Font Size: a A A

Research On Preference Multi-objective Evolutionary Algorithm

Posted on:2014-10-23Degree:MasterType:Thesis
Country:ChinaCandidate:Z Z XieFull Text:PDF
GTID:2268330401490131Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Evolutionary computation is refers to the evolutionary programming, geneticalgorithm, and evolutionary strategy. Evolutionary algorithm has been widely used in theNP-hard problems such as engineering control, function optimization, and machinelearning. It could solve various complex optimization problems of real life by simulationthe process of biological evolution hybrid variation to retain the good genes of fathergeneration and natural selection with―survival of the fittest‖. It originated in1950s,mature in the1980s and now development up to a independent research focus and widelyused in various disciplines. From the perspective of the optimization problem can bedivided into single objective optimization problem (SOP) and the multi-objectiveoptimization problem (MOP). SOP only one goal is optimized and produce one solution,but the MOP need optimize two or more confilicting goals and produce infinite number ofsolutions and with the increasing of optimization goal the difficulty and the number ofoptimal solutions will geometric progression growth.The traditional multi-objective optimization algorithm is concerned on obtain thewide distribution, uniform convergence and optimal solutions, but to the decision makersonly one or few solutions are needed and most others are wasted. So preference-basedmulti-objective evolutionary algorithms (PMOEAs) is proposed. The so-called PMOEA isintroduced the preference information and then to get the solutions which most satisfy thepreference to avoid the unnecessary waste of computing resources.This paper proposed an angle relationship based preference multi-objectiveevolutionary algorithm which is based on the angle relationship to layering with thefitness and use the reference point to guide the population tend to interested region of theDM. According to the comparison between the vector from the pivot to the individual andthe vector from the pivot to the individual which is nearest to the reference point todetermine which solution is good and which solution is bad. Comparing with severalpopular preference multi-objective evolutionary algorithm, the results show that theproposed algorithm in this paper has the following characteristics:1) Flexibly control theextent;2) Supports multiple reference points;3) The reference point ’s position(in thefeasible region, infeasible region or on the Pareto front) does not affect the experimentalresults;4) Has a better adaptability than other algorithms;5) The algorithm can quickly get the solutions which are the DM interested in the high dimensional problems.Preference multi-objective algorithm is a hot research topic now, but there are not goodevaluation methods yet. This is mainly because of the subjectivity of the Preference, sowe can only use these evaluation methods which are for the traditional multi-objectivealgorithm and they are inaccurate. This paper presents a method which is combination ofthe IGD and GD evaluation methods for the evaluate the preference algorithm. Thismethod can objectively reflect the performance of prefenece algorithm which traditionalevaluation evaluation method can not do.
Keywords/Search Tags:MOEA, PMOEA, Decision Maker, Pareto Dominance, Angle Dominance
PDF Full Text Request
Related items