Spacecraft mass information cleaning method based on multi-feature extraction

Abstract

The invention relates to a spacecraft mass information cleaning method based on multi-feature extraction, and the method comprises the steps: carrying out the feature extraction of typical time sequence feature information of different telemetry data distributions, and carrying out the discrimination and screening of outlier data through employing a perfect expert knowledge interpretation method established through employing various indexes. Therefore, massive telemetry data cleaning and effective telemetry data extraction are realized, the problems of misjudgment and missed judgment caused by uneven spacecraft on-orbit data quality in telemetry interpretation are remarkably solved, and powerful support is provided for spacecraft data rear-end processing and subsequent on-orbit decision making.

Description

Technical Field

[0001] The present invention belongs to the field of control engineering and relates to a method for cleaning a large amount of information of a spacecraft based on multi-feature extraction. Background Technique

[0002] During the process of on-orbit telemetry data transmission, storage, and processing of a spacecraft, due to reasons such as attenuation of signals between the ground and space, abnormal acquisition of equipment, and interference from complex environments, the real-time telemetry data obtained usually contains some invalid data with characteristics of occasional anomalies or discrete jumps, which is called wild value anomalies. Although it accounts for a small proportion in the large amount of data, during daily on-orbit monitoring and mission execution, such wild value anomaly data will directly affect the subsequent data analysis effect, seriously hinder the judgment of the health status of the spacecraft by professionals, significantly reduce the accuracy of fault identification and handling, and easily cause problems such as false alarms or missed alarms. With the current increase in on-orbit missions, in the face of the challenge of a large amount of data, improving the quality of spacecraft telemetry data through data screening and cleaning to provide support for in-depth data analysis and subsequent decision-making has become an urgent topic that cannot be ignored.

[0003] For the wild value screening and data cleaning of spacecraft telemetry, the current main methods include manual screening method and upper and lower limit threshold screening method. The manual recognition method has a large amount of operation and low efficiency, and it is difficult to meet the real-time processing of large-scale data and the interpretation requirements of a large number of on-orbit missions; the upper and lower limit threshold method has high customization requirements for parameters and is difficult to process typical invalid wild values within the upper and lower limits of periodic signals. In view of the inherent time series characteristics of telemetry signals, many studies have carried out machine automatic processing of telemetry wild values to improve the processing efficiency and accuracy. However, these methods have poor versatility and are difficult to balance the diversity of inputs, the rapidity of algorithms, and the accuracy of models, and their engineering application value is limited. Patent No. CN201510860785.4 discloses a method for preprocessing the elimination of wild values in satellite telemetry data, which uses parity bits and frame count continuity to judge whether a telemetry frame contains wild value anomalies. The method of eliminating wild values for the entire frame is likely to cause the loss of key information and is powerless for wild values that do not affect parity bits; Patent No. CN201510860785.4 discloses a method for selecting valid data of an on-orbit spacecraft, which eliminates wild values from telemetry data based on membership function modeling and credibility threshold setting and fills in data. It relies on expert prior knowledge, has a high algorithm complexity, is complex to implement and debug, and is difficult to be widely applied on a large scale. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and propose a method for cleaning a large amount of information of a spacecraft based on multi-feature extraction. According to the different data characteristics of each telemetry variable, the telemetry variables are divided into subsets, and a multi-class data validity judgment method is used to construct expert judgment rules for each telemetry variable in the subset one by one, so as to quickly realize the screening and cleaning covering all data types, and be able to perform overall elimination processing on the telemetry data with multiple wild value anomalies occurring simultaneously at a single moment, thus greatly reducing the workload of manual data screening and significantly improving the quality of telemetry data.

[0005] The technical solution of the present invention is: A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction, including:

[0006] Based on the telemetry characteristics of the telemetry variables, the spacecraft telemetry variables are divided into sets to obtain multiple telemetry variable subsets;

[0007] After initializing the variables, for different telemetry variable subsets, extract the telemetry characteristics of each spacecraft telemetry variable in the subset;

[0008] Based on the obtained telemetry characteristics, establish independent expert judgment rules; for the telemetry data at a certain moment, use the expert judgment rules to determine whether the telemetry values corresponding to each telemetry variable are wild value anomalies, and the telemetry values that meet the rules are retained, and the telemetry values that do not meet the rules are marked as wild value anomalies;

[0009] Eliminate the telemetry values marked as wild value anomalies, and complete the traversal processing of the telemetry data containing continuous time periods to realize the cleaning task of the large amount of telemetry data.

[0010] The multiple telemetry variable subsets include the digital quantum subset Θ P , the steady-state analog quantum subset Θ Q , the periodic analog quantum subset Θ R , the aperiodic analog quantum subset Θ S .

[0011] The following criteria are used to divide the telemetry variables into four subsets:

[0012] Telemetry variables θ p1 , θ p2 , …, θ pa whose telemetry values show the characteristics of a finite number of integer enumeration values are included in the digital quantum subset Θ P , where a is the number of elements in Θ P ;

[0013] Telemetry variables θ q1 , θ q2 , …, θ qb whose telemetry values show the characteristics of floating-point data values with small fluctuations in orbit under various working conditions or environments are included in the steady-state analog quantum subset ΘQ inside, where b is Θ Q the number of elements inside;

[0014] The telemetry variable θ whose telemetry value exhibits the characteristic of a floating-point data value changing in a fixed period r1 , θ r2 , …, θ rc Incorporate the periodic simulation quantum set Θ R inside, where c is Θ R the number of elements inside;

[0015] The telemetry variable θ whose telemetry value exhibits the characteristic of a floating-point data value randomly fluctuating within the effective range or changing with working conditions and environment, etc. s1 , θ s2 , …, θ sd Incorporate the aperiodic simulation quantum set Θ S inside, where d is Θ S the number of elements inside.

[0016] For the subsets of different telemetry variables mentioned above, extract the telemetry characteristics of each spacecraft's telemetry variable in the subset, including:

[0017] For each telemetry variable in the digital quantum set Θ P determine the finite enumeration set of the corresponding telemetry value;

[0018] For each telemetry variable in the steady-state simulation quantum set Θ Q determine the time-series statistical characteristic parameters of the corresponding telemetry value, including mean, variance, median, and median absolute deviation;

[0019] For each telemetry variable in the periodic simulation quantum set Θ R determine the main frequency signal period, signal amplitude, phase, and error bound after fast Fourier transform;

[0020] For each telemetry variable in the aperiodic simulation quantum set Θ S determine the upper and lower limit ranges of the corresponding values;

[0021] Extract the telemetry characteristics of the telemetry variable parameters determined for each of the above subsets.

[0022] Initialize the variables mentioned above. The initialization content includes the start time t init for data wild value elimination processing, the time interval Δt between adjacent moments, and the end time t end , and the time variable t = t init , the counter cnt = 0, and the counter threshold N TH .

[0023] Adopt the following criteria to extract the telemetry characteristics of the telemetry variable parameters determined for each subset:

[0024] Digital quantum set Θ P Internal telemetry variable, determining a finite enumerated set of corresponding telemetry values: determining the i-th telemetry variable θ pi ∈Θ P , i = 1, …, a corresponding telemetry value φ pi Finite enumerated set Λ pi , that is, set Λ pi Is the smallest set containing φ pi With normal enumerated values;

[0025] Steady-state simulation quantum set Θ Q Internal telemetry variable, determining the mean, variance, median, and median absolute deviation of the corresponding telemetry values: for the j-th telemetry variable θ qj ∈Θ Q , j = 1, …, b, whose telemetry value at time t = T is Given a set of continuous time series Define the operator Median(·) as the median calculation operator, then the mean μ qj , standard deviation σ qj , median absolute deviation MAD(φ qj ) are defined as:

[0026]

[0027] MAD(φ qj ) = Median(|φ qj - Median(φ qj )|);

[0028] Periodic simulation quantum set Θ R For each telemetry variable in it, determine the main frequency signal period, signal amplitude, phase, and error bound after fast Fourier transform: for the k-th telemetry variable θ rk ∈Θ R , k = 1, …, c, construct the m-order Fourier reconstruction function of the telemetry value φ rk And determine the parameters And determine the parameters B rk And error bound Th rk , where Are the m-order signal frequency, amplitude, and phase respectively, B rk Is a constant offset, and the order m is not greater than 5;

[0029] Aperiodic simulation quantum set Θ S For each telemetry variable in it, determine the upper and lower limit ranges of the corresponding values: determine the l-th telemetry variable θ sl ∈Θ R, for l = 1, …, d corresponding to the telemetry value φ sl upper limit value and lower limit value

[0030] Establish independent expert interpretation rules based on the obtained telemetry features; for the telemetry data at a certain moment, use the expert interpretation rules to judge whether the telemetry variables corresponding to each telemetry value are wild value anomalies. The telemetry values that conform to the rules are retained, and the telemetry values that do not conform to the rules are marked as wild value anomalies, where:

[0031] For the digital quantity parameter set, use a finite enumeration set to judge the validity and mark wild values;

[0032] For the steady-state analog quantity parameter set, use the Shewhart principle and the median absolute deviation index to judge the validity and mark wild values;

[0033] For the periodic analog quantity set, use Fourier transform to reconstruct the signal and then judge the validity and mark wild values;

[0034] For the aperiodic analog quantity set, use the effective upper and lower bound ranges to judge the validity and mark wild values.

[0035] For the digital quantity parameter set, using a finite enumeration set to judge the validity and mark wild values includes:

[0036] Traverse all telemetry variables θ P in the digital quantity subset Θ pi ∈Θ P , i = 1, …, a, establish expert rules and judge the data validity: in the telemetry data at the current moment t, if the telemetry value pi corresponding to the telemetry variable θ does not belong to the finite enumeration set Λ pi , that is then mark it as a wild value anomaly, indicating that the telemetry value of this telemetry variable has a wild value anomaly, and at the same time the counter cnt = cnt + 1.

[0037] For the steady-state analog quantity parameter set, using the Shewhart principle and the median absolute deviation index to judge the validity and mark wild values includes:

[0038] Traverse all telemetry variables θ Q in the steady-state analog quantity subset Θ qj ∈Θ Q , j = 1, …, b, use the Shewhart principle and the median absolute deviation index to establish expert rules and judge the data validity: in the telemetry data at the current moment t, if the telemetry value qj corresponding to the telemetry variable θ simultaneously satisfies the following two conditions:

[0039]

[0040] Then, perform wild value anomaly annotation to indicate that a wild value anomaly has occurred in the telemetry value of the telemetry variable, and at the same time, the counter cnt = cnt + 1.

[0041] For the set of periodic analog quantities, after signal reconstruction using Fourier transform, judge the validity and annotate wild values, including:

[0042] Traverse all telemetry variables θ R in the periodic analog quantity subset Θ rk ∈Θ R , k = 1,..., c, establish an expert rule based on the Fourier reconstruction method and perform data validity judgment: in the telemetry data at the current moment t, if the telemetry value rk corresponding to the telemetry variable θ exceeds the error bound Th of the expected value at the current moment rk outside the range, that is Then, perform wild value anomaly annotation to indicate that a wild value anomaly has occurred in the telemetry value of the telemetry variable, and at the same time, the counter cnt = cnt + 1.

[0043] For the set of aperiodic analog quantities, judge the validity and annotate wild values using the effective upper and lower bound ranges, including:

[0044] Traverse all telemetry variables θ S in the aperiodic analog quantity subset Θ sl ∈Θ S , l = 1,..., d, establish an expert rule and perform data validity judgment: in the telemetry data at the current moment t, if the telemetry value sl corresponding to the telemetry variable θ is not within the upper and lower limit ranges, that is Then, perform wild value anomaly annotation to indicate that a wild value anomaly has occurred in the telemetry value of the telemetry variable, and at the same time, the counter cnt = cnt + 1.

[0045] Eliminate the telemetry values marked as wild value anomalies, complete the traversal processing of the telemetry data containing continuous time periods, and implement the cleaning task of massive telemetry data, including: eliminating the telemetry values marked as wild value anomalies. If the number of wild values at the current moment exceeds the set threshold N TH , that is, the counter cnt > N TH , then eliminate all the telemetry values in the current telemetry data to complete the cleaning of the telemetry data at the current moment; traverse the telemetry data containing multiple consecutive frames to implement the cleaning task of massive telemetry data.

[0046] The advantages of the present invention over the prior art are:

[0047] (1) By partitioning the subsets of telemetry data and establishing item-by-item corresponding expert knowledge interpretation rules based on the time-series characteristics of various types of telemetry data, the present invention can fully adapt to the interpretation requirements of different telemetries under different working conditions, and make up for the problems of insufficient flexibility of a single interpretation method, insufficient utilization of data characteristics, and limited functional performance;

[0048] (2) For typical telemetry variables with steady-state time-series characteristics, the present invention creatively combines the Shewhart principle with the method of median absolute deviation to establish an expert interpretation rule, effectively identify and quickly eliminate outlier data. By combining data distribution characteristics and discrete degree information, it can greatly avoid data misjudgment caused by short-term abnormal telemetry, and significantly improve the robustness of the data cleaning method;

[0049] (3) For typical telemetry variables with periodic time-series characteristics, the present invention pioneeringly adopts the Fourier reconstruction method to establish an expert interpretation rule. By comparing the consistency between the actual telemetry value and the expected telemetry value, it can not only detect conventional outliers, but also identify typical outliers within the signal amplitude range, effectively solving the problem of difficultly processing outliers within the threshold for telemetry data with periodic characteristics, and enhancing the intelligence and automation of interpretation;

[0050] (4) The interpretation rules of the present invention are trained quickly, the outlier anomalies are accurately identified, and the outlier removal process is efficient and precise. It can be widely applied to the processing scenarios of massive information of spacecraft, maximally eliminate the misjudgment and missed judgment of spacecraft states caused by in-orbit outlier anomalies, significantly improve the data quality, and provide strong support for the subsequent data processing and deep mining. Description of the Drawings

[0051] Figure 1 is the flow chart of the method of the present invention. Detailed Embodiment

[0052] As Figure 1 shown, a method for cleaning massive information of spacecraft based on multi-feature extraction of the present invention is a method for cleaning satellite in-orbit data. The implementation steps are as follows:

[0053] First step, based on the telemetry characteristics of telemetry variables, partition the set of spacecraft telemetry variables to obtain multiple telemetry variable subsets.

[0054] The telemetry data of a certain spacecraft contains N telemetry variables θ1, θ2, …, θ N , and the telemetry values corresponding to the telemetry variables in the telemetry data Φ t at time t are According to the time-series data characteristics of the telemetry values, the set Θ = {θ1, θ2, …, θ N} composed of all telemetry variables is divided into four subsets, namely the digital quantum subset ΘP = {θ p1 , θ p2 , …, θ pa}, the steady-state simulation quantum set Θ Q = {θ q1 , θ q2 , …, θ qb}, the periodic simulation quantum set Θ R = {θ r1 , θ r2 , …, θ rc} and the aperiodic simulation quantum set Θ S = {θ s1 , θ s2 , …, θ sd}, where a, b, c, and d are the number of elements of the telemetry variables in each subset respectively, and N = a + b + c + d, and:

[0055] Telemetry variables θ with the characteristic of having a finite number of integer enumeration values for the telemetry values p1 , θ p2 , …, θ pa are included in the digital quantum set Θ P . Such parameters are generally telemetry of the on-orbit status of a single machine or a subsystem, such as subsystem section marks, single machine switch status, and operation modes, etc.;

[0056] Telemetry variables θ with the characteristic of having floating-point data values that fluctuate little under various working conditions or environments q1 , θ q2 , …, θ qb are included in the steady-state simulation quantum set Θ Q . Such parameters are generally telemetry of single machine analog quantities, such as the rotational speed of a momentum wheel, the telemetry of the secondary power supply of a control computer, etc.;

[0057] Telemetry variables θ with the characteristic of having floating-point data values that change in a fixed period r1 , θ r2 , …, θ rc are included in the periodic simulation quantum set Θ R . Such parameters include telemetry parameters related to thermal control, orbit, etc.,

[0058] such as the axle temperature of a momentum wheel, solar ephemeris, etc.;

[0059] Telemetry variables θ with the characteristic of having floating-point data values that randomly fluctuate within the effective range or change with factors such as working conditions and environments s1 , θ s2 , …, θ sd are included in the aperiodic simulation quantum set Θ SAmong them, such parameters include telemetry such as the output of sensors, such as the output of sun sensors.

[0060] In the second step, for different subsets of telemetry variables Θ P , Θ Q , Θ R and Θ S Quickly extract the telemetry features of each telemetry variable. Initialize the algorithm parameters, and set the start time t init , the time interval Δt between adjacent moments and the end time t end . Initialize the time variable t = t init , the counter cnt = 0, and the counter threshold N TH . For each telemetry parameter in the subset, obtain the corresponding time series of telemetry values based on the on-orbit historical operation data or ground test data for a period of time, and respectively determine the normal range of each telemetry value statistically:

[0061] For the digital quantity parameter subset Θ P , determine the i-th telemetry variable θ pi ∈Θ P , i = 1, …, a, and the finite enumeration set Λ pi of the corresponding telemetry value φ pi , that is, the set Λ pi is the smallest set containing the normal enumeration values of φ pi ;

[0062] For the steady-state simulation quantum subset Θ Q , determine the j-th telemetry variable θ qj ∈Θ Q , j = 1, …, b, and the time-series statistical characteristic parameters of the corresponding telemetry value φ qj , including the mean μ qj , the standard deviation σ qj , the median Median(φ qj ) and the median absolute deviation MAD(φ qj ). For a continuous time series containing T data

[0063] Define the operator Median(·) as the median calculation operator, Median(φ qj ) is the median of φ qj , then the mean μ qj , the standard deviation σ qj , and the definition of the median absolute deviation MAD(φ qj ) is:

[0064] (1)

[0065] (2)

[0066] (1) MAD(φ qj ) = Median(|φ qj - Median(φ qj )|)

[0067] Periodic simulation quantum set Θ R , construct the k-th telemetry variable θ rk ∈Θ R , k = 1, …, c corresponding to the telemetry value φ rk 's m-th order Fourier reconstruction function and determine the parameters B rk and the error bound Th rk , where are the m-th order signal frequency, amplitude, and phase respectively, B rk is a constant offset, and the order m is generally not greater than 5.

[0068] Aperiodic simulation quantum set Θ S , determine the upper limit value sl ∈Θ R corresponding to the l-th telemetry variable θ sl of the telemetry value φ and the lower limit value

[0069] Step 3: Establish independent expert interpretation rules based on the telemetry characteristics of the telemetry variables; for the telemetry data at a certain moment, establish expert interpretation rules based on the telemetry characteristics to determine whether the telemetry values corresponding to each telemetry variable are wild value anomalies. Retain the telemetry values that conform to the rules, and mark the telemetry values that do not conform to the rules as wild value anomalies;

[0070] Traverse all telemetry variables θ P in the digital quantum set Θ pi ∈Θ P , i = 1, …, a, establish the effectiveness of the expert rule and mark wild values: in the telemetry data at the current moment t, if the telemetry value pi corresponding to the telemetry variable θ does not belong to the finite enumeration set Λ pi , that is perform wild value anomaly marking, indicating that the telemetry value of this telemetry variable has a wild value anomaly, and at the same time the counter cnt = cnt + 1;

[0071] Step 4: Traverse all telemetry variables θ Q in the steady-state simulation quantum set Θ qj∈Θ Q For \(j = 1,\ldots,b\), the Shewhart principle and the median absolute deviation index are used to establish an expert rule to judge the validity and mark outliers: In the telemetry data at the current moment \(t\), if the telemetry variable \(\theta\) qj corresponding telemetry value If the following conditions are simultaneously satisfied:

[0072] (1)

[0073] (2)

[0074] Perform an outlier anomaly annotation, indicating that the telemetry value of this telemetry variable has an outlier anomaly, and at the same time the counter \(cnt=cnt + 1\);

[0075] In the fifth step, traverse the periodic simulation quantum set \(\Theta\) R for all telemetry variables \(\theta\) rk ∈Θ R For \(k = 1,\ldots,c\), an expert rule is established based on the Fourier reconstruction method to judge the validity and mark outliers: In the telemetry data at the current moment \(t\), if the telemetry variable \(\theta\) rk corresponding telemetry value exceeds the error bound \(Th\) rk outside the range, that is Perform an outlier anomaly annotation, indicating that the telemetry value of this telemetry variable has an outlier anomaly, and at the same time the counter \(cnt=cnt + 1\);

[0076] In the sixth step, traverse the aperiodic simulation quantum set \(\Theta\) S for all telemetry variables \(\theta\) sl ∈Θ S For \(l = 1,\ldots,d\), an expert rule is established to judge the validity and mark outliers: In the telemetry data at the current moment \(t\), if the telemetry variable \(\theta\) sl corresponding telemetry value is not within the upper and lower limits, that is Perform an outlier anomaly annotation, indicating that the telemetry value of this telemetry variable has an outlier anomaly, and at the same time the counter \(cnt=cnt + 1\);

[0077] In the seventh step, the telemetry values marked as outlier anomalies are removed, and at the same time it is judged that: If the number of outliers at the current moment exceeds the threshold \(N\) TH , that is, the counter \(cnt > N\) TH , then all telemetry values are removed. After cleaning the telemetry data at the current moment, update the time variable \(t=t+\Delta t\); if \(t > t\) end , output the cleaned result; otherwise, clear the counter \(cnt = 0\) and return to the third step to clean the telemetry data at the next moment.

[0078] Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solution of the present invention by using the methods and technical content disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction, characterized in that, Including: Based on the telemetry features of telemetry variables, partitioning the spacecraft telemetry variables to obtain multiple subsets of telemetry variables; After initializing the variables, for different subsets of telemetry variables, extracting the telemetry features of each spacecraft telemetry variable in the subset; Based on the obtained telemetry features, establishing independent expert interpretation rules; for the telemetry data at a certain moment, using the expert interpretation rules to judge whether the telemetry values corresponding to each telemetry variable are wild value anomalies, retaining the telemetry values that conform to the rules, and marking the telemetry values that do not conform to the rules as wild value anomalies; Eliminating the telemetry values marked as wild value anomalies, and completing the traversal process for the telemetry data containing continuous time periods, realizing the cleaning task of massive telemetry data.

2. The method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 1, wherein: The multiple subsets of telemetry variables include a digital quantum subset Θ P , a steady-state analog quantum subset Θ Q , a periodic analog quantum subset Θ R , and an aperiodic analog quantum subset Θ S .

3. The method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 2, wherein: Using the following criteria to divide the telemetry variables into four subsets: The telemetry value represents a telemetry variable θ with a finite number of integer enumeration value characteristics p1 , θ p2 , …, θ pa Included in the digital quantum set Θ P , where a is the number of elements in Θ P ; The telemetry variable θ whose telemetry value exhibits the characteristic of a floating-point data value with relatively small fluctuations in orbit under various working conditions or environments q1 , θ q2 , …, θ qb is included in the steady-state simulation quantum set Θ Q , where b is the number of elements in Θ Q ; The telemetry variable θ whose telemetry value exhibits the characteristic of floating-point data values changing in a fixed period r1 , θ r2 , …, θ rc is incorporated into the periodic simulation quantum set Θ R , where c is the number of elements in Θ R ; The telemetry variable θ whose telemetry value exhibits the characteristic of a floating-point data value that randomly fluctuates within the effective range or varies with the working conditions, environment, etc. s1 , θ s2 , …, θ sd Included in the non-periodic analog quantum set Θ S where d is the number of elements in Θ S within.

4. A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 3, characterized in that: For different subsets of telemetry variables, extracting the telemetry features of each spacecraft telemetry variable in the subset, including: Digital quantum set Θ P For each telemetry variable within, determine a finite enumeration set of corresponding telemetry values; Steady-state simulated quantum set Θ Q For each telemetry variable within it, determine the time-series statistical characteristic parameters of the corresponding telemetry value, including the mean, variance, median, and median absolute deviation; Periodic simulation quantum set Θ R For each telemetry variable within it, determine the main frequency signal period, signal amplitude, phase, and error bound after fast Fourier transform; Aperiodic analog quantum set Θ S For each telemetry variable within, determine the upper and lower limit ranges of the corresponding values; Extracting telemetry features for the telemetry variable parameters determined for each of the above subsets.

5. A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 4, characterized in that: Initialize the variable, where the initialization content includes the start time t of data outlier processing init , the time interval Δt between adjacent moments and the end time t end , the time variable t = t init , the counter cnt = 0, the counter threshold N TH .

6. The method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 4, wherein: Using the following criteria to extract telemetry features for the telemetry variable parameters determined for each subset: Digital quantum set Θ P Internal telemetry variable, determining a finite enumeration set of corresponding telemetry values: determining the i-th telemetry variable θ pi ∈Θ P , where i = 1, …, a for the corresponding telemetry value φ pi The finite enumeration set Λ pi , that is, the set Λ pi is the smallest set containing the normal enumeration values of φ pi ; Steady-state simulated quantum set Θ Q Internal telemetry variables, determining the mean, variance, median, and median absolute deviation of the corresponding telemetry values: For the j-th telemetry variable θ qj ∈Θ Q , j = 1, …, b, whose telemetry value at time t = T is Given a set of continuous time series Define the operator Median(·) as the median calculation operator, then the mean μ qj , standard deviation σ qj , and median absolute deviation MAD(φ qj ) are defined as: MAD(φ qj ) = Median(|φ qj - Median(φ qj )|); Periodic simulation quantum set Θ R For each telemetry variable within, determine the main frequency signal period, signal amplitude, phase, and error bound after fast Fourier transform: For the k-th telemetry variable θ rk ∈Θ R , k = 1, …, c, construct the m-th order Fourier reconstruction function of the telemetry value φ rk and determine the parameters and B rk and the error bound Th rk , where are the m-th order signal frequency, amplitude, and phase respectively B rk is a constant offset, and the order m is no greater than 5 Aperiodic analog quantum set Θ S For each telemetry variable within, determine the upper and lower limit ranges of the corresponding values: Determine the l-th telemetry variable θ sl ∈Θ R , where l = 1, …, d, for the corresponding telemetry value φ sl the upper limit value and the lower limit value 7. A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 4, characterized in that: Based on the obtained telemetry features, establishing independent expert interpretation rules; for the telemetry data at a certain moment, using the expert interpretation rules to judge whether the telemetry values corresponding to each telemetry variable are wild value anomalies, retaining the telemetry values that conform to the rules, and marking the telemetry values that do not conform to the rules as wild value anomalies, where: For the digital quantity parameter set, using a finite enumeration set to judge the validity and mark wild values; For the steady-state analog quantity parameter set, using the Shewhart principle and the median absolute deviation index to judge the validity and mark wild values; For the periodic analog quantity set, using Fourier transform for signal reconstruction and then judging the validity and marking wild values; For the non-periodic analog quantity set, using the effective upper and lower bound ranges to judge the validity and mark wild values.

8. A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 7, characterized in that: For the digital quantity parameter set, using a finite enumeration set to judge the validity and mark wild values, including: Traverse the digital quantum set Θ P for all telemetry variables θ pi ∈Θ P , where i = 1, …, a, establish expert rules and perform data validity judgment: In the telemetry data at the current moment t, if the telemetry value pi corresponding to the telemetry variable θ does not belong to the finite enumeration set Λ pi , that is then perform outlier anomaly annotation, indicating that the telemetry value of this telemetry variable has an outlier anomaly, and at the same time the counter cnt = cnt + 1.

9. A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 7, characterized in that: For the steady-state analog quantity parameter set, using the Shewhart principle and the median absolute deviation index to judge the validity and mark wild values, including: Traverse the steady-state simulation quantum set Θ Q All telemetry variables θ qj ∈Θ Q , j = 1, …, b, establish expert rules using the Shewhart principle and the median absolute deviation index and perform data validity judgment: In the telemetry data at the current moment t, if the telemetry variable θ qj The corresponding telemetry value Simultaneously satisfy the following two conditions: Then perform wild value anomaly marking, indicating that the telemetry value of this telemetry variable has a wild value anomaly, and at the same time the counter cnt = cnt + 1.

10. A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 7, characterized in that: For the periodic analog quantity set, using Fourier transform for signal reconstruction and then judging the validity and mark wild values, including: Traverse the periodic simulation quantum set Θ R All telemetry variables θ within rk ∈ Θ R , k = 1, …, c, establish expert rules based on the Fourier reconstruction method and perform data validity judgment: In the telemetry data at the current moment t, if the telemetry variable θ rk The corresponding telemetry value Exceeds the expected value at the current moment Of the error bound Th rk Outside the range, that is Then perform outlier anomaly annotation, indicating that the telemetry value of this telemetry variable has an outlier anomaly, and at the same time the counter cnt = cnt + 1.

11. A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to claim 7, characterized in that: For the non-periodic analog quantity set, using the effective upper and lower bound ranges to judge the validity and mark wild values, including: Traverse the non-periodic analog quantum set Θ S All telemetry variables θ sl ∈Θ S , l = 1, …, d, establish expert rules and perform data validity judgment: In the telemetry data at the current time t, if the telemetry variable θ sl The corresponding telemetry value Is not within the upper and lower limits, that is Then perform outlier anomaly annotation, indicating that the telemetry value of this telemetry variable has an outlier anomaly, and at the same time the counter cnt = cnt + 1.

12. A method for cleaning a large amount of information of a spacecraft based on multi-feature extraction according to any one of claims 8-11, characterized in that: Eliminate the telemetry values marked as outliers, complete the traversal processing of the telemetry data containing continuous time periods, and implement the cleaning task of massive telemetry data, including: eliminating the telemetry values marked as outliers. If the number of outliers at the current moment exceeds the set threshold N TH , that is, the counter cnt > N TH , then eliminate all the telemetry values in the current telemetry data and complete the cleaning of the telemetry data at the current moment; traverse the telemetry data containing multiple consecutive frames to implement the cleaning task of massive telemetry data.

Images