Using RBF-based neural networks for dynamic multidimensional potential field modeling

This paper describes an application of RBF-based artificial neural networks for dynamic multidimensional potential field modeling. Potential field in question is used for route search in a multidimensional state space. We describe requirements to action sequence selection algortihm. Taking these requirements as basis we deduce the applicability of potential field method that we utilize to forecast action sequence. Research that was carried out suggests the effectiveness of RBF neural networks application for multi-dimentional potential field approximation. This method speeds up calculations required to define direction of the field in a certain point. Computer simulation results are presented.