Integrated Control of Wind Farms, Facts Devices and the Power Network Using Neural Networks and Adaptive Critic Designs

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

Intellectual Merit: Building on earlier success with smaller systems, this team will develop general-purpose integrated control systems using brain-like design principles to handle larger and more complex systems than have been ever been controlled in the past using such principles. They will be integrating together the use of adaptive dynamic programming (sometimes called "reinforcement learning" or "adaptive critics"), recurrent neural networks (which provide unique capabilities in approximating nonlinear dynamical systems), learning and adaptation, and particle swarm optimization techniques. They will be developing this integration in the context of managing a large complex real system (initially in computer simulation, and then in the laboratory) dominated by partially observed continuous variables, nonlinearity and random disturbances. Broader benefits: The testbed to be controlled represents large windfarms using the most advanced, affordable and efficient (but hard to manage) systems of wind turbines and electronic power control hardware ("FACTS"). The ability to achieve such reliable control and efficiency, at low cost, will be crucial to the goal of supplying 20 percent of the world's electrical energy by wind. It will be crucial to making intermittent power like wind more valuable to the grid - and hence more deserving of larger payments from the grid to wind generators, in a rational market system. The team also has active partnerships with Africa and with Brazil, which can supply some of the advanced low-cost FACTS technology needed to achieve success - and perhaps also some additional testbeds. This project may be a crucial step in bring the ideals of an intelligent adaptive power grid into the real world.