Font Size: a A A

A Study On Estimation Of Distribution Algorithm For Solving Multi-objective Optimization Problems

Posted on:2018-11-23Degree:DoctorType:Dissertation
Country:ChinaCandidate:M F ShiFull Text:PDF
GTID:1368330563950935Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
In our daily life and science field,most of scientific and engineering problems are multi-objective optimization problems(MOPs).MOPscomprise several mutual effected and conflicting objectives that should be satisfied simultaneously.Hence,no single solution can optimize all the objectives at the same time.How to solve these problems has been becoming a hot topic considered by academy and engineering.However,it is difficult to code the traditional multi-objective optimization algorithms.Besides,the traditional multi-objective optimization algorithms have strict restrictions in mathematically.Thus,the application scope of traditional multi-objective optimization algorithms is greatly limited.It has importantresearch value and practical significance to construct excellent multi-objective EDA to efficiently solve MOPs.Estimation of distribution algorithms(EDAs)is a novel random exploring optimization algorithm based on statistical learning.EDAs build the probabilistic model to describe the population in stead of traditional cross and mutation operator.EDAs can obtain the global statistical information from a macro perspective.For the variable relation is highly correlated to distribution of feasible solutions,the structure of variable relation canbe easily captured by EDAs.Thus,EDAs can solve difficult optimization and search problems using variable relation.In this dissertation,we focus on the study on EDAs for solving MOPs and the application of image registration in the project of Dazu Rock Carvings digital image inpainting research.The main attributions of this dissertation are as follows:?A full variate Gaussian model-based RM-MEDA without clustering process.A regularity model-based multi-objective estimation of distribution algorithm(RM-MEDA)has been proposed recently years for solving continuous multiobjective optimization problems with variable linkages.However,the number of clusters is problem-dependent and has a significant effect on the performance of RM-MEDA.Thus,a full variate Gaussian model-based RM-MEDA without clustering process,named FRM-MEDA is proposed to deal with the situation where the number of clusters K is smaller than required.In FRM-MEDA,the clustering process is removed(K=1)from the original algorithm and the full variate Gaussian model(FGM)is introduced to keep the population diversity and make up the loss of the performance caused by removing the clustering process.The experiments demonstrate that the proposed algorithm significantly outperforms the RM-MEDA without clustering process and the RM-MEDA with K equal to AVE_K.Meanwhile,the introduction of FGM makes the FRM-MEDA faster and more stable when solving all the test instances.?Social reform model and the social reform model based multi-objective optimization framework.The human society development process is that people with knowledge accumulation and historical experience,constantly and rationally to select,planning and design the social model,so as toactively interfering in society and history under the current period of history.SR(Social Reform model)is derived from the evolutionary pattern of human social change considers the interpersonal interactivity and its significant impact to the evolution of the whole population.Different from the common evolution proposed by Darwinian,SR is more focused on the holistic characteristic of the population.The SR is a triple{(IM,FCM),CF}.In which,the IM(Independent Model)is mainly used to desribe the dominant evolutionary direction and guide the population to approximate the Pareto solution set;the FCM(Full Correlation Model)is used to strenghthen the variable relation and keep population diversity;the CF(Catalytic Factor)is used to improve the convergence speed of algorithm by enhancing the influence of dominant evolutionary direction on the population to affect the distribution of evolutionary direction in current population.Then,a SR-based multi-objective optimization framework with?general purpose‘is constructed,under which a variety of multi-objective EDAs with?specific purpose‘can be instantiated to solve various MOPs.?Two instantiated algorithms under the social reform model based multi-objective optimization framework.To study the feasibility and adaptability of the SR-based multi-objective optimization framework,two algorithms for solving different test instances are instantiated under the framework.First algorithm is a SR-based multi-objective EDA for solving MOPs with variable linkage(SR-MEDA-VL),and the second algorithm is a SR-based multi-objective EDA for solving ZDT test instances(SR-MEDA-ZDT).The experiments demonstrate that SR-MEDA-VL and SR-MEDA-ZDTcan achieve good performance on convergence speed,convergence quality and diversity maintain.Moreover,the proposed SR-MEDA-VL and SR-MEDA-ZDT can minimize the influence of the variable dimension when solving the optimization problems.Good performance of SR-MEDA-VL and SR-MEDA-ZDT indicate that SR is easily instantiated as different algorithms to adapt to specific MOPs,which can ensure the feasibility and adaptability of the SR-based multi-objective optimization framework.?A multi-objective optimization-based image registration method.A multi-objective estimation model is built to describe the feature matching pairs(data set).Moreover,FRM-MEDA is presented to solve the established multi-objective model.Thus,a multi-objective optimization-based image registration method(MO-IRM)is proposed.In MO-IRM,FRM-MEDA only requires a few iterations to find out a correct model.FRM-MEDA can not only greatly reduce the computational overhead but also effectively decrease the possibility of false registration.For the sake of practical project,a multiple images registration flowchart is given in the paper.The proposed MO-IRM is applied to the Dazu grottoes image regostration.The experiment results demonstrate that the proposed MO-IRM achieves ideal registration performances on both two images and multiple images,and greatly outperforms the compared algorithms on the runtime.Meanwhile,the increasing time along with the number of feature points is not so obvious in the model estimation process of MO-IRM.
Keywords/Search Tags:Multi-objective optimization problem, Estimation of distribution algorithm, RM-MEDA, Social reform model, Image registration
PDF Full Text Request
Related items