Intelligent system modelling and simulation using hybrid recurrent networks
David Al-Dabass, David J. Evans, Siva Sivayoganathan · Nottingham Trent University's Institutional Repository (Nottingham Trent Repository) · 2002
Numerous intelligent systems in practice exhibit complex behaviour that cannot be easily modelled using simple nets. In this paper we re-cast this problem in terms of hybrid recurrent nets, which consist of combinations of static nodes, either logical or arithmetic, and recurrent nodes. The behaviour of a typical recurrent node is modelled as a second order dynamical system. The causal parameters of such a recurrent node may themselves exhibit temporal tendencies that can be modelled in terms of further recurrent nodes. Layers of recurrent nodes are added until a complete account of the behaviour of the system has been achieved. Algorithms are given to abduct the values of the parameters of these models from behaviour trajectories of intelligent systems. One novel aspect of the work lies in having a simple hierarchical 6th order linear model to represent a fairly complicated behaviour encountered in numerous real examples in finance, biology and engineering.