Mission Planning Of Space Target Inspection Task Based On Deep Neural Network

When an on-orbit service spacecraft need to serve multiple targets,the servespacecraft need to transfer from one target to other targets one by one,the progress of serve-spacecraft transferring between multiple targets is called space target inspection task.Based on the space target inspection task,this paper has carried out relevant research from the aspects of rapid estimation of speed increment for orbit-transfer,inspection sequence optimization of single platform inspection task and task allocation problem of multi-platform inspection task.The main result achieved in this dissertation are summarized as follows:1.A rapid estimation model of orbit-transfer speed increment is established based on deep neural network.The calculation speed of orbit-transfer speed increment would affect the planning efficiency of space target inspection task directly.To improve the planning efficiency,the deep learning method is used to replace the traditional speed increment calculation method.The factors which may influence the learning accuracy are analyzed,such as training data set scale,hyperparameters of the deep neural network,and training method.The simulation results show that the deep neural network can learn the speed increment in a satisfactory accuracy,the RMSE error is about 0.0088,and this error is acceptable for the requirements of preliminary task design.2.A sequence optimization method for single-spacecraft-to-multi-targets inspection task is proposed based on the well-trained deep neural network sub-model.The single-spacecraft-to-multi-targets inspection task is analyzed at first,and a mathematical model of inspection task is established with a goal of reducing the cost of space target inspection task.A hierarchic optimization method is proposed,the deep neural network is used as the inner optimizer to solve the speed increment,and the integer-coding-genetic-algorithm is used as the outer optimizer to solve the inspection sequence.The simulation results show that using deep neural network as the inner optimizer can improve the calculating speed of sequence optimization greatly,for a case with 5 targets,using SQP operator as inner optimizer will take 1.8 hours,while using deep neural network operator as inner optimizer only need 7.784 seconds.3.A target allocation method for multi-spacecrafts-to-multi-targets inspection task is proposed based on the well-trained deep neural network sub-model.When several spacecrafts carry out the space inspecting tasks corporately,it is necessary to allocate targets to each spacecraft.The multi-spacecrafts-to-multi-targets inspection task is analyzed at first,and a mathematical model is established.The well-trained deep neural network model is adopt to calculate the evaluation matrix,which is relevant to the orbit-transfer speed increment.Three kinds of quantitative relationships between the serving spacecraft and the target spacecraft has been discussed,and a modified Hungarian algorithm is adopt to solve the three different circumstance.The simulation results show that the combination of Hungarian algorithm and deep neural network can be used to allocate the space inspection targets efficiently;the improved Hungarian algorithm can effectively allocate the tasks when the number of the server spacecraft does not match the number of the target spacecraft;based on the task target assigned by the Hungarian algorithm,the inspection sequence is further optimized to reduce the cost of orbit transfer.The method using deep learning method to estimate the speed increment has been discussed detailed in this paper.Combined with integer coding genetic algorithm and Hungarian algorithm,the problems of Space inspection sequence optimization and target assignment are solves.The methods and conclusions concluded in this paper can provide valuable reference for mission planning problem of on-orbit service mission.