Research On Many-Objective Evolutionary Algorithm For Objective-Oriented Relationship

Multi-Objective Problems are abound in the real world.According to the different objective numbers,this paper subdivides the problem into Multi-Objective Problems(MOPs),HighMany-Objective Problems(HMOPs)and Many-Objective Problems(Ma OPs).According to the degree of corresponding satisfaction between the optimal solution and the objective,a special Ma OPs,which called Plural-milti/many-Objective Problems(PMOPs)in this paper,is proposed.There is a default competitive relationship between the objectives in traditional M(a)OPs and HMOPs.Other relationships which are common exist among objectives include similarity,redundancy,collaboration,constraint,and irrelevant relationships.Due to the number of objectives,the performance of high and Many-Objective Evolutionary Algorithms(M(a)OEAs)still face challenges.The idea of traditional MOEAs searching compromise solutions is not suitable for solving special optimization problems that some objectives need to be fully satisfied.This paper researches on different types of HMOPs and Ma OPs with different objective relationships.From the perspective of improving the performance of existing algorithms,an improved algorithm for HMOPs with the default competition objectives relationships and an improved algorithm for Ma OPs with similar or redundant objectives relationships are proposed.From the perspective of efficiency,a solving strategy for HMOPs and Ma OPs with constraint relationships and a hierarchical optimization strategy for PMOPs with cooperative relationships are proposed.The main contributions and specific research contents include the following four aspects.(1)Aiming at the problem that the classical algorithm NSGA-III can not definitely seaching the whole solution space and the single-objective optimal solutions can not definitely participate in evolution,an improved algorithm based on a noval two-archives strategy for HMOPs with default competitive objectives relationships is proposed.The experiments on the benchmark functions show that the proposed algorithm can find a better distributed solution set faster.The method is also applicable to classical decomposition-based MOEA and other manyobjective optimization algorithms based on reference points.(2)For the Ma OPs with similarity or redundancy relationships between objectives,an objective reduction method based on adaptive growth tree clustering is proposed,which overcomes the problem that the number of clusters is difficult to determine and the sample points are complex.A reduction method based on aggregation is proposed,which overcomes the problem of missing diversity caused by deleting redundant objectives.Experiments on the benchmark functions show that the proposed method has better performance on HV,IGD and comprehensive indicators.(3)Aiming at HMOPs and M(a)OPs with constraint relationship,a hierarchical optimization method is proposed,which overcomes the problem of traditional MOEA searching for compromise solutions,and can make the searched optimal solution meet the constraint target deterministically.The performance of the hierarchical optimization method is verified by an example of the curling match arrangement arrange.Experiments show that the proposed method can find the optimal solution that meets the practical needs.(4)A general framework for solving special many-objective optimization problems that fully satisfy one objective for each optimal solution deterministic is proposed.It overcomes the traditional multi-objective optimization algorithm to find a compromise solution between competing objectives and can not fully meet each objective.For the PMOPs with collaborative relationship,the optimization framework for analyzing and utilizing the objective relationship information is designed.The path expression method and the specific solving algorithm are given in the software test data generation problem instance.Experiments show that the proposed method can effectively solve PMOPs with collaborative relationships.For different types of Ma OPs with different relationships among objectives,the efficient and practical optimization algorithms are researched according to the objectives relationships and the characteristics of problems,which provides a basis for the subsequent research on different types of Ma OPs solving methods.