Decomposition And Dominance Relation Based Many-objective Evolutionary Algorithm

In recent years,the research on the multi-objective optimization problem has been deepened.The reason is that the multi-objective optimization problems widely exist in engineering and scientific fields,such as submarine propulsion system optimization,aeroengine health management,automobile engine calibration,water flow distribution and so on.With the continuous development of the multi-objective optimization problem,it has more and more complex forms,which leads to the many-objective optimization problem.When dealing with the problem of many-objective evolution,the traditional algorithm has some limitations,such as the non dominated solution is not uniform and the diversity distribution of the solution is not ideal.In recent years,how to balance the convergence and diversity of the population in the many-objective evolutionary algorithm has become a hot and difficult problem in the field of evolutionary optimization.At the same time,efficient algorithm design has important theoretical and practical significance for the many-objective optimization.There are two design criteria for the algorithm: convergence criteria and diversity criteria.It is found that the diversity of solutions is usually guaranteed by the method of decomposing population with reference vector,and the distribution of solution diversity can also be guaranteed by using reference point to guide the search of solutions.The Pareto dominated method is usually used to ensure the convergence of the solution.The non dominated solution is arranged in layers,and the solution in the higher priority level is selected as the solution with good convergence to enter the next generation population.With the increase of the number of objectives,the traditional Pareto dominated method can not sort the non-dominated solutions well,so that the convergence of the many-objective evolutionary algorithm can not be guaranteed.So it is very important to find a new way to ensure the convergence and diversity of the solution.In this thesis,we propose a method of combination of decomposition and dominance to solve the many-objective evolution problem,which aims to ensure the convergence and diversity of the solution at the same time.The main work is as follows:1.In view of the multi peak,linear and discontinuous many-objective evolutionary problems in Pareto front,this thesis proposes Decomposition and angle dominance relation based many-objective evolutionary algorithm(Ddr EA).The algorithm first decomposes the whole population into a group of subpopulations by using the weight vector,then optimizes these subpopulations cooperatively,calculates the value of each solution in the subpopulation by using the angle domination relationship and angle to ensure the convergence and diversity of the solution,and finally selects the elite according to the fitness value to make the elite solution enter the next generation.The simulation results show that compared with the mainstream many-objective evolutionary algorithm,Ddr EA can better balance the convergence and diversity of the population.2.In view of the complicated many-objective evolution problems such as discontinuity and degradation in Pareto front,this thesis proposes Reference point and angle dominance relation based many-objective evolutionary algorithm(Rp Adr).In this algorithm,the non dominated solutions are ranked by the angle dominated criterion to ensure the convergence of the solution,and then the niche reservation strategy guided by the reference point is used to search the solution set to ensure the diversity of the solution.The simulation results show that Rp Adr algorithm has better performance in dealing with DTLZ test problem,and it can balance the convergence and diversity well.