The Multi-Objective Genetic Algorithm Applied to Benchmark Problems An Analysis

The multiobjective genetic algorithm (MOGA) has been applied to various real-world problems in a variety of fields, most prominently in control systems engineering, with considerable success. However, a recent empirical analysis of multi-objective evolutionary algorithms (MOEA's) has suggested that a MOGA-based algorithm performed poorly across a diverse set of two-objective test problems. In this report, it is shown that a conventional MOGA with standard settings can provide improved performance, but this still compares unfavourably to the best-performing contemporary MOEA, the Strength Pareto Evolutionary Algorithm (SPEA). The importance of the MOEA, as a framework is stressed and consequently, a real-coded MOGA for real-parameter multi-criterion problems is developed using modern gudelines for the design of evolutionary algorithms. This MOGA is shown to outperform the "best" MOEA, rather that a considered implementation of the methodology is required in order to reap full rewards. This study also questions the effectiveness of the traditional fitness sharing method of niching, with respect to the current set of multiobjective benchmark problems.