On-orbit anomaly detection and correction method, system and equipment for remote sensing instrument

Abstract

The invention discloses a remote sensing instrument data on-orbit anomaly detection and correction method, system and equipment. The method comprises the following steps: training a neural network by using historical normalized hot mirror temperature time sequence data to obtain an initial network model capable of predicting hot mirror temperature data; dividing the new hot mirror temperature time sequence data into two segments according to a preset rule, and predicting the first segment of hot mirror temperature time sequence data by using the initial network model to obtain a first segment of predicted temperature; combining the hot mirror temperature historical time sequence data with the first section of hot mirror temperature time sequence data to obtain a new training set, and performing fine adjustment on the initial network model parameters by using the new training set to update the network model; using the updated network model to predict the second segment of hot mirror temperature time sequence data to obtain a second segment of predicted temperature; and combining the first section of predicted temperature and the second section of predicted temperature into predicted heat mirror temperature, comparing the predicted heat mirror temperature with new heat mirror temperature sequence data, identifying abnormal data in the new heat mirror temperature sequence, and correcting the abnormal data.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of satellites, and more particularly, relates to an on-orbit anomaly detection and correction method, system and device for a remote sensing instrument. BACKGROUND

[0002] Radiometric calibration is the basis for quantitative application of remote sensing instrument data. For a microwave imager, obtaining accurate and effective brightness temperature at ground depends on the measurement data of each calibration component, such as the temperature of the hot mirror. The microwave imager of Fengyun-3D satellite is affected by the space environment, and the hot mirror temperature measurement module fails after 5 years of on-orbit operation, so the temperature of the hot reflector cannot be effectively obtained, which leads to the inability to obtain effective brightness temperature for ground observation.

[0003] The automatic detection of hot mirror temperature anomaly is mainly performed through threshold setting based on experience, but due to the characteristics of strong nonlinearity, high noise and non-stationarity of the data, the false detection rate and the missed detection rate of the threshold method are high. For the temperature measurement failure problem, a method based on cold mirror temperature estimation was used to restore the brightness temperature calculation, but the annual average deviation of the brightness temperature reached 5K, which seriously affected the data application.

[0004] In recent years, deep learning, transfer learning and time series model technology have provided a new paradigm for telemetry data anomaly detection, and many foreign institutions have carried out research on spacecraft telemetry data anomaly detection. NASA uses clustering algorithms to build a data model for extracting nominal telemetry data, and develops a system that can determine the measurement area of new incoming telemetry data. The German Space Operations Center has developed an automatic telemetry health monitoring system based on a modular data analysis framework. The system uses machine learning techniques to analyze satellite telemetry data and obtain satellite health status information, and has the ability to analyze new behaviors in telemetry data and automatically identify anomalies.

[0005] The remote sensing instrument will generate a large amount of component telemetry parameter data, and the time series data is taken as the research object. By using machine learning to mine the running characteristics and rules of each parameter data, the goal of detecting and correcting anomalies in the data can be achieved. The current related research at home and abroad is still in the stage of direct application of mature machine learning algorithms, and there is no improvement in the adaptability of machine learning algorithms for data characteristics, resulting in high false detection rate and poor adaptability of algorithms. If a deep learning model is used to improve application results, there are problems of large number of parameters and long training time of mainstream deep learning models. At the same time, if the causal relationship between parameters is ignored, it will lead to insufficient detection ability of sudden anomalies and significant error accumulation in long-term prediction. SUMMARY

[0006] In view of the above defects or improvement needs of the prior art, the present application provides a remote sensing instrument on-orbit anomaly detection and correction method, system and device, which aims to improve the usability of data by correcting the abnormal value of the hot mirror temperature and re-performing calibration, thereby solving the influence of hot mirror temperature failure.

[0007] To achieve the above object, according to one aspect of the present application, a remote sensing instrument data on-orbit anomaly detection and correction method is provided, the method comprising:

[0008] training a neural network using the normalized historical hot mirror temperature time series data to obtain an initial network model of predictable hot mirror temperature data;

[0009] dividing the new hot mirror temperature time series data into two segments according to a preset rule, and using the initial network model to predict the first segment of hot mirror temperature time series data to obtain a first segment of predicted temperature;

[0010] merging the historical hot mirror temperature time series data with the first segment of hot mirror temperature time series data to obtain a new training set, and using the new training set to fine-tune the initial network model parameters to update the network model;

[0011] using the updated network model to predict the second segment of hot mirror temperature time series data to obtain a second segment of predicted temperature;

[0012] merging the first segment of predicted temperature and the second segment of predicted temperature into a predicted hot mirror temperature, comparing the predicted hot mirror temperature with the observed value corresponding to the new hot mirror temperature time series data, identifying the abnormal data in the new hot mirror temperature sequence and correcting it.

[0013] As a further improvement and supplement to the above scheme, the present application further includes the following additional technical features.

[0014] Preferably, the method further comprises:

[0015] calculating the root mean square error and the mean absolute error of the predicted hot mirror temperature and the observed value corresponding to the new hot mirror temperature time series data;

[0016] selecting the model parameters corresponding to the minimum value of the root mean square error and the mean absolute error as the parameters of the network model.

[0017] Preferably, the calculation method of the root mean square error of the predicted hot mirror temperature and the observed value corresponding to the new hot mirror temperature time series data is:

[0018] ;

[0019] wherein N is the number of samples, a predicted hot mirror temperature value, an actual hot mirror temperature observation value.

[0020] Preferably, the method for calculating the mean absolute error of the predicted hot mirror temperature and the observation value corresponding to the new hot mirror temperature time series data is:

[0021] ;

[0022] wherein N is the number of samples, a predicted hot mirror temperature value, an actual hot mirror temperature observation value.

[0023] Preferably, the neural network can also select a cold mirror fitting algorithm, an ARIMA algorithm, an LSTM algorithm, or a single-stage optimization LSTM algorithm.

[0024] Preferably, the method for fine-tuning the network model parameters using a new training set to update the network model comprises:

[0025] Fine-tuning the hyperparameters of the model using a parameter-based transfer learning algorithm.

[0026] Preferably, the method for identifying and correcting abnormal data between the predicted hot mirror temperature and the observation value corresponding to the new hot mirror temperature time series data comprises:

[0027] Calculate the difference between the predicted hot mirror temperature and the observation value corresponding to the new hot mirror temperature time series data, mark the observation value whose difference exceeds the error threshold as abnormal, and replace the observation value with the second segment of predicted hot mirror temperature to achieve abnormal data detection and correction.

[0028] Preferably, the method for dividing the new hot mirror temperature time series data into two segments according to a predetermined rule comprises:

[0029] The time length of the first segment of hot mirror temperature time series data and the second segment of hot mirror temperature time series data is consistent or inconsistent.

[0030] According to a second aspect of the present application, a remote sensing instrument data on-orbit anomaly detection and correction system is provided, the system comprising:

[0031] a training module, the training module being configured to train hot mirror temperature time series data;

[0032] a prediction module, the prediction module being configured to predict hot mirror temperature data according to the hot mirror temperature time series data;

[0033] an optimization module, the optimization module being configured to update the prediction module to reduce errors;

[0034] The identification module is configured to compare the predicted hot mirror temperature with an observed value corresponding to the hot mirror temperature time series data, identify and correct abnormal data.

[0035] According to a third aspect of the present application, a remote sensing instrument data on-orbit anomaly detection and correction device is provided, the device comprising:

[0036] one or more processors;

[0037] a memory device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the remote sensing instrument data on-orbit anomaly detection and correction method according to the first aspect.

[0038] Overall, compared with the prior art, the above technical solutions conceived by the present application have the following beneficial effects:

[0039] By merging the input gate and the forgetting gate of the LSTM into the update gate, the model is lightweight, and the training speed is improved; the time series data is segmented for prediction, and dynamic fine-tuning is performed in combination with transfer learning to improve the training accuracy; the hyperparameters in the model are adaptively adjusted in combination with the Bayesian method to reduce the influence of data changes. The model application effect is verified by using independent data in the normal period. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0041] Figure 1 is a remote sensing instrument data on-orbit anomaly detection and correction method flowchart provided by the present embodiment one;

[0042] Figure 2 is a normal state sequence hot mirror temperature graph of FY-3D / MWRI provided by the present embodiment one;

[0043] Figure 3 is an abnormal state sequence hot mirror temperature graph of FY-3D / MWRI provided by the present embodiment one;

[0044] Figure 4 is a remote sensing instrument data on-orbit anomaly detection and correction method on-orbit test effect diagram provided by the present embodiment one;

[0045] Figure 5is a schematic diagram of a one-day test effect of the on-orbit anomaly detection and correction method of remote sensing instrument data provided in embodiment one;

[0046] Figure 6 is a schematic diagram of the result time sequence of the temperature prediction value and the observation value difference of the five methods in embodiment one;

[0047] Figure 7 is a schematic diagram of the result time sequence of the temperature prediction value and the observation value difference of the five methods in embodiment one;

[0048] Figure 8 is a schematic diagram of the statistical result of the one-orbit deviation and standard deviation of the temperature data of the five methods in embodiment one;

[0049] Figure 9 is a schematic diagram of the statistical result of the one-day deviation and standard deviation of the temperature data of the five methods in embodiment one;

[0050] Figure 10 is a schematic diagram of an on-orbit anomaly detection and correction system of remote sensing instrument data provided in embodiment two;

[0051] Figure 11 is a schematic diagram of an on-orbit anomaly detection and correction device of remote sensing instrument data provided in embodiment three. DETAILED DESCRIPTION

[0052] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0053] Embodiment one:

[0054] The embodiment one provides an on-orbit anomaly detection and correction method of remote sensing instrument data, which comprises the following steps, as shown in Figure 1

[0055] S101: training a neural network by using historical normalized hot mirror temperature time sequence data to obtain an initial network model of predictable hot mirror temperature data.

[0056] In the embodiment one, the linear normalization method is adopted to calculate the minimum value and the maximum value of the hot mirror temperature. For each hot mirror temperature x, the normalized hot mirror temperature is obtained by applying the formula The maximum value and the minimum value of the hot mirror temperature in the time sequence data are statistically obtained. ​

[0057] In this embodiment, FY-3D / MWRI hot mirror temperature time series data is obtained, hot mirror temperature data is preprocessed, model training data is formed, an improved LSTM model is established, hot mirror temperature anomaly is identified and corrected, and the correction effect of hot mirror temperature prediction is verified.

[0058] In this embodiment, data from January 2020 to December 2020 (a total of 17568000 samples) are used as the training set, and data from 2021 are used as the test set for model verification. The long time series hot mirror temperature of FY-3D / MWRI is as shown in Figure 2 and Figure 3 , wherein Figure 2 is the normal state, Figure 3 is the hot mirror temperature failure state, as shown in Figure 3 The upper half of the data shows periodic variation characteristics.

[0059] Figure 4 and Figure 5 is the verification result of the data in the non-failure period of 2021, and the blue line is the predicted value and the red line is the observed value. Through calculation, the difference between the predicted value and the measured value is stable within 2K in a long time period, the daily MAE is 0.255K, the monthly MAE is 0.523K, and the annual MAE is 0.637K, indicating that the initial network model for predicting hot mirror temperature data can accurately predict the hot mirror temperature based on historical data. One orbit refers to the time from the North Pole to the North Pole of the polar orbit satellite, and one day refers to 24 hours.

[0060] S102: The new hot mirror temperature time series data is divided into two segments according to a preset rule, and the first segment of the hot mirror temperature time series data is predicted by using the initial network model to obtain the first segment of the predicted temperature.

[0061] S103: The hot mirror temperature historical time series data and the first segment of the hot mirror temperature time series data are merged to obtain a new training set, and the initial network model parameters are fine-tuned to update the network model using the new training set.

[0062] S104: The second segment of the hot mirror temperature time series data is predicted by using the updated network model to obtain the second segment of the predicted temperature.

[0063] S105: The first segment of the predicted temperature and the second segment of the predicted temperature are merged into the predicted hot mirror temperature, and the predicted hot mirror temperature is compared with the observed value corresponding to the new hot mirror temperature time series data to identify and correct the abnormal data in the new hot mirror temperature sequence.

[0064] In the first embodiment, considering the influence of computing power and efficiency, the time series data is divided into two segments, and in actual use, the historical time series data can be divided into N segments, the hot mirror temperature of the first segment is predicted first, the historical data + the first segment data is used as a new training set to train the original model, the original model is optimized, iterated and updated to obtain a first updated model, the hot mirror temperature of the second segment is predicted by using the first updated model, the historical data + the first segment data + the second segment data is used as a new training set to train the first updated model to obtain a second updated model, the hot mirror temperature of the third segment is predicted by using the second updated model, and so on. The model will dynamically change with the increase of data to improve the model effect and make the anomaly detection more accurate.

[0065] In combination with the embodiment of the application, there is also a preferred implementation scheme, specifically, the method for comparing the predicted hot mirror temperature with the observation value corresponding to the new hot mirror temperature time series data comprises:

[0066] The root mean square error and the mean absolute error of the predicted hot mirror temperature and the observation value corresponding to the new hot mirror temperature time series data are calculated;

[0067] The model parameters corresponding to the minimum value of the root mean square error and the mean absolute error are selected as the parameters of the network model.

[0068] In combination with the embodiment of the application, there is also a preferred implementation scheme, specifically, the calculation method of the root mean square error of the predicted hot mirror temperature and the observation value corresponding to the new hot mirror temperature time series data is:

[0069] ;

[0070] Wherein, N is the sample number, is the predicted hot mirror temperature, and is the actual hot mirror temperature observation value.

[0071] In combination with the embodiment of the application, there is also a preferred implementation scheme, specifically, the calculation method of the mean absolute error of the predicted hot mirror temperature and the observation value corresponding to the new hot mirror temperature time series data is:

[0072] ;

[0073] Wherein, N is the sample number, is the predicted hot mirror temperature, and is the actual hot mirror temperature observation value.

[0074] In combination with the embodiment of the application, there is also a preferred implementation scheme, specifically, the neural network can also select a cold mirror fitting algorithm, an ARIMA algorithm, an LSTM algorithm or a single-stage optimization LSTM algorithm.

[0075] The LSTM recurrent neural network comprises a module with a memory cell, which can learn the characteristics in the time domain of data and has good processing performance for time series data. The LSTM comprises three units: an input gate, a forgetting gate and an output gate, which respectively control the input, update and output of information, so that the network has a certain memory function, but also makes the network learning parameters numerous and slows down the network training speed. To shorten the model training time and the real-time performance of data prediction, the LSTM recurrent neural network model is simplified in this embodiment, and a variant LSTM recurrent neural network is proposed. The forgetting gate and the input gate in the traditional LSTM recurrent neural network are combined into an update gate, and the update gate uses a Sigmoid layer to update information, that is, how much proportion of information is retained in the memory cell at t 1 time is supplemented at t time, which ensures that the output at t time is fairly affected by the states at t 1 time and t time, so that the memory cell always maintains an effective state.

[0076] The network model in this embodiment is compared with the cold mirror fitting algorithm, the ARIMA algorithm, the traditional LSTM algorithm and the single-stage optimized LSTM algorithm by using the partially failed FY-3D / MWRI hot mirror temperature data. Among them, the cold mirror fitting algorithm establishes the hot mirror-cold mirror temperature mapping relationship through linear regression; the ARIMA algorithm is an autoregressive moving average model, and the parameters are optimized by AIC criterion; the traditional LSTM algorithm is a standard long short-term memory network, which sets 2 hidden layers (128 units per layer), a learning rate of 0.001 and Dropout=0.2; the single-stage optimized LSTM algorithm is based on the traditional LSTM and introduces Bayesian hyperparameter optimization, which searches for the optimal learning rate (0.0005~0.01) and batch size (32~256) through Gaussian process.

[0077] Figure 6 And Figure 7 The temperature prediction values of the two-stage algorithm, the cold mirror fitting algorithm, the ARIMA algorithm, the traditional LSTM algorithm and the single-stage optimized LSTM algorithm are given, and the difference between the measured values and the result time series graphs of one day and one day are given. Table 1 lists the corresponding RMSE and MAE statistical values. The results in Table 1 show that the RMSE and MSE of the two-stage LSTM algorithm in this embodiment are the smallest for one day, one month and one year, which proves that the calculation result of the two-stage LSTM algorithm in this embodiment has higher accuracy.

[0078] Figure 8 And Figure 9The statistical results of the one-orbit deviation and one-day deviation and standard deviation of the temperature data of each algorithm are given. It can be seen that, on the orbit and day scales, the deviation and standard deviation of ARIMA are larger, while those of the LSTM-based algorithm are relatively smaller; on the day scale, the two-stage model in Embodiment I is better than other LSTM algorithms, showing advantages in long-period prediction.

[0079] In combination with the embodiments of the present application, there is also a preferred implementation scheme, specifically, the method for updating the network model by fine-tuning the network model parameters with a new training set comprises:

[0080] The hyperparameters of the model are fine-tuned by using a parameter-based transfer learning algorithm.

[0081] In combination with the embodiments of the present application, there is also a preferred implementation scheme, specifically, the method for identifying and correcting abnormal data by using the predicted hot mirror temperature and the observation value corresponding to the new hot mirror temperature time series data comprises:

[0082] The difference between the predicted hot mirror temperature and the observation value corresponding to the new hot mirror temperature time series data is calculated, the second segment of hot mirror temperature whose difference exceeds the error threshold is marked as abnormal, and the abnormal value is replaced with the second segment of predicted hot mirror temperature, so as to realize abnormal data detection and correction.

[0083] Considering the instability and variability of actual data and the long period of prediction, in order to improve the accuracy of prediction, the prediction time sequence is divided into two segments. In combination with the embodiments of the present application, there is also a preferred implementation scheme, specifically, the method for dividing the new hot mirror temperature time series data into two segments according to a preset rule comprises:

[0084] The time length of the first segment of hot mirror temperature time series data and the second segment of hot mirror temperature time series data is consistent or inconsistent.

[0085] Embodiment I realizes the detection and repair of hot mirror temperature on-orbit data anomalies, and improves the quality of affected data observation brightness temperature. In addition to model lightweight improvement to improve training efficiency, in order to adapt to the change of on-orbit data over time and ensure the reliability of long-period prediction, a parameter update based on Bayesian network and a segmented iterative prediction mechanism are introduced. The verification results based on independent data show that, compared with other methods, the effect of the model of the present research is the best, which can improve the stability and usability of on-orbit brightness temperature observation data. The method in Embodiment I can also be applied to other fields of abnormal detection and repair based on time series data and data reconstruction, to ensure data quality.

[0086] Embodiment II:

[0087] Embodiment II provides an on-orbit anomaly detection and correction system for remote sensing instrument data, which comprises a data preprocessing module, a model training module, a model prediction module, a data correction module and a data output module.Figure 10 As shown in the figure, the system comprises:

[0088] a training module for training the hot mirror temperature time series data;

[0089] a prediction module for predicting the hot mirror temperature data according to the hot mirror temperature time series data;

[0090] an optimization module for updating the prediction module to reduce the error;

[0091] an identification module for comparing the predicted hot mirror temperature with the observed value corresponding to the hot mirror temperature time series data, identifying and correcting abnormal data.

[0092] Embodiment three:

[0093] In this embodiment three, a remote sensing instrument data on-orbit anomaly detection and correction device is provided, such as Figure 11 As shown in the figure, the device comprises:

[0094] one or more processors;

[0095] a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the remote sensing instrument data on-orbit anomaly detection and correction method as any one of the embodiments one.

[0096] Figure 11 The remote sensing instrument data on-orbit anomaly detection and correction device structure diagram provided in this embodiment three is shown. Figure 11 The example remote sensing instrument data on-orbit anomaly detection and correction device block diagram suitable for realizing the embodiment of the present application is shown. Figure 11 The remote sensing instrument data on-orbit anomaly detection and correction device shown is only an example, and should not bring any limitation to the function and use range of the embodiment of the present application.

[0097] As shown in the figure Figure 11 The remote sensing instrument data on-orbit anomaly detection and correction device is in the form of a general device. The components of the remote sensing instrument data on-orbit anomaly detection and correction device can include but are not limited to one or more processors or processing units, memory, bus connecting different system components including memory and processing unit.

[0098] The term "bus" refers to one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures. Examples of these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0099] The on-orbit anomaly detection and correction device for remote sensing instrument data typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the device for correcting the intelligent logging interpretation model, including volatile and non-volatile media, removable and non-removable media.

[0100] The memory may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory. The remote sensing instrument data on-orbit anomaly detection and correction device may further include other removable / non-removable, volatile / non-volatile computer system storage media. For example only, the storage system may be used to read and write non-removable, non-volatile magnetic media ( Figure 11 Not shown, often called a "hard drive"). Although Figure 11 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical medium) may be provided. In these cases, each drive may be connected to the bus via one or more data medium interfaces. The memory may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of various embodiments of the present invention.

[0101] A program / utility having a set (at least one) of program modules, which may be stored, for example, in a memory, includes, but is not limited to, an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may include an implementation of a network environment. The program modules generally implement the functions and / or methods of the embodiments described herein.

[0102] The remote-sensor-instrument data-on-orbit anomaly detection and correction device can also communicate with one or more external devices such as a keyboard or a pointing device, a display, etc. The remote-sensor-instrument data-on-orbit anomaly detection and correction device can also communicate with one or more devices that enable a user to interact with the remote-sensor-instrument data-on-orbit anomaly detection and correction device, and / or with any devices (e.g., a network card, a modem, etc.) that enable the remote-sensor-instrument data-on-orbit anomaly detection and correction device to communicate with one or more other devices. Such communication can occur via Input / Output (I / O) interfaces. Still yet, the remote-sensor-instrument data-on-orbit anomaly detection and correction device can communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or the Internet) through a network adapter. As Figure 11 illustrated in FIG. 10, a network adapter 1020 communicates with the other components of the remote-sensor-instrument data-on-orbit anomaly detection and correction device via bus. It should be appreciated that although not shown, other hardware and / or software components could be used in conjunction with the remote-sensor-instrument data-on-orbit anomaly detection and correction device. Such components include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0103] The processing unit executes various functions applications and data processing by running programs stored in the memory, such as implementing the remote-sensor-instrument data-on-orbit anomaly detection and correction method provided by any of the embodiments of the present application.

[0104] The above descriptions are only the preferred embodiments of the present application, not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the scope of the present application.

Claims

1. A method for on-orbit anomaly detection and correction of remote sensing instrument data, characterized in that, The method comprises: training a neural network using historical normalized hot mirror temperature time series data to obtain an initial network model for predicting hot mirror temperature data; dividing new hot mirror temperature time series data into two segments according to a preset rule, and using the initial network model to predict the first segment of hot mirror temperature time series data to obtain a first segment of predicted temperature; merging the hot mirror temperature historical time series data and the first segment of hot mirror temperature time series data to obtain a new training set, and fine-tuning the initial network model parameters using the new training set to update the network model; using the updated network model to predict the second segment of hot mirror temperature time series data to obtain a second segment of predicted temperature; merging the first segment of predicted temperature and the second segment of predicted temperature into predicted hot mirror temperature, comparing the predicted hot mirror temperature with the observed value corresponding to the new hot mirror temperature time series data, identifying abnormal data in the new hot mirror temperature sequence and correcting the abnormal data.

2. The method for on-orbit anomaly detection and correction of remote sensing instrument data according to claim 1, wherein, The method further comprises: calculating the root mean square error and the average absolute error of the predicted hot mirror temperature and the observed value corresponding to the new hot mirror temperature time series data; selecting the model parameters corresponding to the minimum values of the root mean square error and the average absolute error as the parameters of the network model.

3. The method for on-orbit anomaly detection and correction of remote sensing instrument data according to claim 2, wherein, The calculation method of the root mean square error of the predicted hot mirror temperature and the observed value corresponding to the new hot mirror temperature time series data is: ； where N is the number of samples, is the predicted hot mirror temperature, is the actual hot mirror temperature observation.

4. The method for on-orbit anomaly detection and correction of remote sensing instrument data according to claim 3, characterized in that, The calculation method of the average absolute error of the predicted hot mirror temperature and the observed value corresponding to the new hot mirror temperature time series data is: ； where N is the number of samples, is the predicted hot mirror temperature, is the actual hot mirror temperature observation.

5. The method for on-orbit anomaly detection and correction of remote sensing instrument data according to claim 4, wherein, The neural network can also use a cold mirror fitting algorithm, an ARIMA algorithm, an LSTM algorithm or a single-stage optimization LSTM algorithm.

6. The method for on-orbit anomaly detection and correction of remote sensing instrument data according to claim 5, wherein, The method of fine-tuning the network model parameters using the new training set to update the network model comprises: using a parameter-based transfer learning algorithm to fine-tune the hyperparameters of the model.

7. The method for on-orbit anomaly detection and correction of remote sensing instrument data according to claim 5, wherein, The method of identifying and correcting abnormal data of the predicted hot mirror temperature and the observed value corresponding to the new hot mirror temperature time series data comprises: calculating the difference between the predicted hot mirror temperature and the observed value corresponding to the new hot mirror temperature time series data, marking the observed value whose difference exceeds the error threshold as abnormal, and replacing the observed value with the second segment of predicted hot mirror temperature to realize abnormal data detection and correction.

8. The method for on-orbit anomaly detection and correction of remote sensing instrument data according to claim 1, wherein, The method of dividing the new hot mirror temperature time series data into two segments according to a preset rule comprises: The time length of the first segment of hot mirror temperature time series data and the second segment of hot mirror temperature time series data is consistent or inconsistent.

9. An on-orbit anomaly detection and correction system for remote sensing instrument data, the system comprising: The system comprises: a training module for training hot mirror temperature time series data; a prediction module for predicting hot mirror temperature data according to hot mirror temperature time series data; an optimization module for updating the prediction module to reduce errors; an identification module for comparing the predicted hot mirror temperature with the observed value corresponding to the hot mirror temperature time series data, identifying and correcting abnormal data.

10. An on-board anomaly detection and correction device for remote sensing instrument data, the device comprising: comprises: one or more processors; a storage device for storing one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method for on-orbit anomaly detection and correction of remote sensing instrument data according to any one of claims 1-8.

Images