Load data stationarity assessment method based on random sampling

Abstract

This invention relates to the field of on-orbit payload data analysis, and particularly to a method for assessing the stationarity of payload data based on random sampling. The method includes: imputing missing values ​​in the payload data and recording the imputation time points corresponding to all imputed data; performing n random samplings on the imputed payload data, and selecting the k random sampling results that use the fewest imputation time points from the n random sampling results; performing nonlinear reconstruction and stationarity analysis on the k random sampling results respectively to obtain k stationarity analysis results; and performing a weighted fusion of the obtained k stationarity analysis results to determine whether the stationarity of the payload data has changed. This invention provides accurate and robust stationarity assessment of payload data under conditions of uncertain downlink cycles and severe data loss, reducing the impact of data loss on analysis results and improving the reliability and efficiency of on-orbit payload status monitoring.

Description

Technical Field

[0001] This invention relates to the field of on-orbit data analysis technology, and in particular to a method for evaluating the stationarity of payload data based on random sampling. Background Technology

[0002] During spacecraft payloads' on-orbit flight missions, the payloads generate massive amounts of operational data. The downlink transmission of this data is the core basis for carrying out payload status monitoring, performance evaluation, and fault early warning. Furthermore, the assessment of payload data stability is a key technical link to ensure the stable operation of the payload in orbit. Its assessment results directly determine the accuracy of the judgment on the payload's working status and are of great significance to the smooth progress of the mission.

[0003] Traditional methods for assessing payload data stability typically rely on continuous, complete payload data throughout its entire lifecycle. These methods are effective in scenarios with stable data acquisition conditions, such as ground testing. However, during actual in-orbit flight, limitations imposed by factors such as the coverage area of ​​the ground control zone and the transmission conditions of the control link result in significant uncertainty in the downlink cycle of payload data. Data cannot be continuously transmitted throughout the entire lifecycle, which greatly limits the applicability of traditional full-cycle stability assessment methods and makes them difficult to directly apply to the analysis of in-orbit payload data.

[0004] To address the issue of missing data, the industry commonly employs conventional missing value completion techniques for preprocessing. These techniques can, to some extent, mitigate the analytical obstacles caused by a small amount of missing data. However, during the downlink process of on-orbit payload data, a large amount of data is often missing. If ordinary completion techniques are continued in this case, a significant amount of invalid noise will be introduced into the load data stability analysis. This noise will not only interfere with the normal data analysis process but may even obscure the key state change characteristics that need to be monitored during load operation, leading to serious deviations in the stability assessment results and failing to accurately reflect the true operating state of the load.

[0005] Therefore, how to accurately and robustly assess the stability of load data even when the downlink cycle of load data is uncertain and severely lacking has become a technical challenge that urgently needs to be solved in this field. Summary of the Invention

[0006] In view of this, the present invention proposes an evaluation method that combines random sampling, selective reconstruction and weighted fusion, aiming to reduce the impact of missing data on the analysis results and improve the reliability and efficiency of on-orbit load status monitoring.

[0007] To achieve the above objectives, the technical solution created by this invention is implemented as follows:

[0008] A method for evaluating the stationarity of load data based on random sampling includes the following steps:

[0009] S1: Fill in the missing values ​​in the load data and record the completion time points corresponding to all the completed data;

[0010] S2: Perform n random samplings on the completed load data, where n is an integer greater than 2. Select the k random sampling results with the fewest completion time points from the n random sampling results, where k is a positive integer less than or equal to n.

[0011] S3: Perform nonlinear reconstruction and stationarity analysis on the k random sampling results respectively to obtain k stationarity analysis results;

[0012] S4: The obtained k stationarity analysis results are weighted and fused, and the weighted fusion results are used to determine whether the stationarity of the load data has changed; the weighting coefficients are determined based on the number of time points of the supplementary data used in the random sampling results.

[0013] Furthermore, in step S1, missing value completion is achieved using a time series prediction model based on a recurrent neural network. The time series prediction model is trained using historical continuous load data from the ground testing phase, and the training objective is to predict the value at time t based on the value at time t-1.

[0014] Furthermore, in step S2, random sampling is performed through n random sparse time matrices, and the sampling rate of random sampling is no greater than 50%.

[0015] Furthermore, the sampling rate for random sampling is 25%.

[0016] Furthermore, in step S2, the specific method for selecting k random sampling results is as follows: sort the n random sampling results in reverse order according to the number of completion time points used, and take the first k random sampling results after sorting.

[0017] Furthermore, in step S3, the nonlinear reconstruction of the k random sampling results is achieved using the orthogonal matching pursuit algorithm, and the stationarity analysis is performed in the frequency domain. The spectral analysis result F corresponding to each random sampling result is obtained through the frequency domain analysis. k .

[0018] Furthermore, in step S4, the weighted fusion of the k stationarity analysis results specifically involves the weighted fusion of the k spectral analysis results F. k The fusion spectrum analysis result F is obtained by weighting the coefficients. fusion .

[0019] Furthermore, in step S4, the formula for calculating the weighting coefficients is as follows:

[0020] W=(NC) / N;

[0021] Where W is the weighting coefficient, N is the total number of data points in a single random sampling result, and C is the number of completion time points used in a single random sampling result.

[0022] Furthermore, in step S4, the specific process of determining whether the stationarity of the load data has changed using the weighted fusion result is as follows:

[0023] Frequency domain analysis was performed on the completed payload data to obtain the baseline spectrum analysis result F; the fused spectrum analysis result F was then analyzed. fusion Compare the results with the baseline spectrum analysis result F. If the fused spectrum analysis result F... fusion If the value of a certain frequency point f is higher than the value of the corresponding frequency point f in the baseline spectrum analysis result F, and the frequency point f is an abnormal frequency point, then it is determined that the stability of the load data has changed.

[0024] Furthermore, in step S4, frequency points that exceed the normal operating frequency of the load are determined to be abnormal frequency points.

[0025] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0026] 1. Effectively suppress noise interference from missing data: By prioritizing the selection of the sample subset with the fewest completion data points from multiple random sampling results and performing dynamic weighted fusion based on the proportion of real data, the impact of systematic noise introduced by the completion algorithm on stationarity analysis is significantly reduced. This avoids the problem of missing data drowning out real signal changes in traditional methods and improves the accuracy of on-orbit anomaly detection.

[0027] 2. Significantly reduced computational complexity: By adopting a random sparse sampling mechanism, only about 25% of the original data needs to be analyzed to achieve effective evaluation. Compared with the traditional full data processing method, the amount of computation is reduced by an order of magnitude, significantly saving on-board or ground computing resources and meeting the timeliness requirements of on-orbit real-time monitoring.

[0028] 3. Enhance the robustness and reliability of the analysis: By performing n independent random samplings and weighted fusion of multiple analysis results, the random misjudgment that may be caused by relying on a single data segment is avoided, the statistical confidence of the evaluation results is improved, and the stationarity judgment is more robust and reliable.

[0029] 4. Adaptive to on-orbit data characteristics: This method does not rely on equally spaced sampling or complete data sequences. It is naturally adapted to the characteristics of uncertain downlink cycle, high missing rate and random distribution of aerospace payload data, providing a new approach with strong engineering feasibility and wide applicability for on-orbit health management.

[0030] 5. Weighting Mechanism Enables Precise Quantitative Evaluation: This invention innovatively transforms the number of supplementary data points used in each random sampling into a weighting coefficient for the analysis result. This design ensures that, during the final fusion judgment, the analysis results of samples with high reliability (i.e., fewer supplementary points) receive greater weight, while the influence of samples with low reliability (i.e., more supplementary points) is suppressed. Compared to simple averaging or single analysis, this data quality-based weighted fusion mechanism provides a more scientific and reasonable final judgment with stronger fault tolerance. Attached Figure Description

[0031] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0032] Figure 1 A schematic flowchart of the load data stationarity assessment method based on random sampling as described in the embodiments of the present invention;

[0033] Figure 2 A schematic diagram of the load data stationarity assessment method based on random sampling as described in the embodiments of the present invention;

[0034] Figure 3 This is a schematic diagram illustrating the completion of missing load data values ​​as described in the embodiments of the present invention;

[0035] Figure 4 This is a schematic diagram of random sampling of complete load data as described in an embodiment of the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0038] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0039] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0040] While traditional missing value imputation methods can address the issue of missing data in a small number of load data points, a large number of missing data points exist during the downlink phase of load data. Using traditional missing value imputation methods introduces significant noise into the stationarity analysis of load data, potentially even obscuring important variations. However, by randomly sampling the imputed load data, selecting those with smaller missing values, and then performing nonlinear reconstruction on these selected samples, the time series of the load data can be accurately restored. This provides a reliable basis for subsequent stationarity analysis of the load data, effectively mitigating the impact of the downlink periodic instability on stationarity analysis. Based on this idea, this invention proposes a load data stationarity assessment method based on random sampling.

[0041] The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0042] like Figure 1 and Figure 2 As shown in the figure, an embodiment of the present invention provides a method for evaluating the stationarity of load data based on random sampling, comprising the following steps:

[0043] S1: Fill in the missing values ​​in the load data and record the completion time points corresponding to all the completed data.

[0044] Missing value completion is achieved using a time series prediction model based on a recurrent neural network. Historical continuous load data from the ground testing phase are input into the recurrent neural network for training to obtain the time series prediction model. The training objective is to predict the value at time t based on the value at time t-1.

[0045] After the payload data is downloaded, the trained time series prediction model is used to fill in the missing values ​​in the payload data, such as... Figure 3 As shown, all the data points that were completed are recorded.

[0046] S2: Perform n random samplings on the completed load data, where n is an integer greater than 2. Select the k random sampling results from the n random sampling results that use the fewest completion time points, where k is a positive integer less than or equal to n.

[0047] Random sampling is performed through n random sparse time matrices, and the sampling rate of random sampling is no greater than 50%. Preferably, the sampling rate of random sampling is 25%.

[0048] The process of performing random sampling on n random sparse time matrices is as follows:

[0049] I. Generating the matrix: First, generate a sequence of all zeros of the same length as the time points of the padded data. Then, according to the set sampling rate (e.g., 25%), randomly flip the corresponding number of 0s to 1s.

[0050] For example: if there are a total of 100 time points, the sampling rate is 25%, and 25 time points are randomly selected, their corresponding matrix positions are set to 1.

[0051] II. Perform sampling: Apply this random sparse time matrix to the completed payload data sequence.

[0052] The rule is: only retain the data values ​​at the time points where the corresponding position in the matrix is ​​1, and discard the data values ​​where the corresponding position is 0. In this way, 25% of the data points are randomly selected from a complete data sequence, resulting in a sparse and non-uniform sampling result.

[0053] III. Multiple sampling: Repeat steps I and II n times, each time generating a new, independent random sparse time matrix (i.e., the positions of 1 are randomly redistributed), thus obtaining n different random sampling results.

[0054] By employing a random sparse sampling mechanism, effective evaluation can be achieved by analyzing only a small portion of the original data. Compared with traditional full data processing methods, the computational load is reduced by an order of magnitude, significantly saving on-board or ground computing resources and meeting the timeliness requirements of real-time on-orbit monitoring.

[0055] Select n random sparse time matrices with a sampling rate of less than 25%, and randomly sample the payload data after imputing missing values. An example of random sampling is shown below. Figure 4 As shown, the horizontal axis represents time, and the vertical axis represents numerical values. Based on the time points where the recorded completed values ​​were located, the completed values ​​used in n random samplings were calculated. Three of these time points fall exactly within the data time points recorded in step S1. Therefore... Figure 4 The random sampling used three padding values.

[0056] The specific method for selecting the k random sampling results with the fewest completion time points from the n random sampling results is as follows: sort the n random sampling results in reverse order according to the number of completion time points used, and take the first k random sampling results after sorting.

[0057] By prioritizing the selection of the sample subset with the fewest completion data points from multiple random sampling results and performing dynamic weighted fusion based on the proportion of real data, the impact of systematic noise introduced by the completion algorithm on stationarity analysis is significantly reduced. This avoids the problem of missing data overwhelming real signal changes in traditional methods and improves the accuracy of on-orbit anomaly detection.

[0058] S3: Perform nonlinear reconstruction and stationarity analysis on the k random sampling results respectively to obtain k stationarity analysis results.

[0059] The k random sampling results are reconstructed nonlinearly using the orthogonal matching pursuit algorithm or other nonlinear reconstruction algorithms.

[0060] Stationarity analysis is a frequency domain analysis, and the spectral analysis result F corresponding to each random sampling result is obtained through frequency domain analysis. k .

[0061] S4: The obtained k stationarity analysis results are weighted and fused, and the weighted fusion results are used to determine whether the stationarity of the load data has changed; the weighting coefficients are determined based on the number of time points of the supplementary data used in the random sampling results.

[0062] The weighted fusion of k stationarity analysis results specifically involves the fusion of k spectral analysis results F k The fusion spectrum analysis result F is obtained by weighting the coefficients. fusion .

[0063] The weighting coefficients for weighted fusion are calculated using the following formula:

[0064] W=(NC) / N;

[0065] Where W is the weighting coefficient, N is the total number of data points in a single random sampling result, and C is the number of completion time points used in a single random sampling result.

[0066] The specific process for determining whether the stationarity of the load data has changed using the weighted fusion results is as follows:

[0067] The frequency domain analysis of the completed payload data was performed using the Fourier transform method to obtain the baseline spectrum analysis result F; the fused spectrum analysis result F was then analyzed. fusion Compare the results with the baseline spectrum analysis result F. If the fused spectrum analysis result F... fusion If the value of a certain frequency point f is higher than the value of the corresponding frequency point f in the baseline spectrum analysis result F, and the frequency point f is an abnormal frequency point, then it is determined that the stability of the load data has changed.

[0068] The load has a normal operating frequency, and frequency points that exceed the normal operating frequency of the load are judged as abnormal frequency points.

[0069] By performing n independent random samplings and weighted fusion of multiple analysis results, the random misjudgment that may result from relying on a single data segment is avoided, the statistical confidence of the evaluation results is improved, and the stationarity judgment is made more robust and reliable.

[0070] This invention innovatively transforms the number of imputation data points used in each random sampling into a weighting coefficient for the analysis result. This design ensures that, during the final fusion judgment, the analysis results of samples with high reliability (i.e., fewer imputation points) receive greater weight, while the influence of samples with low reliability (i.e., more imputation points) is suppressed. Compared to simple averaging or single analysis, this data quality-based weighted fusion mechanism provides a more scientific and reasonable final judgment with stronger fault tolerance.

[0071] This method does not rely on equally spaced sampling or complete data sequences, and is naturally adapted to the characteristics of uncertain downlink cycles, high missing rates and random distribution of aerospace payload data, providing a new approach with strong engineering feasibility and wide applicability for on-orbit health management.

[0072] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0073] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A load data stationarity evaluation method based on random sampling, characterized in that, The method comprises the following steps: S1: missing value completion is performed on the missing part of the load data, and all the completion time points corresponding to the completed data are recorded; S2: n times of random sampling are performed on the completed complete load data, n is an integer greater than 2, and k random sampling results using the least number of completion time points are selected from the n times of random sampling results, k is a positive integer less than or equal to n; S3: performing nonlinear reconstruction and stationarity analysis on the k random sampling results respectively to obtain k stationarity analysis results; in step S3, the nonlinear reconstruction of the k random sampling results is implemented by using an orthogonal matching pursuit algorithm, and the stationarity analysis is frequency domain analysis, and a frequency spectrum analysis result F corresponding to each random sampling result is obtained through the frequency domain analysis k ; S4: the k stationary analysis results obtained are weighted and fused, and the weighted fusion result is used to judge whether the stationarity of the load data changes; wherein the weighting coefficients are determined based on the number of time points of the completion data used in the random sampling result; In step S4, the weighted fusion of the k stationary analysis results is specifically a weighted average of the k spectrum analysis results F k The weighted average is obtained by weighting according to the weighting coefficients, and the fused spectrum analysis result F fusion ; In step S4, the calculation formula of the weighting coefficient is: W=(N-C) / N; Wherein, w is the weighting coefficient, N is the total number of data points of a single random sampling result, and C is the number of completion time points used in a single random sampling result; In step S4, the specific process of judging whether the stationarity of the load data changes by using the weighted fusion result is: The complete load data after the supplement is subjected to frequency domain analysis to obtain a reference spectrum analysis result F; the fusion spectrum analysis result F is compared with the reference spectrum analysis result F fusion If the value of a certain frequency point f in the fusion spectrum analysis result F is higher than the value of the corresponding frequency point f in the reference spectrum analysis result F, and the frequency point f is an abnormal frequency point, it is determined that the stationarity of the load data has changed. fusion ​ 2. The random-sampling-based load data stationarity assessment method according to claim 1, characterized in that, In step S1, the missing value completion is realized by using a time series prediction model based on a recurrent neural network, the time series prediction model is trained by using historical continuous load data in the ground test stage, and the training target is to predict the value at t time according to the value at t-1 time.

3. The random-sampling-based load data stationarity assessment method of claim 1, wherein, In step S2, the random sampling is performed by n random sparse time matrices, and the sampling rate of the random sampling is not greater than 50%.

4. The random-sampling-based load data stationarity assessment method of claim 3, wherein, The sampling rate of the random sampling is 25%.

5. The random-sampling-based load data stationarity assessment method of claim 1, wherein, In step S2, the specific way of selecting k random sampling results is: the n times of random sampling results are sorted in reverse order according to the number of completion time points used, and the first k random sampling results after sorting are taken.

6. The random-sampling-based load data stationarity assessment method of claim 1, wherein, In step S4, the frequency points exceeding the normal working frequency of the load are determined as abnormal frequency points.

Images