Studies on Bilinear Models for the Monthly Riverflow Series
| Autor: | Chen, Cheng-Ping, 陳正平 |
|---|---|
| Rok vydání: | 1998 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 86 Subba Rao and Gabr(1981)針對非線性時間序列模式中的雙線性模式( Bilinear Model,簡稱BL模式)提出完整的參數推估及模式選取方法，且虞 氏(1986)利用雙波譜證明水文序列大多具有非線性，故將雙線性模式運用 於水文時序之探討應有其適用性。本研究採用合成資料探討參數推估之準 確性、模式選取方法之準確性及非線性檢定之檢定能力，而後則採用實測 資料探討其於台灣河川月流量資料之適用性。本研究依據Subba Rao and Gabr(1981)所提出利用最大概似法推估參數並依據判定準則選取模式，其 中於推估參數值時也可利用Marquardt(1963)所提出之演算法得另一趨近 式。研究結果顯示，於合成資料之參數推估上，利用趨近式之參數推估結 果具有較為穩定、不易發散等特性；利用BIC及BIC1準則之模式選取能力 相似且較AIC準則為佳。於非線性檢定上，引用Saikkonen and Luukkonen(1988)提出將Lagrange Multiplier檢定(簡稱LM檢定)運用於非 線性檢定。研究結果顯示，於合成資料之非線性LM檢定上，在樣本數到 達500時可達相當不錯之結果，且其檢定能力隨樣本數增加而增加。另， 本研究利用LM檢定及當原非線性BL項抽離其檢定結果不顯著之LM檢定延伸 觀念，提出一新的BL模式選取方法，模式選取結果顯示，其選取能力與依 據判定準則之模式選取方法相去不遠，可供模式選取上之參考。對於實測 資料適用性之探討，本研究分析台灣地區20站河川月流量資料中有4站檢 定結果顯示其確實存在非線性之BL模式，且BL模式於豐枯流量之預測能力 確實較傳統線性模式有所提昇，然而其提昇幅度並不顯著。 Subba Rao and Gabr''s (1981) was focused on the bilinear model (BL Model) of the nonlinear time series model that construct a method of parameters estimation and model identification. Yu (1986) used bispectrum to prove that most of hydrological series are nonlinear. Therefore, it should be suitable to make use of bilinear model on discussing hydrological time series. The following research adopted the synthetic data to examine the accuracy of the parameters estimation and model identification, and the defecting ability of nonlinearity. Furthermore, this study adopted the real data of the monthly riverflow of Taiwan to examine the suitability of this model.This research used the method of maximum likelihood to estimate the parameters, which proposed by Subba Rao and Gabr (1981), on the other hand, an approachment of algorithm for estimating parameters which proposed by Marquardt (1963) is also used in this study. The result showed that the approachment of algorithm contribute more stability and convergence when estimates the parameter. For comparing the ability of model selection criteria, BIC and BIC1, better than the AIC, has the close abilities.In this study, the lanrange multiplier test (LM Test), which proposed by Saikkonen and Luukkonen (1988), was applied on the nonlinear test. For the synthetic data study, it shows a good result when the samples reached 500; the testing ability as the samples increased. Furthermore, on the subject of model selection, a BL method derived by using LM test has been proposed in the study. For the real data, 20 stations of monthly riverflow in Taiwan area was studied. The 4 stations out of 20 showed that the series are nonlinear. The forecasting ability of the proposed BL method is better than the traditional method, although it is not obvious. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|