Genetic Programming for Lighting Control Using Frequent Trees and Depth Information

Abstract

Genetic Programming (GP) is one of the widely used and practical optimization methods in the problem domain represented by a tree structure; hence, GP has been used to produce instances as a competitive method with human-produced ways, and numerous researchers have applied GP to a wide variety of fields, including electrical circuits, robotics, image filters. Namely, it is important to explore new application areas. In this paper, we define the generalized lighting control problem, and present a novel method using GP for individual lighting. Previous researches have reported that providing each worker's desired illuminance can enhance intellectual productivity and save energy consumption; however, it is difficult to determine an optimum light intensity which satisfies both different desired illuminance values and energy-saving. The proposed method automatically generates lighting control rules, and evaluates rules with observed illuminance data. According to the Building Block Hypothesis, we also propose a new subtree encapsulation method for lighting control based on frequent trees and depth information. First, by using three well-known benchmark problems, we verify that the proposed encapsulation method improves solution quality. Next, we demonstrate its effectiveness for the generalized lighting control problem.