An online generation method of ascent trajectory based on feedforward neural networks

Abstract

With the development of advanced guidance technology and onboard calculation ability, a drawback of poor real-time performance of traditional trajectory optimization methods has emerged. In this paper, a neural network is applied to generate the online ascent trajectory of a solid rocket. With the advantage of its nonlinear mapping capability, the neural network is used to approximate the trajectory computation model to reduce the burden of onboard computation. In particular, according to the different environment characteristics of the atmospheric and exoatmospheric portions, two neural networks (a control-network and a costate-network) are trained offline. The former generates the pitch angle command in real time during the atmospheric portion. The latter provides good initial costate guesses for the indirect method in exoatmosphere, and then the trajectory deviation after flight through the atmosphere is corrected by solving the optimal control of the remaining time online. Finally, numerical simulations are performed to show the effectiveness and robustness of the proposed scheme.