摘 要 在企業中所面臨的決策中,時間通常是一個重要的變數。故管理者實施分析時,亦常以過去的歷史資料為依據,預測相關變數的預測值。過去的歷史資料,我們稱之為時間序列。更明確的定義,時間序列預測是一群統計資料,由某一段時間出現的相關資料實施預測,即為時間序列。 時間序列預測結果的精確將影響企業的規劃、經營的成敗,現代的商業和經濟活動,本質上是動態的,而且是多變化的。當管理者面對所遭遇的環境特性與實際時間序列預測成本效益問題時,如何選取合適的時間序列預測方法是相當重要的課題,如何對時間序列做一可靠的預測,為管理者最重要的工作,但要使用何種工具對以往若干時日連續不斷所產生的時間序列,加以詳細的預測,以明瞭其變動的趨勢。 本論文對時間序列實施分析,分析時先以卡爾曼濾波器能動態處理的能力對資料實施前處理,針對時間序列中之特徵值再結合類神經網路對時間序列實施估測,最後經由專家意見定性知識擷取透過模糊理論產生調整值,對估測值實施調整,使其預測方法更具強健性,結合不同以上之方法,期能建構出定量�定性整合性演算法之時間序列預測,對現代的商業和經濟活動的動態時間序列提供一個完備的預測方法。
ABSTRACT Time is usually an important variable in the decisions of business. Since, when they analyzed, managers often predict the other corresponding variables by historical data, called time series. Time series prediction is a group of statistic data, which was predicted by some corresponding data happened during some time. The exactness of outcome of time series prediction would affect business planning and administration. Nowadays business and economical activities in essence are dynamic variable. In the face of environment characters and cost-benefit problems of real series prediction, how to choose suitable method of time series prediction is a very important issue for managers. How to develop a robust method for prediction of time series is the most important work of managers. In the thesis, we combined the Kalman filter, with artificial neural network, and fuzzy theory. We hope this could build a hybrid method to analyze time series by qualitative and quantitative method, to provide a reliable prediction method for the dynamic and variable time series of business and economical activities.