Genetic algorithm for multimodal continuous optimization problems
Anuradha Maria · 1995
Engineering practice is replete with black-box optimization problems characterized by the absence of well-defined relationships between the input variables and the output(s) of a system. These problems tend to be multi-variable, continuous, multimodal, and multi-objective. Various direct search methods are used to solve such optimization problems. Optimization based on the principles of natural genetics is an emerging area of research. The advantage of genetics-based methods is that they use random choice to guide a highly exploitative search, thereby striking a balance between exploration of the feasible domain and exploitation of good solutions. Because of the essential step of coding, however, genetic algorithms are poorly suited to solve continuous optimization problems. Almost all existing implementations of genetic algorithms operate on discrete domains or continuous domains discretized with a chosen precision. The objective of this dissertation is to develop a genetic algorithm for multimodal continuous optimization problems. The development of the GOCP (Genetic Optimization for Continuous variable unconstrained black-box Problems) algorithm, experimentation on the parameter settings, and the extent to which they affect the algorithm efficiency is discussed. The GOCP algorithm is extended to solve constrained black-box problems and the resulting algorithm (GOCCP) is applied to solve non-linear programming problems. The applicability of the GOCCP algorithm is extended to multi-objective constrained black-box problems. In addition, a hybrid algorithm that combines ideas from interior point methods in linear programming and GOCCP is described.