A satellite health monitoring method, device and medium based on artificial intelligence algorithms

Abstract

The present invention discloses a satellite health monitoring method, device and medium based on artificial intelligence algorithms, relating to the technical field of telemetry parameter monitoring. The method includes: constructing a telemetry parameter image database of a target satellite; obtaining data within the real-time period of multiple telemetry parameters; determining the health status of the characteristic images of the data within the real-time period of each telemetry parameter according to the characteristic images in the telemetry parameter image database; determining the health status scores of the corresponding instruments according to the preset health status scores corresponding to the health status of the characteristic images of the data within the real-time period of each telemetry parameter; and evaluating the health status of the target satellite according to the health status scores of the corresponding instruments and the preset weights of the corresponding instruments. The present invention can improve the accuracy of satellite health monitoring results.

Description

Technical Field

[0001] The present invention relates to the technical field of telemetry parameter monitoring, and particularly to a satellite health monitoring method, device and medium based on an artificial intelligence algorithm. Background Art

[0002] During the on-orbit operation of a satellite, after the sensor parameter information obtained by the internal operation status monitoring system is encoded, it is transmitted to the ground through data transmission. This telemetry data is the only basis for ground operation personnel to understand the on-orbit operation status of the satellite. The telemetry data is large in quantity, high in dimension, complex in relationship, and strong in correlation and professionalism, belonging to a typical application field of industrial big data. It reflects the orbital information, performance changes, working mode switching, and whether there are faults of the spacecraft. Effective analysis and intelligent calculation of the telemetry data will provide an effective basis for ground operation and management personnel to judge the performance of the satellite and instruments and carry out various operation and maintenance management work. Especially the abnormal data in the telemetry data, whose variation law is different from that of normal telemetry data or does not conform to the satellite or instrument working mode setting, can thus reflect problems such as failure of the acquisition device, damage to the transmission link, performance degradation of the corresponding device, quality problems, mechanical and electronic faults, or design deficiencies. Timely and effectively warning of the abnormal patterns existing in the telemetry data, intelligently attributing the abnormal pattern data, and evaluating the health status of the satellite / instrument have significant practical significance for improving the ground service quality and enhancing the maturity, safety and reliability of all links of satellite / instrument design, development, production and maintenance.

[0003] With the complication of satellite functions and the diversification of types, the dimension of real-time monitored telemetry data has developed in an explosive manner. To realize the industrial operation and management of satellite system engineering, it is urgent to analyze satellite telemetry big data, give play to the advantages of artificial intelligence, and accurately and quickly analyze the satellite parameters such as abnormal warning and root cause intelligent positioning, which has significant practical significance for improving the ground service quality and maintaining the safety and reliability of all links of satellite operation. Summary of the Invention

[0004] The purpose of the present invention is to provide a satellite health monitoring method, device and medium based on an artificial intelligence algorithm, which can improve the accuracy of satellite health monitoring results.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A satellite health monitoring method based on an artificial intelligence algorithm, the method includes:

[0007] Construct a telemetry parameter image database for the target satellite; the telemetry parameter image database includes characteristic images of multiple telemetry parameters within a period; the characteristic images include normal characteristic images and multiple abnormal characteristic images, and the abnormal types corresponding to each of the abnormal characteristic images;

[0008] Obtain the real-time period data of multiple telemetry parameters;

[0009] According to the characteristic images in the telemetry parameter image database, determine the health status of the characteristic images of the real-time period data of each of the telemetry parameters; the health status is normal or abnormal; when the health status is abnormal, output the corresponding abnormal type;

[0010] According to the preset health status scores corresponding to the health status of the characteristic images of the real-time period data of each of the telemetry parameters, determine the health status scores of the corresponding instruments;

[0011] Evaluate the health status of the target satellite according to the health status scores of the corresponding instruments and the preset weights of the corresponding instruments.

[0012] Optionally, constructing a telemetry parameter image database for the target satellite specifically includes:

[0013] Obtain the historical period data of each telemetry parameter of the target satellite;

[0014] Judge whether the health status of the characteristic image of the historical period data is normal;

[0015] When the health status of the characteristic image of the historical period data is normal, use the characteristic image of the historical period data as the normal characteristic image of the telemetry parameter;

[0016] When the health status of the characteristic image of the historical period data is abnormal, use the characteristic image of the historical period data as the abnormal characteristic image of the telemetry parameter, and mark the abnormal type corresponding to the abnormal characteristic image;

[0017] According to the normal characteristic images and abnormal characteristic images of each of the telemetry parameters, and the abnormal types corresponding to the abnormal characteristic images, obtain the telemetry parameter image database of the target satellite.

[0018] Optionally, according to the characteristic images in the telemetry parameter image database, determining the health status of the characteristic images of the real-time period data of each of the telemetry parameters specifically includes:

[0019] Taking the characteristic images in the telemetry parameter image database as inputs and the health status of the characteristic images as outputs, train a deep learning model to obtain a health status recognition model;

[0020] Input the characteristic images of the data within the real-time period of each of the telemetry parameters into the health status recognition model to obtain the health status of the characteristic images of the data within the real-time period of each of the telemetry parameters.

[0021] Optionally, determine the health status scores of the corresponding instruments according to the preset health status scores corresponding to the health status of the characteristic images of the data within the real-time period of each of the telemetry parameters, specifically including:

[0022] Use the characteristic images of the data within the historical period of the telemetry parameters corresponding to each of the instruments as the input, and the characteristic images in the corresponding telemetry parameter image database as the output to train the neural network model to obtain the trained neural network model, and use the parameter weights of the trained neural network model as the contribution degrees of the corresponding telemetry parameters of each of the instruments to the health status of the corresponding instrument;

[0023] Multiply the preset health status scores corresponding to the health status of the characteristic images of the data within the real-time period of each of the telemetry parameters by the contribution degrees of the corresponding telemetry parameters to the health status of the corresponding instrument to determine the health status scores of the corresponding instruments.

[0024] Optionally, the neural network model includes a self-attention network layer, a multiplication layer, and a fully connected layer connected in sequence.

[0025] Optionally, use the parameter weights of the fully connected layer as the contribution degrees of the corresponding telemetry parameters of each of the instruments to the health status of the corresponding instrument.

[0026] Optionally, use the characteristic images of the data within the real-time period of the telemetry parameters corresponding to each of the instruments as the input to the trained neural network model to obtain the output data of the attention layer;

[0027] Multiply the output data of the attention layer by the first data of the parameter weights of the fully connected layer to obtain the contribution degrees of the corresponding telemetry parameters of each of the instruments to the health status of the corresponding instrument.

[0028] Optionally, the formula for evaluating the health status of the target satellite is:

[0029] The health status score of the target satellite = the health status score of instrument 1 × weight 1 + the health status score of instrument 2 × weight 2 + … + the health status score of instrument n × weight n.

[0030] A computer device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the steps of the satellite health monitoring method based on the artificial intelligence algorithm described in any one of the above.

[0031] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the satellite health monitoring method based on the artificial intelligence algorithm described in any one of the above are implemented.

[0032] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0033] The present invention adopts a deep learning algorithm, utilizes satellite instrument telemetry data, realizes intelligent early warning of instrument telemetry data, satellite health status assessment, etc., and provides an effective basis for carrying out various operation and maintenance management work. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0035] Figure 1 Schematic diagram of the feature consistency data image provided in Embodiment 1 of the present invention;

[0036] Figure 2 Schematic diagram of the portal data image provided in Embodiment 1 of the present invention;

[0037] Figure 3 Schematic diagram of the broadband data image provided in Embodiment 1 of the present invention;

[0038] Figure 4 Schematic diagram of the mutation data image provided in Embodiment 1 of the present invention;

[0039] Figure 5 Schematic diagram of the generation process of the typical image of the telemetry parameters of the present invention;

[0040] Figure 6 Schematic diagram of the semantic classification outline of the present invention;

[0041] Figure 7 Schematic diagram of the control classification of the present invention;

[0042] Figure 8 Schematic diagram of the calibration source classification of the present invention;

[0043] Figure 9 First schematic diagram of the semantic classification of the microwave thermometer type III of the present invention;

[0044] Figure 10 Schematic diagram of the control quantity parameters of the present invention;

[0045] Figure 11 It is the first schematic diagram of the functions of the controller of the present invention;

[0046] Figure 12 It is the first schematic diagram of the functions of the calibration source of the present invention;

[0047] Figure 13 It is the schematic diagram of the confusion matrix of the present invention;

[0048] Figure 14 It is the schematic diagram of the health scoring criteria for FY-3E satellite of the present invention;

[0049] Figure 15 It is the schematic diagram of the scoring weights for FY-3E of the present invention;

[0050] Figure 16 It is the schematic diagram of the architecture design provided by the present invention;

[0051] Figure 17 It is the actual application flow chart of the satellite health monitoring platform based on artificial intelligence algorithm provided by the present invention;

[0052] Figure 18 It is the flow chart of the satellite health monitoring method based on artificial intelligence algorithm provided by the present invention;

[0053] Figure 19 It is the schematic diagram of the fitting network structure of the present invention;

[0054] Figure 20 It is the internal structure diagram of the computer device. Detailed implementation manners

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0056] The purpose of the present invention is to provide a satellite health monitoring method, device and medium based on artificial intelligence algorithm, which can improve the accuracy of satellite health monitoring results.

[0057] To make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0058] Embodiment 1

[0059] As Figure 18 shown, the satellite health monitoring method based on artificial intelligence algorithm in this embodiment includes:

[0060] Step S1: Construct a telemetry parameter image database for the target satellite; the telemetry parameter image database includes characteristic images of multiple telemetry parameters within a period; the characteristic images include normal characteristic images, multiple abnormal characteristic images, and the abnormal types corresponding to each of the abnormal characteristic images.

[0061] S1 specifically includes:

[0062] Step S11: Obtain the data of each telemetry parameter of the target satellite within the historical period.

[0063] Step S12: Determine whether the health status of the characteristic image of the data within the historical period is normal.

[0064] Step S13: When the health status of the characteristic image of the data within the historical period is normal, use the characteristic image of the data within the historical period as the normal characteristic image of the telemetry parameter.

[0065] Step S14: When the health status of the characteristic image of the data within the historical period is abnormal, use the characteristic image of the data within the historical period as the abnormal characteristic image of the telemetry parameter, and mark the abnormal type corresponding to the abnormal characteristic image.

[0066] Step S15: Obtain the telemetry parameter image database of the target satellite according to the normal characteristic images and abnormal characteristic images of each telemetry parameter, and the abnormal types corresponding to the abnormal characteristic images.

[0067] Step S2: Obtain the data of multiple telemetry parameters within the real-time period.

[0068] Step S3: Determine the health status of the characteristic image of the data of each telemetry parameter within the real-time period according to the characteristic images in the telemetry parameter image database; the health status is normal or abnormal; when the health status is abnormal, output the corresponding abnormal type.

[0069] S3 specifically includes:

[0070] Step S31: Use the characteristic images in the telemetry parameter image database as the input and the health status of the characteristic images as the output to train a deep learning model to obtain a health status recognition model.

[0071] Step S32: Input the characteristic images of the data of each telemetry parameter within the real-time period into the health status recognition model to obtain the health status of the characteristic images of the data of each telemetry parameter within the real-time period.

[0072] Specifically, the deep learning model is a BERT model.

[0073] Step S4: Determine the health state scores of the corresponding instruments according to the preset health state scores corresponding to the health states of the characteristic images of the data within the real-time period of each telemetry parameter.

[0074] S4 specifically includes:

[0075] Step S41: Use the characteristic images of the data within the historical period of the telemetry parameters corresponding to each instrument as the input, and the characteristic images in the corresponding telemetry parameter image database as the output to train the neural network model, obtaining a trained neural network model, and use the parameter weights of the trained neural network model as the contribution degrees of the corresponding telemetry parameters of each instrument to the health state of the corresponding instrument.

[0076] Specifically, the neural network model includes a self-attention network layer, a multiplication layer, and a fully connected layer connected in sequence.

[0077] There are two ways to determine the contribution degrees of the telemetry parameters corresponding to each instrument to the health state of the corresponding instrument:

[0078] The first way: Use the parameter weights of the fully connected layer as the contribution degrees of the telemetry parameters corresponding to each instrument to the health state of the corresponding instrument.

[0079] The second way: Use the characteristic images of the data within the real-time period of the telemetry parameters corresponding to each instrument as the input to the trained neural network model to obtain the output data of the attention layer; multiply the output data of the attention layer by the first data of the parameter weights of the fully connected layer to obtain the contribution degrees of the telemetry parameters corresponding to each instrument to the health state of the corresponding instrument.

[0080] Step S42: Multiply the preset health state scores corresponding to the health states of the characteristic images of the data within the real-time period of each telemetry parameter by the contribution degrees of the corresponding telemetry parameters to the health state of the corresponding instrument to determine the health state scores of the corresponding instruments.

[0081] Step S5: Evaluate the health state of the target satellite according to the health state scores of the corresponding instruments and the preset weights of the corresponding instruments.

[0082] Specifically, the formula for evaluating the health state of the target satellite is:

[0083] The health state score of the target satellite = the health state score of instrument 1 × weight 1 + the health state score of instrument 2 × weight 2 + … + the health state score of instrument n × weight n.

[0084] Furthermore, the formula for evaluating the health state of the target satellite is:

[0085] The health status score of instrument 1 × 1 / n + the health status score of instrument 2 × 1 / n + … + the health status score of instrument n × 1 / n.

[0086] In the present invention, instrument 1, instrument 2, …, instrument n are all payloads. The health status scores of instrument 1, instrument 2, …, instrument n are the health status scores of each instrument.

[0087] The satellite parameter analysis and early warning platform based on artificial intelligence algorithm mainly performs automated analysis on satellite and instrument parameters through various interactive methods, and uses analysis methods based on fault diagnosis, expert system, deep learning, etc. to perform intelligent warning and health status assessment on platform telemetry data, instrument telemetry data, and instrument observation data. Among them, the abnormal detection of various telemetry parameters belongs to the process of fault diagnosis.

[0088] This platform selects the B / S structure as the system architecture, as Figure 16 shown, uses MYSQL as the system database, and combines the file system to store data information. It uses the front-end and back-end separated MVC system for system development and implementation, and the main programming languages are python, nodejs, etc.

[0089] The satellite parameter analysis and early warning platform based on artificial intelligence algorithm evaluates the health status of satellites according to different telemetry parameters of different satellites and different payloads. According to the change law of telemetry parameter data of on-orbit satellites, it selects appropriate health status assessment methods for telemetry parameters to perform health status assessment on telemetry parameters, and on this basis, performs intelligent calculation and health status assessment on telemetry data, which has important practical significance for the whole-life health management of satellites, can be used for prior prediction and active prevention of satellite faults, and has important reference significance and engineering value for ensuring the safe and reliable operation of satellites and mission success.

[0090] The satellite parameter analysis and early warning platform based on artificial intelligence algorithm provides a comprehensive and detailed satellite health status monitoring solution for users through the design of a three-level interface. It can not only help users understand the health status of satellites and their carried instruments in real time, but also discover potential problems in advance through functions such as prediction and deviation analysis, providing a strong guarantee for the safe and stable operation of satellites.

[0091] The first-level interface includes the instrument health status score and the long-term sequence health status score curve. Among them, the instrument health status score shows the comprehensive health status scores of each instrument. Usually, these scores are based on a series of preset monitoring parameters and standards. The scoring system is quantifiable, and the levels of the monitoring parameters are determined according to the quantified scores and the scores corresponding to different preset levels. The long-term sequence health status score curve is a curve showing the change of the health status score of each instrument over time. Users can intuitively understand the historical change trend of the instrument health status, so as to predict its possible future status.

[0092] The second-level interface is the score of the key components of the instrument, including key component identification and component score comparison. Among them, at the level of key component identification, the platform will conduct individual health status scoring for the key components of each instrument. These key components are usually important factors affecting the overall performance or reliability of the instrument, and the key components can be set according to the actual situation. At the level of component score comparison, through scoring, users can understand the health status of each key component, and by comparing the scores of different components, determine which components may require more attention or maintenance.

[0093] The third-level interface includes the actual observation curve and the prediction curve of the instrument parameters. Among them, the actual observation curve part shows the curve of the actual observed values of each instrument parameter changing over time. These observed data directly reflect the operating status and performance of the instrument. The prediction curve is generated by the platform based on historical observation data and other relevant information. These prediction curves can help users predict the future operating status of the instrument, so as to discover potential problems in advance. By comparing the actual observation curve and the prediction curve, users can clearly observe the deviation situation. These deviations may indicate that the instrument is in an abnormal state or has a fault, and further inspection and maintenance are required.

[0094] In practical applications, as Figure 17 shown, the specific implementation process of the present invention is described as follows:

[0095] (1) Telemetry Parameter Feature Recognition: During the on-orbit operation of a spacecraft, in order to obtain its internal operating status and provide real-time data for remote control objects, sensors in the spacecraft telemetry system sense the measured quantities and convert them into electrical signals. After various signals are combined according to a certain system, radio communication technology is used to transmit them to ground telemetry equipment. The ground equipment restores the original parameter information of each path through signal demodulation technology, and stores and displays it. The parameter information obtained in this case is the spacecraft telemetry data. The spacecraft telemetry data is large in quantity, complex in relationship, and strong in professionalism, belonging to a typical application field of industrial big data. The rich spacecraft telemetry data is closely related to the performance status of the measured equipment, reflecting the functions and performance changes of the equipment. Therefore, it is necessary to identify the typical data characteristics of telemetry parameters in order to form the basis for ground operation and management personnel to understand the on-orbit operation status of the spacecraft.

[0096] The typical data characteristics mainly include characteristic consistency data, portal data, broadband data, mutation data, and time-varying attenuation data, as specifically Figures 1 to 4 shown.

[0097] (2) Generation of Typical Images of Telemetry Parameters: Currently, the telemetry parameters of Fengsan generally follow a typical periodic change pattern with an orbit / day cycle. Therefore, it is necessary to explore what kind of periodic change pattern the telemetry parameters belong to, how they change within the cycle, draw the typical image characteristics within the cycle, draw the conventional change images of each telemetry parameter, and the images of the performance forms of telemetry parameters under various different abnormal conditions, in order to facilitate subsequent abnormal recognition and analysis.

[0098] As Figure 5 shown, the specific process of generating typical images of telemetry parameters includes:

[0099] 1. Analyze historical data.

[0100] The main purpose is to identify the change patterns of telemetry parameters. Through the observation and analysis of a large amount of historical data, it can be found that many telemetry parameters follow a certain periodic change pattern. These periods may be related to factors such as the orbit operation period, the Earth's rotation period, and the daily change period. Understanding these periodic patterns is the basis for subsequent generation of typical images.

[0101] 2. Calculate the periods of telemetry parameters.

[0102] After determining that the telemetry parameters have a periodic change pattern, the next step is to calculate the periods of these parameters. This usually involves statistical analysis of time series data, such as using methods like Fourier transform and autocorrelation function to identify and calculate the periods. The calculated periods will be an important basis for drawing the images of telemetry parameters within the cycle.

[0103] 3. Draw the images of telemetry parameters within the cycle according to the calculated periods.

[0104] After obtaining the period of the telemetry parameters, the typical images within the period can be started to be drawn. Specifically, it includes:

[0105] Data screening: Extract the data of each period from the historical data, and then draw the images corresponding to the respective periods. That is, extract the data of each period from the historical data, and then draw the images corresponding to the respective periods to obtain the images of each period.

[0106] Data processing: Perform necessary preprocessing on the screened data, such as abnormal point elimination, denoising, smoothing, etc., to improve the quality of the period images.

[0107] Image drawing: Use a time series graph to display the changes in telemetry parameters within the period.

[0108] Through the process of data screening, data processing, and image drawing, typical images reflecting the periodic change laws of telemetry parameters can be generated. These images are of great significance for understanding the dynamic characteristics of telemetry parameters and monitoring the equipment status. It should be noted that different telemetry parameters may have different periodic change laws, so this process will be adjusted and optimized according to specific situations in practical applications.

[0109] (3) Telemetry parameter image knowledge base.

[0110] Each telemetry parameter will exhibit typical periodic changes. The telemetry parameter image knowledge base will form an expert knowledge base with the periodic change time, the characteristic images of normal performance within the period, and the characteristic images of abnormal performance under specific abnormalities as the basis for subsequent health assessment.

[0111] (4) Telemetry parameter image anomaly recognition.

[0112] Based on the telemetry parameter image knowledge base, the telemetry parameter image anomaly recognition is divided into two categories. One category starts from the known abnormal data, draws the image of the actually downloaded telemetry data according to the periodic change law, and checks whether it is consistent with the characteristic images of normal performance in the image expert knowledge base. If not, it is further compared with the characteristic images of abnormal performance. If the comparison is consistent, it is marked as an abnormal image. The other category is when the abnormal image is unknown, manual intervention is required. It is necessary to check what type of image this is. If it belongs to the characteristic images of normal performance, it is added to the expert knowledge base of characteristic images of normal performance. If it belongs to the characteristic images of abnormal performance, after manually identifying the abnormal type, it is added to the expert knowledge base of characteristic images of abnormal performance.

[0113] (5) Telemetry parameter health status assessment.

[0114] The satellite platform has a large number of telemetry parameters, and each instrument on the satellite also has a large number of telemetry parameters. It is necessary to classify the large number of telemetry parameters according to the parameter semantics according to specific rules to construct a component - telemetry parameter tree. Based on the abnormal images identified by the abnormal recognition of telemetry parameter images, score the current health status of the telemetry parameters and give an assessment of their health status.

[0115] Specific method: First, calculate how many cycles each telemetry parameter will have in a day. According to the abnormal recognition of telemetry parameter images, identify how many cycles are abnormal. The score for the telemetry parameter is the number of normal cycles in a day divided by the total number of cycles in a day. Secondly, according to the supplemented semantic classification tree, give the weight of each telemetry parameter according to the contribution degree of each parameter to the instrument. The health status score of the instrument is obtained by multiplying the score of the telemetry parameter by the weight. Finally, by default, the importance of each instrument to the satellite is the same. Multiply the health status score of each instrument by the weight = the score of the satellite. This score will be graded. 80 - 100 is excellent, 60 - 80 is good, and below 60 is poor, which is the result of the satellite health status assessment.

[0116] Figures 6 to 8 It is the construction algorithm of the semantic classification model. The goal of the BERT model is to train with a large amount of unlabeled corpus to obtain rich semantic information of the text, and complete 13 natural language processing tasks based on this semantic information. Its model structure and input-output structure are as Figure 6 shown. In the present invention, BERT is used to complete the text classification task. The semantic classification model is obtained by training with the training set. The training set and the validation set are randomly allocated in the FY3E annotation dataset at a ratio of 5 to 1. As shown in Table 1 and Table 2. Taking FY-3E / MWTS as an example, the obtained telemetry parameter health assessment results are as Figures 9 to 13 shown.

[0117] Furthermore, as Figure 13 shown, the confusion matrix is used for classification. The abscissa represents the predicted classification category, and each number from 0 to 12 represents a category. The ordinate represents the labeled classification category, and each number from 0 to 12 represents a category, and its meaning is the same as that of the predicted classification category. The value of the element in the matrix represents the number of categories of the classification where the element is located in the horizontal and vertical coordinates (for example, X = 0, Y = 1, representing the number of data with the predicted classification category of 0 and the labeled classification category of 1). The goodness of the evaluation classification is determined by the median value of the diagonal. The larger the value, the more accurate the classification.

[0118] Table 1 Statistical table of test scores of the test set

[0119] Evaluation method Score Accuracy 0.8792 F-value Macro 0.7856 Accuracy Macro 0.7559 Recall Macro 0.9039

[0120] Table 2 Table of classification mapping relationships in the confusion matrix

[0121]

[0122]

[0123] (6) Health status assessment of each payload of FY-3E.

[0124] Establish a relationship model between telemetry parameters and calibration, and evaluate the influence degree of the telemetry parameters of each payload of FY-3E on payload calibration. Assign a certain weight to the corresponding payload telemetry parameters according to the evaluated influence degree, and combine the weight with the health status score of the telemetry parameters to obtain the health status score results of each payload. As shown in Table 3, the labeled quantities of each payload parameter of FY-3E are presented.

[0125] Table 3 Statistical table of labeled quantities of each payload parameter of FY-3E

[0126]

[0127]

[0128] The specific implementation method is as follows:

[0129] Since there are many types of parameters in each satellite instrument, and most of the data parameters have little relationship with the satellite health status, analyzing all parameters will affect the final evaluation result. To solve this problem, the present invention proposes to use a self-attention network to calculate the contribution degree of each instrument parameter to the final satellite measurement data, and use this as the main basis for evaluating the satellite health status from the instrument parameter health assessment. A method for analyzing the contribution degree of satellite instrument parameters based on self-attention, which is divided into four steps: instrument parameter definition and collection, construction of a fitting network, training of weight parameters, and acquisition of contribution degree. The present invention takes the microwave thermometer (Micro-Wave Temperature Sounder, MWTS) of Fengyun-3 satellite 05 (hereinafter referred to as FY-3E satellite) as the experimental object. The main purpose of the present invention is to convert the health status of instrument parameters into the health status of the instrument. Since the main measurement data of the microwave thermometer of FY-3E satellite is the earth observation brightness temperature in the level 1 data product, and the earth observation brightness temperature is calculated from the parameters in the corresponding observation data, the mapping relationship from the initial parameter health status to the instrument health status can be converted into the mapping relationship from the parameters in the observation data to the earth observation brightness temperature. In this way, a linear fitting network from the parameters in the observation data to the earth observation brightness temperature can be constructed and the network model can be trained, and finally the weight parameters can be obtained according to the model.

[0130] Step 1: Instrument parameter definition and collection. The data used in this invention includes the observation data of FY-3E / MWTS. Both types of data are provided by the National Satellite Meteorological Center. Input parameter collection: This invention selects data related to the calculated measurement data (the earth observation brightness temperature data in FY3E-MWTS) from the observation data, and removes data such as longitude, latitude, and time. In addition, there are many two-dimensional or three-dimensional data in the observation data (this is because there are cases where multiple related data in the observation data are merged into one table, such as the merger of multiple temperatures, and the same type of data detected by multiple channels or multiple detectors). These data need to be disassembled into multiple one-dimensional data according to their different actual meanings, and the dimension corresponding to the number of scan lines is retained. After disassembly, these data also need to be named and defined, and the group, data table, and data disassembly index corresponding to each data in the observation data are recorded to facilitate direct reading by the subsequent program. Fitting parameter collection: This invention selects the main measurement data from the L1 data (the corresponding data of the microwave thermometer is the earth observation brightness temperature). For example, the dimension of the earth observation brightness temperature data is 17×604×98, where 17 represents the number of channels, 604 represents the number of scan lines, and 98 is the scan width. The data disassembly method for the input parameter data is the same. Since there are too many arrays to be disassembled for the combination of the number of channels and the scan width, only the middle array (the array at the 49th index) is taken when disassembling the dimension corresponding to the scan width. In this way, 17 one-dimensional arrays with the length of the number of scan lines are obtained, and 17 is the predicted number of MWTS. Special case handling: The number of scan lines of the input parameter must be the same as that of the fitting parameter. If they are different, the scan lines of the two data need to be screened according to the time for the closest match. The specific method is to use the time of the data with fewer scan lines as the benchmark to match the scan line with the closest time of the other data. The screening is completed after all the data with fewer scan lines are matched. A specific example is the FY-3E satellite wind field measurement radar (Wind Radar, WRAD).

[0131] Step 2: Construct a fitting network. The network structure of this invention consists of a self-attention network layer, a multiplication layer, and a fully connected network layer with the number of nodes equal to the predicted number. The network structure diagram is as Figure 19 shown.

[0132] 1. Self-attention network layer:

[0133] The self-attention network (Self-Attention Network) is a deep learning model used to process sequence data. Currently, it is mainly applied in the fields of natural language processing and computer vision, and has not been applied in linear fitting tasks. The data input in this invention satisfies the time sequence and has the law generated by time change, belonging to sequence data, which meets the processing conditions of the self-attention network.

[0134] The self-attention network can calculate a weight for each element in the sequence, indicating the importance of that element to the entire sequence. This enables the network to adaptively focus on and highlight the most relevant and important elements in the sequence. In the self-attention network, the process of attention calculation determines the similarity scores between each element and other elements. These scores can be regarded as the degree of correlation between elements. By normalizing these scores, the weights between each element and other elements can be obtained, and these weights reflect the importance of the elements. The self-attention network can perform global importance calculations on the entire sequence at each time step or each position.

[0135] The reason for using the self-attention network in the present invention is to give a preliminary importance score to each parameter sequence of the input. This score is based on the degree of correlation of each parameter sequence. Since the target parameters of pre-fitting are calculated from each input parameter, the input parameters related to the target parameters will also have some correlations with each other. The self-attention network will increase the weights of these parameters and reduce the weights of the parameters with poor correlations. After training the training parameters in the self-attention network, the weights of the input parameters related to the target parameters will gradually increase, thereby improving the performance of network contribution degree analysis. The calculation formula of the self-attention network is as follows:

[0136]

[0137] In the formula, a i represents the weight obtained by calculating the i-th parameter through the self-attention network, X i represents the input data of the i-th parameter, W i represents the network training weight parameter of the i-th parameter, b i represents the network training bias parameter of the i-th parameter, e represents the exponential operation, · represents matrix multiplication, N represents the number of input parameters, tanh is the hyperbolic tangent activation function, and the formula is:

[0138]

[0139] where x represents the input data, and tanh(x) represents the hyperbolic tangent activation function.

[0140] After calculating the weights of each input parameter through the self-attention network, then use the multiplication layer to multiply the parameter weights with the input parameters. (Point-to-point multiplication, not matrix multiplication, that is, each input parameter has to be multiplied by the corresponding parameter weight).

[0141] Fully connected layer:

[0142] The fully connected layer is one of the most common network layers in deep learning neural networks. It is usually used as the last layer or an intermediate layer of the neural network. The main feature of the fully connected layer is that each neuron is connected to all neurons in the previous layer, that is, each neuron has a weighted connection with all outputs of the previous layer. This means that each neuron in the fully connected layer can receive all the information from the previous layer and process it. The formula is as follows:

[0143]

[0144] In the formula, X i represents the i-th input parameter, Y j represents the j-th output parameter, W i represents the training parameter weight of the i-th input neuron, b j represents the training bias term of the j-th output neuron, ReLU represents the ReLU activation function, and the ReLU activation function sets the parameter less than 0 to 0. The formula is:

[0145] ReLU(x) = max(0, x);

[0146] where ReLU(x) represents the ReLU activation function, x represents the input parameter, and max represents taking the maximum value.

[0147] The matrix with the number of input parameters is converted into a matrix with the number of output parameters through the fully connected layer. The W in the formula can be extracted as the parameter contribution degree after the model is trained.

[0148] Step 3: Train the weight parameters. Match the input parameters (original observation data, instrument parameter data) obtained in Step 1 with the target parameter (terrestrial observation brightness temperature value) data and divide them into a training set and a validation set according to a ratio of 9:1. Since the weight parameters in the model are required, there is no test set; then batch process the training set data and the validation set data.

[0149] Training starts after processing: The training data of each batch is shuffled and then input into the model to start prediction, calculate the error, and the optimizer network optimizes in three steps. When all the training data has completed the above steps, that is, when the prediction, error calculation, and network optimization steps are completed, then predict all the validation sets and calculate the error in two steps. After all the data has completed the above operations, it is considered that one training round is completed. After 100 training rounds are completed, the training ends. The optimizer used in the present invention is the Adam optimizer. Adam (Adaptive Moment Estimation) is a commonly used optimization algorithm for updating the parameters of a neural network in deep learning. It combines the ideas of the Momentum method and adaptive learning rate, and can effectively optimize the training process of the model.

[0150] Step 4: Obtaining contribution degrees. There are two ways to obtain contribution degrees in the present invention. The first way is static contribution degree acquisition, that is, extracting the parameter weights W in the fully connected layer of the next layer of the attention network layer i as the parameter contribution degrees of each parameter. This parameter contribution degree represents the comprehensive contribution degree of the training data of the training model to the training model. The operation method is to read all the weight parameters W in the first fully connected layer of the training model file, and the order of the parameters is the same as the order of the parameters input during the training of the training model; The second way is dynamic contribution degree acquisition. Since the attention network is added to this network, the parameter weights can be fine-tuned for each new input. The specific method is to input the new data into the model to calculate and directly extract the output of the attention layer and multiply it in order with the weight parameters of the first fully connected layer as the final dynamic parameter contribution degree. The formula is as follows:

[0151] C i =a i *W i ;

[0152] Among them, C i represents the i-th dynamic parameter contribution degree, W i represents the i-th static parameter contribution degree, and a i represents the i-th attention weight.

[0153] (7) FY-3E health status assessment.

[0154] Based on the assessment of the payload health status, considering the complex, highly redundant, reconfigurable, and non-linear characteristics of the satellite system, through steps (5) and (6), the on-orbit health status assessment method of the satellite based on the reconfigurability degree and the on-orbit health status assessment method of the satellite based on payload-telemetry parameters are used. The observation residuals are obtained by comparing the measured data with the historical health data, and the observation residuals are obtained by comparing the measured data with the historical health data as the health status characteristics. By combining the satellite health status assessment model with the assessment results of each payload health status, the FY-3E health status assessment result is finally obtained.

[0155] Furthermore, the description of Figure 14 is as follows: The satellite is divided into platform information and instrument information. A total of 11 instruments are carried on the FY-3E satellite, and the right side is the abbreviation of all instruments. Among them, MERSI is the Medium Resolution Spectral Imager; MWTS is the Microwave Temperature Sounder; MWHS is the Microwave Humidity Sounder; GNOS is the Global Navigation Satellite System Occultation Sounder; HIRAS is the Infrared Hyperspectral Atmospheric Sounder; WindRAD is the Wind Field Measurement Radar; X-EUVI is the Solar X-ray Extreme Ultraviolet Imager; SIM is the Solar Radiation Monitor; SSIM is the Solar Irradiance Spectrometer; Tri-IPM is the Ionospheric Photometer; SEM is the Space Environment Monitor.

[0156] As Figure 15 shown, the specific calculation formula for the FY-3E health status assessment is:

[0157] Health status score of instrument 1 × 1 / n + Health status score of instrument 2 × 1 / n + … + Health status score of instrument n × 1 / n.

[0158] The present invention has the following advantages:

[0159] (1) The architecture design of the platform of the present invention includes the BS architecture and the interaction between the front end and the back end.

[0160] (2) The present invention applies the corresponding anomaly detection algorithm to the satellite telemetry data, and optimizes the anomaly detection algorithm model during the anomaly detection process, improving the accuracy of the satellite telemetry data anomaly detection.

[0161] (3) The satellite health assessment method provided by the present invention obtains the score of each instrument according to the contribution degree of each instrument parameter and instrument calibration, and finally obtains the health score of the satellite by using the contribution degree of each instrument to the satellite health.

[0162] Embodiment 2

[0163] A computer device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the satellite health monitoring method based on an artificial intelligence algorithm in Embodiment 1.

[0164] Embodiment 3

[0165] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the satellite health monitoring method based on an artificial intelligence algorithm in Embodiment 1.

[0166] Embodiment 4

[0167] A computer program product, comprising a computer program, and when the computer program is executed by a processor, it implements the steps of the satellite health monitoring method based on an artificial intelligence algorithm in Embodiment 1.

[0168] Embodiment 5

[0169] A computer device, which may be a database, and its internal structure diagram may be as Figure 20 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store transactions to be processed. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements the satellite health monitoring method based on an artificial intelligence algorithm in Embodiment 1.

[0170] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the object or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions.

[0171] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided by the present invention can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided by the present invention can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0172] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0173] Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A satellite health monitoring method based on artificial intelligence algorithm, characterized in that: The method comprises: Constructing a telemetry parameter image database of the target satellite; the telemetry parameter image database includes characteristic images of multiple telemetry parameters within a period; the characteristic images include a normal characteristic image and multiple abnormal characteristic images, and the abnormal type corresponding to each abnormal characteristic image; Get real-time cycle data of multiple telemetry parameters; Determine the health status of the characteristic image of the data in the real-time period of each of the telemetry parameters according to the characteristic image in the telemetry parameter image database; the health status is normal or abnormal; when the health status is abnormal, output the corresponding abnormality type; Determine the health status score of each corresponding instrument according to the preset health status score corresponding to the health status of the characteristic image of the data in the real-time period of each telemetry parameter; Evaluating the health status of the target satellite according to the health status scores of the corresponding instruments and the preset weights of the corresponding instruments; Determining the health status of the characteristic image of the data in the real-time period of each of the telemetry parameters according to the characteristic image in the telemetry parameter image database specifically includes: Taking the characteristic image in the telemetry parameter image database as input and the health status of the characteristic image as output, training the deep learning model to obtain a health status recognition model; Inputting the characteristic image of the data in the real-time period of each telemetry parameter into the health status recognition model to obtain the health status of the characteristic image of the data in the real-time period of each telemetry parameter; The determining of the health status scores of the corresponding instruments according to the preset health status scores corresponding to the health status of the characteristic images of the data in the real-time period of each telemetry parameter specifically includes: Taking the characteristic image of the data in the historical period of the telemetry parameter corresponding to each of the instruments as input and the characteristic image in the corresponding telemetry parameter image database as output, the neural network model is trained to obtain a trained neural network model, and the parameter weight of the trained neural network model is used as the contribution of the telemetry parameter corresponding to each of the instruments to the health status of the corresponding instrument; Multiplying the preset health status score corresponding to the health status of the characteristic image of the data in the real-time period of each telemetry parameter by the contribution of the corresponding telemetry parameter to the health status of the corresponding instrument to determine the health status score of each corresponding instrument; The neural network model includes a self-attention network layer, a multiplication layer and a fully connected layer connected in sequence; Using the parameter weight of the fully connected layer as the contribution of the telemetry parameter corresponding to each of the instruments to the health status of the corresponding instrument; Inputting the characteristic image of the real-time period data of the telemetry parameters corresponding to each of the instruments into the trained neural network model to obtain the output data of the attention layer; The output data of the attention layer is multiplied by the first data of the parameter weight of the fully connected layer to obtain the contribution of the telemetry parameters corresponding to each of the instruments to the health status of the corresponding instrument.

2. The satellite health monitoring method based on artificial intelligence algorithm according to claim 1, characterized in that: Construct the telemetry parameter image database of the target satellite, including: Acquire historical period data of each telemetry parameter of the target satellite; Determining whether the health status of the characteristic image of the data in the historical period is normal; When the health status of the characteristic image of the data in the historical period is normal, taking the characteristic image of the data in the historical period as the normal characteristic image of the telemetry parameter; When the health status of the characteristic image of the data in the historical period is abnormal, the characteristic image of the data in the historical period is used as the abnormal characteristic image of the telemetry parameter, and the abnormal type corresponding to the abnormal characteristic image is marked; According to the normal characteristic images and abnormal characteristic images of each of the telemetry parameters, and the abnormality types corresponding to the abnormal characteristic images, a telemetry parameter image database of the target satellite is obtained.

3. The satellite health monitoring method based on artificial intelligence algorithm according to claim 1, characterized in that: The formula for evaluating the health status of the target satellite is: The health status score of the target satellite=the health status score of instrument 1×weight 1+the health status score of instrument 2×weight 2+…+the health status score of instrument n×weight n.

4. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the satellite health monitoring method based on an artificial intelligence algorithm as described in any one of claims 1 to 3.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the satellite health monitoring method based on artificial intelligence algorithm described in any one of claims 1 to 3 are implemented.

Images