolerance and high accuracy, and the detection algorithm cannot cause false alarms due to jump caused by one or two wild values.
Based on the logic structure diagram used in the star sensor data anomaly detection method described in the embodiments corresponding to fig. 1,2,3, 4, 5, 6, 7, 8, and 9, the embodiments of the present disclosure further provide a computer readable storage medium, for example, a non-transitory computer readable storage medium may be a Read Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like. The storage medium stores computer instructions for executing the star sensor data anomaly detection method described in the embodiments corresponding to fig. 1 and 3, which are not described herein.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any adaptations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
Claims (4)
1. A star sensor data anomaly detection method, the method comprising:
selecting telemetry sample data for preprocessing to obtain target telemetry data;
selecting the detection quantity and obtaining a confidence interval of the detection quantity of the target telemetry data through confidence analysis; detecting that the target telemetry data exceeds the confidence interval on line to be abnormal data;
the selected detection quantity is selected as a single-parameter detection quantity; or, selecting the multi-dimensional telemetry association detection quantity; the confidence interval is the confidence interval of the single parameter model error detection amount; or, a confidence interval of the correlation coefficient detection amount;
the telemetering sample data preprocessing is to reject abnormal values of the telemetering sample data, obtain a target telemetering data segment, and carry out periodic segmentation and diversity on the target telemetering data segment;
the period segmentation of the target telemetry data segment means that the time sequence change period of the target telemetry data is obtained through a machine learning mode, scanning is carried out from the starting point to the end point of the target telemetry data segment in a moving time window, and the period of the telemetry data segment and the starting point and the end point of the period are identified and selected through judging the maximum value or the minimum value in at least one period corresponding to the monotonicity of the data in the moving time window, so that telemetry data is divided according to the period;
the diversity of the target telemetry data segment means that the target telemetry data is divided into 3n groups according to the period, wherein n is more than or equal to 10, and the 3n groups of data are randomly divided into 3 groups: the 3 groups are respectively a training set, a testing set and a detecting set; the training set is used for machine learning of a time sequence model of the star-sensitive quaternion, the testing set is used for rechecking of a quaternion model, and the quaternion model is obtained by learning of the training set; the detection set telemetry data refers to the whole period array with a fault time period or real-time telemetry data of satellites, which are selected in advance in at least one period; the target telemetry data is the test set telemetry data;
the detection set telemetry data obtains a target telemetry data time sequence model through a fitting model;
the fitting model is a time sequence model which is obtained by extracting periodic characteristics and time sequence function characteristics of the target telemetry data by a machine learning method and adopting a computer trigonometric function fitting program;
the fitting model value of the target telemetry data time sequence model is used for selecting the difference value between the fitting model value and the target telemetry data as the detection quantity in at least one complete period of different time periods;
the detection quantity is subjected to confidence analysis to obtain a confidence interval of the detection quantity of the target telemetry data; the confidence interval is obtained by carrying out statistical analysis on detection quantity sequences of N sample telemetry data to obtain the mean value and standard deviation of the detection quantity sequences, presetting a significance test level, and calculating to obtain the confidence interval of error detection quantity, wherein the confidence interval of the error detection quantity is the confidence interval of target telemetry data detection quantity, and N is more than or equal to 1;
detecting a difference value between the simulated model value and the target telemetry data on line at least in a moving time window, and judging that the target telemetry data is abnormal when the difference value exceeds a confidence interval range of the detection quantity of the target telemetry data; and judging that the abnormal value is greater than 10 times, and judging that the target telemetry data is abnormal data when the abnormal value is monotonically increased to 30 times.
2. The anomaly detection method according to claim 1, wherein the outlier rejection is rejection of outliers around zero values or rejection of outliers exceeding upper and lower limits of the telemetry data;
the elimination of abnormal values near the zero value is to scan from the time starting point of the target telemetry data at least in one moving time window period, sort the data in the moving time window according to the size, obtain the median of the ordered sequence, add three times of standard deviation of the data in the moving time window with the median as a reference to obtain an upper limit interval and a lower limit interval, and eliminate the data outside the upper limit interval and the lower limit interval;
the removing of abnormal values exceeding the upper limit and the lower limit of the telemetry data refers to the statistical analysis of the quaternion telemetry data of the satellite star sensor through a computer program, the upper limit and the lower limit of a quaternion telemetry data sequence are determined, and abnormal values, outside the upper limit and the lower limit, of the quaternion telemetry data are removed.
3. The anomaly detection method according to claim 1, wherein the confidence interval for obtaining the target telemetry data detection amount by confidence analysis of the selected detection amount is that correlation coefficients related to each other of the quaternion telemetry data are selected as detection amounts at least in at least one period in the same time period, correlation coefficients of M samples are calculated, wherein M is greater than or equal to 1, a significance test level is preset according to a statistical method, and the confidence interval of the quaternion correlation coefficients under the level is obtained.
4. The abnormality detection method according to claim 3, characterized in that: the confidence interval of the quaternion correlation coefficient under the level is used for judging abnormal data when the target telemetry data exceeds the confidence interval range of the quaternion correlation coefficient under the level through online detection;
and the abnormal data is used for selecting quaternion sample telemetry data of at least one period in the same period to obtain the correlation coefficient of the quaternion, wherein the correlation coefficient of the quaternion is target telemetry data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010361728.2A CN111611294B (en) | 2020-04-30 | 2020-04-30 | Star sensor data anomaly detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010361728.2A CN111611294B (en) | 2020-04-30 | 2020-04-30 | Star sensor data anomaly detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111611294A CN111611294A (en) | 2020-09-01 |
| CN111611294B true CN111611294B (en) | 2023-06-09 |
Family
ID=72201210
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010361728.2A Active CN111611294B (en) | 2020-04-30 | 2020-04-30 | Star sensor data anomaly detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111611294B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112465044B (en) * | 2020-12-03 | 2022-12-27 | 上海卫星工程研究所 | Satellite working condition identification and segmentation method and system based on sensitive parameter multi-evidence fusion |
| CN113435317A (en) * | 2021-06-25 | 2021-09-24 | 中国电子科技集团公司第五十四研究所 | Unmanned aerial vehicle telemetry data correlation analysis method |
| CN115641916B (en) * | 2022-09-29 | 2024-05-03 | 杭州准芯生物技术有限公司 | Molecular biology detection method, computer equipment and readable storage medium |
| CN115617023B (en) * | 2022-12-05 | 2023-03-31 | 中国西安卫星测控中心 | Spacecraft attitude control system abnormity positioning method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103646167A (en) * | 2013-11-22 | 2014-03-19 | 北京空间飞行器总体设计部 | Satellite abnormal condition detection system based on telemeasuring data |
| CN106374993A (en) * | 2016-08-31 | 2017-02-01 | 长光卫星技术有限公司 | Full-automatic unattended satellite telemetering monitoring platform and data processing method |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104156615A (en) * | 2014-08-25 | 2014-11-19 | 哈尔滨工业大学 | Sensor test data point anomaly detection method based on LS-SVM |
| CN108734360B (en) * | 2018-06-19 | 2021-09-07 | 哈尔滨工业大学 | Correction-based ELM prediction model multidimensional telemetering data intelligent interpretation method |
| CN109034191B (en) * | 2018-06-19 | 2021-06-01 | 哈尔滨工业大学 | ELM-based one-dimensional telemetry data abnormal interpretation method |
| US10713079B2 (en) * | 2018-06-25 | 2020-07-14 | Nutanix, Inc. | System and method for managing telemetry data and agents in a telemetry system |
| US11567914B2 (en) * | 2018-09-14 | 2023-01-31 | Verint Americas Inc. | Framework and method for the automated determination of classes and anomaly detection methods for time series |
| CN110909822B (en) * | 2019-12-03 | 2022-11-11 | 中国科学院微小卫星创新研究院 | Satellite anomaly detection method based on improved Gaussian process regression model |
-
2020
- 2020-04-30 CN CN202010361728.2A patent/CN111611294B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103646167A (en) * | 2013-11-22 | 2014-03-19 | 北京空间飞行器总体设计部 | Satellite abnormal condition detection system based on telemeasuring data |
| CN106374993A (en) * | 2016-08-31 | 2017-02-01 | 长光卫星技术有限公司 | Full-automatic unattended satellite telemetering monitoring platform and data processing method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111611294A (en) | 2020-09-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111611294B (en) | Star sensor data anomaly detection method | |
| US11586913B2 (en) | Power equipment fault detecting and positioning method of artificial intelligence inference fusion | |
| Coble et al. | Applying the general path model to estimation of remaining useful life | |
| CN112508105B (en) | Fault detection and retrieval method for oil extraction machine | |
| EP1709499A1 (en) | Trending system and method using window filtering | |
| CN110083803B (en) | Method and system for detecting water taking abnormality based on time sequence ARIMA model | |
| CN112581719B (en) | Semiconductor packaging process early warning method and device based on time sequence generation countermeasure network | |
| US20220004163A1 (en) | Apparatus for predicting equipment damage | |
| CN113449703B (en) | Quality control method and device for environment online monitoring data, storage medium and equipment | |
| CN117270514B (en) | Production process whole-flow fault detection method based on industrial Internet of things | |
| Qin et al. | Remaining useful life prediction for rotating machinery based on optimal degradation indicator | |
| CN112100574A (en) | Resampling-based AAKR model uncertainty calculation method and system | |
| CN116985183A (en) | Quality monitoring and management method and system for near infrared spectrum analyzer | |
| CN117152119A (en) | Profile flaw visual detection method based on image processing | |
| TWI427487B (en) | Method for sampling workpiece for inspection and computer program product performing the same | |
| CN116702005A (en) | Neural network-based data anomaly diagnosis method and electronic equipment | |
| CN113033845B (en) | Construction method and device for power transmission resource co-construction and sharing | |
| CN115659271A (en) | Sensor abnormality detection method, model training method, system, device, and medium | |
| Dzakowic et al. | Advanced Techniques for the verification and validation of prognostics & health management capabilities | |
| CN112069621B (en) | Method for predicting residual service life of rolling bearing based on linear reliability index | |
| CN113205215A (en) | Knowledge-based battlefield situation prediction method | |
| WO2021126591A1 (en) | Method for physical system anomaly detection | |
| Boukra et al. | Contribution in Enhancing the Remaining Useful Life Prediction in Abrupt Failures: Bearing Case. | |
| CN112464146B (en) | Key subsystem based on historical telemetering data and single-machine correlation health baseline construction method | |
| CN116448062B (en) | Bridge settlement deformation detection method, device, computer and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |