High-Dimensional Multi-Objective Optimization Strategy Based on Decision Space Oriented Search
DOI: 10.23977/csoc.2019.11001 | Downloads: 19 | Views: 2613
Wang Peng 1
1 School of Economics and Management, Dalian University, No.10, Xuefu Avenue, Economic & Technical Development Zone, Dalian, Liaoning,The People's Republic of China(PRC)
Corresponding AuthorWang Peng
Traditional multi-objective evolutionary algorithms (MOEAs) have good performance for low-dimensional continuous multi-objective optimization problems, but with the increase of the target dimension of the optimization problem, the optimization difficulty will also increase sharply. The main reasons are: the algorithm itself. Insufficient, the selection pressure becomes smaller when the dimension increases, and the convergence and distribution conflicts are difficult to balance. This paper proposes a directional search strategy in decision space by using the characteristics of continuous multi-objective optimization problem to optimize the high-dimensional multi-objective optimization. The strategy can be combined with the MOEA based on dominance relationship. DS firstly samples and analyzes the problem, and analyzes the problem characteristics to obtain the convergence subspace control vector and the distributed subspace control vector. The algorithm search process is divided into the convergence search phase. And the distributed search phase, which corresponds to the convergence subspace and the distribution subspace respectively, and uses the sampling analysis pair to make a macroscopic influence on the region of the generation of the individual generation in the different stages of the search. Convergence and distribution are considered in stages to avoid convergence. Sexuality and distribution are difficult to balance, and at the same time, search for funds in a certain stage. The relative concentration of the source increases the search ability of the algorithm to some extent. In the experimental part, the NSGA-II and SPEA2 algorithms combined with the DS strategy are compared with the original NSGA-II and SPEA2 algorithms, and DS-NSGA-II is taken as an example. Compared with other high-dimensional algorithms MOEAD-PBI, NSGA-III, Hype, MSOPS and LMEA, the experimental results show that the performance of DS strategy is significantly higher than that of NSGA-II and SPEA2 algorithms. DS-NSGAII has strong competitiveness compared with the existing classic high-dimensional multi-target algorithm.
KEYWORDSHigh-dimensional multi-objective optimization, Decision space, Directed search, Convergence subspace, Distributed subspace
CITE THIS PAPER
Peng Wang, High-Dimensional Multi-Objective Optimization Strategy Based on Decision Space Oriented Search, Cloud and Service-Oriented Computing (2019) Vol. 1: 1-6. DOI: http://dx.doi.org/10.23977/csoc.2019.11001.
 Zheng JH. (2017) Multi-Objective Evolutionary Algorithm and its Application, Beijing: National Defence Industry Press, 108-122
 Deb K. (2016) Real-coded Genetic Algorithms with Simulated Binary Crossover: Studies on Multimodal and dMultiobjective Problems, Complex Systems, 7, 68-89.
 Zitzler E. (2004) A Multiobjective Evolutionary Algorithm Based on Decomposition. IEEE Transactions on Evolutionary Computation, 6, 712-731.