The Research About Preference-based Dominance Strategy

When solving actual multi-objective optimizatoi nproblems(MOP), multi-objective evolutiona ryalgorithms (MOEA) can settle all kinds of encountered problemsbecause they use stochastic, population-based and intelligent search. So MOEA becomesone of the focuses in the research on intelligent computation. Different fromsingle-objective optimizatio nproblems(SOP), a set of trade-off solutions or Paretooptimal solutions will be obtained when solving MOP, rather than a single global optimalsolution. In the past several decades, researchers generate the Pareto optimal solutionsthrough iteratively running MOEA. However, with the increase of the MOP’s targetdimension, the obtained Pareto optimal solutions increase rapidly, which makes it difficultfor decision makers to make right decision. In the past three decades, manypreference-based MOEA have been proposed, which can achieve good result in solvinghigh-dimension problems. Considering from the perspective of decision makes, thepreference degree of each object are different. Therefore, it is unnecessary to search out allthe Pareto optimal solutions. Only Pareto optimal solutions that decision makers are mostinterested in are needed. For this reason, researchers have mingled decision makingstrategies in the optimization methods. In the process of algorithm optimization,integrating the preference information into MOEA, can not only simplify thedecision-making, but also reduce the search costs, and lead algorithm search to the regionof interests.In the paper, the work can be summarized as the following points.1. Learn something about the basic concepts of MOEA and the current researchstatuses of preference-based MOEA systematically. And when designing andimplementing a preference-based MOEA, some important points need to be considered.2. According to the reference points, weights and search range information providedby decision makers, we establish a unified model of the preferences. On the basis of thismodel, we present a improved dominance relationship.3. According to the preferences provided by decision makers and the characteristicsof ε-Pareto dominance relationship, we present a improved dominance relationship.The presented methods first construct new introduction method of preferenceinformation to integrate the angle information, reference points and weights into MOEA.New dominance relationship proposed to distinguish non-dominated solutions, and toimprove the convergence of the algorithm. The effectiveness of the proposed introductionmethod of preference information and dominance relationship is verified according to theprinciple analysis and comparative experiments.