A Concept for Real-Valued Multi-objective Landscape Analysis Characterizing Two Biochemical Optimization Problems

摘要

Landscape analysis is an established method to provide an insight into the characteristic properties of an optimization problem with the aim of designing a suitable evolutionary algorithm for a given problem. However, these conventional landscape structures require sophisticated notions for multi-objective optimization problems. This work presents a real-valued multi-objective landscape analysis concept that allows the investigation of multi-objective molecular optimization problems. Sophisticated definitions for ruggedness, correlation and plateaus on multi-objective real-valued landscapes are introduced and indicators are proposed for this purpose. This landscape concept is realized on a generic three- and four-dimensional biochemical minimization problem and the results of this analysis are discussed regarding the design principles of a multi-objective evolutionary algorithm.