Dynamic Back Propagation based MRAC with Fuzzy Emulator for DCDC Converter

Abstract

Model Reference Adaptive Control (MRAC) is commonly used in traditional neural network based adaptive controller design. Neural network based MRAC often requires plant emulator when neural controller is connected in between the plant and the control input. Several methodologies for constructing the plant emulator have been proposed but they require either off-line training of the plant emulator or an exact mathematical model of the plant. This limits the capability of a neural system to generalize the controller for a nonlinear plant. Authors suggest a simple combination of Nero-Fuzzy technique to address this problem. Newly introduced dynamic back propagation learning framework for the training of feed forward neural networks with a simple fuzzy emulator is proposed. This design work can be extended for the class of nonlinear system with structural uncertainty. This simple new method incorporates the online training algorithm. Simulation studies are carried out in Matlab for the DC-to-DC converter and the prototype of buck converter is implemented using DSP processor, to obtain real time response. The overall simple control methodology requires less memory in DSP. The practical and simulation results are compared and contrasted with PID controller. The proposed method can be used in most of the industrial applications. Ill. 15, bibl. 22 (in English; summaries in English, Russian and Lithuanian).