A phase plane technique for non-linear system modelling

A technique for the time-domain modeling of nonlinear plants is developed. The system learns the plant characteristics by applying step inputs and recording the resultant phase-plane trajectories as a set of temporal vectors. The density of the vectors is controlled by a nonlinear grid in the phase plane. An updating mechanism allows plant changes to be tracked, and an inference strategy is developed to allow behaviour prediction from the stored model. Operation of the system is demonstrated by simulation results.<>