Autonomous Mission Planning Method Of Observing Satellites Based On Learning Decision-making

Due to the enhancement of satellite capabilities and the complexity of user requirements,the traditional mission planning based on ground centralized management mode has been unable to adapt to new needs and challenges.The traditional mission planning mode does not solve the environmental uncertainty problem well,resulting in slow response and low robustness.Satellite autonomous mission planning is an effective way to solve the satellite rapid response in uncertain environments.In this paper,the machine learning method is combined with the satellite mission planning,and the autonomous mission planning method of observing satellites based on learning decision-making is proposed.The main work in this paper is summarized as follows:(1)Through the analysis of the working principle and observing process of agile satellites and non-agile satellites,the differences between them are clarified,and the autonomous mission planning models of agile satellites and non-agile satellites are established respectively.Through the analysis of the shortcomings of the traditional management mode,the management mode and process of autonomous mission planning are proposed.(2)The autonomous task planning method based on learning decision-making is designed: the method framework which includes ground learning and on-board decisionmaking is determined.The ground part is responsible for training the autonomous task decision-making model with the historical planning scheme,and the real-time decisionmaking is carried out on the satellite by using the trained model;Based on the full analysis of the autonomous mission planning model,the ten-dimensional feature variables used to describe the meta-tasks are extracted,the input and output of the autonomous task decision-making model are defined,and the training set and test set are designed.An on-board autonomous mission planning method based on time sequence is designed.The meta-task is used as the unit for planning.Each time the satellite only decides whether the current task is executed,and the planning problem is decomposed into real-time decision-making problems,which fully combines the advantages of machine learning in solving decisionmaking problem.Based on BP neural network,support vector machine and random forest,the autonomous mission decision-making model is realized,and the elements and parameters of each model are clarified.(3)In the experimental part,the evaluation indicators are designed to verify the efficiency of the autonomous mission decision model and our method.Comparing the results of autonomous mission decision with the decision results of historical planning,it is proved that the decision model has high decision accuracy.Comparing the results of our method with the results of the historical planning scheme,it is proved that our method greatly reduces the running time of mission planning on the basis of maintaining the total observation benefit.