|摘要: ||河川治理工程實務上，不同規劃時間因流量資料長度不同，所推求之設計流量亦會有所差異，但規劃單位常以安全為由，不論資料長短均選擇推估值較大者為設計流量，如此不僅與統計上資料長度愈長、推估愈精準之法則相違背，亦可能不經濟。因此本研究藉可靠度分析，探討資料長度對設計流量之可靠度。本文採用二種方法分析可靠度，此兩種方法可用以探討前期(即資料較短者)推估流量、後期(即資料較長者)推估流量、理論值及前後期之推估精準度，其一為點推估法：不論前期推估流量值大於或小於後期推估流量值時，若選用後期為設計流量，其仍然大於理論值或前期推估值之機率；其二為區間推估法：點推估法中僅比較單值，但就統計觀點而言，給予一信賴區間可把與理論值差異極小之值視為相等的，故為解決點推估法只單純比較值的大小所造成的不合理處，本研究以顯著水準α=0.1, 0.5，得出後期之區間推估值，探討對於上述兩種方法之差異。本研究將每組合成資料之前15個值視為前期資料，依序加入一個資料視為後期資料，故後期資料個數依序由16遞增至45。研究中，針對不同分布、不同回歸週期分別探討設計流量之可靠度，結果顯示，使用後期之推估值為設計流量時，其可靠度平均約0.6至0.95，因此，無論後期所推估之設計流量是否較前期者為小，若均採用後期推估值，皆具有相當之安全性。|
In practical planning for river improvements, the estimated design discharge was usually different according to different length of the data that was depended on design timing. For safety reason, the planners usually selected the larger value of estimator for the design discharge without worrying about the data length. In statistics, however, this way is not only in contravention of the method that the longer data length, the estimator is more precise, but also may not be economical. This study uses two methods to analysis the reliabilities of the estimated values from the shorter length of data, the estimated value from the longer length of data, theoretical value and the accuracy of the two estimated values comparing to the theoretical value. The first method is called points estimated method. Whether the estimated value from the shorter length of data is larger than that from the longer length of data or not, the probability of estimated value from the longer length of data in safety design is larger than those of the other two;The second method uses interval estimation. In order to solve the unreason results which is caused from a single value by using the pointed method, we chose the interval a=0.1, 0.5 to calculated the estimated values from the longer length of data. In each set of synthetic data, the first fifteen values are taken as the shorter length of data. The number of longer length of data will be sixteen increasing to forty-five. By means of several kinds of distribution functions and return period to analysis the reliability, the result shows that the reliability is between 0.6 to 0.95 when the longer length of data is chosen. So, no matter the estimated value of design discharge of longer data is smaller than that of the shorter one or not, as long as we chose the longer length of data, we will have safety design of river discharge in planning for the river improvement.