Research On Observation Scheduling Model And Method Of Astronomy Satellite

Observation scheduling has always been the core decision-making issue in astronomical observation satellite missions.This problem mainly solves the contradiction between limited satellite observation resources and the infinite demand for space mystery detection.While ensuring the safe,effective and stable operation of satellites,the maximum efficiency of the satellite observation capabilities can be achieved to obtain the detection data needed and promote the progress of astronomical research.However,when executing the observation plan,it is inevitable that it will be unpredictably interrupted,so that the observations can not be carried out strictly according to the established plan,such as the failure of observation payload and the insertion of unpredictable high priority tasks.Especially since the 21 st century,time domain astronomy is becoming the latest trend in astronomy.Observations of transient sources and bursters account for an increasing proportion of astronomical satellite missions.Transient sources and bursters are unpredictable both in time and space.Observations of these targets are called Target of Opportunity(To O)observations.A large number of To O bring new challenges to the scheduling problem.To meet the unpredictable observation requirements,this paper conducts research work from three aspects: single-satellite proactive scheduling,multi-satellite centralized scheduling and multi-satellite distributed scheduling.The main research results are as follows:(1)Based on the background of SVOM,the mission planning with both the conventional task and To O task of a single astronomical satellite is studied.It solves the problem that the unpredictable interruption of the observation task caused by the transient source delays the observation task in the original plan.In single-satellite mission planning,to maintain the relative stability of observation completion time in the original observation plan while arranging To O,two optimization objective functions,static criticality and dynamic criticality,are proposed considering the observation window information of known tasks,and a NSGA-II-based algorithm is designed and implemented.The proposed algorithm improves the weighted on-time completion rate and the weighted completion rate of the observation tasks in the original plan.It effectively solves the problem of maintaining the stability of pre-scheduled observations' completion time in the scheduling results under the influence of To O.(2)On the background of large-scale satellite observation cluster that provide continuous observations for To O,an on-line scheduling algorithm for multi-satellite cooperative observation for unpredict tasks are studied.Firstly,according to the specific application requirements of the problem,an optimization model for on-line scheduling of multisatellite cooperative observation for unpredict tasks is established.Next,the algorithm for observability of satellites to astronomical objects is studied in detail.In order to solve the problem,whose optimized objective function cannot be calculated directly in the scheduling model,two criteria,Time Coverage and Sky Coverage,are proposed.The first one is to describe the temporal continuity of observation of the target,and the second one is for the distribution of the observation capability of the satellites combination to To O,which may occur in any direction of the universe.Then,optimized the objective function of the focused problem is approximated with the above two criteria.Based on this approximation,a binary tree cut algorithm is presented.Finally,simulation experiments are designed to verify the proposed algorithm.The results show that the overall capability of the satellite observation network can be effectively improved by considering both the two criteria,especially the Sky Coverage.It enables the satellite observation network to reserve resources for the coming To O.(3)In the distributed scheduling of the astronomical satellite network with To O as the primary task,satellites cannot obtain the available status of other satellites.Therefore,it is difficult for satellites to cooperate to accomplish observation tasks.The cooperative observation cluster working mode are designed by adopting the centralized training and distributed execution paradigm.A deep neural network algorithm based on multi-agent policy gradient sharing is proposed.A method to describe the state of satellite using time coverage is also proposed.Compared with the comparison algorithms,this method has better scheduling capability in experimental data.