INTERNATIONAL JOURNAL OF COMPUTING ANTICIPATORY SYSTEMS, Volume 11, 2002 Published by CHAOS, ISSN 1373-5411 ISBN 2-9600262-5-X 1 Optimized Anticipatory Control Applied to Electric Power Systems T. E. FIENO D. T. BARGIOTAS* L. H. TSOUKALAS Purdue University, School of Nuclear Engineering, West Lafayette, IN 47907-1290 – USA fieno@helios.ecn.purdue.edu , tsoukala@helios.ecn.purdue.edu *Technological and Educational Institute of Chalkida Department of Electrical Engineering, Psachna, Evia, 34400 – GREECE bargiotas@teihal.gr Abstract. Electric generation control is performed in a distributed manner to supply power to geographically defined control areas. The goal of generation control is to keep the inadvertent flow of power across a control area's boundary as small as possible. If a difference exists between the power supplied and the power demanded in a control area, the load deficit or surplus would be either borrowed from or stored as the kinetic energy in rotating machines in the grid. This research addresses the challenge of matching the power demand of a local area grid with the power delivered by a pulverized coal-fired power plant. A neural network anticipatory controller for a model power plant coupled with a neural network time-series forecaster is presented to prescribe the power output into the grid. Keywords: Anticipatory Systems, Neural Networks, Deregulation of Power Systems, Applied Intelligent Systems 1. INTRODUCTION The differences between biological organisms and artificial systems are often the subject of debate in engineering and research circles. The ability to incorporate the levels of flexibility, adaptability, and goal-driven action of the biological world into a manmade entity is priceless when considering the potential for an autonomous control system. One of the differences between life and artifact is the anticipatory nature of organisms. Anyone who has ever played ―fetch‖ with a dog knows that after only a few trials the dog will learn to predict when you will throw the object. Even more astounding is that the dog will learn trends about where the object will land based purely on observation rather than on understanding laws of motion. And yet, even though anticipatory systems perfuse our everyday life, nearly all engineering control systems act in a purely feedback mode. Predictions about the future states of the environment are just now being considered for incorporation into advanced control design. One of the barriers to their introduction is the difficulty in quantifying and operating on likely or unlikely events in the future control horizon. Although anticipatory systems have always been an integral part of everyday biological life, the study of anticipatory systems is a relatively new area in the field of science and engineering (Mikulecky, 1996), (Tsoukalas, 1998), (Tsoukalas, 1991). The overall question the study described here seeks to answer is whether or not the notion of anticipation can be exploited to make an engineering system perform better or in a safer, more flexible, and generally more optimal manner. To address this question in a technically sound fashion, an iterative system for designing a combined predictive controller and environmental forecasting routine is utilized. The system is examined through application to the regulation of an electric power grid and comparison with conventional approaches. To describe an anticipatory system in an abstract and more general manner, we view it as a system that utilizes predictions about the future states of itself and of its environment to direct its present actions (Rosen, 1985). For this paper, the system environment is defined as the entity that the controller and plant (the system) are attempting to regulate. In the aforementioned example, the dog knows the likely trajectory of the object being chased and the predicted trajectory of its own body. The dog is trying to match the anticipated future state of its body with the predicted future state of the object and thus forms an anticipatory system. 2. ANTICIPATORY TEST SYSTEM Two systems are needed to study the usefulness of anticipatory regulation applied to the electric grid. The first is the power consumption data for a given control area, and the second is the mathematical simulation of the power- producing facility. These conform to the environment and the controlled machine. 2.1 Data Test System To assess the flexibility and power of the proposed model, data from an electric power grid will provide the testbed. Power consumption data obtained from Commonwealth Edison are used. Fig. 1 shows the power demand recorded for a specific residential feeder on the Commonwealth Edison grid for July of 1999. The power data are sampled every 15 minutes. Two data sets are used to design the prediction system (July and August 1999), and three data sets are used to