The Improvement Of Dominance Relations In High-dimensional Multi-objective Evolutionary Algorithm

Multi-objective evolutionary algorithms (MOEA) are particularly suitable tosolve real life problems, but they have some limitations when dealing with problemswith many objectives, typically more than three. The reason of these limitations is:1)As the increasing of the number of objectives, the proportion of non-dominatedsolutions in a population rises exponentially.2) In conventional multi-objectiveevolutionary algorithm, the diversity maintenance mechanisms have no preferencein extreme solution, so it affect the algorithm’s optimization ability inmany-objective problem (MAOPS). Thus, the current mainstream Multi-objectiveevolutionary algorithms usually consist of two selection operators, the convergenceoperator is based on Pareto dominance and favors non-dominated solutions overdominated ones, the diversity operator is constituted diversity preserving and extremesolution preference mechanism concerning the convergence operator. Inmany-objective problem, studying the relationship of domination is an important partof this paper and is also one of the newer research direction in the area of evolutionarycomputation. So it has a definitely research significance.In order to improve the performance of optimization, starting from themechanism of CDAS, a Multi-objective evolutionary algorithm with expandeddominance area to improve the convergence pressure is proposed. In the conventionalalgorithm with expanded dominance area, non-dominated solutions appear to be cut,to solve this problem, the article propose a new method which have a self-controllingdominance area. In the early evolution, by relaxing the dominance to strengthen theselection pressure, and when solutions approach to the Pareto Front, the dominance ismore and more like Pareto dominance. Thus, the populations can convergent rapidlyin the early evolution, and weaken the problem of reduction of diversity as possible inthe later stage of evolution because of the non-dominated solutions appear to be cut,therefore avoid the solution set into a local optimum. Experimental results show thatcompared whit conventional Multi-objective evolutionary algorithms, the proposedmethod has a better optimization capability, convergence efficiency and optimizeperformance.