Comparison of Guided Random Search Techniques to Minimize Objective Functions of Continuous Variables

DAVID HANSMANN; GUIDO FIER; G. R. TERRANOVA

Bariloche

Congreso; XIV Congreso Regional de Física Estadística y Aplicaciones a la Materia Condensada; 2016

Universidad Nacional de Cuyo

p { margin-bottom: 0.25cm; line-height: 120%; }a.cjk:link { }a.ctl:link { }Single-objectiveoptimization (SOO) deals with the study of those kinds of problems inwhich one has to minimize or maximize functions of real or integervariables. In contrast to multi-objective optimization (MOO)(multi-criteria or multi-attribute optimization), SOO deals with thetask of optimizing simultaneously one or more non-conflictingobjectives with respect to a set of constraints. In this context, inparticular if the function is unimodal, one can choose among manygood alternative algorithms, rangingfrom Newton´smethodtoSimplex Methods to Random Search Methods.In this work wecompare velocity of convergence, precision, robustness and generalperformance of different randomsearch optimizationalgorithms using a set ofproper artificial landscapes(test functions) likeGoldstein?Price function,Ackley´s function, Beale´s function and others. Theoptimization algorithmsinclude classical random search techniques and particleswarm optimization (PSO)techniques, wherethe latter are based on dynamic of swimming E.coli bacteria, known asrunand tumble.In addition we test and compare several modifications andsimplifications of the E.coli dynamicsthatwere used during the last years to model swimming and swarmingbacteria.