CN112464146B - Key subsystem based on historical telemetering data and single-machine correlation health baseline construction method - Google Patents
Key subsystem based on historical telemetering data and single-machine correlation health baseline construction method Download PDFInfo
- Publication number
- CN112464146B CN112464146B CN202011404591.0A CN202011404591A CN112464146B CN 112464146 B CN112464146 B CN 112464146B CN 202011404591 A CN202011404591 A CN 202011404591A CN 112464146 B CN112464146 B CN 112464146B
- Authority
- CN
- China
- Prior art keywords
- working condition
- binary
- remote reference
- reference sequence
- remote
- 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
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computational Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Operations Research (AREA)
- Software Systems (AREA)
- Algebra (AREA)
- Probability & Statistics with Applications (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The application discloses a key subsystem based on historical telemetering data and a single-machine relevance health baseline construction method, which comprises the following steps: selecting a group of binary remote reference sequence combinations in linear correlation or nonlinear correlation; firstly, time calibration is carried out on the selected binary remote reference sequence combination; secondly, carrying out target working condition recognition and cutting on the binary remote reference sequence subjected to time calibration to obtain a binary remote reference sequence under a target working condition period; thirdly, fitting the binary remote parameter sequence data under the target working condition in the normal state to obtain a binary linear correlation health baseline, or converting a nonlinear relation into a linear relation by adopting a discrete integration-based method, and fitting to obtain a binary nonlinear correlation health baseline; the binary linear correlated health baselines and the binary non-linear correlated health baselines together form a correlated health baselines library. And quantitative stable representation is realized on the satellite key subsystem and the single-machine health state through the established relevance health baseline library.
Description
Technical Field
The application relates to a satellite health monitoring technology, in particular to a key subsystem based on historical telemetering data and a single-machine relevance health baseline construction method.
Background
The satellite is a system with complex functions and components, and comprises a plurality of key subsystems and single machines, and a plurality of running state data. Whether the operation state data of the satellite are in an abnormal operation state or not is monitored by analyzing the operation state data of the satellite, so that the operation fault of the satellite can be found and processed in time, and the method has an extremely important significance for ensuring the operation reliability of the satellite.
When anomaly monitoring of a satellite is carried out, firstly, the health state of the satellite needs to be accurately, effectively and quantitatively characterized. The traditional health state characterization methods are mainly divided into 3 types, and the method is characterized in that: a. single parameter health characterization: and setting corresponding single parameter thresholds for certain satellite state sensitive parameters one by one, and judging the current health state of the satellite according to the condition that whether the parameters exceed the limits or not. The single parameter threshold method can only finish the binarization representation (health/abnormity) of the satellite health state, and is easily interfered by the satellite operation condition and random environment fluctuation, thus causing false alarm and false alarm with higher probability; b. qualitative health status characterization: and collecting multiple groups of telemetering data of the satellite, performing qualitative health characterization on one of the telemetering data by using amplitude trend change indexes (abnormal rising, abnormal falling and the like), and comprehensively reasoning to obtain a qualitative health state of the satellite by combining certain logic combination. The method can only carry out qualitative characterization on the health state of the satellite, and is difficult to find early abnormity of the satellite; c. physical model health status characterization: and completing redundant modeling of a satellite key subsystem, namely an observer model, through detailed data such as satellite control logic, physical structures and the like. The health state of the satellite is quantitatively judged through the difference between the theoretical remote parameter data and the actual remote parameter data output by the observer model. Due to the fact that the satellite structure is numerous and complicated, the operation environment is variable, all observer models need complete object information, the models are large in size, the occupied operation resources are high, the requirements of satellite-borne deployment and real-time operation are difficult to match, and meanwhile the problems of high popularization difficulty, low representation accuracy, high expert knowledge dependence and the like exist.
In order to effectively perform the satellite anomaly detection, a set of effective satellite health state characterization methods is necessary to be provided.
Disclosure of Invention
Aiming at the problems of high expert knowledge dependence, insensitivity to early abnormality, poor robustness of abnormality detection, excessively complex model quantity and the like of the current satellite health state characterization method, the application aims to provide a key subsystem based on historical telemetering data and a single-machine relevance health baseline construction method.
The application provides a key subsystem based on historical telemetering data and a single-machine relevance health baseline construction method, which comprises the following steps:
selecting a group of binary remote reference sequence combinations in linear correlation or nonlinear correlation;
firstly, time calibration is carried out on the selected binary remote reference sequence combination, and data points of the binary remote reference sequence combination are matched one by one;
secondly, performing target working condition identification and cutting on the binary remote reference sequence subjected to time calibration to obtain a binary remote reference sequence under a target working condition period, wherein a stable mathematical relation existing between the binary remote reference sequences under the target working condition period is a binary relevance health baseline of the binary remote reference sequence under the target working condition period;
thirdly, if the binary correlation health baseline is in a linear relation, fitting an analytic equation of the linear relation by using the binary remote reference sequence data of the target working condition in the normal state to obtain parameters of a binary linear correlation health baseline model, and obtaining the binary linear correlation health baseline; if the binary correlation health baseline is in a nonlinear relation, converting the nonlinear relation into a linear relation by adopting a discrete integral-based method, and fitting to obtain parameters of a binary nonlinear correlation health baseline model to obtain the binary nonlinear correlation health baseline; the binary linear correlated health baselines and the binary non-linear correlated health baselines together form a correlated health baselines library.
Preferably, the method further comprises a fourth step of performing a goodness-of-fit test on the correlation health baseline.
Preferably, the method further comprises a fifth step of obtaining an interval coefficient of the correlation healthy baseline by a parameter interval estimation method in statistics, so as to obtain a corresponding correlation healthy baseline family.
Preferably, in the first step, the step of performing time calibration is as follows:
determining reference remote reference, wherein in the time calibration pretreatment, one remote reference sequence in the binary remote reference sequences is selected as a reference remote reference sequence, and the other one in the binary remote reference sequences is a remote reference sequence to be calibrated;
performing reference comparison search, namely taking time labels of all points of a reference remote reference sequence to form a reference time axis, and searching the remote reference sequence to be calibrated in an associated time range by using each time point on the reference time axis, wherein if a remote parameter data point exists in the associated time range, the time point is an effective time point, and data corresponding to the effective time point of the remote reference sequence to be calibrated is an effective data point;
and aligning the remote reference sequences, performing reference comparison search on the reference remote reference sequence and the remote reference sequence to be calibrated at all effective time points, wherein remote parameter data points of the reference remote reference sequence at each effective time point form the reference remote reference sequence after time calibration, and remote parameter data points of the remote reference sequence to be calibrated at each effective time point form the remote reference sequence to be calibrated after time calibration, namely realizing one-to-one matching of data points of the binary remote reference sequence combination.
Preferably, in the second step, when the target working condition is identified and cut, the method comprises the following steps:
carrying out time calibration on the working condition sensitive remote reference sequence;
generating a target working condition conversion threshold;
judging and generating a target working condition time period based on the evidence logic combination;
and extracting binary remote parameter sequence data in the target working condition time period.
Preferably, the working condition sensitive remote reference sequence is determined by a plurality of remote reference sequences which determine various correlation relations among binary remote reference sequences in a complete period; the remote parameter sequences are working condition sensitive remote parameter sequences;
and time calibration is carried out on the working condition sensitive remote reference sequence by utilizing the reference time axis.
Preferably, for each sequence in the operating condition sensitive remote reference sequences, fitting multiple operating conditions of the sequence by using multiple normal distributions; for each working condition, generating a working condition segmented threshold interval by adopting a statistical principle to obtain an upper boundary value and a lower boundary value of stable working condition distribution of the working condition; combining the upper and lower boundaries which are distributed adjacently in pairs to obtain a threshold interval of the conversion working condition; and the threshold value intervals of all the conversion working conditions form a target working condition conversion threshold value set.
Preferably, the target working condition switching threshold value set is logically combined through a logical connection symbol and a logical comparison symbol to form a multi-evidence logical combination criterion;
and obtaining a set of all target working condition periods by using the reference time axis, the time-calibrated working condition sensitive remote reference sequence and the target working condition switching threshold value set.
Preferably, the binary remote reference sequence of the target working condition time period under each working condition time period is cut out from the binary remote reference sequence calibrated by time by taking the set of all the target working condition time periods as an index.
Preferably, in the third step, when a discrete integration based method is used,
extracting a time tag by using the target working condition time interval set, and carrying out point-by-point difference processing on the time value of adjacent remote parameter data in the time interval range of each target working condition time interval to obtain a target working condition time interval set; for each target working condition time interval, multiplying the target working condition time interval set by an independent variable remote reference sequence in the binary remote reference sequence of the target working condition time interval point by point to obtain each unit time interval increment value in the target working condition time interval, and accumulating the unit time interval increment values to obtain a unit time interval accumulated value of the target working condition time interval; and the accumulated values of unit time intervals of all the target working condition periods form independent variable remote parameter discrete integral sequences, so that the nonlinear relation fitting between the binary remote parameter sequence combinations is simplified into linear relation fitting.
The invention has the advantages and positive effects that:
according to the key subsystem based on historical telemetering data and the single machine correlation health baseline construction method, the defects of single health characterization method, poor abnormal detection robustness and excessively complex model quantity of the traditional method are overcome through the binary correlation health baseline family under the actual satellite data conditions of multiple operation working conditions and noise interference, and the stable and accurate characterization of the satellite key subsystem and the single machine health state is realized. The health baseline construction method provided by the invention has the advantages of high utilization rate of remote parameter information, low computing resource requirement and less dependence on expert knowledge, and can effectively meet the related requirements of real-time deployment of a satellite-borne terminal while effectively supporting data analysis of a satellite at a ground terminal.
Drawings
FIG. 1 is a flow chart of the associative health baseline construction of the present application;
FIG. 2 is the original 4 remote reference partial sequences without time scaling of example 1;
FIG. 3 is the time-scaled reference sequence of FIG. 2;
FIG. 4 is a distribution histogram of a condition-sensitive remote reference sequence;
FIG. 5 is a sequence of the identified and screened condition-sensitive remote reference;
FIG. 6 shows the sliced binary telemetry sequence data of the target operating condition period;
FIG. 7 is a graph of the binary linear correlation healthy baseline family of the embodiment 1 and its local detail;
FIG. 8 is the original remote reference partial sequence of example 2 without time scaling;
FIG. 9 is a schematic diagram of a remote reference sequence after time scaling;
FIG. 10 is a schematic diagram of the recognition effect of the target condition period;
FIG. 11 is a diagram illustrating the cutting effect of binary remote reference sequence data in a target working condition period;
FIG. 12 is a schematic diagram of a constructed correlation health baseline family curve.
Detailed Description
The invention comprises the following steps:
step one, time calibration pretreatment
Selecting a binary remote reference sequence with potential relevance from the massive satellite remote references:
wherein X is independent variable remote reference sequence, and the total number of points is N1(ii) a Y is a dependent variable remote reference sequence, and the total number of points is N2. The premise of constructing the correlation health baseline of the binary remote reference sequence is to ensure that data points in the two sequences are strictly matched with each other correspondingly. In practice, however, the sampling frequency between the remote references is different,The sampling moments are not consistent, and the data precondition requirements are difficult to meet. And carrying out time calibration pretreatment on the binary remote parameter sequence combination to obtain a binary remote parameter combination with strictly matched data points, and providing a data base for subsequent construction of a correlation health baseline.
Step 101: reference telemetry determination
The binary telemetry sequences X, Y are respectively subjected to differential processing to obtain differential value sequences DX、DY. Respectively taking mode numbers of the difference value sequences of the two remote parameters to obtain a set of the difference value mode numbers of the remote measurement sequences:
M＝{m(DX),m(DY)}
where m (-) represents the mathematical operation of taking the mode on a sequence of values.
And selecting the remote reference sequence (assumed as X) with the maximum mode of the differential value as the remote reference sequence. Thereafter, the telemetry sequence Y is time-scaled with respect to the time axis of the telemetry sequence X.
Step 102: benchmark contrast search
Time labels of points X of reference remote reference sequence are taken to form reference time axis
Taking a first time point t in a time axis1Searching for the remote reference sequence Y to be calibrated at t1The telemetry parameter values within the associated range. t is t1The associated time range is calculated as follows:
m(DY) And (4) regarding the mode of the difference value of the to-be-calibrated remote reference sequence calculated in the step 101, wherein the association time range is a left-open and right-closed interval. The calculation methods of the associated time ranges of the rest of the moments are all as described above, and are not described in detail later.
For the remote reference sequence Y to be calibrated, if the remote reference sequence Y has remote parameter data points Y 'in the associated time range'1Then the point is pointed outAs the first time scale of sequence Y, note
t1Is called the effective time point, is recorded as
For the reference remote reference sequence X, t1Corresponding data point x1The first time calibration value as the remote reference sequence X is recorded as
Step 103: remote reference sequence alignment
Repeating the reference comparison searching operation in step 102 for each time point in the reference remote reference sequence X and the remote reference sequence Y to be calibrated, and finally obtaining all effective time point sets
And m is the total number of effective time points.
The reference remote reference sequence X after time calibration is recorded as
Recording the remote reference sequence Y to be calibrated after time calibration as
The total number of data points contained within both sequences is m.
Data points in the binary remote reference sequences X 'and Y' after time calibration are strictly matched one by one, so that subsequent construction of the associated health baseline can be carried out.
Step two, target working condition recognition and cutting
In the periodic operation process of the satellite, various operation states, namely multiple operation working conditions, often exist in each period. Under different operating conditions, the correlation relationship between the satellite binary remote reference sequence X, Y will change accordingly. In order to establish an accurate correlation healthy baseline between the binary remote parameter sequences X, Y, firstly, a target working condition is identified through a working condition sensitive remote parameter sequence, and the remote parameter data in the binary remote parameter sequence X, Y is cut by utilizing an identified target working condition time interval, so that the establishment of the subsequent correlation healthy baseline is supported. This is achieved by the following procedure.
Step 201: time calibration of working condition sensitive remote reference sequence
Assuming that the correlation between the binary remote reference sequences X, Y shows l states in a complete cycle, that is, there are l operating conditions, and the l operating conditions can be determined by N remote reference sequences, then the N remote reference sequences are regarded as condition-sensitive remote reference sequences, and are recorded as:
wherein N is2+jAnd counting the total number of data contained in the j-th working condition sensitive remote reference sequence.
And (3) carrying out time calibration processing on the binary remote reference sequence and the working condition sensitive remote reference sequence by using the method in the step one to obtain a working condition sensitive remote reference sequence after time calibration, and recording as:
data points between the working condition sensitive remote reference sequence and the binary remote reference sequence after time calibration are strictly matched one by one, and a data basis is provided for subsequent target working condition identification.
Step 202: target operating condition transition threshold generation
If there are l kinds of working conditions in a complete period, there must be l kinds of data distribution types in the data in the working condition sensitive remote reference sequence. For satellite remote parameter data, the data distribution basically presents the characteristic of normal distribution.
For this purpose, the j operating condition sensitive remote reference sequence Z calibrated by the elapsed timej′For example, the condition is sensitive by fitting the above-mentioned I kinds of condition data with a plurality of normal distribution functionsThe remote sensing parameter data can be fitted with I groups of normal distribution parameters, and the distribution parameters are arranged in ascending order according to the mean value to obtain a working condition sensitive remote sensing parameter sequence Zj′Set of parameters of, note
Wherein
Respectively is the mean value and the variance of the ith working condition distribution.
And constructing the correlation health baseline required data in the stable period of the target working condition, identifying the working condition sensitive remote parameter data in the target working condition distribution, and marking the time interval of the target working condition duration. The remote parameters in the rest of the working condition distributions and among the working condition distributions are discarded.
In contrast, a 3 sigma statistical principle is adopted to generate a working condition segmented threshold interval, and the ith working condition distribution is taken as an example to obtain upper and lower boundary values of stable working condition distribution
Similarly condition-sensitive remote reference sequence Zj′The upper and lower boundary values of the group l can be obtained. Combining the upper and lower boundaries distributed adjacently in pairs to obtain l-1 groups of working condition conversion threshold value intervals
Is recorded as:
assuming that the required target working condition serial number is i, two adjacent threshold value conversion intervals of the working condition i
Two working condition switching point threshold values are obtained
Wherein
Similarly, target working condition conversion threshold value generation is carried out on all the N working condition sensitive remote reference sequences, and a target working condition conversion threshold value set can be obtained and recorded as:
step 203: target working condition time interval discrimination generation based on evidence logical combination
AND logically combining the target working condition conversion threshold value set obtained in the step 202 through an AND/OR logical connection symbol AND a >/< logical comparison symbol to form a multi-evidence logical combination criterion so as to complete the identification of the remote reference data of the target working condition time interval. It follows the following operational flow.
Inputting a uniform time tag sequence of N time calibration working condition sensitive remote parameter sequences
N time calibration condition sensitive remote reference sequence Z1′,...,ZN′Target operating condition conversion threshold value set ZthresholdThe initial target working condition state bit Label is 0, and the initial target working condition time period element number i is 1.
The method comprises the steps that the Label (0) represents that data of a previous-time working condition sensitive remote parameter sequence do not belong to a target working condition section, the Label (1) represents that the data of the previous-time working condition sensitive remote parameter sequence belong to the target working condition section, and the i (1) represents that the number of currently identified target working condition time periods is 1.
The process is as follows:
and (3) outputting:
wherein, ttargetIs a set of all the target working condition time periods obtained through discrimination,
the time starting point and the time end point of the ith section of target working condition.
Step 204: target working condition time interval binary remote reference sequence data extraction
After time calibration, the binary remote reference sequences X 'and Y' and the working condition sensitive remote reference sequence Z1′,...,ZN′The time labels between the two are strictly uniform. Thus, the set of target operating condition periods t obtained in step 303 may be utilizedtargetFor indexing, cutting out a target working condition time interval binary remote reference sequence under a working condition time interval from the complete binary remote reference sequences X 'and Y':
step three, construction of binary relevance health baseline
Obtaining a binary remote reference sequence X 'under the target working condition period through the second step'target、Y′target。
In a normal state, a stable mathematical relation Y 'exists in the binary remote reference sequence of the target working condition period'target＝f(X′target) I.e., a binary associative healthy baseline. When the corresponding equipment of the satellite is abnormal, the association rule of the binary remote reference sequence is changed and shows that the satellite is under the target working conditionSegment binary remote reference sequence X'target、Y′targetNo longer satisfies the normal baseline relationship Y'target＝f(X′target) Therefore, the abnormal detection of the health state of the satellite is realized.
According to the difference of a baseline mathematical form f () between target working condition binary remote reference sequences, two basic types of the definite binary relevance health baseline are as follows: a binary linear correlation health baseline, a binary non-linear correlation health baseline. At this stage, a linear or nonlinear model is selected according to the prior judgment of the correlation relation of the binary remote reference sequence, and the construction of a subsequent correlation health baseline model is supported.
Step 301: binary linear correlation health baseline construction
For a binary linear correlation health baseline, binary remote reference sequence data X 'under a normal state target working condition'target、Y′targetAnd an analytical equation is constructed as follows:
Y′target＝f(X′target)＝a+b×X′target
at this time, data X 'is subjected to least square method'target、Y′targetPerforming iterative fitting to obtain parameters of the binary linear correlation healthy baseline model, and marking the parameters as [ a, b ]]。
Step 302: binary nonlinear correlation health baseline construction
For a binary nonlinear correlation healthy baseline, the nonlinear relationship between the satellite remote parameter data is generally represented as a piecewise time integral type, and the analytic equation form is as follows:
wherein the content of the first and second substances,
is the ith target working condition period
Remote control of timeThe data of the data is transmitted to the data receiver,
the remote parameter data of the starting moment in the ith target working condition period,
is the adjacent time in the ith target working condition period
The time interval of (c).
Therefore, before constructing the binary non-linear management-related health baseline, the non-linear remote parameter data needs to be converted to obtain the data in the form described above. A non-linear remote parameter conversion method based on discrete integration is adopted, and the steps are as follows:
step 302-1: time tag acquisition
Extracting the target working condition time interval set obtained in the step 203
Step 302-2: time difference processing
Taking the ith target working condition time interval as an example, the point-by-point difference processing is carried out on the time values of the adjacent remote parameter data in the time interval range
Obtaining a sequence of time intervals
Repeating the above operations for other target working condition time periods to obtain a target working condition time interval set:
step 302-3: discrete integral conversion
Utilizing the target operating condition time interval obtained in step 302-2Set Δ ttargetIs remotely referred to a sequence X 'from an independent variable of a target working condition'targetCarrying out point-by-point multiplication, taking the ith target working condition time interval as an example:
On the basis, the increment value of the unit time interval is accumulated point by point
Obtaining the accumulated value of unit time interval
Repeating the operation for other target working condition periods to obtain an independent variable remote parameter discrete integral sequence sigma:
therefore, the conversion from the original remote parameter data to the discrete integral data is completed, and the original nonlinear relation fitting is simplified into linear relation fitting.
Step 302-4: fitting of baseline parameters
Through the steps, the mathematical relationship between the original binary remote reference sequences is converted into a linear form, taking the ith target working condition section as an example:
at this time, data σ and Y 'are processed by the least square method'targetPerforming iterative fitting to obtain binary nonlinear correlationParameters of the healthy baseline model are denoted as [ a, b ]]。
Step four, testing the goodness of fit
After the binary linear/nonlinear correlation health baseline is constructed in the third step, the fitting effect of the health baseline needs to be quantitatively evaluated, and the goodness of fit R is utilized for the quantitative evaluation2As a metric, the calculation formula is as follows:
SST is the sum of the total squares, and SSR is the sum of the regression squares; y isiThe actual data is remotely referred to by the dependent variable,
the data is remotely estimated for the dependent variable,
is the dependent variable mean value; r is2As a goodness of fit indicator, R2Closer to 1 indicates better fit of the correlation health baseline. Presetting R2Threshold value, if the current baseline R2Within the threshold, the currently constructed associative healthy baseline model is judged to be valid.
Step five, generating the correlation health baseline family
Meanwhile, in order to more robustly represent the relevance between satellite telemetering, the uncertain interference caused by factors such as data noise is eliminated. The interval coefficient of the related healthy baseline can be obtained by a parameter interval estimation method in statistics
And a corresponding family of associative healthy baselines is obtained.
For a binary linear correlation healthy baseline, its family of healthy baselines is characterized as follows:
Y′target＝a+b×X′target
for a binary linear correlation healthy baseline, its family of healthy baselines is characterized as follows:
Y′target＝a+b×σ
thus, the correlation healthy baseline family coefficients
And the healthy baseline coefficients [ a, b ] obtained in the third step]The correlation health characterization of the binary remote reference sequence is completed together, so that the abnormal and noise interference in the telemetry data can be overcome, and the satellite health characterization result has stronger robustness.
The method of the present application will be described below with reference to a satellite power subsystem as an example.
Example I description of a Linear correlation health Baseline construction Process
Step one, time calibration pretreatment
Selecting a group of binary remote reference sequence combinations in linear correlation, wherein the remote reference sequence of 'storage battery 1-9 voltage' is recorded as X, and the remote reference sequence of 'storage battery voltage' is recorded as Y. Selecting the combination of the working condition sensitive remote parameter sequences, and recording the remote parameter sequence of 'charging current of the storage battery' as Z1Recording the 'discharge current of the storage battery' as a remote reference sequence Z2。
The original remote reference partial sequence without time scaling is shown in fig. 2.
As can be seen from fig. 2, the 4 original remote reference sequences without time scaling have different sampling frequencies and sampling moments.
By the difference processing method described in step 101, m (D) can be obtained1)＝1，m(D2)＝1，m(D3)＝5，m(D4)＝2，m(D3)＞m(D4)＞m(D2)＝m(D1) Therefore, the charging current of the storage battery is selected as a reference remote reference sequence, and the remote reference sequence is subjected to time calibration processing by the method described in steps 102 and 103. The remote reference sequence after the time scaling process is shown in fig. 3.
As can be seen from FIG. 3, after time calibration processing, the selected binary-associated remote reference sequence X, Y (storage battery voltage 1-9, storage battery voltage) and the working condition-sensitive remote reference sequence Z are selected1、Z2(storage battery charging current, storage battery discharging current) and realizes data point one-to-one strict matching.
And step two, identifying and cutting the target working condition.
For the working condition sensitive remote reference sequence and the binary remote reference sequence in this case, the two working conditions are included: battery charge state and battery discharge state.
Distribution histograms of the battery charging current and the battery discharging current are respectively drawn, and the effect display is shown in fig. 4.
As shown in FIG. 4, the distribution histograms of the above-mentioned condition-sensitive remote reference sequences all show a distinct mixed normal distribution state. According to the method for generating the target operating condition transition threshold value in step 202, the operating condition transition threshold value set Z of the embodiment can be obtainedthreshold＝{1.3,4.0}。
On the basis, the charging state of the storage battery is set as the target working condition of the embodiment, and the criterion rule of the target working condition of the embodiment is set by using the target working condition time interval discrimination and generation method based on the evidence logical combination in step 203
Wherein
The remote reference sequence data respectively represent the charging current and the discharging current of the storage battery.
According to the method in step 203, the remote reference sequence data sensitive to the working conditions of the embodiment are identified together, and 8 sections of target working conditions are screened out. The working condition recognition effect is shown in fig. 5.
After the target working condition time period set is obtained, the binary remote reference sequence data of the target working condition time period are cut and extracted according to the method in the step 204, so that the construction of a healthy baseline under the subsequent charging working condition is supported. The cutting effect of the binary remote reference sequence data of the target working condition time interval is shown in FIG. 6.
And step three, constructing a correlation health baseline model. By observing the binary remote reference sequence curve of the target working condition time period, the voltage of the storage battery 1-9/the voltage of the storage battery presents a typical linear relation. Therefore, binary linear correlation healthy baseline construction is performed according to the method described in step 301, and the model coefficients are as follows:
| serial number | Binary remote control | Target operating conditions | Base |
| 1 | 1-9 voltage of accumulator/voltage of accumulator | Charging of storage batteries | [3.86,3.72] |
And step four, testing the goodness of fit. And setting the goodness-of-fit valid interval to be [0.6,1.4] according to the method in the fourth step. And (3) carrying out goodness-of-fit test on the binary linear relevance health baseline constructed in the step three, wherein the test result is as follows:
R2＝1.01
the goodness-of-fit is tested to be within a threshold value, so the binary relevance healthy baseline model is effective.
And step five, constructing a correlation health baseline family. And setting the interval estimation confidence coefficient to be 0.95 according to the parameter interval estimation method in the fifth step to obtain an interval coefficient of the healthy baseline family, and taking the interval coefficient and the model coefficient obtained in the third step as the healthy representation of the binary remote parameter sequence.
The interval coefficients for the healthy baseline family are shown below:
in this embodiment, the binary linear correlation healthy baseline family curve and the local detail picture are shown in fig. 7.
Example two, non-Linear correlation health Baseline construction flow description
Step one, time calibration processing. And selecting a group of binary remote reference sequence combinations in nonlinear association, wherein the remote reference sequence of 'charging current of the storage battery' is recorded as X, and the remote reference sequence of 'capacity of the storage battery' is recorded as Y. Selecting the combination of the remote reference sequences sensitive to the working conditions, and recording the remote reference sequence of 'charging current of the storage battery' as Z1Recording 'discharge current of storage battery' as a remote reference sequence Z2。
The original remote reference partial sequence without time scaling is shown in fig. 8.
As can be seen from fig. 8, the 3 original remote reference sequences without time scaling have different sampling frequencies and sampling moments.
By the difference processing method described in step 101, m (D) can be obtained1)＝5，m(D2)＝1，m(D3)＝2，m(D1)＞m(D3)＞m(D2) Therefore, the charging current of the storage battery is selected as a reference remote reference sequence, and the remote reference sequence is subjected to time calibration processing by the method described in steps 102 and 103. The remote reference sequence after the time scaling process is shown in fig. 9.
As can be seen from fig. 9, after time calibration, the binary remote reference sequence X, Y (battery charging current, battery capacity) and the operating-condition-sensitive remote reference sequence Z are selected1、Z2(storage battery charging current, storage battery discharging current), and data points are strictly matched one by one.
Step two, target working condition recognition and cutting
Similarly, the operation condition sensitive remote reference sequence and the binary remote reference sequence in this embodiment also include two operation conditions: battery state of charge and battery state of discharge.
And selecting the storage battery charging as a target working condition, and generating a working condition conversion threshold according to the method in the steps 201 and 203. Since the target condition is the same as the first embodiment, it is not described herein. The working condition recognition effect is shown in fig. 10.
After the target working condition time period set is obtained, the binary remote reference sequence data of the target working condition time period are cut and extracted according to the method in the step 204, so that the construction of a healthy baseline under the subsequent charging working condition is supported. The cutting effect of the binary remote reference sequence data of the target working condition time interval is shown in FIG. 11.
And step three, constructing a correlation health baseline model. Through observation of a target working condition binary remote reference curve, the charging current of the storage battery and the capacity of the storage battery present a typical nonlinear integral relation. The healthy baseline is constructed according to the method described in step 302, and the coefficients of the healthy baseline construction model are as follows:
| serial number | Binary remote control | Target operating conditions | Base |
| 1 | Battery charging current-battery capacity | Charging condition | [-0.54,0.0003] |
And step four, testing the goodness of fit. And D, performing goodness-of-fit test on the constructed healthy base line according to the method in the step four, wherein the test result is as follows:
R2＝0.68
through inspection, the goodness of fit is within a threshold value, and the healthy baseline model is effective.
And step five, constructing a healthy baseline family. And setting the confidence coefficient to be 0.95 according to the parameter interval estimation algorithm in the step five, and taking the interval coefficient of the obtained healthy baseline and the model coefficient as the healthy representation of the binary remote reference sequence.
The interval coefficients for the healthy baseline family are shown below:
the healthy baseline family curve is shown in fig. 12.
The key sub-system based on the historical telemetering data and the single-machine relevance health baseline construction method provided by the invention adopt unsupervised autonomous learning construction to obtain the satellite relevance health baseline family, and effectively complete the joint characterization of the satellite health state. The noise and outlier interference of actual data are overcome through the constructed binary relevance healthy baseline family.
Unless defined otherwise, all technical and/or scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The materials, methods, and examples set forth in this application are illustrative only and not intended to be limiting.
Although the present invention has been described in conjunction with specific embodiments thereof, it is evident that many alternatives, modifications, and variations will be apparent to those skilled in the art in light of the teachings of this application and yet remain within the scope of this application.
Claims (8)
1. A key subsystem based on historical telemetering data and a single-machine correlation health baseline construction method comprise the following steps:
selecting a group of binary remote reference sequence combinations in linear correlation or nonlinear correlation;
firstly, time calibration is carried out on the selected binary remote reference sequence combination, and data points of the binary remote reference sequence combination are matched one by one;
secondly, performing target working condition identification and cutting on the binary remote reference sequence subjected to time calibration to obtain a binary remote reference sequence under a target working condition period, wherein a stable mathematical relation existing between the binary remote reference sequences under the target working condition period is a binary relevance health baseline of the binary remote reference sequence under the target working condition period;
thirdly, if the binary correlation health baseline is in a linear relation, fitting an analytic equation of the linear relation by using the binary remote reference sequence data of the target working condition in the normal state to obtain parameters of a binary linear correlation health baseline model, and obtaining the binary linear correlation health baseline; if the binary correlation health baseline is in a nonlinear relation, converting the nonlinear relation into a linear relation by adopting a discrete integral-based method, and fitting to obtain parameters of a binary nonlinear correlation health baseline model to obtain the binary nonlinear correlation health baseline; the binary linear correlation health baseline and the binary nonlinear correlation health baseline together form a correlation health baseline library so as to realize quantitative stable representation of key subsystems and single machines;
fourthly, checking the construction effect of the correlation health baseline; respectively calculating the regression square sum of the dependent variable remote reference real data and the regression square sum of the dependent variable remote reference baseline estimation data, wherein the ratio of the two is used as a test index of the correlation health baseline construction effect, and the closer the index is, the better the correlation health baseline fitting effect is shown to be; presetting a detection index threshold, and if the current baseline detection index value is within the threshold, judging that the correlation healthy baseline model is valid;
and fifthly, obtaining an interval coefficient of the correlation health baseline by a parameter interval estimation method in statistics, thereby obtaining a corresponding correlation health baseline family.
2. The key subsystem and standalone association health baseline construction method based on historical telemetry data of claim 1, wherein:
in the first step, the time calibration is performed as follows:
determining reference remote reference, wherein in the time calibration pretreatment, one remote reference sequence in the binary remote reference sequences is selected as a reference remote reference sequence, and the other one in the binary remote reference sequences is a remote reference sequence to be calibrated;
the method comprises the steps of reference comparison searching, wherein time labels of all points of a reference remote reference sequence are taken to form a reference time axis, each time point on the reference time axis is used for searching in a relevant time range for the remote reference sequence to be calibrated, if a remote parameter data point exists in the relevant time range, the time point is an effective time point, and data corresponding to the time point of the remote reference sequence to be calibrated are effective data points;
and aligning the remote reference sequences, performing reference comparison search on the reference remote reference sequence and the remote reference sequence to be calibrated at all effective time points, wherein the remote parameter data points of the reference remote reference sequence at each effective time point form the reference remote reference sequence after time calibration, and the remote parameter data points of the remote reference sequence to be calibrated at each effective time point form the remote reference sequence to be calibrated after time calibration, namely realizing one-to-one matching of data points of the binary remote reference sequence combination.
3. The historical telemetry data-based key subsystem and stand-alone correlation health baseline construction method of claim 2, wherein:
in the second step, when the target working condition is identified and cut, the method comprises the following steps:
carrying out time calibration on the working condition sensitive remote reference sequence;
generating a target working condition conversion threshold;
judging and generating a target working condition time period based on the evidence logic combination;
and extracting binary remote reference sequence data in the target working condition time interval.
4. The historical telemetry data-based key subsystem and stand-alone correlation health baseline construction method of claim 3, wherein:
in the time calibration of the working condition sensitive remote reference sequence,
the working condition sensitive remote reference sequence is determined by a plurality of remote reference sequences which determine various correlation relations among the binary remote reference sequences in a complete period; the plurality of remote reference sequences are working condition sensitive remote reference sequences;
and time calibration is carried out on the working condition sensitive remote reference sequence by utilizing the reference time axis.
5. The historical telemetry data-based key subsystem and stand-alone correlation health baseline construction method of claim 4, wherein:
in the generation of the target operating condition changeover threshold value,
for each sequence in the working condition sensitive remote reference sequences, fitting a plurality of working conditions by utilizing a plurality of normal distributions; for each working condition, generating a working condition segmented threshold interval by adopting a 3 sigma statistical principle to obtain upper and lower boundary values of stable working condition distribution of the working condition; combining the upper and lower boundaries which are distributed adjacently in pairs to obtain a threshold interval of the conversion working condition; and the threshold value intervals of all the conversion working conditions form a target working condition conversion threshold value set.
6. The historical telemetry data-based key subsystem and stand-alone correlation health baseline construction method of claim 5, wherein:
in the discrimination generation of the target working condition period based on the evidence logical combination,
performing logical combination on the target working condition conversion threshold value set through a logical connection symbol and a logical comparison symbol to form a multi-evidence logical combination criterion;
and obtaining a set of all target working condition periods by using the reference time axis, the time-calibrated working condition sensitive remote reference sequence and the target working condition switching threshold value set.
7. The historical telemetry data-based key subsystem and stand-alone correlation health baseline construction method of claim 6, wherein:
in the extraction of the binary remote reference sequence data in the target working condition time interval,
and cutting the binary remote reference sequence which belongs to the target working condition time period under each working condition time period for the binary remote reference sequence calibrated by time by taking the set of all the target working condition time periods as indexes.
8. The historical telemetry data-based key subsystem and stand-alone correlation health baseline construction method of claim 7, wherein:
in the third step, when a discrete integration based method is used,
extracting a time tag by using the target working condition time interval set, and carrying out point-by-point difference processing on the time values of adjacent remote parameter data points in the time range of each target working condition time interval to obtain a target working condition time interval set; for each target working condition time interval, multiplying the independent variable remote parameter sequence in the target working condition time interval set and the binary remote parameter sequence in the target working condition time interval point by point to obtain each unit time interval increment value in the target working condition time interval, and accumulating the unit time interval increment values to obtain a unit time interval accumulated value of the target working condition time interval; and the accumulated values of unit time intervals of all the target working condition periods form an independent variable remote parameter discrete integral sequence, so that the nonlinear relation fitting between remote parameter sequence combinations is simplified into linear relation fitting.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011404591.0A CN112464146B (en) | 2020-12-03 | 2020-12-03 | Key subsystem based on historical telemetering data and single-machine correlation health baseline construction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011404591.0A CN112464146B (en) | 2020-12-03 | 2020-12-03 | Key subsystem based on historical telemetering data and single-machine correlation health baseline construction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112464146A CN112464146A (en) | 2021-03-09 |
| CN112464146B true CN112464146B (en) | 2022-06-07 |
Family
ID=74804835
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011404591.0A Active CN112464146B (en) | 2020-12-03 | 2020-12-03 | Key subsystem based on historical telemetering data and single-machine correlation health baseline construction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112464146B (en) |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102270271B (en) * | 2011-05-03 | 2014-03-19 | 北京中瑞泰科技有限公司 | Equipment failure early warning and optimizing method and system based on similarity curve |
| CN103646167B (en) * | 2013-11-22 | 2017-03-15 | 北京空间飞行器总体设计部 | A kind of satellite abnormal state detection system based on telemetry |
| CN103679378B (en) * | 2013-12-20 | 2016-08-31 | 北京航天测控技术有限公司 | Method and device based on telemetry assessment heath state of spacecraft |
| CN104297699B (en) * | 2014-09-26 | 2017-03-15 | 航天东方红卫星有限公司 | A kind of satellite power supply health status tracking and detection method based on intelligent interpretation |
| CN110008565B (en) * | 2019-03-28 | 2020-12-11 | 浙江大学 | Industrial process abnormal working condition prediction method based on operation parameter correlation analysis |
-
2020
- 2020-12-03 CN CN202011404591.0A patent/CN112464146B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112464146A (en) | 2021-03-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111459700B (en) | Equipment fault diagnosis method, diagnosis device, diagnosis equipment and storage medium | |
| Shi et al. | Rolling bearing initial fault detection using long short-term memory recurrent network | |
| CN110222991B (en) | Metering device fault diagnosis method based on RF-GBDT | |
| CN111382542A (en) | Road electromechanical equipment life prediction system oriented to full life cycle | |
| CN111611294B (en) | Star sensor data anomaly detection method | |
| GB2476246A (en) | Diagnosing an operation mode of a machine | |
| CN114167838B (en) | Multi-scale health assessment and fault prediction method for servo system | |
| CN111122162B (en) | Industrial system fault detection method based on Euclidean distance multi-scale fuzzy sample entropy | |
| CN117196066A (en) | Intelligent operation and maintenance information analysis model | |
| CN114004137A (en) | Multi-source meteorological data fusion and pretreatment method | |
| CN114545234A (en) | Method for monitoring battery state based on battery temperature gradient change | |
| CN114118219A (en) | Data-driven real-time abnormal detection method for health state of long-term power-on equipment | |
| CN115600928A (en) | OAO-RVM-based GIS basin-type insulator state evaluation method | |
| CN117368651B (en) | Comprehensive analysis system and method for faults of power distribution network | |
| CN112215307B (en) | Method for automatically detecting signal abnormality of earthquake instrument by machine learning | |
| CN104317778A (en) | Massive monitoring data based substation equipment fault diagnosis method | |
| CN116660761B (en) | Lithium ion battery detection method and system | |
| Wang et al. | Large-scale field data-based battery aging prediction driven by statistical features and machine learning | |
| CN112016193B (en) | Online prediction method and system for lubrication failure of shield tunneling machine system | |
| CN112464146B (en) | Key subsystem based on historical telemetering data and single-machine correlation health baseline construction method | |
| CN117630797A (en) | Ammeter health state detection method, system and storage medium based on working current | |
| CN112345972A (en) | Power failure event-based power distribution network line transformation relation abnormity diagnosis method, device and system | |
| CN112329868A (en) | CLARA clustering-based manufacturing and processing equipment group energy efficiency state evaluation method | |
| CN114167837B (en) | Intelligent fault diagnosis method and system for railway signal system | |
| CN115659271A (en) | Sensor abnormality detection method, model training method, system, device, and 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 |