AnEvolutionary Algorithm forOptimizing Functions withUV
Hiroshi Takeichi, Isao Ono · 2006
Thefunction optimization isoneofthemost important optimization problems. Inapproaches tofunction optimization byevolutionary computation, areal-coded genetic algorithm, UNDX+MGG, showsgoodperformance on multimodal functions withepistasis among parameters. However, UNDX+MGG hasaproblem thatitsperformance is goodonfunctions withbigvalley structures butdeteriorates on those withtheUV structures. On theotherhand,ISMshows goodperformance on functions withtheUV structures. However, ISMhastwoproblems that1)itfails insearch when theregion oftheVvalley including theoptimum isverynarrow and2)its performance deteriorates onfunctions withbigvalley structures. Inthispaper,we propose a new evolutionary algorithm thataimsat overcoming theproblems of UNDX+MGG andISMandexamine itseffectiveness through someexperiments. I.INTRODUCTION Thefunction optimization isoneofthemostimportant optimization problems. Thegenetic algorithm (GA)isoneof thepromising approaches tofunction optimization problems. InGAsforfunction optimization problems, bit-string GAs that employ binary orGraycoding asrepresentations have beenusedtraditionally. However, ithasbeenpointed outthat bit-string GAscannot find accurate solutions compared to other optimization methods (3). GAsthat employ real-number vectors asrepresentations arecalled real-coded GAs.Real-coded GAshavebeen proposed andreported toshowbetter performance than