2007-08-012024-05-17https://scholars.lib.ntu.edu.tw/handle/123456789/689702摘要：動態資源配置雖能有效因應需求不確定與系統可靠度問題，但現有方法受限於運算複雜性（computational complexity），未能在大型系統內考慮前述兩類不確定因素。針對現有方法的缺點，本計畫中的動態資源配置決策將考慮：1.即時的系統可靠度狀態與等候線長度；2.系統的未來狀態變化機率（如：訂單到達時間的機率分配及各資源的損壞、維修發生機率等）。不同於更新時間間隔較長的現有方法，本計畫將決策時間點（decision epoch）細分至所有系統內動態事件（如服務完成、訂單到達與設備損壞等）的發生時刻，所以能夠完整考慮其他方法所無 法描述的系統動態特性。由於動態事件發生頻繁，每天、每週都可以切割為極多的決策時間點，因此將用infinite planning horizon 近似。 本研究以無限規劃時間隨機動態規劃（ infinite planning horizon stochastic dynamic programming）為方法，進行動態資源配置與維修保養排程最佳化。第一年度中將證明最佳解的單調結構特性（monotone structure），並實際撰寫最佳化求解軟體。最佳解的結構特性將使規劃求解時間大幅縮短；第二年度中將根據最佳化求解軟體的結果輸出，開發動態資源配置與預防保養排程決策支援系統，此系統預計將顯著提高平均產出並降低等候成本。本研究將以TFT-LCD 製程為例，使用模擬的方式驗證決策支援系統的強健性（robustness）。除了學術貢獻外，本研究期望成為產業界進行最佳化時的參考。<br> Abstract: dynamic resource allocation should be done. In this research, we propose a new dynamic method for resource allocation and preventive maintenance scheduling problem. This new method explicitly considers all dynamic events in the system, including demand uncertainties, queue length distribution, and real time resource reliability status. The decision epoch of this method is every dynamic event time in the system. Because of the large number of decision epochs, infinite planning horizon is a good approximation. In this research, infinite horizon stochastic dynamic programming will be used to solve the resource allocation problem. First, we plan to show the monotone structure of optimal resource allocation policies. This monotone structure can facilitate the search of optimal resource allocation and preventive maintenance policies in large systems. A new decision support system will be developed for large systems with uncertainties on future demand and reliability. To verify the robustness of proposed method, TFT-LCD manufacturing systems will be studied using computer simulation. The goal of this research is to provide a guideline for resource allocation problem in industry.隨機動態規劃可靠度需求不確定動態資源配置stochastic dynamic programmingreliabilitydemand uncertaintydynamic