A Novel Multi-Agent Controller for Dynamic Systems based on Supervisory Loop Approach

This paper presents a new intelligent agent supervisory loop based approach for dynamic system control. The scheme consists of three software agents that work in an autonomous manner for the precision control of a system that shows multiple modes and drastic parametric ‘jumps’. The first two agents are based on artificial intelligent techniques (fuzzy systems and Radial Basis Function Neural Network (RBFNN)) that monitor a closed loop system using a supervisory loop concept. The third agent acts as a traditional Model Reference Adaptive Controller (MRAC) which controls the system dynamics and helps to keep the error inbound. The scheme efficiently alleviates the burden on soft computing techniques due to the strength of the MRAC and at the same time intelligently controls system when nonlinear and functional changes occur. Theoretical analysis and implementation results show that this approach has immense potential in modular agent based controller design.