A Comparative Study of Diversity Preservation Techniques in Multiobjective Evolutionary Algorithms


Most real world applications found in today’s world necessitate dealing with certain common issues. These competing, often conflicting problems have kept researchers around the globe inquisitive and interested over the years and continue to do so, attributing to several open questions in the area. These problems, which deal with two or more objectives and invariably, involve large and complex search spaces, are referred to as multi-objective optimization problems (MOP’s). Although several traditional methods has been put forth and tested, evolutionary algorithms is being reckoned to be one of the approaches that provide efficient and effective solutions to these challenging problems, mainly because of its ability to deal with problems that are multi-objective in nature. These algorithms are, naturally termed as multiobjective evolutionary algorithms. Evolutionary algorithms are classified into three major forms: Genetic Algorithms (GA), Evolutionary programming (EP) and Evolutionary strategies (ES). Owing to the popularity of this area of research, several new approaches based on evolutionary techniques has evolved over the years. Additionally, many successful attempts towards improvement of these existing methods have emerged too. Moreover, other nature inspired approaches like particle swarm optimization and immune systems are widely researched as well. This thesis attempts to summarize and classify information on these biological inspired approaches, highlighting the importance of analyzing the research techniques followed by them thereby motivating researchers to come up with novel ideas for exploiting the search capabilities of these algorithms. A comparative analysis and study of the main algorithms are also provided based on diversity measures, along with their advantages and disadvantages and application areas. New approaches are proposed through hybridization methods on diversity techniques in multi-objective algorithms. A software toolbox for MOEAs is also developed. Finally, future development in this area and potential path for further research is addressed.