曹承礎臺灣大學:資訊管理學研究所楊琇珊Yang, Hsiu-ShanHsiu-ShanYang2010-05-052018-06-292010-05-052018-06-292008U0001-2407200802550200http://ntur.lib.ntu.edu.tw//handle/246246/179884企業在進行研發及生產等活動皆須仰賴大量資訊,因此,如何將資訊轉化為輔助企業決策的知識,即成為資訊超載時代之下所面對的首要課題。而在獲取知識的途徑中,通常針對特定專業領域,萃取並整合專家意見的作法是為較常見的模式。能有效統整專家意見並將其以結構化模型呈現,則可提供日後分析與應用,並進一步支援相關領域之決策制定。本研究先擷取專家意見與討論中所提及的外生變數及對預測值之可能影響程度,再整合決策樹(Decision Tree)分析與貝氏網路(Bayesian Network)方法,將所得之意見資料建立成完整的專家知識脈絡。使原本僅能得到單一分析者觀點的預測值,現在能保存集合多元專家預測觀點的推論架構,進而作為決策支援系統預測值之微調參考。於電力需求與供給面的整合,可大幅提升電力資源之使用績效,電力負載預測也進而成為重要課題。而透過決策支援系統(Decision Support System, DSS)的輔助,已能達到以長期歷史資料佐以專家意見來進行電力負載量之預測,再由專家根據決策支援系統的預測值作微調,然而此傳統作法並無保存專家意見背後據以判斷的龐大知識架構。本研究乃選定電力產業為實例應用研究之標的,以研究中所提出之IDTBN(Integrated Decision Tree and Bayesian Network)方法,期能整合專家諮詢會議中所提出之意見並建立出分析模型,使專家對決策支援系統之結果值進行微調時,能以此反映出未來發展趨勢的模型並作為參考,使電力負載預測結果更具準確性。The enterprise rely on a lot of information to lead production, research and development activities; therefore, how to transfer the information into knowledge which assists enterprises in making decision, becomes one of the most important issue in the age of information overload. Usually, extracting and integrating expert opinions of specific fields are the common rule to acquire knowledge.ffectively converging expert opinions and showing the structure of the knowledge model can be used for analysis and application in the future and the decision-making of related areas. This thesis aims to extract the extraneous variables and their impact degree toward predicting value. The goal is to integrate Decision Tree analysis and Bayesian Network and construct a complete knowledge profile of experts with those variables and data. Not only will the predicting value of single analyst’s view but also an inferring structure including multiple views of experts be acquired.ecause the integration of demand and supply side for electric power could enhance the utility performance of electric power resource, the prediction of power load becomes more and more important. With the aid of Decision Support System (DSS), the power load could be predicted by making use of historical records and experts’ opinions and slightly adjusting the predicting value of DSS by experts, but the great knowledge structure used for judgment was not kept. In this study, we select the power industry as our research target, and use IDTBN (Integrated Decision Tree and Bayesian Network) method presented in this paper, to integrate the opinions given in the expert meeting and build an analysis model. As a result, it can improve the prediction of electric power load and cope with the changing trend of the future.誌謝辭 I文摘要 IIHESIS ABSTRACT III錄 V目錄 VII目錄 VIII一章 、緒論 1一節 研究動機 1二節 研究目的 2三節 研究步驟與流程 4四節 論文章節安排 6二章 、文獻探討 8一節 決策樹方法 8.1.1 決策樹分析介紹 9.1.2 決策樹之構建與修剪 10二節 貝氏網路模式 11.2.1 建構貝氏網路 12.2.2 貝氏網路之數學模式與架構 13.2.3 貝氏網路之學習模式 17三節 類神經網路 19.3.1 類神經網路理論及架構 19.3.2 倒傳遞類神經網路 22四節 案例式推理 24.4.1案例式推理概念 25三章 、研究方法 28一節 應用實例—台電公司電力長期負載預測 28二節 IDTBN模式 32三節 實證研究方法與設計 35四節 研究架構 37五節 專家意見內容過錄 37四章 、研究過程與結果 39一節 資料前置處理 39二節 效益衝量指標 42三節 決策樹方法結果 43四節 貝氏網路方法結果 47五節 IDTBN方法應用結果 51六節 IDTBN方法效益分析 55五章 、結論與建議 60一節 結論 60二節 建議 62.2.1研究限制 62.2.2未來研究方向 63錄一:專家諮詢會議資料過錄表(前5筆資料) 64錄二:資料過錄格式表 70錄三:原始資料檔案 72考文獻 74application/pdf2007802 bytesapplication/pdfen-US貝氏網路決策樹資料探勘專家意見整合電力長期負載預測Bayesian NetworkDecision TreeData MiningIntegration of Expert OpinionsLong-term Electric Load PredictionIDTBN方法應用於整合專家意見之實證研究——以電力長期負載預測為例An Empirical Study on IDTBN Applied to the Integration of Expert Opinions -- The Case of Long-term Electric Load Predictionhttp://ntur.lib.ntu.edu.tw/bitstream/246246/179884/1/ntu-97-R95725004-1.pdf