This paper proposes an adaptive PBI for massively parallel MOEA/D in a distributed memory environment. Massively parallelization in a distributed memory environment effectively speeds up evolutionary multi-objective o...
详细信息
ISBN:
(数字)9781665467087
ISBN:
(纸本)9781665467087
This paper proposes an adaptive PBI for massively parallel MOEA/D in a distributed memory environment. Massively parallelization in a distributed memory environment effectively speeds up evolutionary multi-objective optimization algorithms for practical application problems. On the other hand, when MOEA/D is divided for parallelization by focusing on the reference vector in the objective function space, the T-neighbor is divided and the problem that the solution distribution becomes sparse near the boundary of the divided region arises. Here, we propose a method to improve the problem that the T-neighbor is divided and the solution distribution becomes sparse by adaptively controlling the penalty value in the PBI function according to the distance from the reference vector using a distribution function such as Laplace distribution. The effectiveness of the proposed method is shown by comparison with execution using a single CPU.
The use of Fast Multipole Methods (FMM) combined with embedded Krylov solvers preconditioned by a sparse approximate inverse is investigated for the solution of large linear systems arising in industrial acoustic and ...
详细信息
The use of Fast Multipole Methods (FMM) combined with embedded Krylov solvers preconditioned by a sparse approximate inverse is investigated for the solution of large linear systems arising in industrial acoustic and electromagnetic simulations. We use a boundary element integral equation method to solve the Helmholtz and the Maxwell equations in the frequency domain. The resulting linear systems are solved by iterative solvers using FMM to accelerate the matrix-vector products. The simulation code is developed in a distributed memory environment using message passing and it has out-of-core capabilities to handle very large calculations. When the calculation involves one incident wave, one linear system has to be solved. In this situation, embedded solvers can be combined with an approximate inverse preconditioner to design extremely robust algorithms. For radar cross section calculations, several linear systems have to be solved. They involve the same coefficient matrix but different right-hand sides. In this case, we propose a block variant of the single right-hand side scheme. The efficiency, robustness and parallel scalability of our approach are illustrated on a set of large academic and industrial test problems.
This paper proposes the partitioning method with edge weight vectors sharing for parallel distributed MOEA/D in a distributed memory environment. Massively parallelization in a distributed memory environment effective...
详细信息
ISBN:
(纸本)9798400701207
This paper proposes the partitioning method with edge weight vectors sharing for parallel distributed MOEA/D in a distributed memory environment. Massively parallelization in a distributed memory environment effectively speeds up evolutionary multi-objective optimization algorithms for practical application problems. On the other hand, when MOEA/D is divided for parallelization by focusing on the weight vector in the objective function space, the T-neighborhood is divided and the problem that the solution distribution becomes sparse near the boundary of the divided region arises. Here, we propose a partitioning method to share edge weight vectors among all partitions, and then assign other weight vectors uniformly to each partition. Using the constrained knapsack problem as a benchmark problem, we show that it is possible to eliminate the need for migration processing to correct the value of the ideal point and improve the problem that the T-neighborhood is divided, and the distribution of the solution becomes sparse.
暂无评论