Research On Satellite Formation Reconfiguration Planning Based On Surrogate Model

With the continuous maturity of microsatellite technology and the continuous development of distributed collaborative control technology,the collaborative work of multiple small satellites to complete complex space exploration tasks has become a research hotspot in the international aerospace field.According to the different requirements of the mission,the formation configuration can be reconfigured to meet the requirements by controlling the satellite formation.This paper takes the formation of microsatellites in low earth orbit as the object,aiming at the shortcomings of the traditional optimization method,such as high computational overhead and low optimization efficiency,when solving the reconfiguration planning problem of satellite formation considering the optimal fuel consumption and collision avoidance constraints,the optimization method based on surrogate model is used to study the two-impulse reconfiguration planning of satellite formation.Firstly,the technical methods needed to build the surrogate model are studied and analyzed,which are experiment design,model training and model accuracy evaluation index.The basic principles and construction methods of four commonly used surrogate models,namely Polynomial response surface(PRS),Radial basis function(RBF),Kriging(KRG)and Artificial neural network(ANN),are given.Two low dimensional functions are used to visually demonstrate and compare the predictive ability of each surrogate model to unknown functions.Then,the surrogate model method for fast prediction of the shortest distance between any two satellites in the process of satellite formation reconfiguration is studied.Aiming at the reconfiguration of two-impulse maneuver orbit change under Clohessy-Wiltshire(CW)equation of relative motion,PRS,RBF,KRG and ANN surrogate models for the shortest distance between satellites in the process of space circular formation size adjustment and space circular formation reconfiguration from elliptical formation to space circular formation are constructed respectively,and the comparison is made in terms of prediction accuracy and modeling efficiency.The results show that when the training set size is the same,when the design space dimension of the shortest distance surrogate model between satellites is low,such as the space circular formation size adjustment with 8-dimensional input,the accuracy and efficiency of the ANN model are comprehensively the best,When the design space dimension of the surrogate model is high,such as the ellipse formation with 16-dimensional input is reconfigured into a space circular formation,the accuracy of the KRG model is the best.Finally,the problem of collision avoidance planning for twoimpulse reconfiguration of satellite formation based on surrogate model is studied to reduce the computational cost caused by dealing with collision avoidance constraints in the optimization process.Two surrogate optimization algorithms,EGO global optimization algorithm and online modified the shortest distance surrogate model optimization algorithm,are used respectively,and compared with the results of direct optimization using the shortest distance model with high fidelity.The results show that the two optimization methods based on surrogate model are superior to the direct optimization method based on the shortest distance model with high fidelity in both efficiency and accuracy;For the size adjustment of the space circular formation with a prediction error rate of only 2% of the surrogate model,the optimization accuracy and efficiency of the online correction surrogate model optimization algorithm are better than the EGO algorithm,while for the reconfiguration of the elliptical formation with a prediction error rate of7.4% into the space circular formation,the optimization accuracy is better than the EGO algorithm,but the optimization efficiency is slightly lower than the EGO algorithm.