傳統繁衍合成資料方法，均直接以原始資料為分析對象。因此於資料轉換或建立模式過程中，勢將無法避免受到資料樣本值變異之影響，造成資料轉換失真或偏估參數。本研究提出之「時間序列分離理論」，將時間序列視為由「邊際分布」與「級值序列」兩部份資料所組成。因此，可應用現有許多理論機率分布及序率模式，分別處理此兩部份資料，毋須轉換非常態資料或建立非線性序率模式。根據研究結果顯示，本研究所提出之序列分離理論，突破傳統線性自迴歸模式無法繁衍同時具高偏態且高相關性合成資料之限制。 Most of conventional methods for generating hydrologic time series areapplied to raw data directly. The process of data transformation ormodel building is influenced by sample variation; consequently, itresults in incorrect data transformation, biased-estimated parametersor misjudged distribution of noise, etc. In order to improve thesedefections of conventional methods, 'time series separation theory'and 'order series generation method' are proposed in this study. Intime series separation theory, time series are decomposed intomarginal distribution and data order series so that time series can beanalyzed with theoretical distributions and stochastic modelsseparately. Therefore, it is no need to use non-Gaussian or nonlinearstochastic models to generate non-Gaussian time series. It is shown inthe results that the order series generation method can generate bothhigh-autocorrelated and high-skewed time series, overcoming thelimitation of conventional linear autoregressive models.
關聯:
第十一屆水利工程研討會論文集(下冊)=Proceedings of the 11th Hydraulic Engineering Conference，頁L143-L148
