本文以灰色理論為基礎對時間數列資料結構性轉變的問題進行探討。運用灰色理論累加生成的數據資料建構GM(1,1)模式，再藉由灰色模式的誤差分析來了解整體資料的變動軌跡，並提出相關的檢定法則，以判定資料的離群值(outliers)、轉折點(change point)或轉折區間(change period)。 首先，我們先選取部分資料建構適合的灰色GM(1,1)動態模式，以了解資料變動的軌跡，並掌握資料規律性。隨著資料的替換，我們可以建構整體動態模式，若發現資料替換造成軌跡產生了劇烈的變動，這表示此資料或整段資料並不在先前資料的運行軌跡中，亦即此資料可能是離群值或是轉折點。若為整段資料不在先前資料的運行軌跡中，而轉換為另一模式，即為轉折區間。 我們將所提出來的方法與其他找時間數列轉折點及轉折區間的方法做比較。最後，以1979~2012年的匯率月平均資料為例來驗證吾人所提出的方法。結果有10 個轉折區間，最長的轉折區間為1986/03~1988/05 共27 個月。最短的轉折區間為2003/10~2004/04 共7 個月。

We want to use the gray theory in the structure change of time series. According the datum by the accumulated generating operation (AGO) to construct the gray model of first derivate and one input variable, that is GM(1,1), then use the error analysis of gray model to get the datum fluctuant trace and propose the detecting method for the outliers, change points, or change periods. First, we choose some datum to fit a dynamic model of GM(1,1). To get the datum fluctuant trace and its rule of datum. Along with data change we can construct the global dynamic model of the datum. If we find the violent fluctuant in the trace of data, that display this data is not in the circulating trace, that show this data is an outliers or change point. If there are many data not in the trace and become another model, that reveal a change period. Last, I’ll use the model construction process to construct the dynamic model and find the outliers, change point, or change period in the future for the month’s average exchange rate of US/NT dolor from 1979 to2012, and to prove the efficiency and practicality of this method. There are ten change periods. The longest change period has 27 months from 1986/03 to 1988/05. The shortest change period has 7 months from 2003/10 to 2004/04.