Learning classifiers shape reactive power to decrease losses in power distribution networks

Abstract

Energy is continuously dissipated in power systems due to electrical resistances in transmission and distribution lines. Part of the losses is due to reactive power that travels back and forth in power lines, all the way from power sources to load points. Capacitors can provide local complementary reactive power that decrease losses. As energy loads vary in intensity and characteristics with time, better results are achieved when capacitors are controlled to match changing reactive power profiles. This paper explores the possibilities of adopting a learning classifier systems framework to control capacitors reactive power output in power distribution networks. A "corps" of classifiers keeps the distribution network under permanent surveillance against change in reactive power profiles that may increase losses. Whenever a significant change occur, a classifier steps in and suggests new capacitor taps. Classifiers improve their performance through permanent evolution. Case studies evaluate the methodology with applications to networks available in the literature.