Adaptive decoupling control of multivariable nonlinear non-minimum phase systems using neural networks

We develop an adaptive neural decoupler for discrete-time multivariable nonlinear non-minimum phase systems. Using Taylor's formula, the nonlinear system can be viewed as a linear non-minimum phase system with measurable disturbances. The feedforward decoupling strategy which was used in linear systems is employed and static decoupling can be achieved. For unknown systems, one group of neural networks are trained off-line to estimate the Jacobian matrix, another group are used to approximate the nonlinear terms online. Adaptive decoupling is thus developed.