Neural net based mrac of nonlinear plants with unknown structure

Abstract EN

Model reference adaptive control (MRAC) for nonlinear plants of unknown structure is considered.

Amulti-layered feed-forward neural network (MFNN) is used as the controller.

Dynamics are introduced in the controller by injecting the setpoint, plant input, and the plant output through the dynamics generating filters (DGF).

The on-line adaptation of the MFNN weights is carried out using the recently-developed concept of block partial derivatives (BPD).

Two algorithms: one based on the MIT rule and the other based on the strictly positive real (SPR) rule are developed.

When a linear controller replaces the MFNN, the proposed algorithms reduce to the well-established linear MRAC algorithms.

Results of simulation studies are also reported.