國立臺灣大學電機工程學系暨研究所張時中2006-07-252018-07-062006-07-252018-07-062004-07-31http://ntur.lib.ntu.edu.tw//handle/246246/7990本三年計畫的目標如下﹕(1) 為排序佳化 模擬設計削減搜尋空間的新方法，(2) 將 上述方法開發成工具模組，作為整合系統 佳化平台的一部份，(3) 將模擬進行排序 佳化應用在次世代半導體廠的有效生產排 程。為達成這些目標，第一年中我們發展 一個以模擬為基礎的排序策略疊代 (OOBPI)法，用來處理平穩(Stationary)馬可 夫決策。我們利用策略疊代的架構，以模 擬為基礎的排序佳化來估算每一狀態的 cost-to-go 函數值與與最佳決策。初步模 擬結果顯示這個方法較傳統模擬為基礎的 策略疊代法的計算效能可快百倍。另外正 進以合約演算法的觀念架構來設計對狀態 的最佳計算資源分配，以進一步提昇 OOBPI 法的計算效能。In this proposed three-year research project, objectives are (1) to design search space reduction method for simulationbased OO, (2) to develop these methods into tool modules as part of an integrated system optimization platform, and (3) to apply simulation-based OO to effective production scheduling of 300mm foundry fabs. In the first year, we have designed an OO-Based Policy Iteration (OOBPI) method to handle the combinatorial complexity of decisions over the time axis for Stationary Markov decision problems. Utilizing the framework of policy iteration, we approximate the optimal cost-to-go and optimal decision of each state by simulation-based OO. The OOBPI method demonstrates, in preliminary numerical studies, two orders of speed-up in than policy iteration using traditional simulation for evaluating cost-to-go values. To efficiently handle the large state space under computing processor capacity and run time limits, we have been investigating the notion of contract algorithms in general and ordinal computing budget allocation in specific to further speed up the convergence of OOBPI.application/pdf69289 bytesapplication/pdfzh-TW國立臺灣大學電機工程學系暨研究所ordinal optimizationpolicy iterationsimulation-based policy iterationMarkov decision processcontract algorithmcomputing budget allocationreporthttp://ntur.lib.ntu.edu.tw/bitstream/246246/7990/1/922213E002099.pdf