Research On Many-Objective Evolutionary Algorithms And Their Applications

So far the study in many-objective optimization algorithms and applications has become a hot topic in the field of information,which has already deep into the production and scientific research fields.The parameters of existing many-objective evolutionary algorithms(MaOEAs),which based on relaxed Pareto-dominance(MOEA/RP),are difficult to set.And the relaxtion will lead the migration of searching.When the number of objectives is very large,Pareto approximate solutions obtained after even more complicated computation have poor convergence,uneven distribution and incomplete covering.Compared with MOEA/RP,MaOEAs based on decomposition(MOEA/D)has a large advantage on convergence.However,MOEA/D can not obtain a set of uniformly distributed optimal solutions in the cases that the PF is complex.Therefore,research more effective and practical MOEA has great theoretical significance and potential practical value.The aim of this dissertation is to explore the theories and mechanisms of many-objective optimization,to design modified strategies to overcome the shortcomings of existing MaOEAs,to improve the performance of MaOEAs on all kinds of many-objective optimization problems(MaOPs),and apply the improved algorithms to practical engineering optimization problems.The main work in this dissertation contains the following four aspects:In order to better deal with MaOPs with a smaller number of objectives and irregularly complex PF,the fuzzy theory is introduced into the elitist selection.After that,a many-objective evolutionary algorithm based on fuzzy dominance(MFEA)is proposed.Firstly,using fuzzy logic to improve dominance to realize relaxed dominant condition,and then increase environment selective pressure.Secondly,use the niching and the k-closest distance to improve the Harmonic distance.The sparse level of a solution among the current non-dominated population is evaluated within the neighborhood.Estimate individual packed density quickly and effectively with small amount of calculation.At last,non-dominated solutions are fast ranked by cutting.The solutions in the first layer are selected to maintain the convergence,meanwhile,the solutions in other layers with great sparse level are selected to improve the distribution.The numerical experiment results confirm the effectiveness of the improved methods.In order to better deal with MaOPs with large number of objectives and regularly complex PF,an adaptive direction vector is designed for every subproblem to control the movement and search direction of the individual.Then,an improved MOEA/D based on adaptive direction vector adjustment(MOEA/D-AD)is proposed.The direction vector index can be selected properly to structure suitable front curve for the complex PF.The direction vectors are adaptive adjusted periodically in the evolution process.Firstly,the subproblems in the discontinuous regions are removed by the reassignment of direction vectors according to the location of solutions.Secondly,some new direction vectors mapped by non-dominated solutions are added into real sparse regions.At last,the set of subproblems is reconstructed according to different measures in different stages to obtain uniformly distributed set in the feasible region.The numerical experiment results confirm the effectiveness of the improved methods.In order to better deal with MaOPs with a great many objectives and extremely complex objective functions and PFs,decomposition and elite strategy are combined in this project.Then,a many-objective hybrid evolutionary algorithm based on adaptive multi-population decomposition(MOHEA-AMD)is proposed.The Tchebycheff decomposition scheme is used to deal with the large number of objectives.The elitist selection is used to deal with the complex PF.A multi-population hybrid evolutionary mechanism is designed to improve the search capacity in the fixed directions and between different directions.In this way,its ability to solve complex objective is improved.The numerical experiment results confirm the effectiveness of the improved methods.The three modified algorithms are applied on the practical many-objective optimization problems,such as the design of ship's principal parameters,the design of orthogonal waveform of MIMO radar,and the design of a shape-shifting robot,to improve the existing design methods.These expand the application field of many-objective optimization algorithm while validating the effectiveness of the improved algorithms.