Two New Many-objective Evolutionary Algorithms Based On Reference Vector Guidance

In real life,many optimization problems are composed of multiple objectives.For most optimization problems,the objectives are usually contradictory and conflicting.Therefore,it is very difficult to find a set of solutions to satisfy all the objectives at the same time.Generally speaking,when the number of objectives is less than or equal to 3,we call this kind of problem as a multi-objective optimization problem;and when the number of objectives is more than 3,it is called many-objective optimization problem.In recent years,with the development of intelligent optimization algorithms,evolutionary algorithms have attracted the attention of more and more researchers in different research fields.Because of traditional optimization methods are difficult to obtain ideal results when solving complex multi-objective optimization problems,many researchers have been inspired by biological evolution,and many swarm intelligent evolutionary algorithms are developed to solve multi-objective optimization problems.However,the main difficulty in solving the many-objective optimization problem is that as the objective dimension increases,the proportion of non-dominated solutions in the population also increases,leading to an increase in selection pressure.It is difficult to maintain the diversity of the population and visualize the solution set in the high-dimensional objective space.To solve many difficulties faced by many-objective optimization problems,this thesis introduces a reference vector into many-objective evolutionary algorithms to explore the performance of reference vector in guiding the population evolution and improving the convergence and diversity of the algorithm.The first algorithm in this thesis was to study the effect of reference vector on the convergence and diversity of the population in the many-objective evolutionary algorithm,thus,an evolutionary many-objective algorithm based on decomposition and hierarchical clustering selection(EA-DAH)is proposed;based on the improvement of population diversity by reference vector in the first algorithm,the second proposed algorithm was a twophase many-objective evolutionary algorithm with elite archive-based adjustment for reference vectors(TPEA-EAARV)to further improve the convergence.In the numerical experiments,we verified the performance of the two proposed algorithms EA-DAH and TPEA-EAARV on the DTLZ benchmark functions.The experimental results showed that the new algorithms have good performance in solving many-objective optimization problems with irregular Pareto front.