查詢結果分析
來源資料
頁籤選單縮合
題名 | 平行計算應用於成本相關性模擬之誤差校正演算法=Using Parallel Programming Paradigms to Reduce Errors of Correlated Simulation in Cost Estimation and Time Scheduling |
---|---|
作者 | 楊亦東; 李品毅; Yang, I-tung; Lee, Pin-yi; |
期刊 | 營建管理季刊 |
出版日期 | 20120300 |
卷期 | 90 2012.03[民101.03] |
頁次 | 頁28-38 |
分類號 | 494.7 |
語文 | chi |
關鍵詞 | 估價; 不確定分析; 相關係數矩陣; 平行計算; 質群最佳化; Cost estimation; Uncertainty analysis; Correlation matrix; Parallel computing; Particle swarm optimization; |
中文摘要 | 營建工程專案中作業成本常具統計上的相關情況。忽略作業成本相關性將扭曲估價的評估結果,使得營建工程專案無法如價地完成,進而影響工期與品質。過往研究已可將作業之間的相關性進行量化作為評估之依據;在確定作業成本之間的相關係數後,使用相關性模擬NORmal To Anything (NORTA)及Iman and Conover(IC)來進行成本預估。但使用NORTA及IC進行相關性模擬時必須使用Cholesky分解,如果遭遇原始相關係數矩陣為非正定特徵值為負的情況下,Cholesky分解將無法進行。雖然已經有學者提出將矩陣修正為正定之近似方法,但經過正定化的矩陣將會偏離原本相關係數矩陣,造成成本的預估錯誤。因此,本研究將利用質群最佳化搜尋一個正定且近似原始之相關係數矩陣,藉此校正因正定化矩陣所產生的誤差。另外,由於無論是演算法或者相關性模擬都必須耗費大量的時間進行運算,因此本研究也將使用電腦叢集控制計算核心之溝通,同時開發平行計算策略(主從式、滲透式以及島嶼式),以降低系統計算時間。最後並將本研究所提出之架構應用於實務案例之上,以瞭解其應用可能。結果驗證本研究之架構可有效率地降低相關性模擬的誤差。 |
英文摘要 | In a construction project, the costs of individual activities are often correlated as they are affected by many common factors: inflation, and costs of transportation and raw materials. The neglect of the inherent correlations leads to misleading assessment of total cost. It is therefore indispensible to incorporate the correlations into the cost estimation at the presence of uncertainties. Modern approaches embrace NORmal To Anything (NORTA) and Iman and Conover (IC), both of which require Cholesky factorization of the correlation matrix. Yet, Cholesky factorization would fail when the correlation matrix is not positive definite. Although several adjustment procedures are available to modify the correlation matrix to make it positive definite, the modified correlation matrix may be very different from the original one and therefore causing significant errors in cost estimation. This study uses particl swarm optimizstion (PSO) to search for a feasible correlation matrix, which after correlation simulation will lead to minimum error. Since PSO and correlation simulation are both computationally expensive, this study also investigates the use of a computer cluster and develops three parallel programming strategies (Master-slave, Island, and Diffusion) in reducing computational time. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。