Shuttlecock flight trajectory modeling and analysis using linear and neural network ARX

Abstract

In this paper, a shuttlecock flight trajectory model using Linear ARX and Neural Network ARX (NARX) is proposed and compared. Several ARX and Neural Network ARX model configurations were created. Every Linear ARX and Neural Network ARX model configurations differ in term of number of autoregressive and exogenous components. Every model configurations used the horizontal trajectory as inputs and flight trajectory as output. The experiments showed that, ARX and Neural Network ARX could predict the flight trajectories 80 - 90% and simulate the flight trajectories 60 - 70%. In addition, it also shows that if the regressor was chosen properly, the Neural Network ARX would outperform the Linear ARX. Nonetheless, the wrong choice of autoregressive and exogenous components will lower the Neural Network ARX model configuration performance significantly. On the contrary, although still affected as well, Linear ARX models were not as vulnerable as Neural Network ARX models in term of choice of autoregressive and exogenous components they have.