径流预报误差的混合t Location-Scale分布模型及应用

T Location-Scale Distribution Model of Runoff Prediction Errors and Its Application

基于混合t Location-Scale分布良好的自适应性,利用k均值聚类法挖掘数据隐含结构、良好的聚类效果特点,建立了不同预见期径流预报误差混合t Location-Scale分布模型。将模型应用于雅砻江流域官地水库,对预见期分别为6、12、18 h和24 h的区间径流预报误差进行了分析与建模,结果发现混合t Location-Scale分布模型弥补了单一分布难以描述径流预报误差特征多样性的局限,能更准确地描述不同预见期径流预报误差的统计特征,误差分布与实测径流预报误差的变化规律一致,可为水库水电站的径流预报和调度运行提供更加准确、可靠的来水数据。

Based on the good adaptability of mixed t Location-Scale distribution and the features of k-means clustering method that can mine the hidden structure of data and present better clustering effect,a model of runoff forecasting errors mixed t Location-Scale distribution in different forecast periods is proposed.The model is applied to the reservoir of Guandi Hydropower Station in Yalong River Basin,and the runoff forecast errors of reservoir with forecast periods of 6,12,18 h and 24 h are analyzed and modeled respectively.The results show that the mixed t Location-Scale distribution model can make up the limitation that single distribution cannot describe the diversity of runoff forecasting error characteristics,and describe the characteristics of runoff forecast error distribution in different forecast periods more accurately.The calculated error distribution is consistent with the variation law of the measured runoff prediction error,which can provide more reliable and accurate data for runoff forecast and reservoir operation.