Research On Discrimination Method Of On-orbit Satellite Operation State Based On Machine Learning

With the continuous development of China's aerospace industry,more and more satellites are launched.In order to ensure the normal operation of the satellite and obtain the operating state of satellite in orbit,it is necessary to monitor the satellite in real time.At present,the ground monitoring personnel mainly judges the operating state of the satellite in orbit by monitoring the threshold of important telemetry parameters that is summarized by experts.This method has many disadvantages.For example,the system needs to set a large number of thresholds in advance,the setting of thresholds depends on expert experience,and the system's scalability is poor.When the satellites that need to be monitored change the parameter threshold due to performance changes,these thresholds need to be reset.Aiming at these problems,this thesis takes the satellite of Quantum Experiments at Space Scale as an example to carry out the research on the discrimination method of on-orbit satellite operation state based on machine learning,which makes it possible to automatically identify the on-orbit state of satellite.The research content of this thesis has important research significance and practical value.The main work and innovations of this thesis are as follows:1.Taking the satellite of Quantum Experiments at Space Scale as an example,analyzing the composition and characteristics of telemetry data.The thesis investigates the research status of satellite telemetry parameters analysis,analyses the composition and statistical characteristics of satellite telemetry parameters,and analyses and mines the correlation between single-dimensional telemetry parameters of the satellite of Quantum Experiments at Space Scale.2.Analysis and Modeling of the problem of Discriminating the orbital Operation State of Quantum Experiments at Space Scale.The Quantum Experiments at Space Scale generates a large amount of telemetry data during its orbital operation.These telemetry data can reflect the on-orbit operation states of the Quantum Experiments at Space Scale,including the type of optical experiment completed by the satellite,whether the successful completion of optical experiment and the working mode of payload.Firstly,according to the correspondence between the telemetry data and the type of optical experiment on the satellite,the mathematical model of the correspondence relationship is established.This allows the model can judge the type of optical experiment completed by the satellite based on telemetry data transmitted to the ground.Secondly,for the experiment of Star-ground Quantum Entanglement Distribution,there are two working modes for the payload: one-way entanglement distribution and two-way entanglement distribution.The thesis establishes a mathematical model to discriminate which mode the payload is in.Combined with machine learning theory,the discriminant model of optical experiment type of satellite can be abstracted as multi-classification problem,and the discriminant model of payload working mode can be abstracted as two-classification problem.In this thesis,different machine learning algorithms are selected to train the above models.3.Integration and preprocessing of experimental data sets,and training and verification of satellite on-orbit operational state discriminant models based on different machine learning algorithms.According to the characteristics of telemetry data of Quantum Experiments at Space Scale,many pre-processing methods of telemetry data is designed.In order to obtain the types of optical experiments and payload modes during the on-orbit operation of the Quantum Experiments at Space Scale,the thesis parse the short-term scientific experiment plan documents,and extract the relevant content.Finally,the telemetry data of the satellite is integrated with optical experiment type and payload working mode data according to the time correspondence,and the data set for model training is obtained.This thesis investigates the existing machine learning algorithm,selects the machine learning algorithm suitable for the mathematical model built in this thesis,and trains and verifies the model on the historical telemetry data set of the Quantum Experiments at Space Scale,and analyzes the experimental results.The experimental results show that the proposed method has a discriminative accuracy of more than 99% without the expert prior knowledge,and the experimental results have a strong credibility.In addition,the application of machine learning technology to satellite ground monitoring of satellite can improve the efficiency of ground surveillance personnel.The proposed method has strong scalability and can be conveniently used in the monitoring of other satellites.