A Scalarization Dominance Scheme With Decomposition for Higher Dimensional Multi-Objective Evolutionary Algorithm
Multi objective problems require solutions which can produce optimal results even when most of the objectives are inversely proportional to one another. Multi Objective Evolutionary Algorithms (MOEAs) are one of several approaches taken to solve such problems. The goal of these algorithms is to maximize convergence and minimize divergence from the Pareto Frontier (PF). There are several MOEAs that deal with multi objective problems, but most of them perform poorly as the number of objectives increases. In this paper, we use decomposition along with scalarization to optimize the algorithm for higher dimensions. Scalarization is used to scale down the dimensions in a way that does not affect the outcome. Once the space state is partitioned and normalized, a new algorithm is applied for each generation. The proposed algorithm works by rectifying the downfalls of local update and reduces the overall computation cost by using the best features of both local update along with global update in the generational loop. This change in update strategy gives a huge boost to overall performance of the system when combined with Decomposition and scalarization. The experimental result shows that the population never enters a state of degeneration unlike most algorithms which follow local update strategy. This makes sure that diversity is always ensured as variation is among the three major important factors in a genetic algorithm. The proposed algorithm also accommodates hybrid criteria for comparing solutions. Competency and Correctness of the algorithm is tested by running through proven benchmarks like WFG.