CN111967501B - Method and system for judging load state driven by telemetering original data - Google Patents
Method and system for judging load state driven by telemetering original data Download PDFInfo
- Publication number
- CN111967501B CN111967501B CN202010710292.3A CN202010710292A CN111967501B CN 111967501 B CN111967501 B CN 111967501B CN 202010710292 A CN202010710292 A CN 202010710292A CN 111967501 B CN111967501 B CN 111967501B
- Authority
- CN
- China
- Prior art keywords
- telemetry
- load
- data
- single machine
- machine equipment
- 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
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000012545 processing Methods 0.000 claims abstract description 16
- 239000013598 vector Substances 0.000 claims description 29
- 238000012549 training Methods 0.000 claims description 25
- 238000012216 screening Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 10
- 238000005259 measurement Methods 0.000 claims description 10
- 238000012795 verification Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 description 8
- 230000006872 improvement Effects 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000013145 classification model Methods 0.000 description 3
- 238000013075 data extraction Methods 0.000 description 3
- KJONHKAYOJNZEC-UHFFFAOYSA-N nitrazepam Chemical compound C12=CC([N+](=O)[O-])=CC=C2NC(=O)CN=C1C1=CC=CC=C1 KJONHKAYOJNZEC-UHFFFAOYSA-N 0.000 description 3
- 238000012163 sequencing technique Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012847 principal component analysis method Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000012850 discrimination method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012856 packing Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- 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
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radio Relay Systems (AREA)
Abstract
The invention discloses a method for judging the state of load single machine equipment driven by telemetry original data, which comprises the following steps: receiving telemetry original data downloaded by a satellite; extracting the sub-package telemetry data corresponding to the load single machine equipment from the telemetry original data according to the telemetry parameter configuration file; inputting the sub-package telemetering data into a pre-trained load single machine equipment discrimination model to obtain a load single machine equipment state result; the load single machine equipment discrimination model comprises: the device comprises a telemetering parameter formatting processing module, a outlier rejection module and a load single machine equipment discrimination module; the telemetry parameter formatting processing module is used for formatting the packetized telemetry data according to a processing strategy to obtain telemetry parameter formatting data; the outlier removing module is used for removing outliers in the telemetry parameter formatted data; and the load single machine equipment judging module is used for formatting data according to the telemetry parameters and outputting a state result of the load single machine equipment.
Description
Technical Field
The invention relates to the application fields of satellite control, instrument monitoring and the like. In particular to a method and a system for judging the state of load single machine equipment driven by telemetry original data.
Background
The telemetry parameter data is the only basis for expert scholars and spacecraft ground management personnel to know the on-orbit state of a spacecraft instrument, and the on-orbit instrument equipment state judgment is an important working link in the spacecraft task operation control and supervision process. On one hand, compared with the conventional subsystem of the spacecraft platform, the payload equipment is more personalized, precise and complex along with tasks; on the other hand, with the enhancement of national comprehensive strength and the improvement of technological level, on-orbit tasks are increasingly increased, the functional design is complex, the downlink telemetry parameter format is various, and the telemetry parameter dimension reaches thousands of dimensions. The method brings great difficulty to judging the running state of the load single machine equipment, and provides challenges for limited arc tracking resources and human resources.
The on-orbit running state of the load single machine equipment directly relates to task arrangement and scheduling execution, and further relates to success and failure of the task. The discrimination of the state of the load equipment is more complex and personalized than the discrimination of telemetry parameters and the discrimination of the state of the spacecraft platform. The traditional parameter interpretation method depends on expert knowledge, cannot meet increasingly complex task requirements, has low interpretation efficiency and data utilization rate, and has the problems of misinterpretation, missed interpretation and the like.
With the development of aerospace industry and space exploration tasks, on-orbit operation of the aerospace tasks can be more guided by scientific targets, load equipment is used as a core, and the state judgment of the load equipment is different from the parameter judgment. At present, expert scholars at home and abroad deeply study telemetry parameters through spacecraft platform state monitoring, equipment parameter interpretation and fault interpretation, and obtain some results, but lack telemetry original data discrimination learning methods for the states of various task load single-machine equipment instruments.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a method and a system for judging the state of load single machine equipment driven by telemetry original data.
In order to achieve the above object, the present invention provides a method for discriminating a state of a load stand-alone device driven by telemetry raw data, the method comprising:
receiving telemetry original data downloaded by a satellite;
extracting the sub-package telemetry data corresponding to the load single machine equipment from the telemetry original data according to the telemetry parameter configuration file;
inputting the sub-package telemetering data into a pre-trained load single machine equipment discrimination model to obtain a load single machine equipment state result;
the input of the load single machine equipment distinguishing model is the sub-package telemetering data corresponding to the load single machine equipment, and the output is the state result of the load single machine equipment, and the load single machine equipment distinguishing model comprises: the device comprises a telemetering parameter formatting processing module, a outlier rejection module and a load single machine equipment discrimination module; wherein,
the telemetry parameter formatting processing module is used for formatting the packetized telemetry data according to a processing strategy to obtain telemetry parameter formatting data;
the outlier removing module is used for removing outliers in the telemetry parameter formatted data;
the load single machine equipment judging module is used for formatting data according to the telemetry parameters and outputting a state result of the load single machine equipment.
As an improvement of the method, the method further comprises the step of training a load single machine equipment discrimination model, and specifically comprises the following steps:
extracting sub-package telemetry data corresponding to the load single machine equipment from a spacecraft telemetry original database according to the task model of the spacecraft and the load single machine equipment, and extracting load single machine operation plan data from a spacecraft task operation plan database;
formatting the packetized telemetry data according to a processing strategy to obtain telemetry parameter formatted data, removing outliers by adopting a 3 sigma principle first-order data difference method, and establishing a telemetry parameter formatted data set;
matching the telemetry parameter formatted data set with the load single machine equipment state data according to a time scale to obtain a telemetry feature set;
screening the telemetry feature set, sorting information gain and converting multi-label features to obtain a screened feature set with reduced dimensions, and dividing the feature set into a training set and a verification set;
training the model by using a training set through an integrated learning method to obtain a trained load single machine equipment discrimination model;
and verifying and optimizing parameters of the model by using the verification set to obtain the final load single machine equipment discrimination model.
According to the spacecraft mission model and the load single machine equipment, extracting sub-package telemetry data corresponding to the load single machine equipment from a spacecraft telemetry original database, and extracting load single machine operation plan data from a spacecraft mission operation plan database; the method specifically comprises the following steps:
configuring remote measurement parameter library table access information and task load single machine operation plan library table access information according to the spacecraft task model and load single machine equipment;
extracting load single machine operation plan data from a spacecraft task operation plan database according to the task load operation plan database table access information;
analyzing the single load operation plan data to obtain a single load operation plan and single load equipment state data;
and extracting the sub-package telemetry data corresponding to the single load equipment from the spacecraft telemetry original database according to the telemetry parameter library table access information and the single load equipment state data.
As an improvement of the method, the telemetry parameter formatted data set and the load single machine equipment state data are matched according to a time scale to obtain a telemetry feature set; the method specifically comprises the following steps:
defining telemetry parameter data record vectorsAnd a load device state tensor U (i) ：
Wherein c (i) When the data is acquired from a certain star, the corresponding star is used as a time mark corresponding to the data record vector,for an n-dimensional telemetry parameter data vector at the corresponding time instant, tm= { TM j I j=1, 2,..n } is the set of load telemetry parameters, tm j Telemetry parameters for the j-th dimension;
wherein,for the l-dimensional load state vector at the corresponding time, p= { P k I k=1, 2, l is the set of task payload devices, p k Representing a kth load device status;
matching according to a time scale to obtain a telemetry feature set D= { (TM) (i) ,P (i) )|1≤i≤s}
Wherein, (TM) (i) ,P (i) ) Record for each sample, TM (i) ∈Ω TM For the n-dimensional feature vector to be recorded,to record TM (i) A corresponding tag.
As an improvement of the above method, the screening specifically includes:
computing pearson correlation coefficient ρ for telemetry parameter formatted data in a telemetry feature set 2 (a,b)：
Wherein a and b respectively represent two telemetry parameter formatted data in a telemetry feature set, T represents transposition, and E represents expected value;
if meeting |ρ 2 (a, b) -1|ε, then preserving telemetry parameter formatted data, ε representing a threshold;
and calculating variance for telemetry parameter formatted data in telemetry feature set, and screening out constant value or slowly-varying value irrelevant to classification and discrimination of the state of the load single machine equipment.
As an improvement of the above method, the information gain ordering is specifically: information gain is calculated for the load stand-alone device status data in the telemetry feature set,
wherein X represents a random variable of a telemetry parameter, H (X) represents information entropy of the random variable X, p i Representing the probability of a random event X, and n represents the number of random variable set elements formed by a telemetry parameter set;
conditional entropy H (y|x) of random variable (X, Y):
wherein, (X, Y) is a two-dimensional random variable which represents the composition of the telemetry parameter X and the telemetry parameter Y, and H (Y|X) represents the conditional entropy of the random variable Y under the given priori condition of the random variable X;
the information gain g (Y, X) is:
g(Y,X)＝H(Y)-H(Y|X)
wherein H (Y) represents the information entropy of the random variable Y;
and traversing the information gain of each telemetry parameter feature to the state of the single machine equipment, obtaining the information gain sequence, and selecting the feature with large information gain to leave.
As an improvement of the method, the training set is utilized to train the model through an integrated learning method, so as to obtain a trained load single machine equipment discrimination model; the method specifically comprises the following steps:
initializing weak classifier h 0 (tm)：
Where tm represents the telemetry parameter, arg represents the optimal solution, p i Represents the i-th real sample value, namely the load single machine equipment is in the state i, c represents the classification result judged by the classifier, and L (p i C) represents a sample value and a classification result loss function, and N represents the total number of samples;
the M weak classifiers are iterated, and the internal iteration process is as follows:
for sample feature set TM ", calculate negative gradient residual r mi ：
Where p represents a true sample value of the load stand-alone device state, h (tm i ) Representing the current weak classifier, i.e. for telemetry parameter tm i Judging the state of the load single machine equipment, wherein L represents a loss function, x represents the input parameters of the classifier, h (x) represents the selection of the iterative classifier, and h m-1 (x) A weak classifier representing a previous residual iteration;
taking the residual error as a new sample value to obtain a data set { TM', r of the next class tree mi New regression tree h m (tm) the corresponding leaf node region R jm J=1, 2 …, J being the number of leaf nodes of the regression tree;
computing a best fit c to the leaf area of the regression tree based on empirical risk minimization criteria jm ：
Updating a learner:
wherein h is m-1 (tm) represents the previous regression tree, I (tm ε R) jm ) Representing an indication function;
until the negative gradient residual error converges, obtaining a trained load single machine equipment discrimination model:
a telemetry original data driven single load device status discrimination system, the system comprising: the system comprises a receiving module, an extracting module, a judging module and a trained load single machine equipment judging model; wherein,
the receiving module is used for receiving telemetry original data downloaded by a satellite;
the extraction module is used for extracting the sub-package telemetry data corresponding to the load single machine equipment from the telemetry original data;
the judging module is used for inputting the sub-package telemetry data into a pre-trained load single machine equipment judging model to obtain a load single machine equipment state result.
Compared with the prior art, the invention has the advantages that:
1. the method and the system for judging the state of the load single machine equipment driven by the telemetering original data can be widely applied to the judgment, learning and automatic state identification of the state of the load single machine equipment facing the task target of a spacecraft, and improve the self-adaptability and the automation level;
2. according to the method and the system for judging the state of the load single machine equipment driven by the telemetering original data, provided by the invention, models and algorithms are selected in the aspects of accuracy, interpretability and execution efficiency, and the identification with higher accuracy according to the telemetering original data and the state of the load single machine equipment is realized by adopting integrated learning;
3. the method has the advantages that the information gain parameters are adopted to carry out remote measurement parameter screening, the interpretability requirement of spacecraft task analysis is met, and the high execution efficiency is realized by adopting the parameter dimension reduction methods such as the pearson coefficient, the variance, the information gain and the like.
Drawings
FIG. 1 is a flow chart of a telemetry raw data driven single load status discrimination method of embodiment 1 of the present invention;
FIG. 2 is a schematic diagram showing the constitution of a telemetry original data-driven load stand-alone status discrimination system according to embodiment 2 of the present invention;
FIG. 3 is a schematic diagram showing improvement in operation efficiency of the information gain dimension reduction method in example 1 of the present invention compared with the principal component analysis method;
fig. 4 is a schematic diagram of a confusion matrix of the result of judging the state of the single load unit for the original data of the telemetry parameters of the load in other time periods by using the judging model of the state of the single load unit obtained by training.
Detailed Description
The invention provides a method for rapidly identifying the state of a single load unit based on information gain parameter feature selection and an integrated learning method based on telemetry original data aiming at the problems that the single load unit is high in telemetry parameter dimension, large in data volume, unbalanced in category, incapable of intuitively distinguishing the operation condition of the single load unit and the like, and considering the requirement of space missions on interpretability. The method of the invention comprises the following steps:
the load single machine equipment state discrimination learning problem is classified into a multi-label classification problem model;
extracting load sub-package telemetry original data and extracting a load plan;
formatting the telemetry original data set, fusing sub-table sub-libraries and analyzing a load operation plan file;
constructing a load telemetry parameter characteristic sample set and a load state multi-label sample set, performing time scale matching and outlier rejection on the telemetry parameter sample set and the load state multi-label sample level segment;
the remote measurement parameter characteristic screening dimension reduction and load state multi-label characteristic conversion is realized, the interpretability dimension reduction is realized, and statistic properties such as variance, pelson coefficient and the like, and the remote measurement parameter characteristic and the target value are adopted;
an ensemble learning model training and evaluation method is adopted, and a logarithmic loss function is used.
To complete the model building, the multi-label classification problem model further includes defining telemetry parameter data and load state tensors as follows:
TM＝{tm j i j=1, 2,..n } is the set of load telemetry parameters, tm j Telemetry parameters for the j-th dimension.
Telemetry parameter data record vector at a certain momentExtracting telemetry original data acquired by a certain load or a certain stand-alone device at a certain moment on the satellite from the subpackage telemetry:
wherein, c (i) is the time scale corresponding to the data record vector when the data acquisition on a certain star is on the star corresponding to the data record vector,for an n-dimensional telemetry parameter data vector for the corresponding time instant.
P＝{p k I k=1, 2, l is the set of task payload devices, p k Representing the kth load device status.
Load device state vector U (i) ，For the l-dimensional load state vector at the corresponding time:
to complete the model building, the multi-label classification problem model is expressed as:
given a givenMulti-label training sample set d= { (TM) (i) ,P (i) ) I 1.ltoreq.i.ltoreq.s, for each sample record (TM (i) ,P (i) )，TM (i) ∈Ω TM For the n-dimensional feature vector to be recorded,to record TM (i) A corresponding tag. The multi-label classifier h (·) is learned in the given sample record dataset D as:
h(TM (i) )＝P (i)
load-packing telemetry raw data extraction and load planning, comprising: and selecting a corresponding telemetry source package and a corresponding load operation plan according to the task load and the sub-package telemetry description.
Telemetry original dataset formatting and sub-table sub-library fusion and load operation plan file parsing, comprising: and (5) performing original data digitizing, merging, analyzing and the like.
Constructing a load telemetry parameter feature sample set and a load state multi-label sample set, performing time scale matching and outlier rejection on the telemetry parameter sample set and the load state multi-label sample level segment, and comprising the following steps:
performing outlier rejection on the parameter data digitized by the telemetry original data by adopting a 3 sigma principle first-order data difference;
and performing time scaling and segmentation screening according to the time and the load state of the telemetry data on the satellite to obtain a load telemetry parameter characteristic sample set and a load state multi-label sample set.
And the remote measurement parameter characteristic screening dimension reduction and load state multi-label characteristic conversion is realized, so that the interpretable dimension reduction is realized.
By adopting statistic properties such as variance and pearson coefficient, the spacecraft mission telemetry data pearson correlation coefficient formula is as follows:
calculating a correlation coefficient between the two telemetry data characteristic sample sets, wherein a and b respectively represent two telemetry parameter formatted data in the telemetry characteristic sets, T represents transposition and E represents expected values;
if meeting |ρ 2 (a, b) -1 is less than or equal to epsilon, and one of the characteristics is reserved; epsilon represents a threshold value.
The variance of the telemetry data characteristic sample set is calculated, a constant value or a slowly-varying value irrelevant to the classification and discrimination of the state of the load single machine equipment is screened out, and the calculation amount of a classification model and the interference item of the model accuracy are reduced;
judging information gain of the single load state by calculating the information gain of the single load state, traversing the information gain of the labels of the single load state by each parameter characteristic, obtaining gain sequencing Rank, and selecting the characteristic with large information gain, wherein the formula is as follows:
wherein X represents a random variable of a telemetry parameter, H (X) represents information entropy of the random variable X, p i Representing the probability of a random event X, and n represents the number of random variable set elements formed by a telemetry parameter set;
conditional entropy H (y|x) of random variable (X, Y):
wherein, (X, Y) is a two-dimensional random variable which represents the composition of the telemetry parameter X and the telemetry parameter Y, and H (Y|X) represents the conditional entropy of the random variable Y under the given priori condition of the random variable X;
the information gain g (Y, X) is:
g(Y,X)＝H(Y)-H(Y|X)
wherein H (Y) represents the information entropy of the random variable Y;
and traversing the information gain of each telemetry parameter feature to the state of the single machine equipment, obtaining the information gain sequence, and selecting the feature with large information gain to leave.
An ensemble learning model training and evaluation method is adopted, and a logarithmic loss function is used. The algorithm flow and formula are as follows:
1) Initializing weak classifiers
Where tm represents the telemetry parameter, arg represents the optimal solution, p i The i-th real sample value is represented, namely the load single machine equipment is in a state i, c represents the classification result judged by the distributor, L represents the sample value and the classification result loss function, and N represents the total number of samples;
2) The M weak classifiers are iterated, and the internal iteration process is as follows:
for sample feature set TM ", calculate negative gradient residual
Where p represents a true sample value of the load stand-alone device state, h (tm i ) Representing the current weak classifier, i.e. for telemetry parameter tm i Judging the state of the load single machine equipment, wherein L represents a loss function, x represents the input parameters of the classifier, h (x) represents the selection of the iterative classifier, and h m-1 (x) A weak classifier representing a previous residual iteration;
taking the residual error as a new sample value to obtain a data set { TM', r of the next class tree mi New regression tree h m (tm) the corresponding leaf node region R jm J=1, 2 …, J being the number of leaf nodes of the regression tree;
computing a best fit c to the leaf area of the regression tree based on empirical risk minimization criteria jm ：
3) Updating learning device
Until the negative gradient residual error converges to obtain a final learner
The technical scheme of the invention is described in detail below with reference to the accompanying drawings and examples.
Example 1
The principle of the invention is as follows: and establishing a remote-measurement original data-driven load single machine equipment state discrimination problem model, training a remote-measurement original data set and a load single machine equipment state data set by using an integrated learning method, and automatically discriminating the load single machine equipment state according to the remote-measurement data by using the fitted model.
As shown in fig. 1, embodiment 1 of the present invention provides a method for determining a state of a load stand-alone device driven by telemetry original data, the method using satellite telemetry data to automatically determine the state of the load stand-alone device, the method comprising:
step 1), configuring remote measurement parameter library table access information and task load operation plan library table access information according to satellite tasks to be interpreted and load single machine equipment states;
step 2) initializing system parameters, which mainly comprise real-time and time-delay telemetry parameter table information and task load operation plan file information of the task stand-alone equipment to be accessed;
step 3) selecting and extracting telemetry original data according to the sub-package telemetry and the task load, analyzing the task load operation file to obtain the state information of the single machine equipment, and establishing an original sample set;
step 4) digitizing the telemetry original data and the single load equipment state data according to a processing strategy, and establishing a characteristic sample set of the telemetry data and the single load equipment state information according to data timestamp information;
the method specifically comprises the steps of telemetering parameter data and load state tensor definition, and establishing a multi-label classification problem model of telemetering data and load single machine equipment states.
Telemetry parameter data and load state tensor definitions include:
TM＝{tm j i j=1, 2,..n } is the set of load telemetry parameters, tm j Telemetry parameters for the j-th dimension;
telemetry parameter data record vector at a certain momentExtracting telemetry original data acquired by a certain load or a certain stand-alone device at a certain moment on the satellite from the subpackage telemetry:
wherein, c (i) is the time scale corresponding to the data record vector when the data acquisition on a certain star is on the star corresponding to the data record vector,n-dimensional telemetry parameter data vector for corresponding time;
P＝{p k i k=1, 2, l is the set of task payload devices, p k Representing a kth load device status;
load device state vector U (i) ，For the l-dimensional load state vector at the corresponding time:
the multi-label classification problem model establishment of telemetry data and load single machine equipment states comprises the following steps:
given a multi-label training sample set d= { (TM) (i) ,P (i) ) I 1.ltoreq.i.ltoreq.s, for each sample record (TM (i) ,P (i) )，TM (i) ∈Ω TM For the n-dimensional feature vector to be recorded,to record TM (i) A corresponding tag. The multi-label classifier h (·) is learned in the given sample record dataset D as:
h(TM (i) )＝P (i)
analyzing the characteristics of telemetry data, screening a telemetry data characteristic set, reducing the dimension, analyzing the characteristics of state data of load single-machine equipment, converting the problem, establishing a multi-label classification problem model, and dividing a training set and a verification set to carry out integrated learning model training;
the telemetry data characteristic analysis of this step mainly includes: the method comprises the steps of formatting an original data set, fusing sub-table sub-library and analyzing a load operation plan file, removing telemetry data wild values, analyzing telemetry data statistical characteristics, and calculating information gain to obtain a load telemetry parameter characteristic sample set and a load state multi-label sample set.
The telemetry data outlier rejection is completed by adopting a 3 sigma principle first-order data difference;
the statistical characteristic analysis of the telemetry data adopts variance and Pelson coefficient to calculate, and the Pelson related coefficient formula of the telemetry data of the spacecraft task is as follows:
calculating the correlation coefficient between the two of the telemetry data characteristic sample set and if |ρ is satisfied 2 (a, b) -1 is less than or equal to epsilon, and one of the characteristics is reserved;
the variance of the telemetry data characteristic sample set is calculated, a constant value or a slowly-varying value irrelevant to the classification and discrimination of the state of the load single machine equipment is screened out, and the calculation amount of a classification model and the interference item of the model accuracy are reduced;
judging information gain of the single load state by calculating the information gain of the single load state, traversing the information gain of the labels of the single load state by each parameter characteristic, obtaining gain sequencing Rank, and selecting the characteristic with large information gain, wherein the formula is as follows:
g(Y,X)＝H(Y)-H(Y|X)
the learning algorithm flow and formula for judging the state of the load single machine equipment driven by the telemetering original data are as follows:
a) Initializing weak classifiers
b) The M weak classifiers are iterated, and the internal iteration process is as follows:
1. for sample feature set TM ", calculate negative gradient residual
2. Taking the residual error as a new sample value to obtain a data set { TM', r of the next class tree mi New regression tree h m (tm) the corresponding leaf node region R jm J=1, 2 …, J being the number of leaf nodes of the regression tree;
3. computing a best fit to leaf areas of a regression tree based on empirical risk minimization criteria
c) Updating learning device
d) Obtaining the final learner
And 6) applying the trained model to the adjustment parameters of the verification set, and further applying the training model to the state discrimination of the actual task load single-machine equipment.
And (3) effect analysis:
as shown in FIG. 3, the information gain dimension reduction method is compared with the principal component analysis method to improve the operation efficiency, and the operation efficiency is obviously improved.
Fig. 4 is a schematic diagram of a confusion matrix of the result of judging the state of the single load unit for the original data of the telemetry parameters of the load in other time periods by using the judgment model of the state of the single load unit obtained by training.
Example 2
As shown in fig. 2, embodiment 2 of the present invention provides a telemetry original data driven load stand-alone device status discrimination system, which includes a satellite telemetry original database, a satellite mission operation plan database, a telemetry parameter database table (real-time, time-delay) configuration module, a mission operation plan database table configuration module, a packetized telemetry load data extraction module, a telemetry parameter original data merging and formatting module, a telemetry parameter formatted data set construction module, a outlier rejection module, a segmented time scale alignment module, a feature screening module, a telemetry data and load status multi-label classification integrated learning model training module, a model tuning parameter and an application module.
All raw telemetry data for the data satellite downstream stored in the satellite telemetry data raw database.
The data stored in the satellite mission operation plan database is load operation plan data, including operation plan and status information of the load stand-alone equipment.
The satellite task telemetry parameter library table configuration module: and setting a corresponding telemetry parameter library table according to the satellite task model, and setting a database table and access information of satellite telemetry parameter variables.
Task load operation plan database configuration module: and setting access parameter information of corresponding task load operation plan data according to the task model. And reading load plan data from the task load operation plan database, and analyzing load state information.
A sub-packet telemetry payload data extraction module: and selecting telemetry data under the corresponding sub-package telemetry number according to the load single machine equipment to be judged in the state.
Telemetry parameter raw data merging and formatting module: and merging and numerically converting the corresponding sub-package telemetry data of the selected load single machine equipment state.
Task data analysis module: and analyzing the task load operation plan file to obtain the state data of the load single-machine equipment.
Telemetry parameter formatting dataset construction module: and constructing a telemetry parameter formatted data sample set from the parameter data obtained by the processing of the preamble module.
And the outlier removing module is used for: and finding out and eliminating the outliers in the telemetry data sample set by adopting the 3 sigma principle first-order data difference.
Segment time scale alignment module: and matching the load telemetry parameter data sample set with the load single machine equipment state data according to a time scale, establishing a telemetry data-driven load single machine equipment state discrimination problem model, and dividing the data set into a test set and a verification set. The method comprises the following steps:
telemetry parameter data and load state tensor definitions include:
TM＝{tm j i j=1, 2,..n } is the set of load telemetry parameters, tm j Telemetry parameters for the j-th dimension;
telemetry parameter data record vector at a certain momentExtracting telemetry original data acquired by a certain load or a certain stand-alone device at a certain moment on the satellite from the subpackage telemetry:
wherein c (i) is the time corresponding to the data acquisition on a certain satellite, and is used as the time corresponding to the data record vectorThe number of the mark is set to be equal to the number of the mark,n-dimensional telemetry parameter data vector for corresponding time;
P＝{p k i k=1, 2, l is the set of task payload devices, p k Representing a kth load device status;
load device state vector U (i) ，For the l-dimensional load state vector at the corresponding time:
the multi-label classification problem model establishment of telemetry data and load single machine equipment states comprises the following steps:
given a multi-label training sample set d= { (TM) (i) ,P (i) ) I 1.ltoreq.i.ltoreq.s, for each sample record (TM (i) ,P (i) )，TM (i) ∈Ω TM For the n-dimensional feature vector to be recorded,to record TM (i) A corresponding tag. The multi-label classifier h (·) is learned in the given sample record dataset D as:
h(TM (i) )＝P (i)
and a feature screening module: feature filtering and dimension reduction are performed from the telemetry parameter formatted dataset. The main method is ordered according to the statistical properties and information gain of the sample dataset.
The calculation formula and the feature screening method of the pearson coefficient of the telemetry parameter data are as follows:
computing both telemetry data feature sample setsCorrelation coefficient between them, if meeting |ρ 2 (a, b) -1 is less than or equal to epsilon, and one of the characteristics is reserved;
the variance of the telemetry data characteristic sample set is calculated, a constant value or a slowly-varying value irrelevant to the classification and discrimination of the state of the load single machine equipment is screened out, and the calculation amount of a classification model and the interference item of the model accuracy are reduced;
judging information gain of the single load state by calculating the information gain of the single load state, traversing the information gain of the labels of the single load state by each parameter characteristic, obtaining gain sequencing Rank, and selecting the characteristic with large information gain, wherein the formula is as follows:
/>
g(Y,X)＝H(Y)-H(Y|X)
telemetry data and load state multi-label classification integrated learning model training module: the learning algorithm flow and formula for judging the state of the load single machine equipment driven by the telemetering original data are as follows:
a) Initializing weak classifiers
b) The M weak classifiers are iterated, and the internal iteration process is as follows:
1. for sample feature set TM ", calculate negative gradient residual
2. Taking the residual error as a new sample value to obtain a data set { TM', r of the next class tree mi New regression tree h m (tm) the corresponding leaf node region R jm J=1, 2 …, J being the leaf node of the regression treeNumber of;
3. computing a best fit to leaf areas of a regression tree based on empirical risk minimization criteria
c) Updating learning device
d) Obtaining the final learner
Model tuning parameters and application modules: and verifying the integrated learning model obtained through training in a verification set, adjusting corresponding model parameters, and applying the integrated learning model to an actual running task to judge the state of the load single machine equipment.
Example 3
Based on the method of embodiment 1, embodiment 3 of the present invention proposes a system for discriminating the status of a load stand-alone device driven by telemetry raw data. The system comprises a receiving module, an extracting module, a judging module and a trained load single machine equipment judging model; wherein,
the receiving module is used for receiving telemetry original data downloaded by a satellite;
the extraction module is used for extracting the sub-package telemetry data corresponding to the load single machine equipment from the telemetry original data;
and the judging module is used for inputting the sub-package telemetry data into a pre-trained load single machine equipment judging model to obtain a load single machine equipment state result.
The input of the load single machine equipment discrimination model is the sub-package telemetering data corresponding to the load single machine equipment, and the output is the state result of the load single machine equipment, and the load single machine equipment discrimination model comprises: the device comprises a telemetering parameter formatting processing module, a outlier rejection module and a load single machine equipment discrimination module; wherein,
the telemetry parameter formatting processing module is used for formatting the packetized telemetry data according to a processing strategy to obtain telemetry parameter formatting data;
the outlier removing module is used for removing outliers in the telemetry parameter formatted data;
and the load single machine equipment judging module is used for formatting data according to the telemetry parameters and outputting a state result of the load single machine equipment.
Finally, it should be noted that the above embodiments are only for illustrating the technical solution of the present invention and are not limiting. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the appended claims.
Claims (6)
1. A method for discriminating the state of a load stand-alone device driven by telemetry original data, the method comprising:
receiving telemetry original data downloaded by a satellite;
extracting the sub-package telemetry data corresponding to the load single machine equipment from the telemetry original data according to the telemetry parameter configuration file;
inputting the sub-package telemetering data into a pre-trained load single machine equipment discrimination model to obtain a load single machine equipment state result;
the input of the load single machine equipment distinguishing model is the sub-package telemetering data corresponding to the load single machine equipment, and the output is the state result of the load single machine equipment, and the load single machine equipment distinguishing model comprises: the device comprises a telemetering parameter formatting processing module, a outlier rejection module and a load single machine equipment discrimination module; wherein,
the telemetry parameter formatting processing module is used for formatting the packetized telemetry data according to a processing strategy to obtain telemetry parameter formatting data;
the outlier removing module is used for removing outliers in the telemetry parameter formatted data;
the load single machine equipment judging module is used for formatting data according to telemetry parameters and outputting a state result of the load single machine equipment;
the method also comprises the step of training the load single machine equipment discrimination model, and specifically comprises the following steps:
extracting sub-package telemetry data corresponding to the load single machine equipment from a spacecraft telemetry original database according to the task model of the spacecraft and the load single machine equipment, and extracting load single machine operation plan data from a spacecraft task operation plan database;
formatting the packetized telemetry data according to a processing strategy to obtain telemetry parameter formatted data, removing outliers by adopting a 3 sigma principle first-order data difference method, and establishing a telemetry parameter formatted data set;
matching the telemetry parameter formatted data set with the load single machine equipment state data according to a time scale to obtain a telemetry feature set;
screening the telemetry feature set, sorting information gain and converting multi-label features to obtain a screened feature set with reduced dimensions, and dividing the feature set into a training set and a verification set;
training the model by using a training set through an integrated learning method to obtain a trained load single machine equipment discrimination model;
using the verification set to verify the model and optimizing parameters to obtain a final load single machine equipment discrimination model;
training the model by using a training set through an integrated learning method to obtain a trained load single machine equipment discrimination model; the method specifically comprises the following steps:
initializing weak classifier h 0 (tm)：
Where tm represents the telemetry parameter, arg represents the optimal solution, p i Represents the i-th real sample value, namely the load single machine equipment is in the state i, c represents the classification result judged by the classifier, and L (p i C) represents a sample value and a classification result loss function, and N represents the total number of samples;
the M weak classifiers are iterated, and the internal iteration process is as follows:
for sample feature set TM ", calculate negative gradient residual r mi ：
Where p represents a true sample value of the load stand-alone device state, h (tm i ) Representing the current weak classifier, i.e. for telemetry parameter tm i Judging the state of the load single machine equipment, wherein L represents a loss function, x represents the input parameters of the classifier, h (x) represents the selection of the iterative classifier, and h m-1 (x) A weak classifier representing a previous residual iteration;
taking the residual error as a new sample value to obtain a data set { TM', r of the next class tree mi New regression tree h m (tm) the corresponding leaf node region R jm J=1, 2 …, J being the number of leaf nodes of the regression tree;
computing a best fit c to the leaf area of the regression tree based on empirical risk minimization criteria jm ：
Updating a learner:
wherein h is m-1 (tm) represents the previous regression tree, I (tm ε R) jm ) Representing an indication function;
until the negative gradient residual error converges, obtaining a trained load single machine equipment discrimination model:
2. the method for judging the state of the load single machine equipment driven by telemetry original data according to claim 1, wherein the method is characterized in that the sub-package telemetry data corresponding to the load single machine equipment is extracted from a spacecraft telemetry original database according to the task model of a spacecraft and the load single machine equipment, and the load single machine operation plan data is extracted from a spacecraft task operation plan database; the method specifically comprises the following steps:
configuring remote measurement parameter library table access information and task load single machine operation plan library table access information according to the spacecraft task model and load single machine equipment;
extracting load single machine operation plan data from a spacecraft task operation plan database according to the task load operation plan database table access information;
analyzing the single load operation plan data to obtain a single load operation plan and single load equipment state data;
and extracting the sub-package telemetry data corresponding to the single load equipment from the spacecraft telemetry original database according to the telemetry parameter library table access information and the single load equipment state data.
3. The method for judging the state of the load single machine equipment driven by telemetry original data according to claim 1, wherein the telemetry parameter formatted data set and the load single machine equipment state data are matched according to a time scale to obtain a telemetry feature set; the method specifically comprises the following steps:
defining telemetry parameter data record vectorsAnd a load device state tensor U (i) ：
Wherein c (i) When the data is acquired from a certain star, the corresponding star is used as a time mark corresponding to the data record vector,for an n-dimensional telemetry parameter data vector at the corresponding time instant, tm= { TM j I j=1, 2,..n } is the set of load telemetry parameters, tm j Telemetry parameters for the j-th dimension;
wherein,for the l-dimensional load state vector at the corresponding time, p= { P k I k=1, 2, l is the set of task payload devices, p k Representing a kth load device status;
matching according to a time scale to obtain a telemetry feature set D= { (TM) (i) ,P (i) )|1≤i≤s}
Wherein, (TM) (i) ,P (i) ) Record for each sample, TM (i) ∈Ω TM For the n-dimensional feature vector to be recorded,to record TM (i) A corresponding tag.
4. The method for discriminating conditions of a telemetry-driven load stand-alone device according to claim 1 wherein said screening comprises:
computing pearson correlation coefficient ρ for telemetry parameter formatted data in a telemetry feature set 2 (a,b)：
Wherein a and b respectively represent two telemetry parameter formatted data in a telemetry feature set, T represents transposition, and E represents expected value;
if meeting |ρ 2 (a, b) -1|ε, then preserving telemetry parameter formatted data, ε representing a threshold;
and calculating variance for telemetry parameter formatted data in telemetry feature set, and screening out constant value or slowly-varying value irrelevant to classification and discrimination of the state of the load single machine equipment.
5. The method for discriminating states of telemetry original data driven load stand-alone devices according to claim 1 wherein said information gain ordering is specifically: information gain is calculated for the load stand-alone device status data in the telemetry feature set,
wherein X represents a random variable of a telemetry parameter, H (X) represents information entropy of the random variable X, p i Representing the probability of a random event X, and n represents the number of random variable set elements formed by a telemetry parameter set;
conditional entropy H (y|x) of random variable (X, Y):
wherein, (X, Y) is a two-dimensional random variable which represents the composition of the telemetry parameter X and the telemetry parameter Y, and H (Y|X) represents the conditional entropy of the random variable Y under the given priori condition of the random variable X;
the information gain g (Y, X) is:
g(Y,X)＝H(Y)-H(Y|X)
wherein H (Y) represents the information entropy of the random variable Y;
and traversing the information gain of each telemetry parameter feature to the state of the single machine equipment, obtaining the information gain sequence, and selecting the feature with large information gain to leave.
6. A telemetry original data-driven load stand-alone device status discrimination system implemented in accordance with the method of any one of claims 1-5, the system comprising: the system comprises a receiving module, an extracting module, a judging module and a trained load single machine equipment judging model; wherein,
the receiving module is used for receiving telemetry original data downloaded by a satellite;
the extraction module is used for extracting the sub-package telemetry data corresponding to the load single machine equipment from the telemetry original data;
the judging module is used for inputting the sub-package telemetry data into a pre-trained load single machine equipment judging model to obtain a load single machine equipment state result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010710292.3A CN111967501B (en) | 2020-07-22 | 2020-07-22 | Method and system for judging load state driven by telemetering original data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010710292.3A CN111967501B (en) | 2020-07-22 | 2020-07-22 | Method and system for judging load state driven by telemetering original data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111967501A CN111967501A (en) | 2020-11-20 |
| CN111967501B true CN111967501B (en) | 2023-11-17 |
Family
ID=73363892
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010710292.3A Active CN111967501B (en) | 2020-07-22 | 2020-07-22 | Method and system for judging load state driven by telemetering original data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111967501B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116502516B (en) * | 2023-01-15 | 2023-12-01 | 北京控制工程研究所 | Identification method and device for degradation stage of spacecraft component |
| CN116522467B (en) * | 2023-01-16 | 2023-10-31 | 北京控制工程研究所 | Prediction method and device for service life of spacecraft component |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103761532A (en) * | 2014-01-20 | 2014-04-30 | 清华大学 | Label space dimensionality reducing method and system based on feature-related implicit coding |
| CN104217273A (en) * | 2013-05-31 | 2014-12-17 | 北京空间技术研制试验中心 | Characteristic management method for spacecraft three-dimensional digital model |
| CN105809190A (en) * | 2016-03-03 | 2016-07-27 | 南京邮电大学 | Characteristic selection based SVM cascade classifier method |
| CN110210534A (en) * | 2019-05-21 | 2019-09-06 | 河海大学 | High score remote sensing images scene multi-tag classification method based on more packet fusions |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9641238B1 (en) * | 2016-10-19 | 2017-05-02 | Vector Launch Inc. | Virtualization-enabled satellite platforms |
-
2020
- 2020-07-22 CN CN202010710292.3A patent/CN111967501B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104217273A (en) * | 2013-05-31 | 2014-12-17 | 北京空间技术研制试验中心 | Characteristic management method for spacecraft three-dimensional digital model |
| CN103761532A (en) * | 2014-01-20 | 2014-04-30 | 清华大学 | Label space dimensionality reducing method and system based on feature-related implicit coding |
| CN105809190A (en) * | 2016-03-03 | 2016-07-27 | 南京邮电大学 | Characteristic selection based SVM cascade classifier method |
| CN110210534A (en) * | 2019-05-21 | 2019-09-06 | 河海大学 | High score remote sensing images scene multi-tag classification method based on more packet fusions |
Non-Patent Citations (3)
| Title |
|---|
| "Spacecraft attitude fault-tolerant control based on iterative learning observer and control allocation";Qinglei Hu 等;《Aerospace Science and Technology》;第245-253页 * |
| "基于集成学习的遥测数据互相关结构知识发现";石梦鑫 等;《北京航空航天大学学报》;第46卷(第1期);第181-188页 * |
| "航天器遥测数据异常检测综述";彭喜元 等;《仪器仪表学报》;第1929-1945页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111967501A (en) | 2020-11-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107122375B (en) | Image subject identification method based on image features | |
| CN111967501B (en) | Method and system for judging load state driven by telemetering original data | |
| CN109522221A (en) | A kind of method and system improving fuzz testing efficiency | |
| CN113837238A (en) | Long-tail image identification method based on self-supervision and self-distillation | |
| CN104978569B (en) | A kind of increment face identification method based on rarefaction representation | |
| CN111126820A (en) | Electricity stealing prevention method and system | |
| CN112860676B (en) | Data cleaning method applied to big data mining and business analysis and cloud server | |
| CN111104242A (en) | Method and device for processing abnormal logs of operating system based on deep learning | |
| CN116795886B (en) | Data analysis engine and method for sales data | |
| CN112756759A (en) | Spot welding robot workstation fault judgment method | |
| CN117056834A (en) | Big data analysis method based on decision tree | |
| CN112367303A (en) | Distributed self-learning abnormal flow cooperative detection method and system | |
| CN113778894A (en) | Test case construction method, device, equipment and storage medium | |
| CN116257759A (en) | Structured data intelligent classification grading system of deep neural network model | |
| CN115757103A (en) | Neural network test case generation method based on tree structure | |
| CN117351659B (en) | Hydrogeological disaster monitoring device and monitoring method | |
| CN117078007A (en) | Multi-scale wind control system integrating scale labels and method thereof | |
| CN117372144A (en) | Wind control strategy intelligent method and system applied to small sample scene | |
| CN113420733B (en) | Efficient distributed big data acquisition implementation method and system | |
| CN114897193A (en) | Airplane structure maintenance decision method and decision system based on man-in-the-loop | |
| CN113191569A (en) | Enterprise management method and system based on big data | |
| CN114627333A (en) | Zinc flotation froth image classification algorithm and system for improving deep active learning | |
| CN112559589A (en) | Remote surveying and mapping data processing method and system | |
| CN114077663A (en) | Application log analysis method and device | |
| CN117892091B (en) | Data intelligent analysis method and system based on artificial intelligence |
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 |