Model reference adaptive neuro-controller based on HMLP network

Abstract

In this paper, an Adaptive Neuro-Controller (ANC) based on HMLP Network is proposed as an alternative to a conventional adaptive controller. The ANC will be adapted such that the plant should follow a desired output, which is given by a reference model. The Model Reference Adaptive Control (MRAC) has been used to generate the desired output path and to ensure the output of the controlled system follows the output of the reference model. A neural network model, called Hybrid Multi Layered Perceptron (HMLP) network is used for this ANC as a basic network structure. The convergence rate of the ANC is further improved by using recursive least square (RLS) as the training algorithm. The capability of the ANC trained using recursive least square was demonstrated using two simulated data sets. The proposed ANC technique has been tested using a linear and a nonlinear plant with some variations in operating conditions. The simulation results indicated that this controller is adequate to control the system with unpredictable conditions and disturbances.