Neural Control of Nonlinear Non-minimum Phase Dynamical Systems

This paper argues that the 'optimization along trajectories' formulation of the distal learning approach [1] doesn't work on non-minimum phase dynamical plants [2] and the 'local optimization' one doesn't work on plants with variable or badlyestimated time-delay. It then proposes a modification of the procedure and illustrates improvements on the control of a simulated irrigation canal. This paper was published in the proceedings of the International Conference on Artificial Neural Networks and Genetics Algorithms (ICANNGA'95, Alès, France, April, 18th-21st, 1995). A.Toudeft. Neural Control of Non-minimum Phase Dynamical Systems NEURAL CONTROL OF NONLINEAR NON-MINIMUM PHASE DYNAMICAL SYSTEMS Abdelmoumène TOUDEFT CEMAGREF BP 5095 34033 Montpellier cedex 1, France e-mail: [email protected] LAFORIA Université Paris 6 France Abstract This paper argues that the 'optimization along trajectories' formulation of the distal learning approach [1] doesn't work on non-minimum phase dynamical plants [2] and the 'local optimization' one doesn't work on plants with variable or badly-estimated time-delay. It then proposes a modification of the procedure and illustrates improvements on the control of a simulated irrigation canal.This paper argues that the 'optimization along trajectories' formulation of the distal learning approach [1] doesn't work on non-minimum phase dynamical plants [2] and the 'local optimization' one doesn't work on plants with variable or badly-estimated time-delay. It then proposes a modification of the procedure and illustrates improvements on the control of a simulated irrigation canal.