An explicit guidance system based on neural networks is proposed. The steering law is a function of parameters that are set up according to some optimal criterion. Optimization can be carried out either by semi-analytical approximated methods or by numerical methods. By optimization a guidance function is defined (p) under bar = G ((X) under bar,t) which transforms the current state of the system (X) under bar into the optimal parameters value e. In general calculation of (p) under bar is rather long and involved, and it is not allowed by an on-board computer. Neural networks can fit any nonlinear function with very high accuracy [1], so the basic idea is to train a neural network in order to fit some tabulated data of the optimal guidance function G. If the algorithm for finding the optimal function is replaced on-board by the sub-optimal ''neural'' approximation (p) under bar = N ((X) under bar,t), this system can reduce the on-board computer memory storage and the computing loads for the optimal steering program. In this article the neural guidance is described in some detail. The performances of this guidance system are investigated by application to some test cases. Even if the neural guidance is sub-optimal, the launcher gets to the final state with very high accuracy in the presence of relatively high disturbances. (C) 1997 Elsevier Science Ltd.
Guidance of small launchers using neural networks / Santoni, Fabio; Graziani, Filippo. - In: SPACE TECHNOLOGY. - ISSN 0892-9270. - STAMPA. - 16:5-6(1996), pp. 303-306. (Intervento presentato al convegno IFAC Symposium on Automatic Control in Aerospace tenutosi a BEIJING, PEOPLES R CHINA nel 1995).
Guidance of small launchers using neural networks
SANTONI, Fabio;GRAZIANI, Filippo
1996
Abstract
An explicit guidance system based on neural networks is proposed. The steering law is a function of parameters that are set up according to some optimal criterion. Optimization can be carried out either by semi-analytical approximated methods or by numerical methods. By optimization a guidance function is defined (p) under bar = G ((X) under bar,t) which transforms the current state of the system (X) under bar into the optimal parameters value e. In general calculation of (p) under bar is rather long and involved, and it is not allowed by an on-board computer. Neural networks can fit any nonlinear function with very high accuracy [1], so the basic idea is to train a neural network in order to fit some tabulated data of the optimal guidance function G. If the algorithm for finding the optimal function is replaced on-board by the sub-optimal ''neural'' approximation (p) under bar = N ((X) under bar,t), this system can reduce the on-board computer memory storage and the computing loads for the optimal steering program. In this article the neural guidance is described in some detail. The performances of this guidance system are investigated by application to some test cases. Even if the neural guidance is sub-optimal, the launcher gets to the final state with very high accuracy in the presence of relatively high disturbances. (C) 1997 Elsevier Science Ltd.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.