Satellite anomaly detection system and method

Abstract

The invention provides a satellite anomaly detection system and method, and relates to the technical field of satellite telemetry, the system comprises a data dimension reduction module and an anomaly detection module, the data dimension reduction module and the anomaly detection module are deployed on a satellite-borne detector of a satellite; the data dimension reduction module is used for acquiring a plurality of telemetering parameters of the satellite subsystem and screening n optimal characteristic parameters from the telemetering parameters; and the anomaly detection module is used for predicting the parameter values of the n optimal characteristic parameters through a first time sequence prediction model locally deployed by the satellite-borne detector to obtain respective predicted values of the n optimal characteristic parameters, and carrying out anomaly detection on the satellite based on the predicted values to obtain an anomaly detection result. Through localization processing and intelligent analysis, the precision, real-time performance and reliability of satellite anomaly detection are greatly improved, and safe operation of the satellite is effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of satellite telemetry technology, and in particular to a satellite anomaly detection system and method. Background Art

[0002] With the rapid development of commercial aerospace, the number of satellites and the amount of data have increased dramatically, especially the deployment scale of low-orbit satellite networks has continued to expand. This trend has put forward higher requirements for satellite health management and fault diagnosis.

[0003] However, in low-orbit satellite systems, due to the limited computing power of satellites, complex real-time analysis cannot be performed. Traditional satellite anomaly detection methods usually rely on ground data processing and analysis. It is difficult to meet the needs when faced with huge and complex telemetry data. The satellite's telemetry data needs to be transmitted from the satellite to the ground station for analysis, and then the ground station feeds the analysis results back to the satellite. This process not only has a long data transmission time, but is also limited by the communication bandwidth between the satellite and the ground station. Due to these delays, the occurrence of anomalies cannot be discovered and processed in time, resulting in the satellite may not be able to get the necessary adjustment or repair instructions in the first time, which may cause more serious system failures and even affect the normal operation of the entire satellite. Summary of the invention

[0004] The present invention provides a satellite anomaly detection system and method, aiming to solve the problems existing in the above-mentioned background technology.

[0005] In order to solve the above-mentioned technical problems, the present invention is achieved as follows: In a first aspect, the present invention provides a satellite anomaly detection system, the system comprising a data dimension reduction module and an anomaly detection module, the data dimension reduction module and the anomaly detection module being deployed on an onboard detector of a satellite; The data dimension reduction module is used to obtain multiple telemetry parameters of the satellite subsystem, and screen out n optimal feature parameters from the multiple telemetry parameters, wherein the correlation between the n optimal feature parameters and the preset abnormal type parameters is the largest, and the redundancy between any two optimal feature parameters among the n optimal feature parameters is the smallest, and n is an integer greater than or equal to 2; The anomaly detection module is used to predict the parameter values ​​of the n optimal characteristic parameters through the first time series prediction model locally deployed by the onboard detector, obtain the predicted values ​​of each of the n optimal characteristic parameters, and perform anomaly detection on the satellite based on the predicted values ​​to obtain anomaly detection results.

[0006] Optionally, the anomaly detection module is used to: Determining the prediction error of each optimal feature parameter according to the difference between the current actual value and the predicted value of each of the n optimal feature parameters; Based on a preset smoothing coefficient, a current error threshold is calculated according to the prediction error and the error threshold at a previous moment; When the prediction error of any optimal characteristic parameter exceeds the current error threshold, the satellite anomaly is determined and a corresponding anomaly detection result is output.

[0007] Optionally, the anomaly detection module is used to use the current actual value of each of the n optimal feature parameters and the historical actual value within a preset time window as input data of the first time series prediction model, and obtain the predicted value of each of the n optimal feature parameters output by the first time series prediction model according to the input data, and the first time series prediction model is a long short-term memory network; Outputting predicted values ​​of the n optimal feature parameters based on their respective current actual values ​​and historical actual values ​​through the forget gate, input gate and output gate of the long short-term memory network; The system further comprises a model updating module, which is deployed on a ground computing platform; the satellite-borne detector is used to send the abnormality detection result and the current actual value of each of the multiple telemetry parameters of the satellite's subsystem to the computing platform; The model updating module is used to use the current actual values ​​of each of the multiple telemetry parameters and the anomaly detection results as training data sets to train the second time series prediction model locally deployed on the computing platform until a preset training batch is reached, wherein the anomaly detection results are used as labels of the training data sets; and send model parameters of the second time series prediction model trained to the training batch to the satellite; The anomaly detection module is used to update the model parameters of the first time series prediction model according to the model parameters of the second time series prediction model.

[0008] Optionally, the model updating module is used to use mean square error as a loss function to quantify the deviation between the predicted value of the second time series prediction model for each of the multiple telemetry parameters and the current actual value of each of the multiple telemetry parameters, and adjust the model parameters of the second time series prediction model through the Adam optimization algorithm to minimize the loss function.

[0009] Optionally, the satellite subsystem includes an attitude control subsystem, and the abnormality type parameters include parameters corresponding to communication abnormalities and parameters corresponding to power supply abnormalities; The data dimension reduction module is used to calculate the correlation between the telemetry parameters of the various components of the attitude control system, and to calculate the redundancy between the telemetry parameters of the various components of the attitude control system and other telemetry parameters except the telemetry parameters; and to take the n telemetry parameters with the maximum correlation and the minimum redundancy as the optimal feature parameters.

[0010] Optionally, the system further comprises an onboard computer and a control module deployed on the satellite; The onboard computer is used to obtain the abnormality detection result and generate a corresponding control instruction according to the abnormality detection result, wherein the control instruction is used to instruct the satellite to perform abnormality processing; The control module is used to execute the control instruction.

[0011] In a second aspect, the present invention provides a satellite anomaly detection method, which is applied to the satellite anomaly detection system as described in the first aspect, and the method comprises: Acquire multiple telemetry parameters of the satellite subsystem, and select n optimal characteristic parameters from the multiple telemetry parameters, wherein the correlation between the n optimal characteristic parameters and the preset abnormal type parameters is the largest, and the redundancy between any two optimal characteristic parameters among the n optimal characteristic parameters is the smallest, and n is an integer greater than or equal to 2; The first time series prediction model locally deployed by the satellite's onboard detector is used to predict the parameter values ​​of the n optimal characteristic parameters to obtain the predicted values ​​of the n optimal characteristic parameters, and the satellite is subjected to anomaly detection based on the predicted values ​​to obtain anomaly detection results.

[0012] Optionally, the performing anomaly detection on the satellite based on the predicted value includes: Determining the prediction error of each optimal feature parameter according to the difference between the current actual value and the predicted value of each of the n optimal feature parameters; Based on a preset smoothing coefficient, a current error threshold is calculated according to the prediction error and the error threshold at a previous moment; When the prediction error of any optimal characteristic parameter exceeds the current error threshold, the satellite anomaly is determined and a corresponding anomaly detection result is output.

[0013] Optionally, the first time series prediction model locally deployed by the onboard detector of the satellite predicts the parameter values ​​of the n optimal characteristic parameters, including: Using the current actual value of each of the n optimal feature parameters and the historical actual value within a preset time window as input data of the first time series prediction model, and obtaining the predicted value of each of the n optimal feature parameters output by the first time series prediction model according to the input data, wherein the first time series prediction model is a long short-term memory network; Outputting predicted values ​​of the n optimal feature parameters based on their respective current actual values ​​and historical actual values ​​through the forget gate, input gate and output gate of the long short-term memory network; The method further comprises: Sending the anomaly detection result and the current actual value of each of the plurality of telemetry parameters of the satellite subsystem to a computing platform deployed on the ground; Using the current actual values ​​of each of the multiple telemetry parameters and the anomaly detection results as training data sets, training the second time series prediction model locally deployed on the computing platform until a preset training batch is reached, wherein the anomaly detection results serve as labels of the training data sets; sending model parameters of the second time series prediction model trained to the training batch to the satellite; According to the model parameters of the second time series prediction model, the model parameters of the first time series prediction model are correspondingly updated.

[0014] Optionally, the training of the second time series prediction model locally deployed on the computing platform includes: The mean square error is used as the loss function to quantify the deviation between the predicted value of the second time series prediction model for each of the multiple telemetry parameters and the current actual value of each of the multiple telemetry parameters, and the model parameters of the second time series prediction model are adjusted through the Adam optimization algorithm to minimize the loss function.

[0015] The technical solution provided by the present invention brings at least the following beneficial effects: The present invention performs dimensionality reduction processing on the satellite's telemetry parameters based on correlation and redundancy, so that the abnormality prediction work can be carried out and performed on the satellite, avoiding the delay problem of data transmission to the ground station, and realizing the real-time abnormality monitoring and fault handling of the satellite. By screening out the optimal characteristic parameters most relevant to the abnormality type and minimizing redundancy, the utilization of computing resources is optimized, the computational complexity is reduced, and the detection efficiency is improved. At the same time, the changing trend of the satellite's key parameters can be accurately predicted, the accuracy of abnormality detection is improved, and false alarms and missed alarms are reduced. Overall, the present invention greatly improves the accuracy, real-time and reliability of satellite abnormality detection through localized processing and intelligent analysis, and effectively ensures the safe operation of the satellite. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 is a schematic diagram of the architecture of a satellite anomaly detection system provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of an application framework of a satellite anomaly detection system provided by an embodiment of the present invention; Figure 3 The figure is a schematic diagram of the steps of a satellite anomaly detection method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] With the rapid increase in the number of satellites and the amount of data, traditional satellite anomaly detection methods mainly rely on real-time monitoring and manual analysis by ground personnel, and have problems such as low automation level and insufficient efficiency. At the same time, related anomaly detection technologies often ignore the correlation between the parameters of various satellite subsystems, resulting in insufficient anomaly detection accuracy and difficulty in timely detection of potential hidden faults. In addition, traditional methods rely on manually set thresholds, which are prone to false alarms or missed alarms and cannot cope with complex abnormal patterns. In response to the above problems, the present invention proposes an intelligent satellite anomaly detection system based on data dimensionality reduction and time series prediction models. By deploying data dimensionality reduction modules and anomaly detection modules locally on the satellite, the key parameters most relevant to the anomaly type can be automatically screened out, and real-time anomaly detection can be performed through a time series prediction model, thereby effectively improving the detection accuracy, real-time performance and the system's adaptability, and solving the above defects.

[0020] Figure 1 FIG. 1 is a schematic diagram of the architecture of a satellite anomaly detection system provided by an embodiment of the present invention. Figure 1 As shown, the system includes a data dimension reduction module and an anomaly detection module, and the data dimension reduction module and the anomaly detection module are deployed on a satellite-borne detector.

[0021] The onboard detector is an integrated hardware and software system installed on the satellite, responsible for real-time monitoring and analysis of telemetry data during satellite operation. The onboard detector is used to perform abnormal detection of various satellite subsystems and generate control instructions based on the detection results to ensure the healthy status of the satellite in orbit.

[0022] The data dimension reduction module is used to obtain multiple telemetry parameters of the satellite subsystem and screen out n optimal feature parameters from the multiple telemetry parameters, wherein the correlation between the n optimal feature parameters and the preset abnormal type parameters is the largest, and the redundancy between any two of the n optimal feature parameters is the smallest, and n is an integer greater than or equal to 2.

[0023] The data dimension reduction module is a module on the satellite-borne detector that processes the numerous telemetry parameters collected by the satellite-borne detector. The data dimension reduction module selects a set of the most relevant and less redundant parameters from multiple telemetry parameters. Telemetry parameters refer to real-time data collected by sensors during the satellite's in-orbit operation. Telemetry parameters reflect the health status and operation of each satellite subsystem. Telemetry parameters include but are not limited to physical quantities such as temperature, pressure, voltage, speed, attitude angle, and communication signal strength. Suppose the satellite subsystem set is , each subsystem The corresponding telemetry parameter set is . For example, the thermal control system on the satellite is one of the subsystems responsible for regulating the temperature of various components inside the satellite to prevent overheating or overcooling. Then, temperature, voltage and current can be the telemetry parameters corresponding to the thermal control system. The abnormal type parameter refers to the indicative data or fault state corresponding to the telemetry parameter in the satellite subsystem when an abnormality occurs. Taking the thermal control system as an example, the abnormal type parameter can be a value that characterizes temperature abnormality or sensor failure. When the abnormal type parameter value is 1, it can characterize temperature abnormality or sensor failure. When the abnormal type parameter value is 0, it can characterize the normal state, that is, no abnormality is detected.

[0024] Since the amount of telemetry parameter data is very large and complex, it is not practical to process all the parameters. For multiple telemetry parameters of the target subsystem, an embodiment of the present invention proposes to use a data dimension reduction module to select the most important set of telemetry parameters from all telemetry parameters, which is equivalent to dimensionality reduction processing for a large number of telemetry parameters. This embodiment calculates the correlation between each telemetry parameter and the abnormal type parameter by using mutual information, with the goal of selecting those telemetry parameters that can effectively indicate the occurrence of anomalies, thereby improving the accuracy of anomaly detection. In addition, the redundancy between these telemetry parameters must also be considered. If some telemetry parameters are highly correlated, they may convey the same information, thereby reducing the effect of feature selection. At this stage, by calculating the redundancy of each telemetry parameter to other telemetry parameters (i.e., mutual information), the telemetry parameter with the least redundancy is selected, effectively reducing the computational overhead and improving the efficiency of model training.

[0025] In an optional embodiment, the satellite subsystem includes an attitude control subsystem, and the abnormality type parameters include parameters corresponding to communication abnormalities and parameters corresponding to power supply abnormalities; the data dimension reduction module is used to calculate the correlation between telemetry parameters of each component of the attitude control system, and to calculate the redundancy between the telemetry parameters of each component of the attitude control system and other telemetry parameters except the telemetry parameters; the n telemetry parameters with the largest correlation and the smallest redundancy are used as the optimal characteristic parameters.

[0026] The correlation between telemetry parameters can be measured by mutual information (MI). Mutual information measures the dependency between two random variables, i.e., the For another variable The amount of information. The formula is:

[0027] In the formula, and is a variable and The feature set to which each belongs; yes and The joint probability distribution of , that is, the probability of both occurring at the same time; and They are and The marginal probability distribution of is the probability of each occurrence separately.

[0028] In this embodiment, the greater the mutual information, the stronger the correlation between the two telemetry parameters, that is, they share more information. Conversely, the smaller the mutual information, the weaker the information dependence between the two. The mutual information between each pair of telemetry parameters in can be used to find the telemetry parameter pairs with greater correlation.

[0029] Redundancy refers to a set of parameters where some parameters provide very little additional information or are repeated. Redundancy is measured by the mutual information between parameters. When the mutual information between two parameters is high, there may be redundancy between them. Redundancy can be expressed by the following formula:

[0030] In the formula, Representation feature set Redundancy; is a parameter and parameters The mutual information between It is a feature set The higher the redundancy, the more repeated information there is between different parameters in the feature set. The smaller the redundancy, the more independent information each parameter provides.

[0031] In this embodiment, the goal of the data dimension reduction module is to select those telemetry parameters with strong correlation but small redundancy as the optimal feature parameters. The selection method is to use the Maximum Relevance Minimal Redundancy (mRMR) algorithm. Specifically, the optimization goal of the mRMR algorithm is to find the optimal feature subset in the telemetry parameter set of the satellite subsystem, so that the correlation between the optimal feature subset and the abnormal type parameter is the largest, and the redundancy between any two telemetry parameters in the optimal feature subset is the smallest, so as to obtain The specific formula is as follows:

[0032] In the formula, is a parameter With the target variable The mutual information between is a parameter and The mutual information between It is a feature set The number of telemetry parameters in .

[0033] In this embodiment, the parameters That is the Telemetry parameters, parameters That is the telemetry parameters, target variables Exception type parameter.

[0034] According to the above steps, n optimal feature parameters are finally selected from multiple telemetry parameters, which is equivalent to obtaining an optimal feature subset. , yes A subset of the optimal feature subset The value of is the largest. These optimal feature parameters not only have a strong correlation with the anomaly type parameters, but also have the least redundancy with each other, and are the most important set of telemetry parameters. The value of n is an integer greater than or equal to 2, and at least 2 optimal feature parameters need to be selected for subsequent anomaly detection.

[0035] The anomaly detection module is used to predict the parameter values ​​of the n optimal characteristic parameters through the first time series prediction model locally deployed by the onboard detector, obtain the predicted values ​​of each of the n optimal characteristic parameters, and perform anomaly detection on the satellite based on the predicted values ​​to obtain anomaly detection results.

[0036] The anomaly detection module is used to monitor the operating status of the satellite in real time and perform anomaly detection based on the determined optimal feature subset. The first time series prediction model is a model deployed locally by the onboard detector to predict future values ​​based on historical data. The first time series prediction model has been pre-trained and can predict the expected value of each optimal feature parameter at a future time point. During the satellite anomaly detection process, the n optimal feature parameters are input into the first time series prediction model in the form of time series data, and their changing trends are closely related to the health status of each satellite subsystem.

[0037] The time series prediction model is used to predict n optimal characteristic parameters, and the predicted value of each optimal characteristic parameter at a future moment is obtained. By comparing the current actual value of the actually collected telemetry parameter with the predicted value, it is detected whether there is a significant deviation to determine whether the subsystem has a fault or abnormality.

[0038] The present invention performs dimensionality reduction processing on the satellite's telemetry parameters based on correlation and redundancy, so that the abnormality prediction work can be carried out and performed on the satellite, avoiding the delay problem of data transmission to the ground station, and realizing the real-time abnormality monitoring and fault handling of the satellite. By screening out the optimal characteristic parameters most relevant to the abnormality type and minimizing redundancy, the utilization of computing resources is optimized, the computational complexity is reduced, and the detection efficiency is improved. At the same time, the changing trend of the satellite's key parameters can be accurately predicted, the accuracy of abnormality detection is improved, and false alarms and missed alarms are reduced. Overall, the present invention greatly improves the accuracy, real-time and reliability of satellite abnormality detection through localized processing and intelligent analysis, and effectively ensures the safe operation of the satellite.

[0039] In an optional implementation, the anomaly detection module is used to: The prediction error of each optimal feature parameter is determined according to the difference between the current actual value and the predicted value of each of the n optimal feature parameters.

[0040] The current actual value refers to the parameter value of the telemetry parameter actually measured at a certain moment, such as the current temperature, voltage, pressure, etc. For each optimal characteristic parameter, the difference between the current actual value and the predicted value is calculated as the prediction error. The calculation formula is as follows:

[0041] In the formula, for The prediction error at the time (in this embodiment, Time refers to the current time); For the The optimal characteristic parameters are The current actual value at the moment; For the The optimal characteristic parameters are The predicted value at time.

[0042] The prediction error obtained by calculating the mean square error is a numerical value that measures the size of the error. The larger the value, the more significant the difference between the current actual value and the predicted value, which means that there may be an abnormal risk.

[0043] Based on a preset smoothing coefficient, a current error threshold is calculated according to the prediction error and the error threshold at a previous moment.

[0044] After the prediction error is calculated, the error threshold at the current moment is dynamically adjusted based on the preset smoothing coefficient and the error threshold at the previous moment. The calculation formula for the error threshold is as follows:

[0045] In the formula, is the smoothing coefficient; for Moment (i.e. The prediction error of the previous moment (the moment before the moment); For the The optimal characteristic parameters are The predicted value at time.

[0046] The previous moment error threshold refers to the error threshold of the previous moment, that is, the error threshold calculated at the previous moment. The current prediction error refers to the mean square error calculated at the current moment (as shown in the previous formula). Through the adaptive threshold adjustment mechanism, the error threshold can be updated in a continuous time cycle to make appropriate responses during the monitoring process.

[0047] When the prediction error of any optimal characteristic parameter exceeds the current error threshold, the satellite anomaly is determined and a corresponding anomaly detection result is output.

[0048] The prediction error of each optimal feature parameter is compared with the current error threshold. If the prediction error of an optimal feature parameter exceeds the current error threshold, the optimal feature parameter is judged to be abnormal, and the corresponding abnormality detection result is generated, including the abnormality type (such as abnormal temperature, battery voltage is too high, etc.), location (for example: a specific subsystem of the satellite fails) and recommended treatment measures (such as starting the backup system, alarming, requesting further diagnosis, etc.).

[0049] In an optional embodiment, the anomaly detection module is used to use the current actual value of each of the n optimal feature parameters and the historical actual value within a preset time window as input data of the first time series prediction model, and obtain the predicted value of each of the n optimal feature parameters output by the first time series prediction model based on the input data, and the first time series prediction model is a long short-term memory network.

[0050] Through the forget gate, input gate and output gate of the long short-term memory network, based on the current actual value and historical actual value of each of the n optimal feature parameters, the predicted value of each of the n optimal feature parameters is output.

[0051] Long Short-Term Memory (LSTM) is a special recursive neural network used to process and predict time series data. The LSTM controls the flow of information through forget gates, input gates, and output gates, solving the gradient vanishing and explosion problems of traditional recursive neural networks in long time series prediction. The following is a detailed description of the core structure of the LSTM: The forget gate determines which information should be discarded (forgotten). Its calculation formula is:

[0052] In the formula, is the output of the forget gate (a value between 0 and 1), indicating how much past information is retained; is the weight matrix of the forget gate; is the current actual value of the optimal feature parameter; is the hidden state at the previous moment; is the bias term; is the sigmoid activation function.

[0053] The input gate controls what new information is stored in the cell state. Its calculation formula is:

[0054] In the formula, is the output of the input gate; is the weight matrix of the input gate; is the bias term.

[0055] The candidate cell state determines which new information will be added to the cell state. The calculation formula is:

[0056] In the formula, is the candidate cell state (representing new information); is the weight matrix; is the bias term; is the hyperbolic tangent activation function.

[0057] Combine the output of the forget gate, the output of the input gate and the candidate cell state to calculate the new cell state. The update formula is:

[0058] In the formula, It is the cell state at the current moment, which contains all the memory information in the long short-term memory network; is the cell state at the previous moment.

[0059] The output gate determines which information will be output to the hidden state at the next moment. Its calculation formula is:

[0060] In the formula, is the output of the output gate; is the weight matrix of the output gate; is the bias term.

[0061] Through the cell state and the output of the output gate, the hidden state at the current moment is obtained , and output the optimal feature parameters The predicted value of . Its calculation formula is:

[0062] In the formula, It is the hidden state at the current moment (i.e., the output of the LSTM network), which indicates the LSTM network’s understanding of the input information.

[0063] In this embodiment, the current actual values ​​of n optimal feature parameters and their historical actual values ​​are used as inputs to the long short-term memory network. The historical actual values ​​are the actual values ​​of the telemetry parameters collected in the past, within a preset time window. Assuming that there are multiple historical actual values ​​of the optimal feature parameters, these data can be expanded by time step and input into the long short-term memory network as a time series for training and prediction. Specifically, the current actual value and the historical actual value of each optimal feature parameter are Right now ,in, is the current actual value of the optimal characteristic parameter, is the time step of the preset time window. The predicted value of the optimal feature parameter is also the hidden state The value of .

[0064] In an optional embodiment, the system also includes a model update module, which is deployed on a ground computing platform; the satellite-borne detector is used to send the anomaly detection result and the current actual value of each of multiple telemetry parameters of the satellite's subsystems to the computing platform.

[0065] See also Figure 1 The present invention is an anomaly detection and model updating system based on the cooperation between the ground and the satellite. The satellite's time series prediction model is continuously trained and updated through the communication between the satellite-borne detector and the ground computing platform. The satellite-borne detector is used to collect the satellite's status information (including the current actual values ​​of multiple telemetry parameters) in real time, and perform preliminary anomaly detection through the first time series detection model.

[0066] The model update module is used to use the current actual values ​​of each of the multiple telemetry parameters and the anomaly detection results as training data sets to train the second time series prediction model locally deployed on the computing platform until a preset training batch is reached, wherein the anomaly detection results are used as labels of the training data sets; and the model parameters of the second time series prediction model trained to the training batch are sent to the satellite. The anomaly detection module is used to update the model parameters of the first time series prediction model according to the model parameters of the second time series prediction model.

[0067] The model update module is responsible for receiving the current actual values ​​of the telemetry parameters and anomaly detection results sent back by the satellite. The anomaly detection results serve as the label of each piece of data, that is, the target for training. Based on the collected data, the computing platform performs data analysis and data preprocessing on the data to construct a time series data set, in which the current actual values ​​of multiple telemetry parameters of the satellite subsystem are used as input features, and the corresponding anomaly detection results, that is, whether an anomaly is detected, are used as labels. Specifically, preprocessing includes noise removal and standardization. For outliers in the time series data set , use median filtering for smoothing, and set the time window length to , then Filtered data at time It can be expressed as:

[0068] Subsequently, the dimension difference was eliminated by Z-score standardization, and the standardized parameters Defined as:

[0069] in, is the data mean, The preprocessed data is divided into training set, validation set and test set in chronological order, and the ratio can be set to 7:2:1 to ensure the generalization ability of the time series prediction model.

[0070] The ground computing platform uses the constructed time series data set to train the second time series prediction model deployed locally. The model parameters of the second time series prediction model are updated through the feedforward and back-propagation algorithms. The training process continues until the preset training batch is reached and the training is stopped. After the training is completed, the model parameters of the second time series prediction model will be saved by the computing platform and sent to the satellite for subsequent use on the satellite. This process can be achieved through a satellite communication link. In the case where the time series prediction model is a long short-term memory network, the model parameters include the various weight matrices and bias items of the long short-term memory network.

[0071] After receiving the updated model parameters, the anomaly detection module deployed on the satellite applies them to the first time series prediction model, so that the prediction ability of the first time series prediction model is improved. After the update, the model parameters of the first time series prediction model and the second time series prediction model of the ground computing platform are consistent. It can be understood that there is no essential difference in model type between the second time series prediction model and the first time series prediction model. For example, when the first time series prediction model is a long short-term memory network, the second time series prediction model should also be a long short-term memory network. The two are consistent in model structure.

[0072] In an optional embodiment, the model updating module is used to use mean square error as a loss function to quantify the deviation between the predicted value of the second time series prediction model for each of the multiple telemetry parameters and the current actual value of each of the multiple telemetry parameters, and adjust the model parameters of the second time series prediction model through the Adam optimization algorithm to minimize the loss function.

[0073] The embodiment of the present invention proposes to use mean squared error (MSE) as the loss function to quantify the deviation between the predicted value of each of the multiple telemetry parameters and the current actual value of each of the multiple telemetry parameters by the second time series prediction model, and adjust the model parameters through the Adam optimization algorithm to minimize the loss function. The calculation formula of the loss function is as follows:

[0074] In the formula, For the The optimal characteristic parameters are The current actual value at the moment; The second prediction model is for The optimal characteristic parameters are The predicted value at time. is the number of samples in the time series dataset (i.e., the number of telemetry parameters); It is The mean square error loss of the optimal feature parameters.

[0075] The Adam (Adaptive Moment Estimation) optimization algorithm is the optimization algorithm used in the updating process of the model parameters in this embodiment. It combines the ideas of momentum optimization (Momentum) and adaptive learning rate (AdaGrad), and can effectively adjust the learning rate of each parameter to avoid the problem of gradient explosion or gradient disappearance.

[0076] First, the first-order moment estimate of the gradient is calculated, that is, the average value of the gradient, which is used to smooth the update process. , the formula for the first-order moment estimation is:

[0077] In the formula, It is the first-order moment estimate, which represents the average value of the gradient at the current moment; It is the decay rate of the first-order moment, which can be set close to 1 (such as 0.9), indicating that more historical gradients will participate in the current average calculation; is the gradient at the current moment, indicating the loss function with respect to the model parameters The partial derivative of .

[0078] Then calculate the second-order moment estimate of the gradient, that is, the average value of the square of the gradient, which is used to adaptively adjust the parameters The learning rate is:

[0079] In the formula, is the second-order moment estimate, which represents the average of the squared gradient; is the decay rate of the second-order moment and can be set close to 1 (such as 0.999), indicating that the square of the gradient over a longer period of time has a greater impact on the current estimate.

[0080] Since the first-order moment estimate and the second-order moment estimate are biased toward zero when initialized, these estimates are then corrected for the bias. The correction formula is:

[0081] In the formula, and is the revised estimate, representing the correction for the gradient deviation; is the current time step.

[0082] Finally, the corrected moment estimates are used to update the model parameters , the parameter update formula is:

[0083] In the formula, are the updated model parameters; are the current model parameters; is the learning rate, which controls the step size of parameter update; is a small constant used to prevent the denominator from being zero.

[0084] Through the above steps, the model parameters of the second time series prediction model are gradually updated to minimize the loss function, thereby improving the prediction ability of the model.

[0085] In an optional embodiment, the system further includes an onboard computer and a control module deployed on the satellite.

[0086] The onboard computer is used to obtain the abnormality detection result and generate a corresponding control instruction according to the abnormality detection result, wherein the control instruction is used to instruct the satellite to perform abnormality processing.

[0087] See also Figure 1, The onboard computer (OBC) is a module inside the satellite that is responsible for processing various satellite data, performing complex computing tasks, and controlling various operations of the satellite. The onboard computer obtains the results of the anomaly detection output by the anomaly detection module, that is, it detects which subsystems have anomalies (such as attitude control anomalies, power supply anomalies, communication failures, etc.). Based on the anomaly detection results, the onboard computer analyzes the anomalies and generates corresponding control instructions. These instructions include how to handle abnormal operations, for example, if an attitude control failure is detected, adjust the attitude control system. If a power supply anomaly is detected, switch to the backup power supply. The control instructions generated by the onboard computer can directly affect the behavior of the satellite to ensure that it can cope with various failures and ensure the smooth completion of the mission.

[0088] The control module is used to execute the control instruction.

[0089] See also Figure 1 The control module is a module inside the satellite that is used to execute control instructions. When the onboard computer generates control instructions, the control module receives the control instructions and directly executes related operations. For example, if the control instruction requires the satellite to adjust its attitude, the control module will instruct the attitude control system to make the adjustment. If the control instruction requires the satellite to switch to the backup battery, the control module will interact with the power subsystem to start the backup power supply.

[0090] A satellite anomaly detection system provided by an embodiment of the present invention enables a satellite to automatically handle anomalies in orbit without ground intervention, thereby improving the reliability and autonomy of the satellite.

[0091] Figure 2 is a schematic diagram of an application framework of a satellite anomaly detection system provided by an embodiment of the present invention. Figure 2 The process shown is explained in two parts: The first part is the autonomous detection and control of the satellite. During the on-orbit operation of the satellite, the on-board detector will collect various telemetry data in real time. The on-board detector performs preliminary processing on the collected telemetry data, selects the most important telemetry parameters, and screens out the optimal feature subset consisting of n optimal feature parameters through feature selection algorithms (such as mRMR-FS). Then, the long short-term memory network (LSTM) is used for time series prediction to process the time series characteristics of the telemetry data and capture the time series dependencies in the data, so as to predict the future change trend of each optimal feature parameter. The predicted value generated by the long short-term memory network is compared with the current actual value of the telemetry data to monitor the operating status of the satellite subsystem and output the abnormality detection result. The on-board computer OBC analyzes the real-time monitoring results and decides whether control measures need to be taken based on the detected anomalies. OBC will generate real-time instructions (i.e. control instructions) based on the abnormal situation and transmit the control instructions to the corresponding subsystems of the satellite to realize automatic adjustment of the satellite. Finally, the control module performs adjustment operations on each satellite subsystem according to the received control instructions, thereby ensuring the stability and safety of the satellite's on-orbit operation.

[0092] The second part is the model parameter update on the ground. The satellite transmits the current actual values ​​of all telemetry parameters and the anomaly detection results back to the ground through the satellite-to-ground link. After receiving the data, the computing platform on the ground downloads and stores it. The downloaded data will be processed by data analysis and converted into a format suitable for subsequent processing, including noise removal and missing data filling, so as to provide clean input data for subsequent model updates. After data analysis, it is processed by the model update module, including data preprocessing, batch training and parameter update of the second time series prediction model, so that the model can better predict future telemetry data and enhance anomaly detection capabilities. The updated model parameters and related results will be saved in the knowledge base on the ground, providing a reference basis for future anomaly detection and model updates. The updated model parameters are serialized into a transmittable format so that they can be uploaded to the satellite through the satellite-to-ground link. The serialization process ensures that the parameters can be transmitted between the satellite and the ground in a suitable format. Finally, the serialized updated parameters are uploaded to the satellite. After receiving the updated model parameters, the satellite is redeployed to the onboard detector to update the first time series prediction model accordingly, ensuring that the satellite can use the latest time series prediction model for real-time anomaly detection and prediction.

[0093] Figure 3 This is a schematic diagram of the steps of a satellite anomaly detection method provided by an embodiment of the present invention. Figure 3 , the method is applied to the satellite anomaly detection system as described above, the method comprising: Step S101, obtain multiple telemetry parameters of the satellite subsystem, and screen out n optimal characteristic parameters from the multiple telemetry parameters, wherein the correlation between the n optimal characteristic parameters and the preset abnormal type parameters is the largest, and the redundancy between any two of the n optimal characteristic parameters is the smallest, and n is an integer greater than or equal to 2.

[0094] During the satellite's in-orbit operation, sensors collect telemetry data from each subsystem in real time. Telemetry parameters include but are not limited to physical quantities such as temperature, pressure, voltage, rotation speed, attitude angle, and communication signal strength. Each subsystem has a set of telemetry parameters associated with it. For example, the thermal control system is responsible for regulating the temperature of various components inside the satellite to prevent overheating or overcooling. The corresponding telemetry parameters may include temperature, voltage, current, etc. The set of satellite subsystems includes multiple subsystems, and each subsystem can contain a different set of telemetry parameters.

[0095] According to the anomaly detection task of the target subsystem, the parameters with the strongest correlation with the abnormal type parameters are screened out from all telemetry parameters. The abnormal type parameters reflect the abnormal state of the subsystem. For example, for the thermal control system, the abnormal type parameters can be temperature anomalies or sensor failures. The value of 1 indicates abnormality and the value of 0 indicates normality. In order to calculate the correlation between each telemetry parameter and the abnormal type parameter, the concept of mutual information can be used. Although the telemetry parameters with the strongest correlation with the abnormal type parameters are selected, there may be a certain redundancy between these telemetry parameters. Therefore, it is also necessary to calculate the redundancy between each pair of telemetry parameters to ensure that the telemetry parameters in the selected parameter set are as independent as possible. Through the above steps, the mutual information between each telemetry parameter and the abnormal type parameter is first calculated, and the telemetry parameters with the strongest correlation are selected. Then, the redundancy between these selected telemetry parameters is calculated, and those telemetry parameters with the smallest redundancy are screened out, and finally n optimal feature parameters are obtained.

[0096] Step S102, predicting the parameter values ​​of the n optimal characteristic parameters through the first time series prediction model locally deployed by the satellite's onboard detector, obtaining the predicted values ​​of the n optimal characteristic parameters respectively, and performing anomaly detection on the satellite based on the predicted values ​​to obtain anomaly detection results.

[0097] The first time series prediction model deployed locally on the satellite's onboard detector predicts the n optimal characteristic parameters selected to obtain the predicted value of each characteristic parameter. The time series prediction model analyzes historical telemetry data, learns the time variation trend of each characteristic parameter, and predicts the value of these characteristic parameters at future moments. When the predicted value is generated, it is compared with the actual telemetry parameters currently collected.

[0098] In an optional embodiment, the current actual value of each of the n optimal feature parameters and the historical actual value within a preset time window are used as input data of the first time series prediction model, and the predicted value of each of the n optimal feature parameters output by the first time series prediction model according to the input data is obtained, and the first time series prediction model is a long short-term memory network; through the forgetting gate, input gate and output gate of the long short-term memory network, the predicted value of each of the n optimal feature parameters is output based on the current actual value and the historical actual value of each of the n optimal feature parameters.

[0099] In this embodiment, the current actual values ​​of n optimal feature parameters and their historical actual values ​​will be used as the input of the long short-term memory network. Assuming that there are multiple historical actual values ​​of the optimal feature parameters, these data can be expanded by time step and input into the long short-term memory network as a time series for training and prediction. Specifically, the current actual value and its historical actual value of each optimal feature parameter are Right now ,in, is the current actual value of the optimal characteristic parameter, is the time step of the preset time window. The predicted value of the optimal feature parameter finally predicted is the hidden state of the first time series prediction model The value of .

[0100] If the difference between the actual value and the predicted value exceeds the set threshold, it will be determined that there is an abnormality. Through this prediction and comparison process, the health status of each satellite subsystem can be monitored in real time, and possible faults or abnormalities can be discovered in time to ensure the normal operation of the satellite. In an optional implementation, step S102 includes steps S1021-S1023: Step S1021, determining the prediction error of each optimal feature parameter according to the difference between the current actual value and the predicted value of each of the n optimal feature parameters.

[0101] This embodiment uses the mean square error as the prediction error of each optimal feature parameter. The prediction error obtained by calculating the mean square error is a numerical value that measures the size of the error. The larger the value, the more significant the difference between the current actual value and the predicted value, which means that there may be an abnormal risk.

[0102] Step S1022: Based on a preset smoothing coefficient, a current error threshold is calculated according to the prediction error and the error threshold at the previous moment.

[0103] After the prediction error is calculated, the error threshold at the current moment is dynamically adjusted based on the preset smoothing coefficient and the error threshold at the previous moment. The error threshold at the previous moment refers to the error threshold at the previous moment, that is, the error threshold calculated at the previous moment. The current prediction error refers to the mean square error calculated at the current moment (as shown in the previous formula). Through the adaptive threshold adjustment mechanism, the error threshold can be updated in a continuous time cycle so that an appropriate response can be made during the monitoring process. The specific calculation method has been described in detail in conjunction with the relevant formula in the previous system embodiment, and will not be repeated here.

[0104] Step S1023: When the prediction error of any optimal feature parameter exceeds the current error threshold, the satellite anomaly is determined and the corresponding anomaly detection result is output.

[0105] The prediction error of each optimal feature parameter is compared with the current error threshold. If the prediction error of an optimal feature parameter exceeds the current error threshold, the optimal feature parameter is judged to be abnormal, and the corresponding abnormality detection result is generated, including the abnormality type (such as abnormal temperature, battery voltage is too high, etc.), location (for example: a specific subsystem of the satellite fails) and recommended treatment measures (such as starting the backup system, alarming, requesting further diagnosis, etc.).

[0106] The method further comprises: Step S201, sending the abnormality detection result and the current actual value of each of the multiple telemetry parameters of the satellite's subsystem to a computing platform deployed on the ground.

[0107] The present invention is an anomaly detection and model updating method based on the cooperation between the ground and the satellite. Through the communication between the onboard detector and the ground computing platform, the satellite's time series prediction model is continuously trained and updated. The onboard detector is used to collect the satellite's status information (including the current actual values ​​of multiple telemetry parameters) in real time, and perform preliminary anomaly detection through the first time series detection model.

[0108] Step S202, using the current actual values ​​of each of the multiple telemetry parameters and the anomaly detection results as training data sets, training the second time series prediction model locally deployed on the computing platform until a preset training batch is reached, wherein the anomaly detection results are used as labels for the training data sets.

[0109] When the computing platform on the ground receives the current actual values ​​of the telemetry parameters and the anomaly detection results sent back by the satellite, the anomaly detection results are used as the labels for each piece of data, i.e., the target for training. Based on the collected data, the computing platform performs data analysis and data preprocessing on the data to construct a time series data set, in which the current actual values ​​of multiple telemetry parameters of the satellite subsystem are used as input features, and the corresponding anomaly detection results, i.e., whether an anomaly is detected, are used as labels. The ground computing platform uses the constructed time series data set to train the second time series prediction model deployed locally. The model parameters of the second time series prediction model are updated through the feedforward and backpropagation algorithms. The training process continues until the preset training batch is reached and the training is stopped.

[0110] In an optional embodiment, step S202 includes: using mean square error as a loss function, quantifying the deviation between the predicted value of the second time series prediction model for each of the multiple telemetry parameters and the current actual value of each of the multiple telemetry parameters, and adjusting the model parameters of the second time series prediction model through the Adam optimization algorithm to minimize the loss function.

[0111] The embodiment of the present invention adopts mean square error as the loss function, quantifies the deviation between the predicted value and the actual value of the second time series prediction model for multiple telemetry parameters, and adjusts the model parameters through the Adam optimization algorithm to minimize the loss function. The mean square error is used to calculate the difference between the predicted value and the actual value, thereby measuring the prediction performance of the model. The Adam optimization algorithm combines the advantages of momentum optimization and adaptive learning rate. It smoothes the update process and adaptively adjusts the learning rate by calculating the first-order and second-order moment estimates of the gradient, effectively avoiding the problem of gradient explosion or gradient disappearance. During the update process, after the deviation correction, the model parameters are optimized, thereby improving the prediction accuracy. This process gradually minimizes the loss function and improves the prediction ability and accuracy of the second time series prediction model.

[0112] Step S203: sending the model parameters of the second time series prediction model trained to the training batch to the satellite; and updating the model parameters of the first time series prediction model accordingly according to the model parameters of the second time series prediction model.

[0113] After training, the model parameters of the second time series prediction model will be saved by the computing platform and sent to the satellite for subsequent use on the satellite. This process can be achieved through a satellite communication link. In the case where the time series prediction model is a long short-term memory network, the model parameters include the weight matrices and bias items of the long short-term memory network.

[0114] After receiving the updated model parameters, the anomaly detection module deployed on the satellite applies them to the first time series prediction model, so that the prediction ability of the first time series prediction model is improved. After the update, the model parameters of the first time series prediction model are consistent with those of the second time series prediction model of the ground computing platform.

[0115] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, devices, electronic devices and storage media. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0116] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods and apparatuses according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing terminal device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded into a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0117] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0118] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0119] The satellite anomaly detection system and method provided by the present invention are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technicians in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A satellite anomaly detection system, characterized in that: The system comprises a data dimension reduction module and an anomaly detection module, wherein the data dimension reduction module and the anomaly detection module are deployed on a satellite-borne detector; The data dimension reduction module is used to obtain multiple telemetry parameters of the satellite subsystem, and screen out n optimal feature parameters from the multiple telemetry parameters, wherein the correlation between the n optimal feature parameters and the preset abnormal type parameters is the largest, and the redundancy between any two optimal feature parameters among the n optimal feature parameters is the smallest, and n is an integer greater than or equal to 2; The anomaly detection module is used to predict the parameter values ​​of the n optimal characteristic parameters through the first time series prediction model locally deployed by the onboard detector, obtain the predicted values ​​of each of the n optimal characteristic parameters, and perform anomaly detection on the satellite based on the predicted values ​​to obtain anomaly detection results.

2. The system according to claim 1, characterized in that The anomaly detection module is used to: Determining the prediction error of each optimal feature parameter according to the difference between the current actual value and the predicted value of each of the n optimal feature parameters; Based on a preset smoothing coefficient, a current error threshold is calculated according to the prediction error and the error threshold at a previous moment; When the prediction error of any optimal characteristic parameter exceeds the current error threshold, the satellite anomaly is determined and a corresponding anomaly detection result is output.

3. The system according to claim 1 or 2, characterized in that: The anomaly detection module is used to use the current actual value of each of the n optimal feature parameters and the historical actual value within a preset time window as input data of the first time series prediction model, and obtain the predicted value of each of the n optimal feature parameters output by the first time series prediction model according to the input data, wherein the first time series prediction model is a long short-term memory network; Outputting predicted values ​​of the n optimal feature parameters based on their respective current actual values ​​and historical actual values ​​through the forget gate, input gate and output gate of the long short-term memory network; The system further comprises a model updating module, which is deployed on a ground computing platform; the satellite-borne detector is used to send the abnormality detection result and the current actual value of each of the multiple telemetry parameters of the satellite's subsystem to the computing platform; The model updating module is used to use the current actual values ​​of each of the multiple telemetry parameters and the anomaly detection results as training data sets to train the second time series prediction model locally deployed on the computing platform until a preset training batch is reached, wherein the anomaly detection results are used as labels of the training data sets; and send model parameters of the second time series prediction model trained to the training batch to the satellite; The anomaly detection module is used to update the model parameters of the first time series prediction model according to the model parameters of the second time series prediction model.

4. The system according to claim 3, characterized in that The model updating module is used to use the mean square error as the loss function to quantify the deviation between the predicted value of the second time series prediction model for each of the multiple telemetry parameters and the current actual value of each of the multiple telemetry parameters, and adjust the model parameters of the second time series prediction model through the Adam optimization algorithm to minimize the loss function.

5. The system according to claim 1, characterized in that The satellite subsystem includes an attitude control subsystem, and the abnormality type parameters include parameters corresponding to communication abnormalities and parameters corresponding to power supply abnormalities; The data dimension reduction module is used to calculate the correlation between the telemetry parameters of the various components of the attitude control system, and to calculate the redundancy between the telemetry parameters of the various components of the attitude control system and other telemetry parameters except the telemetry parameters; and to take the n telemetry parameters with the maximum correlation and the minimum redundancy as the optimal feature parameters.

6. The system according to claim 1, characterized in that The system also includes an onboard computer and a control module deployed on the satellite; The onboard computer is used to obtain the abnormality detection result and generate a corresponding control instruction according to the abnormality detection result, wherein the control instruction is used to instruct the satellite to perform abnormality processing; The control module is used to execute the control instruction.

7. A satellite anomaly detection method, characterized in that: Applied to the satellite anomaly detection system according to any one of claims 1 to 6, the method comprising: Acquire multiple telemetry parameters of the satellite subsystem, and select n optimal characteristic parameters from the multiple telemetry parameters, wherein the correlation between the n optimal characteristic parameters and the preset abnormal type parameters is the largest, and the redundancy between any two optimal characteristic parameters among the n optimal characteristic parameters is the smallest, and n is an integer greater than or equal to 2; The first time series prediction model locally deployed by the satellite's onboard detector is used to predict the parameter values ​​of the n optimal characteristic parameters to obtain the predicted values ​​of the n optimal characteristic parameters, and the satellite is subjected to anomaly detection based on the predicted values ​​to obtain anomaly detection results.

8. The method according to claim 7, characterized in that The performing abnormality detection on the satellite based on the predicted value comprises: Determining the prediction error of each optimal feature parameter according to the difference between the current actual value and the predicted value of each of the n optimal feature parameters; Based on a preset smoothing coefficient, a current error threshold is calculated according to the prediction error and the error threshold at a previous moment; When the prediction error of any optimal characteristic parameter exceeds the current error threshold, the satellite anomaly is determined and a corresponding anomaly detection result is output.

9. The method according to claim 7 or 8, characterized in that: The first time series prediction model locally deployed by the satellite-borne detector predicts the parameter values ​​of the n optimal characteristic parameters, including: Using the current actual value of each of the n optimal feature parameters and the historical actual value within a preset time window as input data of the first time series prediction model, and obtaining the predicted value of each of the n optimal feature parameters output by the first time series prediction model according to the input data, wherein the first time series prediction model is a long short-term memory network; Outputting predicted values ​​of the n optimal feature parameters based on their respective current actual values ​​and historical actual values ​​through the forget gate, input gate and output gate of the long short-term memory network; The method further comprises: Sending the anomaly detection result and the current actual value of each of the plurality of telemetry parameters of the satellite subsystem to a computing platform deployed on the ground; Using the current actual values ​​of each of the multiple telemetry parameters and the anomaly detection results as training data sets, training the second time series prediction model locally deployed on the computing platform until a preset training batch is reached, wherein the anomaly detection results serve as labels of the training data sets; sending model parameters of the second time series prediction model trained to the training batch to the satellite; According to the model parameters of the second time series prediction model, the model parameters of the first time series prediction model are correspondingly updated.

10. The method according to claim 9, characterized in that The training of the second time series prediction model locally deployed on the computing platform includes: The mean square error is used as the loss function to quantify the deviation between the predicted value of the second time series prediction model for each of the multiple telemetry parameters and the current actual value of each of the multiple telemetry parameters, and the model parameters of the second time series prediction model are adjusted through the Adam optimization algorithm to minimize the loss function.

Images