Spacecraft power supply system anomaly detection and recovery method based on time sequence regression

Abstract

The invention provides a spacecraft power supply system anomaly detection and recovery method based on time sequence regression, belongs to the technical field of satellite telemetering, and aims to solve the problems that a traditional time sequence data modeling method is difficult to fully mine data relevance under a long time span, the capacity in capturing the overall evolution trend of a system is limited, and the data relevance cannot be fully mined. The method comprises the following steps: acquiring operation data of a power supply system of a spacecraft, and carrying out preprocessing on the operation data of the power supply system of the spacecraft; carrying out data mean filtering; constructing a spacecraft power supply system data recovery model and performing training; fragment residuals are calculated, and an anomaly detection threshold value is set based on the mean value and variance of the fragment residuals; and inputting the test set into the trained model, obtaining a recovery result of the test set, carrying out anomaly detection on a residual error between the recovery result of the test set and a real measurement value of the sensor based on an anomaly detection threshold value, and obtaining output of key parameters of the spacecraft power supply system according to the recovery result of the test set.

Description

TECHNICAL FIELD

[0001] The application relates to a spacecraft power system anomaly detection and recovery method based on time series regression, and belongs to the technical field of satellite telemetry. BACKGROUND

[0002] The power system is the energy core of the spacecraft in-orbit operation, and its stability and reliability are directly related to the mission execution capability and life cycle of the spacecraft platform. The system provides continuous and stable power support for each subsystem by real-time monitoring of power parameters and dynamic adjustment of power distribution. In the running process, the key components in the power system (such as solar cell array, battery pack, power control module, etc.) need to work cooperatively in the complex space environment for a long time to ensure the balance of the whole satellite power supply. However, affected by space particle radiation, alternating extreme temperature difference, device aging and other factors, the system running process is often accompanied by abnormal phenomena of multiple data such as voltage, current, temperature, etc., such as measurement deviation, signal drift, sudden noise and data interruption. If these abnormalities cannot be identified and processed in time, it may lead to power distribution imbalance, load control failure, and even cause systemic failure. Therefore, it is of great significance to build an efficient and intelligent spacecraft power system anomaly detection and data recovery method to identify and correct missing or abnormal information, thereby improving the fault tolerance and autonomous operation capability of the whole system, and ensuring the safety and reliability of the spacecraft long-term in-orbit operation.

[0003] Defects and deficiencies of the prior art: At present, the anomaly detection and data recovery method of the spacecraft power system usually relies on the system modeling technology based on historical time series data. This kind of method builds a prediction model, compares the model output with the actually collected data, identifies the abnormal behavior in the system running, and repairs or reconstructs the abnormal data. This kind of method has been widely used in chemical process control, Internet of Things system, distributed cloud computing and other fields. However, in the application scene of the spacecraft power system, due to its long-term in-orbit operation, complex and diverse task mechanism and the extreme nature of the space environment (such as high-energy particle radiation, severe temperature difference alternation, etc.), the system running data shows significant long-time dependence, periodic evolution law and random disturbance characteristics. These data characteristics not only increase the difficulty of anomaly detection, but also make high-precision data recovery more challenging. The specific challenges include: (1) Due to the long on-orbit operation cycle and rhythmicity of the spacecraft, the operation parameters of the power system often present obvious periodic dependence on a long time scale. In this context, if the traditional recurrent neural network (RNN) is used for modeling, the problems such as gradient disappearance or gradient explosion often occur, which not only increases the complexity of model training, but also limits the effective capture of periodic characteristics, thereby affecting the accuracy of anomaly detection and data recovery.

[0004] (2) The spacecraft task mechanism is complex and variable, and the power system is long-term operated in the space extreme environment (such as space ion radiation, temperature sudden change, etc.), which causes a lot of random noise and non-Gaussian disturbance in the operation data. These factors will seriously interfere with the modeling process, reduce the robustness of anomaly detection, and weaken the reliability and credibility of the data recovery results.

[0005] (3) The current mainstream time series data modeling method mostly uses a serial processing mechanism and a local feature extraction strategy, which is difficult to fully mine the data correlation under a long time span, not only has low training and reasoning efficiency, but also has limited ability to capture the overall evolution trend of the system. This limitation will lead to incomplete perception of the power system state in practical applications, thereby affecting the accuracy of anomaly identification and the overall effect of data recovery. SUMMARY

[0006] To solve the problem that the traditional time series data modeling method is difficult to fully mine the data correlation under a long time span, and has limited ability to capture the overall evolution trend of the system, thereby affecting the accuracy of anomaly identification and the overall effect of data recovery, a spacecraft power system anomaly detection and recovery method based on time series regression is proposed.

[0007] The technical solution adopted by the present application to solve the above problems is as follows: Step 1: Obtain spacecraft power system operation data and perform preprocessing and filtering processing; Step 2: Divide the filtered spacecraft power system operation data into a training set and a test set according to a proportion, and construct a training set based on the training set; Step 3: Construct a spacecraft power system data recovery model based on Transformer, and train the model using the constructed training set to obtain the recovery result of the spacecraft power system operation data ; Step 4: Calculate the segment residual between the recovery result of the spacecraft power system operation data and the spacecraft power system operation data , and based on the segment residual Set the anomaly detection threshold using the mean and variance. TH ; Step 5: Input the test set into the trained spacecraft power system data recovery model and obtain the recovery results of the test set. Based on anomaly detection threshold TH Recovery results of the test set Perform anomaly detection and recover based on the test set results. and spacecraft power system operating data Obtain the output of key parameters of the spacecraft power system .

[0008] Furthermore, step 1 specifically includes: Acquire spacecraft power system operating data The sampling frequency during the acquisition process is... Data on the operation of the spacecraft's power system Missing values ​​are imputed to obtain preprocessed data. ,in, This represents the total number of sampling points for the spacecraft power system operation data. The preprocessed data is then subjected to mean filtering to obtain the dataset. ; The expression for preprocessing spacecraft power system operating data is: (1); In formula (1), , For the first i Preprocessed spacecraft power system operating data NULL For missing values; The expression for mean filtering is: (2); In formula (2), The length of the filter. For the first i Spacecraft power system operating data after sub-mean filtering.

[0009] Furthermore, step 2 specifically includes: Dataset The training set was divided into two parts at a ratio of 9:1. and test set , , Based on the training set Training set for building a power system data recovery model ; Training set of power system data recovery model The expression of the spacecraft power system data recovery model is: (3); (4); In the formula (3) and the formula (4), is the input and output dimension of the spacecraft power system data recovery model.

[0010] Further, the input and output dimensions of the spacecraft power system data recovery model in step 3 are , and the recovery result of the spacecraft power system operation data is , .

[0011] Further, in step 4, the abnormality detection threshold TH is set. Step 4.1: Calculate the mean of the obtained spacecraft power system operation data , and based on the mean , calculate the recovery result of the spacecraft power system operation data and the mean square error, root mean square error and determination coefficient of the spacecraft power system operation data , if the mean square error, root mean square error and determination coefficient do not meet the preset range, retrain the spacecraft power system data recovery model until the output recovery result of the spacecraft power system operation data and the mean square error, root mean square error and determination coefficient of the spacecraft power system operation data meet the preset range. Step 4.2: Set the length of the data anomaly detection window to , and calculate the segment residual of the recovery result of the spacecraft power system operation data and the spacecraft power system operation data . Step 4.3: Calculate the mean and standard deviation of the segment residual , and set the abnormality detection threshold according to the mean and the standard deviation TH . The calculation formula of the mean square error is: (5); The calculation formula of the root mean square error is: (6); The calculation formula of the determination coefficient is: (7); Segment residual The calculation formula is: (8); Spacecraft power system operation data The mean value of The calculation formula is: (9); Segment residual The mean value of And the standard deviation The calculation formula is: (10); (11); Abnormal detection threshold TH The calculation formula is: (12); In formula (12), Denotes the quantile under the standard normal distribution, which can be obtained by looking up the table.

[0012] Further, step 5 specifically includes: Input the test set into the trained spacecraft power system data recovery model to obtain the recovery result of the test set , Calculate the difference between the recovery result And the test set data When , it is determined that the spacecraft power system data is abnormal, and the corresponding recovery result of the test set is used to replace the abnormal data to obtain the output of the spacecraft power system key parameter

[0013] The expression of the output of the spacecraft power system key parameter Is: (13); In formula (13), .

[0014] The beneficial effects of the present application are: 1. The present application aims at the random data noise problem caused by the extreme nature of space environment and complex electromagnetic interference between devices, introduces time series mean filtering strategy to smooth the original operation data and reduce the interference of noise on abnormal detection and data recovery accuracy.

[0015] 2. This invention employs a mean filtering method, based on filtered system runtime sequence data, and combines it with the Transformer large model architecture to construct and train a spacecraft power system data recovery model based on a self-attention mechanism. This model can effectively capture the dynamic evolution of the power system over long time scales, improve the model's feature extraction capability, parallel processing efficiency, and overall interpretability, effectively suppress random noise in the spacecraft power system runtime data, flexibly adapt to the modeling needs of different types of data features, significantly reduce the complexity of feature engineering, and enhance the transfer and application capability of the spacecraft power system data recovery model in multiple scenarios. This reduces the interference of extreme disturbances in the space environment and complex changes in mission modes on system data quality, effectively improving the accuracy and robustness of subsequent anomaly detection and data recovery processes.

[0016] 3. In the anomaly detection stage, this invention employs a threshold-based strategy, comparing the error between the recovery model output and the actual monitored value. When the error exceeds a set threshold, the system is determined to be in an abnormal state, and the abnormal data is replaced and restored using the recovery model's output. When the error does not exceed the threshold, the system is considered to be in a normal state. Ultimately, this invention achieves effective identification and high-precision data recovery of abnormal states in the power system through anomaly detection and recovery processing of spacecraft power system operating data. The mean square error (MSE) is approximately 0.006, the root mean square error (RMSE) is approximately 0.083, and the fitting accuracy reaches approximately 95.5%. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a spacecraft power system anomaly detection and recovery method based on time-series regression provided by the present invention; Figure 2 This is a schematic diagram of the structure of the spacecraft power system data recovery model provided by the present invention; Figure 3 A comparative diagram of the spacecraft power system data recovery before and after the present invention; Figure 4 A comparison diagram between the output of the spacecraft power system data recovery model provided by this invention and the actual data of the spacecraft power system. Detailed Implementation

[0018] Combination Figures 1-4 This implementation method is described as follows: Figure 1 As shown in the figure, the steps of the spacecraft power system anomaly detection and recovery method based on time-series regression described in this embodiment include: S1: Acquire spacecraft power system operating data; Acquire spacecraft power system operating data The sampling frequency during the acquisition process is... sampling frequency ; S2: Data preprocessing; operational data of spacecraft power systems Missing values ​​are imputed to obtain preprocessed data. ,in, In this embodiment, the total number of sampling points for the spacecraft power system operation data is set to... N For 20000, The preprocessing expression is: (1); In formula (1), , For the first i Preprocessed spacecraft power system operating data NULL For missing values; S3: Data mean filtering; The preprocessed data is then subjected to mean filtering to obtain the dataset. ; The expression for mean filtering is: (2); In formula (2), In this embodiment, the length of the filter is set as follows: , For the first i Spacecraft power system operating data after sub-mean filtering.

[0019] This invention addresses the problem of random data noise caused by the extreme nature of the space environment and complex electromagnetic interference between devices by introducing a time-series mean filtering strategy to smooth the original operating data and reduce the interference of noise on the accuracy of anomaly detection and data recovery.

[0020] S4: Divide the training set and the test set, construct a spacecraft power system data recovery model based on Transformer, and use the constructed training set to train the model to obtain the recovery results of the spacecraft power system operation data; S401: Transfer the dataset The training set was divided into two parts at a ratio of 9:1. and test set , , Based on the training set Training set for building a power system data recovery model ; S402: On the basis of S401, set the input dimension and output dimension of the constructed spacecraft power system voltage data recovery model , and based on , construct a training set for the model , wherein , ; S403: Construct a training set , train the established spacecraft power system voltage data recovery model based on Transformer as shown in Figure 2 , and obtain the spacecraft voltage data recovery result: , the data recovery structure is as shown in Figure 3 , the mean value of the spacecraft voltage acquisition numerical value .

[0021] The present application is based on the filtered system runtime sequence data, combined with the Transformer large model architecture, to construct and train a spacecraft power system data recovery model based on the self-attention mechanism. The model can effectively capture the dynamic evolution law of the power system in a long time scale, improve the feature extraction capability, parallel processing efficiency and overall interpretability of the model, effectively suppress the random noise in the spacecraft power system operation data, adapt to the modeling needs of different types of data characteristics, significantly reduce the complexity of feature engineering, enhance the migration application ability of the spacecraft power system data recovery model in multiple scenarios, thereby reducing the interference of extreme disturbances in space environment and complex changes in task mode on system data quality, effectively improving the precision and robustness of the subsequent anomaly detection and data recovery process.

[0022] S5: Calculate the segment residual between the recovery result of the spacecraft power system operation data and the spacecraft power system operation data, and set the anomaly detection threshold based on the mean and variance of the segment residual TH ; S501: Calculate the mean of the obtained spacecraft power system operation data , calculate the recovery result of the spacecraft power system operation data based on the mean , and the mean square error, root mean square error and determination coefficient of the spacecraft power system operation data , if the mean square error, root mean square error and determination coefficient do not meet the preset range, retrain the spacecraft power system data recovery model until the recovery result of the output spacecraft power system operation data and the mean square error, root mean square error and determination coefficient of the spacecraft power system operation data meet the preset range; The calculation formula of the mean square error is: (5); The calculation formula of the root mean square error is: (6); The calculation formula of the coefficient of determination is: (7); After calculation, the mean of the residual error in the embodiment is about 0.006, the root mean square error is about 0.083, and the coefficient of determination is about 95.5%.

[0023] S502: Set the length of the data anomaly detection window as , calculate the recovery result of the spacecraft power system operation data and the segment residual error of the spacecraft power system operation data , calculate the mean and standard deviation of the segment residual error , set the anomaly detection threshold according to the mean and standard deviation TH ; The calculation formula of the segment residual error is: (8); The calculation formula of the mean of the spacecraft power system operation data is: (9); The calculation formula of the mean and standard deviation of the segment residual error is: (10); (11); The calculation formula of the anomaly detection threshold TH is: (12); In formula (12), represents the quantile under the standard normal distribution, which can be obtained by looking up the table, and is set in the embodiment.

[0024] After calculation, the mean and standard deviation of the residual error in the embodiment are , the abnormality detection threshold TH = 0.031.

[0025] S6: inputting the test set into the trained spacecraft power system data recovery model, obtaining the recovery result of the test set, performing abnormality detection on the recovery result of the test set based on the abnormality detection threshold TH , and obtaining the output of the key parameters of the spacecraft power system according to the recovery result of the test set and the spacecraft power system operation data; S6: inputting the test set into the trained spacecraft power system data recovery model, obtaining the recovery result of the test set , , calculating the difference between the recovery result and the test set data , wherein , the spacecraft power system abnormal voltage data is recovered by using formula (11), that is, when it is determined that the data is abnormal (data missing or data failure, ), the output of the spacecraft power system voltage data recovery model is used as the output of the spacecraft power system voltage, and the final output is obtained. Figure 4 .

[0026] The output of the key parameters of the spacecraft power system is expressed as: (13).

[0027] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the technical solution of the present application, and can make equivalent embodiments with equivalent changes. Any simple modification, equivalent replacement and improvement of the above embodiments, as long as it does not depart from the technical solution of the present application, is within the scope of protection of the present application.

Claims

1. A method for anomaly detection and recovery of spacecraft power systems based on time-series regression, characterized in that, include: Step 1: Acquire spacecraft power system operating data And perform preprocessing and filtering; Step 2: Divide the filtered spacecraft power system operation data into training set and test set according to the proportion, and build training set based on training set; Step 3: Construct a Transformer-based data recovery model for the spacecraft power system, and train the model using the constructed training set to obtain the recovery results of the spacecraft power system operating data. ; Step 4: Calculate the recovery results of the spacecraft power system operating data and spacecraft power system operating data Fragment residuals between Based on fragment residuals Set the anomaly detection threshold using the mean and variance. TH ; Step 5: Input the test set into the trained spacecraft power system data recovery model and obtain the recovery results of the test set. Based on anomaly detection threshold TH Recovery results of the test set Perform anomaly detection and recover based on the test set results. and spacecraft power system operating data Obtain the output of key parameters of the spacecraft power system .

2. The method for anomaly detection and recovery of spacecraft power systems based on time-series regression according to claim 1, characterized in that, Step 1 specifically includes: Acquire spacecraft power system operating data The sampling frequency during the acquisition process is... Data on the operation of the spacecraft's power system Missing values ​​are imputed to obtain preprocessed data. ,in, This represents the total number of sampling points for the spacecraft power system operation data. The preprocessed data is then subjected to mean filtering to obtain the dataset. ; The expression for preprocessing spacecraft power system operating data is: (1)； In formula (1), , For the first i Preprocessed spacecraft power system operating data NULL For missing values; The expression for mean filtering is: (2)； In formula (2), The length of the filter. For the first i Spacecraft power system operating data after sub-mean filtering.

3. The method for anomaly detection and recovery of spacecraft power systems based on time-series regression according to claim 1, characterized in that, Step 2 specifically includes: Dataset The training set was divided into two parts at a ratio of 9:

1. and test set , , Based on the training set Training set for building a power system data recovery model ; Training set of power system data recovery model The expression is: (3)； (4)； In formulas (3) and (4), The input and output dimensions of the spacecraft power system data recovery model.

4. The method for anomaly detection and recovery of spacecraft power systems based on time-series regression according to claim 1, characterized in that, Step 3: The input and output dimensions of the spacecraft power system data recovery model are both... The recovery results of the spacecraft power system operating data are as follows: , .

5. The method for anomaly detection and recovery of spacecraft power systems based on time-series regression according to claim 1, characterized in that, In step 4, set the anomaly detection threshold. TH The steps include: Step 4.1: Calculate the acquired spacecraft power system operating data mean Based on the mean Calculate the recovery results of spacecraft power system operating data and spacecraft power system operating data The mean square error, root mean square error, and coefficient of determination are calculated. If these parameters do not fall within the preset range, the spacecraft power system data recovery model is retrained until the recovered spacecraft power system operating data is output. and spacecraft power system operating data The mean square error, root mean square error, and coefficient of determination are within the preset range; Step 4.2: Set the length of the data anomaly detection window to [value]. The results of recovering operational data of the spacecraft's power system. and spacecraft power system operating data Fragment residual ; Step 4.3: Calculate the fragment residuals mean and standard deviation According to the mean and standard deviation Set anomaly detection threshold TH ; The formula for calculating the mean square error is: (5)； The formula for calculating the root mean square error is: (6)； The formula for calculating the coefficient of determination is: (7)； Fragment residual The calculation formula is: (8)； Spacecraft power system operating data mean The calculation formula is: (9)； Fragment residual mean and standard deviation The calculation formula is: (10)； (11)； Anomaly detection threshold TH The calculation formula is: (12)； In formula (12), The quantiles represent the quantiles under the standard normal distribution and are obtained by looking up a table.

6. The method for anomaly detection and recovery of spacecraft power systems based on time-series regression according to claim 1, characterized in that, Step 5 specifically includes: Input the test set into the trained spacecraft power system data recovery model and obtain the recovery results of the test set. , Calculate the recovery result Difference between the test set data and the test set data ,when When an anomaly is detected in the spacecraft power system data, the recovered results of the corresponding test set are used to replace the anomaly data, and the output of key parameters of the spacecraft power system is obtained. Output of key parameters of spacecraft power system The expression is: (13)； Formula (13) .

Images