碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 / The circuit switching and packet switching network have provided two kinds of telecommunication services for people for a long time. In the recent years, it’s necessary to include several traffic classes in the network in order to support each application from different requirements. Therefore, the traffic control of Broadband Integrated Service Digital Network (B-ISDN) becomes an important issue to investigate methods. Under B-ISDN, a range of voice, video, data, image and multimedia services can be supported using the common available resources. Because of the innovation of transmission, the new mechanism must be reconstructed on the traffic management and control under B-ISDN environment. Due to the routing management is one part of the traffic management and control mechanisms, this article will concentrate on the routing mechanisms for B-ISDN. Routing in B-ISDN networks supports real and multimedia traffic by providing enough resources and different guaranteed qualities for each class of traffic. The routing processes are the following: first, according to the traffic demand estimated from historical events, a routing table is constructed by our proposed Genetic Algorithm (GA). Second, based on the derived routing table, a better route is then decided for some real traffic. The algorithm used in the second step is simple and called the Least Loaded Routing method. The proposed GA includes the characteristics of time and state dependence, which are the reasons to conquer the algorithm including only one of them. Since the proposed GA is not affected by network states and its objective functions, Quality of Service (QoS) factors can be appropriately considered and the proposed GA is very flexible in different applications. Finally, the proposed GA is compared with a greedy algorithm. As a result of these experiments, the proposed GA has better performances on the traffic distribution and on the bandwidth utilization of virtual path. 1.1 前言 1 1.2 研究動機與目的 2 1.3 研究範圍 3 1.4 研究架構與方法 4 第二章 文獻回顧 6 2.1 寬頻傳送網路與路由選擇 6 2.2 動態路由技術的比較 7 2.3 路由選擇方法回顧 9 2.4 QoS與路由選擇 11 第三章 寬頻傳送網路之路由規劃 15 3.1 路由表與路由選擇 15 3.1.1 路由選擇法則 16 3.1.2 路由表產生模式 16 3.1.3 最大流量界限 17 3.2 穩定狀態路由法 18 3.2.1 問題定義 18 3.2.2 穩定狀態路由模式 22 3.3 基因搜尋啟發式方法 24 3.3.1 基因演算法簡介 24 3.3.2 流程說明及參數設定 27 3.3.3 路由表產生之範例說明 35 3.4 即時路由法則 36 第四章 演算法實驗及效能分析 38 4.1 GA與貪婪法的比較 38 4.2 擴張問題的討論 39 4.2.1 網路連結度與GA計算結果 40 4.2.2 訊務數與GA計算結果 41 4.3 GA與非線性目標函數 45 4.3.1 最小化平均細胞延遲 45 4.3.2 最小化平均細胞損失 47 4.3.3 最小化最大鏈路使用率 47 4.3.4 最大化細胞接受機率 48 第五章 結論與建議 50 5.1 結論 50 5.2 建議 52 參考文獻 53 附錄一 GA演算法程式碼 56 附錄二 十個問題的最佳解、GA及貪婪法計算結果 62 附錄三 要求品質下，不同問題的終止條件及求解時間比較 63

碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 / Estimating the variance of the sample mean is a lassical problem of sto hastic simulation.Traditional batch means estimators require that the storage space in random access memory or disk space is propor- tional to the sample size.Dynami bat h means (DNBM)estimators proposed by Yeh and S hmeiser (1999)are known to be the onlytype of estimators requiring a onstant storage space for any sample size. In this paper we propose dynamic partial batch means (DPBM)estimators requiring also a onstant storage space for any sample size.We show that DPBMworks better than DBMwith the same storage space for many processes in terms of the mean-squared-error riteria. 1.1 Background............................ 1.2 Motivation.............................5 2 Literature Review 6 2.1 BatchMeansEstimators.....................6 2.2 DynamicNon-OverlappingBM .................9 2.3 DynamicPartial-OverlappingBM................11 3 Some Results on PBM Estimators 15 3.1 PBMEstimators .........................15 3.2 1-2-1PBM.............................15 3.3 Some Observation on Dynamic Batch Means Estimtors ....16 4 The Revision of DPBM 17 .1 TheFirstRevisionofDPBM...................17 .2 InterpretationofSimulationResult ...............21 .3 TheSecondRevisionofDPBM .................22 5 Performance Comparison 6 Conclusion

碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 / Most semiconductor manufacturing companies in Taiwan own several wafer fabrication factories. If a factory need extra capacities and another factory has surplus capacities, the backup plan has to be determined through negotiation. However, they are unable to obtain a better capacity backup plan because of non-technical issues. This lowers the overall productivity and profitability of a multiple factories semiconductor manufacturer. Therefore, this study presents a systematic method capable of solving capacity backup plans for multiple factories semiconductor manufacturers. By using the “Production planning model” proposed by Hung and Leachman [1996], the production plan of each individual factory can be calculated by optimally allocating its available capacities and maximizing its cash flows. After listing the tight or the surplus of all the resources in various periods among all factories, the surplus and tight between two factories of the same resource could be matched. We continue the capacity matching process in period order until no surplus and tight items can be matched. Experiment results show that the proposed method can not only significantly improve the overall profitability but also solve the whole capacity backup plan quickly. It increases the profit margin by 8.7%. The average profit increase is 80% of the gap between the original profit and the optimal one. In addition, over 35% of the problems can obtain optimal production plans when the minimal capacity movement is zero. The average time it takes to solve the whole capacity backup plan is about 1 minute. It takes less than 10 seconds and less than 3 minutes for solving problems with fewer and more factories, respectively. The results of this study demonstrate that the method proposed herein allows multiple factories semiconductor manufacturers to objectively and efficiently draw up better capacity backup plans to enhance the overall productivity and profitability. Abstract ii 第一章 緒論 1 1.1 研究背景 1 1.2 研究動機 3 1.3 研究目的 4 1.4 研究方法 5 1.5 論文架構與研究流程 7 第二章 文獻探討 9 第三章 多廠區生產計劃 13 3.1 問題假設與名詞定義 13 3.1.1 問題假設 13 3.1.2 Event-based模型與Rate-based模型 14 3.2 線性規劃生產計劃模型 15 3.3 產能支援方法 17 第四章 實驗設計與結果分析 24 4.1 實驗設計 24 4.1.1 實驗問題基本設定 24 4.1.2 實驗設計因子 25 4.1.3 實驗績效評估方式 26 4.2 實驗結果與分析 28 4.2.1 因子檢定分析 29 4.2.2 綜合結果與圖表分析 33 第五章 結論與未來研究方向 45 5.1 結論 45 5.2 未來研究方向 45 參考文獻 47

碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 / Tabu search is a widely used heuristic search method. One of the main components of tabu search is tabu list, which makes several latest moves forbidden in order to escape from a small loop. The algorithm starts from an initial solution, and moves the current solution to the best neighborhood which is not forbidden. These iterations will be repeated until the terminating condition is reached. There are two main drawbacks in the traditional tabu search. First, tabu search only provides an approximate solution. There is no way to know the quality of the obtained solutions. Second, although tabu list helps the search avoiding a small cycling problem, it cannot prevent a previously searched area to be searched again or forming a large cycle. The computation efficiency of a tabu search can be improved by implementing a parallel tabu search, which uses multiple processors to search in parallel. In the study, we propose an enhanced parallel tabu search, which attempts to compute more efficiently and conquer the two drawbacks mentioned above. The ratio of finding the old local optimum solutions may suggest the confidence level of the best solution we find at the end of search. Also, the historical memory of local optimum solutions helps us to avoid searching old areas again. To validate the efficiency of the new approach, we will use a series of number sequencing problems. In addition, we apply statistical method to test the suggested relationship between the ratio of old local optimum and the confidence level of the obtained solution. The result of experiments shows that it takes less time for the enhanced parallel tabu search to find global optimum solutions than conventional one. The ratio of finding the old local optimums does not precisely estimates the confidence level of the best solution. However, our results show that the probability of finding the global optimal solutions should be larger than the computed ratio. 1.1研究背景 1 1.2研究動機 1 1.3研究架構 3 第2章 文獻回顧 4 2.1塔布搜尋法 4 2.2平行塔布搜尋法 9 2.3塔布搜尋效率的改善方法 11 第3章 方法構建 13 3.1改進之平行塔布搜尋法 13 3.1.1子空間的定義 13 3.1.2隨機起始解 14 3.1.3長期記憶結構 15 3.1.4落入涵蓋區域的比率 16 3.2改進之平行處理演算法 19 3.3改進之平行塔布搜尋法的優點 23 第4章 實驗設計與分析 24 4.1問題描述 24 4.1.1多個處理器的模擬方法 24 4.1.2數字排序問題 25 4.2 求解數字排序問題的改進之平行塔布搜尋法的設定 26 4.3 離開舊區域最佳解後脫離方向的比較 27 4.4涵蓋比率的實驗設計 29 4.4.1落入涵蓋區域的比率的實驗 29 4.4.2驗證子空間是否為均勻的分佈 32 4.5比較搜尋速度的實驗結果與分析 35 第5章 結論與未來展望 38 參考文獻 39

碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 /

碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 / Automobile industry has a great influence on the economical development of a country. The scale of domestic automobile market in Taiwan is not large; however, there are too many automakers in Taiwan and they have been very competitive already. Further, as Taiwan is about to joining the WTO, the domestic manufacturers are facing the competition from foreign manufacturers. Domestic automakers have to improve their productivity and lower their costs in order to survive under global compotation. Therefore, this study starts with understanding the manufacturing techniques of the automobile industry. Then, we develop a proper scheduling methodology and control methods for production process. Automobile manufacturing is a typical “paced assembly system” and can be divided into three main processes: white body shop, paint shop and assembly line. Due to the different considered factors of different processes, we first compute net demands, and then determine the schedules for paint shop and assembly lines by a backward scheduling, which is followed by a forward scheduling. This study solves the production schedules by using four search rules, rule sa, rule tabu, rule sa-tabu and rule sa-time, which are developed from simulated annealing and tabu search. Because of the high changeability of auto manufacturing processes, we also design some control rules for the key control points, such as white body stock and painted body stock. Hopefully, by using these controlling rules, the actual production sequences can be adjusted to meet the calculated schedule as much as possible. In this way, the satellite factories of a automaker can perform just-in-time deliveries of spare parts for assembly line and thus reduce the inventory cost of work-in-process materials. Experiment results show that rule sa-time has the best performance. In addition, we can obtain proper schedules for assembly lines and paint shop by a backward scheduling. 圖目錄 IV 表目錄 V 第一章 概論 1 1.1研究背景與動機 1 1.2研究目的 2 1.3研究方法 3 1.4本文架構 3 第二章 汽車廠生產方式簡介與問題描述 4 2.1汽車工廠生產方式 4 2.1.1車身熔組線（Body Shop） 4 2.1.2塗裝線（Paint Shop） 5 2.1.3組裝線（Assembly Line） 5 2.2本研究對象的生產概況 6 2.2.1車身工廠生產情形 7 2.2.2油漆工廠生產情形 8 2.2.3車裝工廠生產情形 9 2.2.4整廠概況 9 2.3本研究對象的生產問題 10 2.3.1車身熔組線生產問題 11 2.3.2塗裝廠生產問題 11 2.3.3組裝線問題 12 2.3.4該廠目前希望解決的生產問題 13 第三章 文獻回顧 15 3.1 Paced Assembly Line 15 3.2汽車廠排程相關文獻 18 第四章 方法構建 19 4.1生產排程方法 20 4.1.1淨需求的計算（Netting） 23 4.1.2模擬退火法求解生產排程 29 4.1.3塔布搜尋法求解生產排程 38 4.2控制法則 42 4.2.1車身暫存區（WBS）控制法則 42 4.2.2塗裝完成暫存區（PBS）控制法則 43 第五章 實驗結果與分析 47 5.1生產排程搜尋法則 47 5.2結果與分析 49 第六章 結論與未來研究建議 59 參考文獻 60 附錄 62 A. 輸入檔格式 62 B. 輸出檔格式 64

碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 / Fuzzy Sets Theory was introduced by L. A. Zadeh in 1965. Up to now, fuzzy sets have been applied to many fields such as Decision Analysis, System Theory, Artificial Intelligence, Economics and Control Theory. However, until 1993, Q. Song and B.S Chissom proposed a fuzzy time series method which provides an alternative approach for some special dynamic process. This paper presents two methods to forecast secular trend and seasonal variation time series problems respectively. The revised fuzzy time series method uses Song and Chissom’s first-order time-invariant model to predict such linguistic historical data problems and we illustrate the forecasting process by the enrollments of the University of Alabama. This method obtains a better average error than the error in Song and Chissom’s method. The method using fuzzy regression theory solves the shortcoming that fuzzy time series method could not work in dealing with seasonal variation time series problems. Under different confidence level the resultant forecasting interval would provide more flexibility for a decision maker in making decisions. ABSTRACT ii ACKNOWLEDGEMENTS iii CONTENTS iv TABLE CAPTIONS iv FIGURE CAPTIONS iv LIST OF NOTATIONS iv Chapter 1 INTRODUCTION 1 Chapter 2 LITERATURE REVIEW 3 2.1 Fuzzy Time Series 3 2.2 Fuzzy Regression 8 2.3 Conclusions 11 Chapter 3 A FUZZY TIME SERIES MODEL FOR SECULAR DATA 12 3.1 S&C Fuzzy Time Series Method 13 3.2 A Revised Fuzzy Time Series Method 20 3.3 Evaluation and Discussion 25 Chapter 4 FUZZY REGRESSION METHOD FOR SEASONAL TREND 30 4.1 Fuzzy Regression Model 31 4.2 Conclusion and Discussion 36 Chapter 5 SUMMARY AND CONCLUSION 38 REFERENCE 40

碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 / This thesis proposes a bandwidth allocation procedure for single broadband network traffic source. In recent years, broadband traffic source have been proofed that it has self-similarity and self-similar model is well to fit the broadband network traffic sources. As the traffic model developed, the bandwidth estimations were proposed. The bandwidth allocation problem is an important problem in network programming. The network programming problems such as call admission control and routing have to consider the requirement bandwidth of network traffic. We use the equivalent bandwidths of exponential model and self-similar model to estimate the bandwidth of the traffic sample. The traffic samples are generated from the Poisson model and the self-similar model. In order to utilize the network resources efficiency, we analyze what kind of traffic sources and network condition is suitable for different of bandwidth estimations as well as the influence of buffer sizes. Abstract ii 誌謝辭 iii 目錄 iv 圖目錄 vii 表目錄 viii 第一章 緒論 1 1.1 前言 1 1.2 ATM網路介紹 1 1.3 研究動機與目的 3 1.4 研究範圍 4 1.5 研究架構 4 第二章 文獻回顧 6 2.1 寬頻網路服務 6 2.1.1 服務類型 6 2.1.2 訊務類型 8 2.2 訊務模式 8 2.2.1 卜瓦松模式 9 2.2.2 流體流量模式 9 2.2.3 自我相似模式 10 2.2.4 訊務產生 11 2.3 VP頻寬計算方法 13 2.3.1 有效頻寬與對等頻寬 13 2.3.2 指數模式 15 2.3.3 自我相似模式 15 2.3.4 Le Boudec頻寬估計式 16 2.4 訊務參數估計 16 第三章 頻寬估計程序 18 3.1 簡介 18 3.2 產生訊務資料 19 3.3 訊務參數估計 21 3.4 細胞流失率 23 第四章 實驗分析與說明 26 4.1 實驗組一：卜瓦松模式訊務與H：[ 0.5, 0.6 ]自我相似訊務 26 4.1.1 Le Boudec頻寬估計式 27 4.1.2 卜瓦松模式訊務頻寬估計討論 28 4.1.3 自我相似模式訊務頻寬估計討論 30 4.1.4 卜瓦松訊務與自我相似訊務頻寬估計比較 32 4.2 實驗組二：H：[ 0.5, 1.0 ]自我相似模式訊務 33 4.2.1 H：[ 0.6 , 0.7 ] 自我相似訊務 33 4.2.2 H：[ 0.7 , 0.8 ]自我相似訊務 35 4.2.3 H：[ 0.8 , 0.9 ] 自我相似訊務 37 4.2.4 H：[ 0.9 , 1.0 ] 自我相似訊務 38 4.2.5 不合適估計與最佳估計個數統計 40 4.2.6 結論 41 4.3 實驗組三：多媒體網路訊務 43 4.3.1 視訊會議 44 4.3.2 TV分配 44 4.3.3 MPEG1 46 4.3.4 MPEG2 48 4.3.5 多媒體訊務頻寬估計不合適與最佳個數 48 4.3.6 結論 50 4.4 結論 51 第五章 結論與建議 53 5.1 結論 53 5.2 建議 56 參考文獻 57 附錄A 59 附錄B 62 附錄C 75 附錄D 85

碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 / ABSTRACT This paper proposes a hierarchical procedure for solving multiple objective decision problems in which only partial information is given in the decision process. The procedure consists of two levels, a top-level and a base-level. The main idea is that the top-level provides partial preference information to reduce the non-dominated solution set；then the base-level determines a compromise solution from the reduced set. Furthermore, we also derive a procedure to find the tolerance regions of cost coefficients or RHS such that either the compromise solution or the final extreme solution set will maintain. In this study we also consider that when DM is fuzzy about the tradeoff questions so that the DM may not be able to make exact tradeoffs among the objectives or sometimes, in a maximization (minimization) problem there exists some objectives for which DM may state to achieve substantially more (less) than or equal to some values. In such cases, we consider a fuzzy programming structure and construct an interactive fuzzy programming to find a satisfactory solution. CHAPTER 2 LITERATURE REVIEW . . . . 5 CHAPTER 3 MULTI-OBJECTIVE ANALYSIS WITH PARTIAL PREFERENCE INFORMATION . . . 8 3.1 Preference Presentation . . . . 8 3.2 Properties Derived From the Partial Information . 10 3.3 Conclusion . . . . . 12 CHAPTER 4 THE INTERACTIVE PROCEDURE FOR HIERARCHICAL DECISION MAKING . . . . . 13 4.1 The Zionts-Wallenius Algorithm . . . 13 4.2 The Interactive Procedure for Hierarchical Decision Makers . . . . . 14 4.3 Conclusion . . . . . 23 CHAPTER 5 TOLERANCE ANALYSIS OF AN MOLP WITH HIERARCHICAL DECISION MAKERS . . . . 25 5.1 The Revised Algorithm for Perturbation Analysis . 35 5.2 Tolerance Analysis of the Cost Coefficients for Hierarchical Decision Makers . . . 43 5.3 Tolerance Analysis of the RHS . . . 52 5.4 Tolerance Analysis of Both the Cost Coefficients and RHS . . . . . 53 5.5 Conclusion . . . . . 54 CHAPTER 6 INTERACTIVE FUZZY PROGRAMMING . . 55 6.1 Problem Formulation . . . . 55 6.2 An Interactive Fuzzy Programming for Hierarchical Decision Makers . . . . 58 6.3 Conclusion . . . . . 64 CHAPTER 7 SUMMARY AND CONCLUSIONS . . . 65 REFERENCES . . . . . 67

碩士 / 國立清華大學 / 工業工程與工程管理學系 / 89 /