車載網路受到各方廣泛討論，激發出許多新的議題，由於車載網路的通訊品質不穩定，速度快、節點多，封包傳送不易，因此許多人都採用分群式架構增進效能，以集中式管理群組，避免封包被重複傳送，降低封包碰撞的機會。然而，現有的分群機制只能用在即時方面的應用，在檔案傳輸方面效能不足。本篇論文擬改善C. Shea等人[1]所提出的分群機制File-based Affinity Propagation Cluster, FAPC，建立兼具動態性和檔案相關性的叢集架構，並且提出改善失去叢集管理員的重建機制，以提升分群的穩定性及吞吐量（throughput）。最後，我們以模擬證明所提出的方法優於C. Shea [1]的方法，以query hit ratio、retrieve file ratio、average number of clusters及average cluster head duration為效能指標，觀察在不同時間、車輛數目及車輛速度時效能表現。 / Vehicular Ad-hoc Network (VANET) has been widely discussed and many issues have been proposed. Due to VANET’s unstable quality, varying speed, lots of mobility nodes, it’s not easy to deliver packets. Thus many researchers suggested using cluster architecture to enhance performance. Because of the central management, we can avoid duplication of packets in the same cluster and decrease the probability of packet collision. However, we find most of the cluster architectures are suitable for real-time applications, but not for file transfer. In this research, we improve C. Shea’s [1] method by adding file-similarity by classifying into groups and reselecting cluster head, when the group of nodes have not cluster head. This cluster architecture can enhance stability and throughput. Finally, we use simulation to prove that our method outperforms Chen’s [1] cluster method in terms of query hit ratio, retrieve file ratio, average number of clusters and average cluster head duration.

關係行銷或是一對一行銷是目前行銷領域上廣泛被討論的議 題，企業要如何透過有效的辨識、區隔、互動以及客制化來量身打造 顧客專屬的個人化產品與服務內容，並強化其重複消費動機及忠誠， 為目前各種產業爭相積極追求的目標，此外，由於微利時代風暴，各 產業無不希望透過顧客價值的辨識與經營，實現以更有效、更低的成 本的差異化行銷策略來創造高收益的企業經營目標，以航空產業如此 資本密集，高固定成本，低變動成本以及不對稱的供需平衡，誰掌握 低成本領導與差異化策略優勢，便能決戰存續於二十一世紀超競爭時 代之中。 由於資訊科技、網際網路以及資料探勘技術的臻於成熟， 充份 發揮了跨國、即時、深度滲透與互動的特性，使得關係行銷、一對一 行銷的實現變得更加有效而可行。本研究希望從顧客價值的認定、顧 客忠誠策略以及資料探勘技術的探討，來思考如何運用於航空公司會 員系統的顧客區隔，同時，希能透過航空公司產業通路架構、全球旅 行社訂位系統(CRS)的發展現狀、微妙的航空公司間策略聯盟以及不 同航空公司所提供的會員酬賓計劃內容的探討與陳述，初略地對個案 公司的所在環境進行策略性分析，以建議其所需採取投入關係行銷的 主要焦點客層。 緊接著， 利用資料探勘工具中的分群技術， 選定有效的指標變 數，針對某一區間的會員交易資料進行分群，藉由研究各群會員所蘊 含的特殊屬性，如營收貢獻、產品特性、通路喜好以及消費行為等等， 依據前述所定義的目標客層，以創造顧客價值為目標，精確建立目標 客戶群，並據以設計不同的行銷策略與產品組合，逐步深耕建立完整 會員關係行銷資料庫。 最後， 對於本研究所無法觸及的研究議題， 概略指出後續可能 的研究方向與建議。 / Customer Relationship Management and data mining in this hyper-competitive era have revealed a lot of interesting and innovative opportunities to enrich the capability of company to realize and provide customer value. They touch the most critical issue of the enterprise, “How can we create and sustain successful advantage, and maximize profitability by leveraging new technologies ?＂In this thesis, we will focus on the application of data mining in the FFP of the airlines industry, and look over the differences among FFP members to discover the implicative needs of FFP customers. First of all, we start discussion on literature review in chapter two, which was divided into three parts: customer loyalty strategy, customer value and data mining. In this chapter, we put emphasis on the concepts and definitions of above topics, and they would be helpful to us to select and decide key variables in the following data mining practice. Chapter three of this thesis is to introduce the structure and characteristics of the airlines industry, the history of Computerized Reservation System(CRS), the airlines strategy alliance and the FFP system, and to figure out the way to understand the existing threats and opportunities. Chapter four, which was abode by the steps of data mining process, defines business issues and collects around one year's FFP historical transaction data to establish the target data and perform an actual data mining practice. In this real practice, we use the demographic cluster function of IBM Intelligent Mining tool to do member clustering. We select net revenue, first and business class spending rate, reservation booking designator and customer activation as analytical variables to perform FFP member clustering. Each variable has been well equipped with weight and method to produce best cluster pattern. Finally, according to the mining results we have explored and interpreted, we provide our draft recommendations about marketing planning and mix activities from the perspectives of FFP members clustering.

第四代行動通訊系統(The Fourth Generation of Mobile Communications System，簡稱4G)LTE-A(Long Term Evolution-Advanced)利用載波聚合(Carrier Aggregation)與多天線MIMO(Multi-Input Multi-Output)通道技術大幅提升上傳與下載的傳輸速率，並加入協同多點協調傳輸(Co-ordinated Multi-Point Transmission)技術加強基地台服務的覆蓋率。透過LTE-A的CoMP聯合運作(Joint Processing)方式，藉由鄰近基地台之間的互相協助，有助於位於細胞邊緣處之使用者裝置(User Equipment，UE)訊號傳輸品質提升，將周圍鄰近之基地台訊號的干擾化為有益之訊號來源。中繼技術(Relay)則能將來自基地台之無線電訊號接收後經過解碼與編碼再送出，提升周遭UE接收的訊號強度。 基於行動網路環境中使用者的移動性，細胞邊緣使用者的人數與位置分布隨時間改變，傳統CoMP傳輸多屬靜態的叢集演算法事先定義CoMP傳輸叢集，導致傳輸叢集不符合細胞邊緣使用者的分布與需求，細胞邊緣使用者的傳輸增益有限。動態的CoMP傳輸雖然較靜態的CoMP傳輸符合邊緣使用者的需求與分布，然而，因其屬於分散式的架構缺乏管理控制中心，規劃傳輸叢集的過程需仰賴基地台之間頻繁的控制訊號溝通。 本論文提出一個動態的CoMP傳輸叢集演算法-階層式動態CoMP傳輸叢集演算法(Hierarchical Adaptive Clustering for CoMP ，HACC)，透過階層式架構，不但具備靜態CoMP傳輸演算法集中式系統的優點，也保有動態CoMP傳輸演算法隨使用者分布調整傳輸叢集的特點。首先於系統定義之叢集中選出上層叢集代表(top cluster head，TCH)，由基地台收集服務範圍內UE分布與通訊品質，篩選出細胞邊緣使用者並傳遞此資訊給TCH，由TCH選出較多細胞邊緣使用者的區域為CoMP傳輸叢集之子代表(sub-cluster head)，以CoMP傳輸叢集之子代表為中心點尋找相鄰的區域形成CoMP傳輸叢集。除此之外，再搭配Relay延伸來自基地台之訊號，強化基地台服務範圍內非邊緣區域之訊號，提供UE更佳的傳輸品質。 透過實驗模擬證實，本論文提出的方法在系統整體UE的資料吞吐量比傳統靜態以及Hongbin et al.[10]提出之以UE需求為主的動態CoMP叢集演算法來的優異，特別是對位於細胞邊緣通訊不良處之UE資料吞吐量有更顯著之改善，系統整體的頻譜效率也有所提升。 / The fourth-generation mobile communications system (4G) LTE-A (Long Term Evolution-Advanced) uses carrier aggregation and multi-antenna MIMO channel technology dramatically to increase the speed in both uplink and downlink, and use coordinated multi-point transmission(CoMP) and relay to improve the coverage of base station. Through joint processing(JP) in CoMP, base station(BS) communicates with adjacent BSs and then some of them build up a CoMP cluster helping the user equipment(UE) which is located at the edge of cell by enhancing the signal strength. CoMP-JP is able to transform interference from adjacent cells into useful signals. Relay technology receives radio signals and then amplifies signals before re-transmission to strengthen signals. The number of cell-edge users and their locations change with time due to the mobility of users in mobile communications system. Most traditional static CoMP transmission clustering algorithm are predefined CoMP clusters. As the distribution of cell-edge users in the system changes, the transmission clusters may not meet the needs of cell edge UEs so that the transmission gain is limited. Compared with static CoMP clustering, dynamic CoMP clustering changes with time to meet the needs of cell-edge UEs, providing an appropriate service to cell-edge UEs. However, dynamic system belongs to distributed system and lacks management control center, it highly depends on frequent communication signals among base stations during the process of clustering generation. This paper proposes a dynamic clustering algorithm for CoMP-JP - Hierarchical Adaptive Clustering for CoMP (HACC). By hierarchical structure, HACC not only has the advantages of static CoMP centralized system, but also maintains the characteristics of dynamic CoMP adjusting the clustering with cell-edge users. At the first step, we define an upper cluster representative of the group (top cluster head). Then, depending on the number of cell-edge UEs in every sector, the system chooses sub-cluster head. Sub-cluster head chooses neighboring sectors to generate a CoMP-JP transmission cluster. In addition, relay stations amplify the signal from BS providing better transmission quality for non-cell-edge UEs. Simulation results show that the proposed method outperforms traditional static CoMP clustering and UE-specific CoMP clustering method proposed by Hongbin et al.[10] in data throughput, particularly for cell-edge UEs, and spectrum utilization.

在這份研究中，我們提出一個類分散式神經網路框架，此框架為高頻交易系統研究下之子專案。在系統中，我們透過資料探勘程序發掘財務時間序列中的模式，其中所採用的資料探勘演算法之一即為類神經網路。我們實作一個在分散式平台上訓練類神經網路的框架。我們採用Apache Spark來建立底層的運算叢集，因為它提供高效能的記憶體內運算（in-memory computing）。我們分析一些分散式後向傳導演算法（特別是用來預測財務時間序列的），加以調整，並將其用於我們的框架。我們提供了許多細部的選項，讓使用者在進行類神經網路建模時有很高的彈性。 / In this research, we introduce a distributed framework of artificial neural network (ANN) as a subproject under the research of a high-frequency trading (HFT) system. In the system, ANNs are used in the data mining process for identifying patterns in financial time series. We implement a framework for training ANNs on a distributed computing platform. We adopt Apache Spark to build the base computing cluster because it is capable of high performance in-memory computing. We investigate a number of distributed backpropagation algorithms and techniques, especially ones for time series prediction, and incorporate them into our framework with some modifications. With various options for the details, we provide the user with flexibility in neural network modeling.