基于MPI的大规模变量结构优化技术研究

针对大规模变量结构优化问题,研究了基于虚载荷法的位移敏度求解理论,提出了惩罚函数随优化迭代次数自适应变化的二次扩展内罚函数法,并结合牛顿法作为优化算法.采用MPI并行技术实现了敏度求解的局部并行计算,解决了大规模变量结构优化技术中的规模和效率问题.采用FORTRAN语言在HAJIF2013系统上对上述方法编程实现,实例表明,优化方法能够快速、稳定收敛到最优解,以及基于虚载荷法的位移敏度解析求解理论的高效性和准确性.

Investigation of Large-scale Optimization Technique Based on Parallel Computing

The scale of the optimization problem becomes larger and larger. This paper focused on solving this difficulty by a serious of efforts. The analytical method for sensitivity analysis that based on pseudo load has been investigated deeply,and the interior penalty function method whose penalty function can be changed following the iteration time adaptively has been introduced into the Mathematical Program meth-od. The sensitivity analysis has been solved by using the MPI parallel computing technique in order to re-fine the effectiveness of the whole process. At last, this project has been programmed with FORTRAN based on the HAJIF system. Validations showed that the method in this paper has a high reliability and ef-ficiency.