Neural Network for Preliminary Orbit Determination

摘要

This paper proposes the use of Neural Network (NN) as an alternative to Extended Kalman Filter (EKF) for preliminary orbit determination of satellite using short span simulated observational data. Various levels of noise and biases have been included in the simulated observational data to make the results more realistic. This method is not confined for any particular inclination of the orbit. The standard backpropagation and Lambda-Gamma algorithms made use to train the NN are shown to provide good estimate of the six orbital elements characterising the satellite orbit. Moreover, the proposed method does not suffer from any divergence problem. Thus, the feasibility of satellite orbit determination using data over limited time-span makes the proposed NN approach attractive.