营建工程专案中作业成本常具统计上的相关情况。忽略作业成本相关性将扭曲估价的评估结果，使得营建工程专案无法如价地完成，进而影响工期与品质。过往研究已可将作业之间的相关性进行量化作为评估之依据；在确定作业成本之间的相关系数後，使用相关性模拟NORmal To Anything (NORTA)及Iman and Conover(IC)来进行成本预估。但使用NORTA及IC进行相关性模拟时必须使用Cholesky分解，如果遭遇原始相关系数矩阵为非正定特徵值为负的情况下，Cholesky分解将无法进行。虽然已经有学者提出将矩阵修正为正定之近似方法，但经过正定化的矩阵将会偏离原本相关系数矩阵，造成成本的预估错误。因此，本研究将利用质群最佳化搜寻一个正定且近似原始之相关系数矩阵，藉此校正因正定化矩阵所产生的误差。另外，由於无论是演算法或者相关性模拟都必须耗费大量的时间进行运算，因此本研究也将使用电脑丛集控制计算核心之沟通，同时开发平行计算策略(主从式、渗透式以及岛屿式)，以降低系统计算时间。最後并将本研究所提出之架构应用於实务案例之上，以了解其应用可能。结果验证本研究之架构可有效率地降低相关性模拟的误差。
In this thesis, the validity of the adopted MRT lattice Boltzmann model [14] is examined by computing two-dimensional Poiseuille flow, lid-driven cavity flow, and three-dimensional Poiseuille flow in a square duct. We...
详细信息
In this thesis, the validity of the adopted MRT lattice Boltzmann model [14] is examined by computing two-dimensional Poiseuille flow, lid-driven cavity flow, and three-dimensional Poiseuille flow in a square duct. We also use the density fluctuation and assume the mean density to reduce effects due to compressibility [48]. The present simulation indicates that SRT method fails to simulate higher Reynolds number flows due to the limitation of tau value. In contrast to the SRT method, the tau value for MRT method can be close to 0.5 for 2-D and 3-D simulations owing to the different relaxation rates. It should be noted that 3D MRT computations are very sensitive to boundaries, especially the treatment of the corners and edges. As adopting smaller value of tau combined with MRT method, higher Reynolds number flow can be realized. Finally, parallel-MRT and parallel-SRT models are implemented for large grid density simulations, and simulated results are presented.
暂无评论