Abstract
排程在生產過程中扮演極為重要之角色。然而﹐隨著生產系統日益複雜﹐大部份的排程問題已證明為NP-Complete。傳統的作業研究工具已無法有效率地解決上述問題﹐因此本研究針對動態排程問題提出一套結合傳統分析工具與現代人工智慧方法具有學習能力之智慧型排程系統。 SAOSS 採用以特徵值為導向之彈性排程策略。亦即在不同的系統狀態下使用當時最適切之派工法則 (Dispatching rule) 。基於以上特性，第一章首先闡述有關排程之基本概念;製造過程中所發生之排程問題 SAOSS 必需具有學習動態排程知識、組織相關資訊的能力。因此，SAOSS採用一套連續型二分法算則 (A continuous ID3 Algorithm)，構建一由 ADALINES 所組成之類神精網路系統。網路節點間之加重權數則隱含了動態排程所需之法則。由此方法產生之排程法則十分精簡， SAOSS 的運作效率因而大為提昇。以及本文中使用之研究方法。第二章則回顧以往排程之方法。第三章則介紹智慧型排程系統之基本架構。第四章則陳述此智慧型排程系統的發展步驟。第五章則應用一彈性製造單元作評估。第六章結論與未來發展方向。經由實証，顯示 SAOSS 能處理繁雜之動態排程問題。此外，由 CID3算則所產生之生產排程法則亦較其它方法簡潔。Scheduling, as part of production planning and control, playsan inportant role in the entire manufacturing process. Mostsche duling problems have been proven to be NP-complete whichdegrade s the performance of a conventional OR techniques.Hence, a new approach which can deal with sophisticated,especially dynamic s scheduling problems is strongly desired.In this study, a system attributes oriented knowledge based scheduling system (SAOSS) with inductive learning capability isi ntroduced. SAOSS takes a multi-algorithm paradigm so that itis able to tackle a variety of scheduling problems. In otherwords, different strategies are are inferred by SAOSS withrespect to t he scheduling conditions which makes SAOSS moreintelligent, fle xible, and suitable than others in tacklingcomplicated, dynamic scheduling problems. The embeddedknowledge acquisition mechanism enables SAOSS to acquirescheduling heuristics from both expertise's experience a ndexperiments through inductive learning. An efficient inductiv elearning method, a continuous ID3 (CID3) algorithm, induces decision rules for scheduling by converting correspondingdecision trees into hidden layers of a self-generated neuralnetwork. Con nection weights between hidden units imply thescheduling heuris tics which are interpreted into schedulingrules later. An FMC scheduling problem is given forillustration and justification.