本研究欲提出一COMEX黃金期貨價格的類神經網路模型，期此一模型能預測出當期的黃金期貨價格。在類神經網路模型方面，採用倒傳遞類神經網路；而其輸入層共有九個處理單元，即影響黃金期貨價格的九個變數，輸出層為一個處理單元，即黃金期貨價格，至於隱藏層則採二層，因黃金期貨價格有波動大、難預測且為非線性的特性。 　　為證明類神經網路是否有較傳統統計學方法在此一方面有較強的預測能力，所以以此模型與單變量時間數列模型及迴歸分析模型做比較，並以MSE及MAPE作為評估的準則。 　　在實作方面，研究資料以西元1987年1月至西元1991年12月60筆月資料為訓練樣本；而西元1992年1月至1995年12月48筆月資料為測試樣本。研究結果顯示不論是MSE或MAPE類神經網路模型皆優於迴歸分析模型及時間數列模型。

本論文使用非監督式新細胞認知機（Unsupervised neocognitron）神經網路來便是印刷體中文字。 關於非監督式新細胞認知機，本論文提出兩項修改。第一項，Us1子層的結點不進行學習，而是直接套用人為方式所指定的12個區域特徵，而Us1之後的S子層仍然使用非監督式學習的方式決定其所要偵測的區域特徵。第二項修改則是，在學習過中設定一個上限值來限制代表節點（representative）產生的個數。如此設計的目的是為了避免模板（cell-planes）分配不均的問題。在本研究，採用這兩項修改的新細胞認知機稱為模式一，而使用第二項修改的新細胞認知機稱為模式二。 本論文裡的所有實驗分為兩部分。在第一部分有四個實驗，這些實驗都使用相同的訓練範例與測試範例。訓練範例有兩組，第一組包含“川”,“三”,“大”,“人”,“台”等五個中文字。而第二組包含“零”,“壹”,“貳”,“參”,“肆”等中文字。訓練範例都市採用細明體，而測試範例則是採用其他九種不同字體。第一個實驗的主要目的是測試模式一的績效。實驗結果顯示，模式一很容易學習成功而且辨識率可以接受。另外三個實驗的目的是想要了解某些參數值與系統績效的關係。這些參數包含S-欄的大小（the size of S-column），模板樹（the number of cell-planes），以及節點的接收場大小（the size of cells’ receptive field）。這三個實驗所使用的網路系統是模式一。 第二部分有二個實驗，主要的目的是比較模式一與模式二的系統績效。在第一個實驗，所使用的訓練範例與第一部分實驗相同。實驗結果顯示模式一比較容易成功地學習，而且系統有不錯的表現。第二個實驗，使用17個中文字做為訓練範例。這17個字包括“零”,“壹”,“貳”,“參”,“肆”,“伍”,“陸”,“柒”,“捌”,“玖”,“拾”,“佰”,“仟”,“萬”,“億”,“圓”,“角”。實驗結果顯示，模式一仍然是一個不錯的系統。 / In this study, we are investigating the feasibility of applying the unsupervised neocognitron to the recognition of printed Chinese characters. Two propositions for the unsupervised neocognitron are mentioned. The first on proposes that the input connections of the first layer are manually given, and all subsequent layers are trained unsupervised. The second one concerns the selection of representatives. During the process of learning, the number of cell-planes that send representatives for each training pattern has an upper bound. The unsupervised neocognitron with implementing these two propositions is named as Model 1, and the unsupervised neocognitron with implementing only the second proposition is named as Model 2. Experiment in this study are grouped into two parts, called Part I and Part II. In Part I, four experiments are conducted. For each experiment, two sets of training patterns will be conducted respectively. The first one, called the simple training set, consists of five printed Chinese characters“川”,“三”,“大”,“人”, and “台” with size of 25*25 in MingLight font. The second one, called the complex training set, contains another five printed Chinese characters“零”,“壹”,“貳”,“參”, and “肆” in the some font and size. After training, these characters of other nine different fonts are presented to test the generalization of the network. The objective of the first experiment of Part I is to investigate the performance of Model 1. Simulation results shot that Model 1 demonstrates a good ability to achieve a successful learning. In other three experiments, the effect of choosing different value for some parameters in investigated. The parameters include the size of S-column, the number of cell-planes, and the receptive field of cells. In Part II, a comparison of the performance of Model 1 and Model 2 is made. In the first experiment, Model 1 and Model 2 are trained to recognize the simple and complex training sets described above. Experimental results show that Model 1 shows higher ability to achieve a successful learning, and performance of Model 1 is acceptable. In the second experiment, 17 training patterns are presented during the learning process. These training patterns include “零”,“壹”,“貳”,“參”,“肆”,“伍”,“陸”,“柒”,“捌”,“玖”,“拾”,“佰”,“仟”,“萬”,“億”,“圓”,, and “角”. From the simulation results, Model 1 is a promising approach for the recognition of printed Chinese characters.

財務報表記錄可說是企業經營績效良窳的反映指標，而其中所衍生出來的財務比率，向 來均是管理者、投資者進行企業診斷或未來經營績效預測的重要資訊來源。然而，相關 的研究發現，由於產業間經濟環境與市場結構特性的不同，所呈現出來的財務報表資訊 內涵亦將有所差別。因此，若進一步運用個別產業之報表資訊預測公司未來盈餘時，將 能夠提供產業間結果進行分析與比較的基礎。 如何自報表中獲取與公司經營績效相關之會計資訊，進而建構出優良的盈餘預測模式， 是近幾年來學者感興趣的研究課題之一。鑑於人工智慧之類神經網路系統擁有多項的特點，因此，對於盈餘預測會計資訊萃取的應用上，無非是提供了我們一個新的選擇途徑。 本研究即根據此項概念，以民國70年第一季至民國82年第三季為止共十五項大小產業之 股票上市公司財務報表以及股價報酬等資料作為研究樣本，進行盈餘預測模式的建構以 及投資超額報酬的計算。 進一步地說，本研究的內容可以分成三個部份，第一部份是以整體市場樣本為例，對類 神經網路主要參數如輸入變數組合、隱藏層節點數等進行調整及測試，以從中選取出盈 餘預測效果較佳之模式設定；在第二部份則是運用此一盈餘預測模式，分別對整體市場 以及紡織、電子電機 兩項產業樣本進行網路的訓練與測試，並根據模式所獲得之區別及 預測能力評估指標，探討不同產業特性樣本所建構的模式之間，其預測結果上的差異性 ；而第三部份則是利用各類產業模式預測結果的資訊，從利潤與風險兩種角度，定義"總 體"、"高利潤"、"低風險"、 "高利潤低風險"等四種不同類型投資策略，並以事件研究 法計算各項策略所能獲取之累積超額報酬，最後，則根據各策略之獲利績效，進行產業 間的分析比較，以找出本研究各類特定產業之最適投資策略。 本研究根據前述方式所進行的實驗研究中，獲得了以下三點結論： 一、類神經網路盈餘預測模式之建構 (一)以整體市場樣本為對象所進行之網路的測試中，發現模式整體區別能力大致介於五 到七成之間；而整體預測能力則介於四到六成之間。 (二)本研究所找出盈餘預測效果較佳之網路模式設定如下：1.輸入變數組合：單因子多變量變異數分析之22項顯著性財務比率 2.網路架構(輸入層-隱藏層-輸出層)：22-22-1 3.連結權數初始值設定範圍：-0.1~0.1 二、產業盈餘預測結果之分析 (一)整體而言，產業間模式測試結果的差異並不大，其中以紡織產業的模式區別及預測 能力最好(70%以上)，電子電機產業次之，而整體市場模式的結果均不及兩項單一性產業。 (二)模式預測能力穩定性方面，各產業於五個年度間預測率的波動大致還算穩定，其中 就紡織產業而言，其年度之間模式預測能力的差別不大，但電子電機產業年度間的變化 則要比前者來得明顯。 三、產業投資策略績效之分析 (一)各類型投資策略的整體結果中，紡織與電子電機兩項產業的獲利績效相當，且均要 比整體市場來得好，其中，紡織產業之"高利潤低風險"策略所獲得的累積超額報酬(43.28%) 更居全體之冠。 (二)本研究所找出之個別產業最適投資策略分別為： 1.整體市場：總體策略、低風險策略 2.紡織產業：高利潤低風險策略、高利潤策略 3.電子電機產業：高利潤低風險策略、低風險策略 / Financial Statements are very important information indicating performance of corporations. Managers and investors use financial ratios as vital indexes to evaluate and predict operating results of corporations, and make their decisions. ategy, and compute CAR for each investment strategies. At last, I analyze the investing results of the four strategies for individual industry. ANN ( Artificial Nerual Network) shoot a new direction on researching application of abstracting accounting information which can efficiently predict earnings. According to results of relative researches, financial statements from different industries present and implicate different accounting information. If we further apply ANN on financial statement information to predict earnings of corporations, we can use the results as bases of analyses and comparisons among industries. Because ANN model has many advantages, in this research, I use financial statements and return on stocks from corporations as researching samples to construct prediction models and compute CAR(Cumulative Abcdrmal Return) on investments. These samples are chosen from 15 different industries and covered from the first quarter of 1981 to the third quarter of 1993. This research consists of three parts: 22 financial ratios selected by MANOVA First, I use the general market samples to adjust and predict the vital parameters of ANN models, such as the selection of input variable, the number of hidden node, and finally pick better setups for the prediction model. Second, I use this model to train and test samples from the general market, the textile, and the electron industry, and research the variation of predicting results by different models made up different industries by means of evaluation indexes . Third, I use the results predicted by the three different industry models, inspect of risk and return, to define four types of investment strategies -- "the general", "the high return", "the low risk", and "the high return - low risk" strategy, and compute CAR for each investment strategies. At last, I analyze the investing results of the four strategies for individual industry. After researching, I find:s of the textile and electron industry are better than the general markets'. 1.The better setups of ANN predition models are :industries are: (1)the selection of input variable:the 22 financial ratios selected by MANOVA (2)the ANN model topology(input node - hidden node - output node):22-22-1 rategy (3)the range of initial connection weights:-0.1~0.1 return - low risk strategy 2.The analyses of results predicted by the three different industry models are: (1)the predicting abilities of the textile and electron industry are better than the general markets'. 3.The proper investment strategies of individual industries are: (1)the general market:the general and the low risk strategy (2)the textile industry:the high return and the high return - low risk strategy (3)the electron industry:the low risk and the high return - low risk strategy

大部的類神經網路均為解決特定問題而設計，並非真正去模擬人腦的功能 ，在本論文中介紹一個模擬人類學習方式的類神經網路，稱為推理類神經 網路（The Reasoning Neural Network），其主要兩個組成為強記（ cram -ming）及推理（reasoning）部份，透過彈性的組合這兩個部份可 使類神經網路具有類似人類的學習程序。在本論文中介紹其中一個學習程 序並用四個實驗來評估推理類神經網路的績效，從實結果得知，推理類神 經網路能以合理的隱藏節點數（hidden nodes）達到學習的目標，並建立 一個網路內部表示方式（internal representation），及具有好的推理 能力（g eneralization ability）。 / Most of artification Neural Networks are designed to resolve spe -cific problems, rather than to model the brain. The Reasoning N -eural Network (RNN) that imitates the way of human learning is presented here. Two key components of RNN are the cramming and t -he reasoning. These components coulds be arranged flexibly to a -chieve the human-like learning procedure. One edition of the RNN used in experiments is introduces, and four different proble -ms are used to evaluate the RNN's performance. From simulation results, the RNN accomplishes the goal of learning with a reason -able number of hidden nodes, and evolves a good internal repres -entation and a generalization ability.

依時間順序出現之一連串觀測值，通常會呈現某一型態，而根據所產生的 型態可以作為判斷事件發生的基礎。例如，震波形成原因的判斷﹔追查環 境污染源﹔以及在醫學方面，辨識一個正常人心電圖的型態與患有心臟病 的病人其心電圖的型態…等。對於這些問題，傳統之辨識方法常因前提假 設的限制而失去其準確性。在本文中，我們應用神經網路中的逆向傳播演 算法則來訓練網路，並利用此受過訓練的網路來辨別線性時間數列ARIMA 及非線性時間數列 BL(1,0,1,1)。結果發現，網路對於模擬資料中雙線性 係數介於0.2至$0.8$之間的資料有高達$80\%$以上的辨識能力。而在實例 研究中，我們訓練網路來判斷震波形成的原因，其正確率亦高達80\%以上 。同時，我們也將神經網路應用在環境保護方面，訓練網路來判斷二地區 空氣品質的型態。 / A series of observations indexed in time often produces a pattern that may form a basis for discriminatingetween different classes of events. For instance, in theeology, what are the causes of seismic waves? a earthquakesr the nuclear explosions ?in the eathenics, we can use theethod to inquire the source which pollutes the air in somelace, and in the medicine, to distinguish the difference oflectrocardiograms between a health person and an a patient..ect. In this paper, we utilize the back-propagation to trainnetwork and use of the trained networks to judge the linearRIMA(1,0,0) model between the nonlinear BIL(1,0,1,1) model,e can find that the trained network has a good recognitionhose accurate rate is above 80\% for the coefficient of the bilinear model being equal to 0.5 or 0.8. In a living example, we have trained a network to decidehich is the cause of seismic wave, and the trained networkhose accurate rate is larger than 80\%. At the same time, e also applied neural network in environmental protection.

長久以來,人類在萬物中獨具的高智慧特質吸引了無數的哲學家和科學家 投入對其研究,除了醫學的原因之外,由於人腦所具有卓越的辨識系統及學 習能力,為數不少的科學家們相信人腦存在許多最適化系統與設計,因此如 何模仿人類腦神經的組織與運作,一直是很多人努力及夢寐以求的.因此類 神經網路就是依據這些理念而在各研究領域上廣為發展與應用,其中本文 所探討的倒傳遞神經網路模型更是目前類神經網路模型中最具代表性,應 用最廣的模型.而結構性時間數列模型則是將可被觀察的變數分解成趨勢, 季節性,不規則性等不可被觀察項,故其對經濟意義的解釋是相當明當明顯 的,藉由狀態空間模式的轉換,我們將很容易地利用卡門濾器來作估計與預 測.而本文所欲探的重點在於比較有學習機能的倒傳遞神經網及可利用最 新的資訊更新之結構性時間數列何者之預測能利較佳,藉此瞭解二者之一 些特性.

本論文主要研究以類神經網路模式預測季節性時間序列之有效性。利用適 當地建構樣本訓練集,網路經訓練後可作為季節性時間序列之預測工具。 文中亦提出移動學習法以期提高預測之準確度。並以台灣地區每季進口商 品與勞務總值則作為實證之研究。此季節性時間序列因受離群值之影響而 增加其預測困難度。實證結果顯示類神經網路模式之預測表現較傳統之統 計方法優異,即使此序列受到離群值之干擾。 / We investigate the effectiveness of neural networks for predicting the future behavior of seasonal time series. Utilizing the training set constructed properly, we can train the network who can be used to predict the future of seasonal time series. A shifting-learning method is also employed in order to obtained a better forecasting performance. The quarterly imports of goods and services of Taiwan between the first quarter of 1968 and the fourth quarter of 1990 are studied in the research. The series are contaminated with outliers, which will increase the difficulty of forecasting. Empirical results exhibit that neural networks model free approach have better prediction performance than the classical Box-Jenkins approach, even the series are contaminated with outliers.

摘要 分類是研究的第一步，服務業理必須基於有意義的分類，才 可能提出規範性的結論。此外，服務業的無形產出，必須透過細緻的 服務傳送系統設計來傳遞給顧客。因此服務業的研究首要工作，在從 抽象的層次中提出對服務業無形特性的有效分類。經由文獻檢討及實 務觀察，本文提出三項服務業的產出分類構面:經濟性、社會性及心 理性利益，稱之為服務核心。以此三構面將台北市的服飾零售業分為 四種類型:經濟心理性、心理性、混雜性及經濟性零售店。各類型零 售店中較績優者，相互之間服務傳送系統的差異非常明顯，顯示績效 的殊途同歸性是明顯的。用一類型零售店組內的比較分析中，以類神 經網路求得影響績效最重要的服務傳送系統項目:經濟心理性最重要 的活動與商品無直接的關連;心理性零售店最重要的是人員專業性與 商品風格與品味;經濟性零售店的服務傳送系統則是愈簡單愈好。文 末並提出相關的討論是建議。 / ABSTRACT keywords: service industry, service core, service delivery system, retail industry， neural network Classification is the first step for research. Normative suggestions cannot be provided unless meaningful classification is available in service management. Meanwhile, intangible output in service organizations usually is transferred to customers through delicate service delivery system. Therefore the primary task in service management research is to devise efficacious, theoretical classifications to govern inherent intangibility in service management. Through literature review and field study, this paper proposed three classifying dimensions for fashion retailing, which were termed service cores consisting of economic , social , and psychological benefits. Based on these three dimensions, four types of retailing firms were derived with clustering analysis: eco-psychological, psychological, and economic types. Between groups, better performers were extracted to compare with each other， which demonstrated the significances of equifinality towards performance and differences between these four retailing types. Within groups, neural network analysis was employed to determine important factors in service delivery system. In eco-psychological type, important factors were irrelevant to merchandises. Professional salespersons and special merchandise were critical to psychological stores. Better economic stores were all rated low in most delivery activities. Relevant suggestions and discussions were given to conclude the findings.

GPS 單點定位或稱絕對定位，傳統上使用虛擬距離觀測量，容易受到 電離層延遲影響，導致定位精度較差。因此，本文的目的為構建即時的區 域性電離層模型，以便能夠即時減弱電離層延遲量，提高單頻GPS 單點定 位的精度。 構建電離層模型的方法有很多種，而運用類神經網路為可能方法之一， 但是， 國內較少人探討。本研究嘗詴使用倒傳遞類神經網路(Back-propagation Artificial Neural Network)，構建即時的區域電離層模型，藉由選擇適當的神經訓練函數及隱藏層神經元，利用過去收集的已知參考站的雙頻GPS 資料，計算電離層延遲量，訓練類神經網路，直到精度合乎要求；再以檢核站GPS 資料，檢驗類神經網路預測電離層延遲的功效。 採用的實驗資料為臺南市政府e-GPS 系統所提供六個測站，2008 年1 月3 日到1 月5 日的GPS 資料，計算測站與GPS 衛星連線中假想的電離層 薄殼交點—電離層穿透點(Ionosphere Pierce Point, IPP）之地理位置(緯度φ、經度λ)，及太陽黑子數(sunspot numbers)等當作輸入值，IPP 的垂直電離層延遲當作輸出值，測詴包含單日、兩日以及不同的資料型態(IPP 點、網格點)等情況訓練類神經網路，藉由相對應的驗證資料，檢驗類神經網路的功效，最後將類神經網路的預估成果與全球電離層改正模型、雙頻GPS 資料計算的電離層延遲相比較，並根據改正率與統計特性，評估類神經網 路構建出的區域性電離層模型的成效。 由實驗成果顯示，構建的即時區域性電離層模型的標準差可小於±3TECU，並可改正約80%的電離層延遲誤差，故以類神經網路可有效的構 建出區域性的電離層模型。 / The conventional single point positioning using GPS pseudo rangemeasurements, are vulnerable to ionospheric errors, leading to poor positioningaccuracy. Constructing a real-time ionospheric model is one of the methods that can reduce the ionospheric errors and improve the single point positioning accuracy. Although there are many methods to construct regional ionosphere model,using artificial neural network (ANN) to construct a real-time ionospheric model is less to be mentioned. This study used back-propagation artificial neural network to estimate a regional real-time ionospheric model by selecting the appropriate training functions and the number of hidden layers and its’ nodes. The neural network had to be ‘trained’ by the computed TECs from reference stations’ duel-frequency GPS data until the required accuracy was achieved. The experimental data are collected from 6 e-GPS stations of Tainan city government on January 3 to January 5, 2008. The input values for the ANN includ the geographical location of the ionosphere pierce point (IPP) and solar activity (sunspot number). The output value are those IPPs’ vertical total electron content (VTEC). Different times range and data types (IPPs’ or raster data) for the impact of the ANN are tested. And then compared to Klobuchar model and global ionopheric model, according to the correct rate and the ΔTEC statistic table decide the effectiveness of ANN. According to the test results, the regional ionopheric model constructed by ANN can corrected 80% of the ionospheric errors, the standard deviation of ΔTEC is less than ±3TECU.

國父 孫中山先生曾說：『民生的需要，從前經濟學家都說是衣、食、住三種。照我的研究，應該有四種：於衣、食、住之外，還有一種就是行。』，在各種交通工具中，最普及的就是汽車。汽車由貴族地位的象徵，發展至福特汽車公司一家獨大，最後演變為各大汽車品牌的競爭。更因消費者意識的改變，購買汽車時考慮的不再僅是量產速度、購買價格。在現今生產技術成熟，沒有一家汽車公司具壓倒性優勢的情況下，品牌的因素將會是消費者進行購買決策時一個重要的指標。 本研究欲透過國內六大汽車品牌之顧客關係資料，利用資料採礦模型，瞭解品牌形象、廣告印象及人口統計變數與購買意願之關係，進一步探討各汽車品牌之消費者忠誠度、客群分布與品牌差異，期能在汽車品牌公司百家爭鳴情況下，分析出消費者於不同汽車品牌之品牌知覺，提供汽車品牌之購買意願模型與後續研究參考。