Evaluating Gravity-Assist Range Set Based on Supervised Machine Learning

2018 2nd International Conference on Aerospace Technology, Communications and Energy Systems (ATCES 2018)

航空航天工程;无线电电子学;能源学

Zhang, K.^1 ; Shang, H.^1 ; Chen, Q.^1 ; Qin, X.^1

Beijing Institute of Technology, Beijing

100081, China^1

关键词: Gaussian processes regressions (GPR);    Minimum mean squared error;    Percentage error;    Predicting models;    Regression method;    Strong nonlinearity;    Supervised machine learning;    Velocity increments;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/449/1/012021/pdf
DOI  :  10.1088/1757-899X/449/1/012021

学科分类：航空航天科学

来源: IOP

【 摘 要 】

The dynamics of gravity-assist (GA) trajectories contain strong nonlinearity, which makes the traditional methods for impulse transfer range set (RS) are intractable to deal with the gravity-assist RS. This paper develops a novel method to evaluate the gravity-assist RS based on regression methods in supervised machine learning (SML) field. The performances of three powerful regression methods with several common kernel functions are assessed. The Gaussian Processes Regression (GPR) method with Matérn 3/2 kernel is selected because of the minimum mean squared error (1.11×10-3 km2/s2). The predicting model based on GPR is constructed to make prediction form the orbital elements of destination orbits to the total velocity increment of corresponding optimal GA trajectories. The percentage error of predicting model is no more than 2%. Millions pairs of sample points are generated by the trained predicting model. The points with specified value of total velocity increment are extracted, of which the envelope constitutes the gravity-assist RS. Both of Venus GA and Mars GA trajectories are considered in this paper.