Fuzzy Identification in Time Series

時間數列的模式識別在近年來逐漸受到注意。因為根據時間數列所產生的走勢型態可以作為判斷事件發生與預測未來的基礎。雙線性模式是由ARMA模式所延伸，所以不易與ARMA做一區別。本文就針對這類的問題，提出解決的方法。在本文中，我們應用統計檢定結合模糊理論，建構一個整合式的識別過程。由特徵擷取，找出各種模式之間的差異，再藉由其中的異同建立模糊規則庫。接下來計算出時間數列相對應的特徵屬性，最後由模糊規則庫做出判斷。我們以台積電與聯電的每日收盤價格與成交張數為例，識別的結果與一般的認知相同。Identification of time series model gets more and more attention, because we can analyze the events happened and forecast what will occur in the future based on the accurate model. Bilinear time is extended by ARMA model, so it is hard to distinguish bilinear model and ARMA model. This paper focuses on this type of subject and proposes some possible way to solve.In this paper, we combine statistical tests and fuzzy methods to build a "composite" identification process. First, we try to find out differences between each model by featuring and building the fuzzy rule bases by the differences. Then, we calculate the membership of feature according the time series data. Finally, we make our decision according to the fuzzy rule bases