傳統上,對於非定常性時間序列轉換成定 常性時間序列,其差分階數之判定,並無明確的 判斷準則,往往須藉由經驗加以判定。本研究 利用自相關函數之95%信賴區間判斷序列所需之 差分階數。根據合成資料及實測資料分析結果 顯示,本研究所提之方法能準確判定具有趨勢 或週期性之非定常性時間序列所需之差分階數 。 In the conventional approach, there does not exist a theoretical criterion to determine the differenced order for transforming the nonstationary time series into stationary one. It is usually determined by the personal judgement. In this study, the 95% confidence interval of the autocorrelation function was proposed to detect the differenced order. On the basis of the results obtained by the synthetic data and real data studies, it indicates that the approach proposed by this study can successfully detect the exact differenced order for a given nonstationary time series with trend or periodicity.
第六屆水利工程研討會論文集(下冊)=Proceedings of 6th Conference on Hydraulic Engineering，頁 896-908