High Performance Intelligent Controller for Systems with Unknown Dynamics

The first generation of intelligent control applications was for systems with unknown dynamics, with on-line learning of the neural network weights. The second generation of intelligent control applications has generally provided off-line design methods that first solve the optimal design equations, and then apply the resulting control to the system with no on-line learning. The system dynamics must generally be known. We propose a novel, high performance, and third generation intelligent control framework, which is applicable to systems with unknown dynamics, including air vehicles, Unmanned Air Vehicles (UAVs), Micro-Air Vehicles (MAVs), robots, motors, and power systems. The framework consists of two cooperative on-line learning modules. First, an adaptive critic network is used to approximate the cost function (performance index) of the system. Second, another action network uses the critic information to adjust certain control parameters in the action network. Major advantages include: 1) system dynamics can be unknown; 2) controller achieves optimal performance in the steady-state; 3) the overall system performance is robust to disturbances and unknown changes in the system dynamics.