Font Size: a A A

Research On Algorithms For Large Scale Multi-objective Optimization Problem

Posted on:2022-06-10Degree:MasterType:Thesis
Country:ChinaCandidate:D D TangFull Text:PDF
GTID:2518306605468074Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Multi-objective optimization problems widely exist in industrial manufacturing,production and life.With the development of science and technology and the progress of society,multiobjective optimization problems are with the large-scale characteristics.Large scale multiobjective optimization problems have important applications in railway station path planning,water supply system planning and other scenarios.Although the results of multi-objective optimization algorithms are very fruitful,the research on large-scale multi-objective optimization problem is still in its infancy.The remarkable feature of large-scale multiobjective optimization problem is that the dimension of decision variable space is large,which makes it difficult to search the solutions.For large-scale multi-objective optimization problems,this paper mainly studies the following points: 1.How to transform the highdimensional decision variable space into a series of low dimensional subspaces through some strategies,and how to search in multiple directions in the subspace.2.How to find one or more reliable search directions in the decision variable space to promote the search process.Based on the above two points,the main work of this paper is as follows.(1)An algorithm based on symmetric point search and grouping strategy is proposed.First of all,in order to search for a wider range of solutions in the decision space,a multidirectional search strategy in the decision space is proposed.The main idea is to use the grouping strategy to group the decision variables,and divide the high-dimensional decision space into a series of low dimensional subspaces.In each subspace,the symmetric points of population points in the space are defined,and then the corresponding search directions are constructed.Then a certain number of search directions are selected to guide the evolution of the population,so as to search the solution more comprehensively in the decision variable space.Secondly,in order to reduce the dimension of the problem better,a new space dimension reduction transformation function is proposed,which can better transform the optimization of high-dimensional decision space into the optimization of low dimensional weight vector space.The experimental results show that our algorithm is very potential in solving large-scale multi-objective optimization problems.(2)A potential direction cross search algorithm is proposed.Under the premise of multidirectional search,we should find the high-quality search direction.Therefore,we propose a method to construct the search direction by non-dominated solution.The main idea is to group the decision variables,and divide the high-dimensional decision space into a series of low dimensional subspaces.In the decision subspace after grouping,we rank the population points by non-dominated sorting,and keep non-dominated solutions.Then the nondominated solutions are sorted by crowding distance from the largest to the smallest order.Two adjacent solutions can form a pair to construct a direction,which is called a potential direction.The origin of the decision space and the midpoint of the pair of solutions can also construct a direction,which is called original direction.The lines along the two directions intersect each other in the decision space,so the excellent solutions can be found more comprehensively in the decision space along the two directions.The experimental results show that our algorithm has good performance and superiority in solving large-scale multiobjective problems.
Keywords/Search Tags:Large-scale Multi-objective Optimization, Grouping Strategy, Dimension Reduction, Problem Transformation, Multi-directional Search
PDF Full Text Request
Related items